Introduction to Harmonic Analysis (Student Mathematical Library) 147047199X, 9781470471996

This book gives a self-contained introduction to the modern ideas and problems of harmonic analysis. Intended for third-

123 79 4MB

English Pages 279 [296] Year 1996

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Introduction to Harmonic Analysis
Half-title Page
Title Page
Copyright
Contents
IAS/Park City Mathematics Institute
Preface
Chapter 1. Motivation and preliminaries
1.1. The heat equation in equilibrium
1.2. Holomorphic functions
1.3. Know thy calculus
1.4. The Dirichlet principle
Exercises
Chapter 2. Basic properties
2.1. The mean value property
2.2. The maximum principle
2.3. Poisson kernel and Poisson integrals in the ball
2.4. Isolated singularities
Exercises
Notes
Chapter 3. Fourier series
3.1. Separation of variables
3.2. Fourier series
3.3. Abel means and Poisson integrals
3.4. Absolute convergence
3.5. Fejรฉrโ€™s theorem
3.6. Mean-square convergence
3.7. Convergence for continuous functions
Exercises
Notes
Chapter 4. Poisson kernel in the half-space
4.1. The Poisson kernel in the half-space
4.2. Poisson integrals in the half-space
4.3. Boundary limits
Exercises
Notes
Chapter 5. Measure theory in Euclidean space
5.1. The need for an integration theory
5.2. Outer measure in Euclidean space
5.3. Measurable sets and measure
5.4. Measurable functions
Exercises
Notes
Chapter 6. Lebesgue integral and Lebesgue spaces
6.1. Integration of measurable functions
6.2. Fubiniโ€™s theorem
6.3. The Lebesgue space ๐ฟยน
6.4. The Lebesgue space ๐ฟยฒ
Exercises
Notes
Chapter 7. Maximal functions
7.1. Indefinite integrals and averages
7.2. The Hardyโ€“Littlewood maximal function
7.3. The Lebesgue differentiation theorem
7.4. Boundary limits of harmonic functions
Exercises
Notes
Chapter 8. Fourier transform
8.1. Integrable functions
8.2. The Fourier inversion formula
8.3. Mean-square convergence
Exercises
Notes
Chapter 9. Hilbert transform
9.1. The conjugate function
9.2. Mean-square convergence
9.3. The Hilbert transform of integrable functions
9.4. Convergence in measure
Exercises
Notes
Chapter 10. Mathematics of fractals
10.1. Hausdorff dimension
10.2. Self-similar sets
Exercises
Notes
Chapter 11. The Laplacian on the Sierpiล„ski gasket
11.1. Discrete energy on the interval
11.2. Harmonic structure on the Sierpiล„ski gasket
11.3. The Laplacian on the Sierpiล„ski gasket
Exercises
Notes
Chapter 12. Eigenfunctions of the Laplacian
12.1. Discrete eigenfunctions on the interval
12.2. Discrete eigenfunctions on the Sierpiล„ski gasket
12.3. Dirichlet eigenfunctions
Exercises
Notes
Chapter 13. Harmonic functions on post-critically finite sets
13.1. Post-critically finite sets
13.2. Harmonic structures and discrete energy
13.3. Discrete Laplacians
13.4. The Laplacian on a PCF set
Exercises
Notes
Appendix A. Some results from real analysis
A.1. The real line
A.2. Topology
A.3. Riemann integration
A.4. The Euclidean space
A.5. Complete metric spaces
Acknowledgments
Bibliography
Index
Published Titles in this Subseries
Recommend Papers

Introduction to Harmonic Analysis (Student Mathematical Library)
 147047199X, 9781470471996

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Introduction to Harmonic Analysis

S T U D E N T M AT H E M AT I C A L L I B R A R Y IAS/PARK CITY MATHEMATICAL SUBSERIES Volume 105

Introduction to Harmonic Analysis Ricardo A. Sรกenz

American Mathematical Society Institute for Advanced Study

EDITORIAL COMMITTEE John McCleary Rosa C. Orellana (Chair)

Paul Pollack Kavita Ramanan

2020 Mathematics Subject Classi๏ฌcation. Primary 31B05, 31B10, 31B25, 42A16, 42A20, 42B10, 42B25, 28A20, 28A80.

For additional information and updates on this book, visit www.ams.org/bookpages/stml-105

Library of Congress Cataloging-in-Publication Data Cataloging-in-Publication Data has been applied for by the AMS. See http://www.loc.gov/publish/cip/. DOI: https://doi.org/10.1090/stml/105

Copying and reprinting. Individual readers of this publication, and nonpro๏ฌt libraries acting for them, are permitted to make fair use of the material, such as to copy select pages for use in teaching or research. Permission is granted to quote brief passages from this publication in reviews, provided the customary acknowledgment of the source is given. Republication, systematic copying, or multiple reproduction of any material in this publication is permitted only under license from the American Mathematical Society. Requests for permission to reuse portions of AMS publication content are handled by the Copyright Clearance Center. For more information, please visit www.ams.org/publications/pubpermissions. Send requests for translation rights and licensed reprints to [email protected]. c 2023 by the American Mathematical Society. All rights reserved.  The American Mathematical Society retains all rights except those granted to the United States Government. Printed in the United States of America. โˆž The paper used in this book is acid-free and falls within the guidelines 

established to ensure permanence and durability. Visit the AMS home page at https://www.ams.org/ 10 9 8 7 6 5 4 3 2 1

28 27 26 25 24 23

A Irene

Contents

IAS/Park City Mathematics Institute Preface Chapter 1.

xi xiii

Motivation and preliminaries

1

ยง1.1. The heat equation in equilibrium

1

ยง1.2. Holomorphic functions

3

ยง1.3. Know thy calculus

4

ยง1.4. The Dirichlet principle

9

Exercises Chapter 2.

11 Basic properties

13

ยง2.1. The mean value property

13

ยง2.2. The maximum principle

20

ยง2.3. Poisson kernel and Poisson integrals in the ball

24

ยง2.4. Isolated singularities

29

Exercises

31

Notes

34

Chapter 3.

Fourier series

35

ยง3.1. Separation of variables

35

ยง3.2. Fourier series

38 vii

viii

Contents

ยง3.3. Abel means and Poisson integrals

43

ยง3.4. Absolute convergence

48

ยง3.5. Fejรฉrโ€™s theorem

54

ยง3.6. Mean-square convergence

59

ยง3.7. Convergence for continuous functions

62

Exercises

68

Notes

70

Chapter 4.

Poisson kernel in the half-space

71

ยง4.1. The Poisson kernel in the half-space

71

ยง4.2. Poisson integrals in the half-space

74

ยง4.3. Boundary limits

77

Exercises

80

Notes

82

Chapter 5.

Measure theory in Euclidean space

83

ยง5.1. The need for an integration theory

83

ยง5.2. Outer measure in Euclidean space

85

ยง5.3. Measurable sets and measure

89

ยง5.4. Measurable functions

96

Exercises

102

Notes

103

Chapter 6.

Lebesgue integral and Lebesgue spaces

105

ยง6.1. Integration of measurable functions

105

ยง6.2. Fubiniโ€™s theorem

120 1

124

2

ยง6.4. The Lebesgue space ๐ฟ

130

Exercises

134

Notes

135

ยง6.3. The Lebesgue space ๐ฟ

Chapter 7.

Maximal functions

137

ยง7.1. Indefinite integrals and averages

137

ยง7.2. The Hardyโ€“Littlewood maximal function

138

Contents

ix

ยง7.3. The Lebesgue differentiation theorem

143

ยง7.4. Boundary limits of harmonic functions

145

Exercises

148

Notes

150

Chapter 8.

Fourier transform

151

ยง8.1. Integrable functions

151

ยง8.2. The Fourier inversion formula

156

ยง8.3. Mean-square convergence

159

Exercises

164

Notes

165

Chapter 9.

Hilbert transform

167

ยง9.1. The conjugate function

167

ยง9.2. Mean-square convergence

168

ยง9.3. The Hilbert transform of integrable functions

172

ยง9.4. Convergence in measure

179

Exercises

181

Notes

183

Chapter 10.

Mathematics of fractals

185

ยง10.1. Hausdorff dimension

185

ยง10.2. Self-similar sets

191

Exercises

201

Notes

202

Chapter 11.

The Laplacian on the Sierpiล„ski gasket

203

ยง11.1. Discrete energy on the interval

203

ยง11.2. Harmonic structure on the Sierpiล„ski gasket

207

ยง11.3. The Laplacian on the Sierpiล„ski gasket

212

Exercises

219

Notes

220

Chapter 12.

Eigenfunctions of the Laplacian

ยง12.1. Discrete eigenfunctions on the interval

223 224

x

Contents ยง12.2. Discrete eigenfunctions on the Sierpiล„ski gasket

227

ยง12.3. Dirichlet eigenfunctions

233

Exercises

241

Notes

242

Chapter 13.

Harmonic functions on post-critically finite sets

243

ยง13.1. Post-critically finite sets

243

ยง13.2. Harmonic structures and discrete energy

245

ยง13.3. Discrete Laplacians

250

ยง13.4. The Laplacian on a PCF set

255

Exercises

257

Notes

258

Appendix A. Some results from real analysis

259

ยงA.1. The real line

259

ยงA.2. Topology

261

ยงA.3. Riemann integration

262

ยงA.4. The Euclidean space

265

ยงA.5. Complete metric spaces

267

Acknowledgments

271

Bibliography

273

Index

277

IAS/Park City Mathematics Institute

The IAS/Park City Mathematics Institute (PCMI) was founded in 1991 as part of the โ€œRegional Geometry Instituteโ€ initiative of the National Science Foundation. In mid-1993 the program found an institutional home at the Institute for Advanced Study (IAS) in Princeton, New Jersey. The annual PCMI summer programs take place in Park City, Utah. Each yearโ€™s PCMI summer program is an intensive three-week session where several different activities take place in parallel. These include individual programs for researchers, graduate students, undergraduate faculty, undergraduate students, and K-12 teachers, as well as a workshop devoted to issues surrounding equity in the mathematics classroom. Over 300 people are in attendance each year. A main goal of PCMI is to make all participants aware of the broad spectrum of mathematical activities and to promote interactions between these groups, often leading to new collaboration and new mentoring arrangements. Each summer a different research topic is chosen as the focus of the Research Program and Graduate Summer School. The Undergraduate Program typically focuses on closely related material. Lecture notes from the Graduate Summer School are published each year in the IAS/Park City Mathematics Series. Course material for the Undergraduate Program at PCMI is published intermittently under the IAS/Park City Mathematical Subseries in the Student Mathematical Library. We are very

xi

xii

IAS/Park City Mathematics Institute

pleased to make available to a wide audience the expanded versions of these undergraduate lectures, which we believe should be of great value to students everywhere. Rafe Mazzeo, Series Editor March 2023

Preface

This text grew out of the lecture notes for the course Introduction to Harmonic Analysis given at the Undergraduate Summer School during the 2018 Park City Mathematics Institute (IAS/PCMI). The twelve-hour course contained the basic properties of harmonic functions and the explicit solutions to the Laplace equation in special cases. It also contained a study of the behavior of harmonic functions at the boundary of domains, and introduced the Hilbert transform. The last lectures were dedicated to the construction of harmonic functions in fractals. The lectures were then extended to a full semester course at the University of Colima, aimed to junior and senior undergraduate mathematics majors. Besides a more detailed discussion on the previous topics, the course also included a discussion on Fourier series and their convergence, an introduction to Lebesgue measure and integration, the Hardyโ€“Littlewood maximal function, the Fourier transform and a more extended study of analysis on fractals, in particular the Laplacian on the Sierpiล„ski gasket and the construction of its eigenfunctions. The purpose of both the minicourse at IAS/PCMI and the course at Colima is to introduce the modern ideas and problems of harmonic analysis to undergraduate students, from the point of view of harmonic functions. Most books on harmonic and Fourier analysis are too advanced to be appropriate at the undergraduate level, and undergradute textbooks, as those by Thomas William Kรถrner [K8ฬˆ8] and Elias M. Stein and Rami Shakarchi [SS03], focus on Fourier series and their convergence, rather

xiii

xiv

Preface

than on harmonic functions and their behavior at boundary domains. So this text starts with a motivation to study harmonic functions and the Dirichlet problem. The first chapter discusses the solution to the heat equation in equilibrium, the real and imaginary parts of holomorphic functions, and the minimizing functions of energy, all of which are harmonic functions. This book is intended for junior and senior undergraduates with a basic knowledge of real analysis. It requires familiarity with the properties of complete metric and normed spaces, uniform convergence and density. The needed results are reviewed in Appendix A. It does not require knowledge of measure theory, as the text includes two chapters on measure theory, Lebesgue integration, and approximation theorems. It does requires knowledge of linear algebra, in particular familiarity with vector spaces, subspaces, linear operators and orthogonality. Complex analysis is not required but it is recommended, as a couple of calculations of the Fourier transform of some functions are easily done using line integrals and the residue theorem. The text can be roughly partitioned in three parts. The first part, Chapters 1โ€“4, discusses the basic properties of harmonic functions and the problem of their behavior at the boundary of their domains. Basic properties as the mean value property, the maximum principle or the classification of singularities are discussed in Chapter 2. Explicit solutions to the Dirichlet problem, in terms of Poisson integrals, are discussed for the ball (Chapter 2) and the half-space (Chapter 4). We also discuss the problem of the behavior at the boundary of these domains in those chapters, for the case of continuous boundary values. In Chapter 3, we discuss the solution of the Dirichlet problem in the disk using Fourier series, so in this chapter we also discuss the problem of their convergence. In particular, we discuss Abel means, Fejรฉrโ€™s theorem and mean-square convergence. We end Chapter 3 with the construction of an example of a continuous function with divergent Fourier series at a point. The second part, Chapters 5โ€“9, discusses the problem of the behavior of harmonic functions at the boundary of their domains, in particular the half-space, for noncontinuous functions. This requires the use of measure theory, and thus is developed in Chapters 5 and 6. This is not intended to be a comprehensive course in measure theory, but a brief introduction to the main ideas of Lebesgue integration and the basic

Preface

xv

properties of the Lebesgue spaces ๐ฟ1 and ๐ฟ2 . In Chapter 7 we discuss the Hardyโ€“Littlewood maximal function and the problem of almost everywhere convergence of Poisson integrals of integrable functions at the boundary of the half-space. In Chapter 9 we introduce the Hilbert transform, which describes the limits at the boundary of the conjugate function to the Poisson integral. We discuss the ๐ฟ1 theory of the Hilbert transform and the concept of operators of weak type. In order to study the ๐ฟ2 theory of the Hilbert transform, we introduce the Fourier transform in the previous Chapter 8, where we discuss its basic properties, as the Riemannโ€“Lebesgue lemma for integrable functions and the Plancherel theory of square integrable functions. In the third part, Chapters 10โ€“13, we discuss the theory of harmonic functions on fractals. We start with the fundamental ideas of self-similarity and Hausdorff dimension in Chapter 10, and then proceed to the study of harmonic analysis, the Laplacian and its eigenfunctions, on the Sierpiล„ski gasket, in Chapters 11 and 12. We discuss in these chapters the construction of a harmonic structure and the harmonic functions by interpolation. We describe the construction of the Laplacian and an algorithm to construct its eigenfunctions. In particular, we explicitly describe the Dirichlet eigenfunctions on the Sierpiล„ski gasket. We discuss in Chapter 13 harmonic functions on more general self-similar sets. This text can be used in an introductory course on Harmonic Analysis in several ways. In Colima, semesters are sixteen weeks long, so one has enough time to cover almost all of the material,1 but for shorter semesters we can choose accordingly to the interests and previous knowledge of the audience. Chapters 1โ€“9 provide an introduction to classical harmonic analysis, and students who already took a course on measure theory may skip Chapters 5 and 6. For a course on analysis on fractals, one may choose Chapters 1โ€“3 and then move on to Chapters 10โ€“13, as the first chapters serve as motivation for the study of harmonic functions and eigenfunctions of the Laplacian. Each chapter has a list of exercises and bibliographic and historical notes. Ricardo A. Sรกenz Colima, Mexico, October 2022

1 The Fall 2020 course was transmitted online and is available at the page https://www. facebook.com/HarmonicAnalysis

Chapter 1

Motivation and preliminaries

1.1. The heat equation in equilibrium In this chapter we discuss a number of motivations for the study of harmonic functions, with examples taken from physics to complex analysis. We start in this section with a deduction of the heat equation in equilibrium, using the original argument given by Joseph Fourier in his seminal work Analytical Theory of Heat [Fou55]. Consider the propagation of heat through a solid in space. For example, you can consider a potato in the oven, receiving heat in part of its peel. If you wait sufficiently long, the temperature inside the potato will be in equilibrium; though not necessarily constant in its interior, it will not depend on time. Let ๐‘„ be a small cube inside this solid, which we describe with edges parallel to the axes in โ„3 . Suppose two of its opposite vertices are given by (๐‘ฅ0 , ๐‘ฆ0 , ๐‘ง0 ) and (๐‘ฅ0 +๐œ€, ๐‘ฆ0 +๐œ€, ๐‘ง0 +๐œ€) for some small ๐œ€ > 0, and we consider the propagation of heat in ๐‘„, with temperature function ๐‘ข(๐‘ฅ, ๐‘ฆ, ๐‘ง, ๐‘ก). Since we are assuming the system is in equilibrium, the temperature does not depend on time, so it is then a function ๐‘ข(๐‘ฅ, ๐‘ฆ, ๐‘ง) in ๐‘„. We also assume ๐‘ข is a smooth function in a neighborhood of ๐‘„ (that is, an open set that contains ๐‘„).

1

2

1. Motivation and preliminaries

Figure 1.1. The small cube ๐‘„, with heat propagating in the ๐‘ฅ direction.

By Newtonโ€™s law of heat flow, the amount of heat that enters through the side ๐‘ฅ = ๐‘ฅ0 of ๐‘„ (the left side in Figure 1.1) is proportional to the change in temperature, from hotter to colder, in the ๐‘ฅ direction on this side, so it is given by ๐œ•๐‘ข โˆ’๐พ๐œ€2 (๐‘ฅ0 , ๐‘ฆ0 , ๐‘ง0 ), ๐œ•๐‘ฅ where ๐œ€2 is the surface area of the left side and the proportionality constant ๐พ > 0, which depends only on the material of the solid, is called the conductivity constant. The amount of heat that exits through the side ๐‘ฅ = ๐‘ฅ0 + ๐œ€ of ๐‘„ is then given by ๐œ•๐‘ข (๐‘ฅ + ๐œ€, ๐‘ฆ0 , ๐‘ง0 ). ๐œ•๐‘ฅ 0 The quantity of heat accumulated in ๐‘„ as a consequence of propagation in the ๐‘ฅ direction is the difference between these two quantities, โˆ’๐พ๐œ€2

๐œ•๐‘ข ๐œ•๐‘ข (๐‘ฅ0 , ๐‘ฆ0 , ๐‘ง0 ) โˆ’ ( โˆ’ ๐พ๐œ€2 (๐‘ฅ0 + ๐œ€, ๐‘ฆ0 , ๐‘ง0 )) ๐œ•๐‘ฅ ๐œ•๐‘ฅ ๐œ•๐‘ข 2 ๐œ•๐‘ข (๐‘ฅ , ๐‘ฆ , ๐‘ง )). = ๐พ๐œ€ ( (๐‘ฅ0 + ๐œ€, ๐‘ฆ0 , ๐‘ง0 ) โˆ’ ๐œ•๐‘ฅ ๐œ•๐‘ฅ 0 0 0 By the mean value theorem, there exists 0 < ๐›ฟ < ๐œ€ such that โˆ’ ๐พ๐œ€2

๐œ•๐‘ข ๐œ•๐‘ข ๐œ•2 ๐‘ข (๐‘ฅ0 + ๐œ€, ๐‘ฆ0 , ๐‘ง0 ) โˆ’ (๐‘ฅ0 , ๐‘ฆ0 , ๐‘ง0 ) = ๐œ€ 2 (๐‘ฅ0 + ๐›ฟ, ๐‘ฆ0 , ๐‘ง0 ), ๐œ•๐‘ฅ ๐œ•๐‘ฅ ๐œ•๐‘ฅ so the propagation of heat through ๐‘„ in the ๐‘ฅ direction is then ๐พ๐œ€3

๐œ•2 ๐‘ข (๐‘ฅ + ๐›ฟ, ๐‘ฆ0 , ๐‘ง0 ). ๐œ•๐‘ฅ2 0

1.2. Holomorphic functions

3

Similarly, there exist 0 < ๐œ‚, ๐œƒ < ๐œ€ so that the propagation of heat through ๐‘„ in the ๐‘ฆ and ๐‘ง directions is given by ๐พ๐œ€3

๐œ•2 ๐‘ข (๐‘ฅ , ๐‘ฆ + ๐œ‚, ๐‘ง0 ) ๐œ•๐‘ฆ2 0 0

and

๐พ๐œ€3

๐œ•2 ๐‘ข (๐‘ฅ , ๐‘ฆ , ๐‘ง + ๐œƒ), ๐œ•๐‘ง2 0 0 0

respectively, and the total propagation is then given by ๐พ๐œ€3 (

๐œ•2 ๐‘ข ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข (๐‘ฅ + ๐›ฟ, ๐‘ฆ , ๐‘ง ) + (๐‘ฅ , ๐‘ฆ + ๐œ‚, ๐‘ง ) + (๐‘ฅ , ๐‘ฆ , ๐‘ง + ๐œƒ)). 0 0 0 0 0 0 ๐œ•๐‘ฅ2 ๐œ•๐‘ฆ2 ๐œ•๐‘ง2 0 0 0

Since the system is in equilibrium, the total propagation must be equal to 0. As we are assuming that ๐‘ข is a smooth function, its partial derivatives are continuous, so we obtain, as ๐œ€ โ†’ 0, the equation ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข + + 2 =0 ๐œ•๐‘ฅ2 ๐œ•๐‘ฆ2 ๐œ•๐‘ง

(1.1) at the point (๐‘ฅ0 , ๐‘ฆ0 , ๐‘ง0 ).

Equation (1.1) is called the Laplace equation. We can also write it as ฮ”๐‘ข = 0, where the differential operator ฮ” is given by ฮ”๐‘ข =

๐œ•2 ๐‘ข ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข + + 2. ๐œ•๐‘ฅ2 ๐œ•๐‘ฆ2 ๐œ•๐‘ง

ฮ”๐‘ข is called the Laplacian of ๐‘ข. The solutions of equation (1.1) are called harmonic functions.

1.2. Holomorphic functions In this section we observe that harmonic functions also appear in complex analysis. Recall that ๐‘“ is holomorphic (or analytic) in an open set ๐ท โŠ‚ โ„‚ if, for each ๐‘ง โˆˆ ๐ท, its derivative (1.2)

๐‘“(๐‘ง + โ„Ž) โˆ’ ๐‘“(๐‘ง) โ„Ž โ„Žโ†’0

๐‘“โ€ฒ (๐‘ง) = lim

exists. If we write the holomorphic function ๐‘“ as ๐‘“(๐‘ง) = ๐‘ข(๐‘ฅ, ๐‘ฆ) + ๐‘–๐‘ฃ(๐‘ฅ, ๐‘ฆ), where ๐‘ง = ๐‘ฅ+๐‘–๐‘ฆ and ๐‘ข and ๐‘ฃ are its real and imaginary parts, respectively, then ๐‘ข and ๐‘ฃ satisfy the Cauchyโ€“Riemann equations (1.3)

๐œ•๐‘ข ๐œ•๐‘ฃ = ๐œ•๐‘ฅ ๐œ•๐‘ฆ

and

๐œ•๐‘ฃ ๐œ•๐‘ข =โˆ’ . ๐œ•๐‘ฅ ๐œ•๐‘ฆ

4

1. Motivation and preliminaries

These equations follow directly from the differentiability of ๐‘“. Indeed, if we take the limit in (1.2) by approaching โ„Ž โ†’ 0 with real numbers, we obtain ๐œ•๐‘ข ๐œ•๐‘ฃ ๐‘“โ€ฒ (๐‘ง) = (๐‘ฅ, ๐‘ฆ) + ๐‘– (๐‘ฅ, ๐‘ฆ). ๐œ•๐‘ฅ ๐œ•๐‘ฅ Meanwhile, if we approach โ„Ž โ†’ 0 with purely imaginary numbers, we get 1 ๐œ•๐‘ข ๐œ•๐‘ฃ ๐œ•๐‘ฃ ๐œ•๐‘ข ๐‘“โ€ฒ (๐‘ง) = (๐‘ฅ, ๐‘ฆ) + (๐‘ฅ, ๐‘ฆ) = (๐‘ฅ, ๐‘ฆ) โˆ’ ๐‘– (๐‘ฅ, ๐‘ฆ). ๐‘– ๐œ•๐‘ฆ ๐œ•๐‘ฆ ๐œ•๐‘ฆ ๐œ•๐‘ฆ As these two expressions for ๐‘“โ€ฒ (๐‘ง) must be equal, we obtain (1.3). Assuming ๐‘ข and ๐‘ฃ are smooth functions,1 we can differentiate the Cauchyโ€“Riemann equations and get ๐œ•2 ๐‘ข ๐œ•2 ๐‘ฃ = ๐œ•๐‘ฅ๐œ•๐‘ฆ ๐œ•๐‘ฅ2

and

๐œ•2 ๐‘ฃ ๐œ•2 ๐‘ข = โˆ’ 2. ๐œ•๐‘ฆ๐œ•๐‘ฅ ๐œ•๐‘ฆ

All mixed derivatives are continuous, so they must be equal and thus ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข = โˆ’ . ๐œ•๐‘ฅ2 ๐œ•๐‘ฆ2 Therefore ๐‘ข is a harmonic function. We can similarly verify that ๐‘ฃ is harmonic. As ๐‘ฃ is the imaginary part of a holomorphic function of which ๐‘ข is the real part, we say that the function ๐‘ฃ is a conjugate harmonic function to ๐‘ข. Note that conjugate harmonic functions are not unique, because adding any constant to ๐‘ฃ will give another conjugate harmonic function. Also, observe that โˆ’๐‘ข is the conjugate harmonic function to ๐‘ฃ. Under appropiate conditions on the set ๐ท, one can prove that every harmonic ๐‘ข has a conjugate harmonic function ๐‘ฃ. See Exercise (3) for the special case of the complex plane ๐ท = โ„‚. We will dedicate Chapter 9 to the study of the properties of conjugate harmonic functions in the upper half-plane.

1.3. Know thy calculus Before moving on, letโ€™s dedicate a section to set the notation used in this text, and review some of the results from advanced calculus that weโ€™ll need later on. This will just be a quick summary of these results, so we 1 It is a fact, proven in any basic complex analysis text (see [Gam01], for example), that both แต† and ๐‘ฃ are smooth functions whenever ๐‘“ is holomorphic.

1.3. Know thy calculus

5

invite the reader to consult advanced calculus texts, such as [Fle77] or [Spi65] for the details and proofs. We denote the ๐‘‘-dimensional Euclidean space by โ„๐‘‘ . Thus โ„๐‘‘ = {๐‘ฅ = (๐‘ฅ1 , ๐‘ฅ2 , . . . , ๐‘ฅ๐‘‘ ) โˆถ ๐‘ฅ๐‘– โˆˆ โ„}. We will usually denote the points in the plane โ„2 and in the space โ„3 by (๐‘ฅ, ๐‘ฆ) and (๐‘ฅ, ๐‘ฆ, ๐‘ง), respectively. We denote the Euclidean norm of a vector ๐‘ฅ โˆˆ โ„๐‘‘ by |๐‘ฅ|. Thus |๐‘ฅ| = โˆš๐‘ฅ12 + ๐‘ฅ22 + . . . + ๐‘ฅ๐‘‘2 . For ๐‘ฅ โˆˆ โ„๐‘‘ , ๐‘ฅโ€ฒ is the point in โ„๐‘‘โˆ’1 formed by the first ๐‘‘ โˆ’ 1 coordinates of ๐‘ฅ. We can thus write ๐‘ฅ = (๐‘ฅโ€ฒ , ๐‘ฅ๐‘‘ ). If we need to explicitly distinguish the last coordinate, then we refer to โ„๐‘‘+1 = {(๐‘ฅ, ๐‘ก) โˆถ ๐‘ฅ โˆˆ โ„๐‘‘ , ๐‘ก โˆˆ โ„}. We denote by โ„๐‘‘+1 the upper half-space of points (๐‘ฅ, ๐‘ก) โˆˆ โ„๐‘‘+1 with + ๐‘ก > 0. The open ball of radius ๐‘Ÿ > 0 centered at ๐‘ฅ0 is given by ๐ต๐‘Ÿ (๐‘ฅ0 ) = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘ฅ โˆ’ ๐‘ฅ0 | < ๐‘Ÿ}. If ๐‘ฅ0 = 0, we simply denote it by ๐ต๐‘Ÿ . If, in addition, ๐‘Ÿ = 1, we denote it by ๐”น. The sphere of radius ๐‘Ÿ > 0 centered at ๐‘ฅ0 is given by ๐‘†๐‘Ÿ (๐‘ฅ0 ) = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘ฅ โˆ’ ๐‘ฅ0 | = ๐‘Ÿ}, and we denote it by ๐‘†๐‘Ÿ if ๐‘ฅ0 = 0, and by ๐•Š if we also have ๐‘Ÿ = 1. For an open set ฮฉ โˆˆ โ„๐‘‘ , ๐ถ(ฮฉ) is the space of continuous functions ฬ„ the space of continuous functions on its closure. We in ฮฉ, and ๐ถ(ฮฉ) ๐‘˜ denote by ๐ถ (ฮฉ) the space of ๐‘˜-continuously differentiable functions in ฮฉ, and by ๐ถ โˆž (ฮฉ) the space of smooth functions. Note that ๐ถ โˆž (ฮฉ) =

โ‹‚

๐ถ ๐‘˜ (ฮฉ).

๐‘˜โ‰ฅ1

We denote by ๐ถ๐‘โˆž (ฮฉ) the space of smooth functions with compact support in ฮฉ. That is, ๐‘“ โˆˆ ๐ถ๐‘โˆž (ฮฉ) if ๐‘“ is smooth in ฮฉ and there exists a compact subset ๐พ โŠ‚ ฮฉ such that ๐‘“(๐‘ฅ) = 0 for all ๐‘ฅ โˆ‰ ๐พ. In particular, we say that ๐‘“ is zero โ€œclose to the boundary.โ€

6

1. Motivation and preliminaries

We denote the partial derivative of ๐‘“ with respect to ๐‘ฅ๐‘– either by ๐œ•๐‘“ , by ๐œ•๐‘ฅ๐‘– ๐‘“ or simply by ๐œ•๐‘– ๐‘“, if there is no confusion. The gradient of a ๐œ•๐‘ฅ๐‘– function ๐‘“ is given by the vector โˆ‡๐‘“ = (๐œ•1 ๐‘“, ๐œ•2 ๐‘“, . . . , ๐œ•๐‘‘ ๐‘“). Note that its norm is given by |โˆ‡๐‘“| = (|๐œ•1 ๐‘“|2 + |๐œ•2 ๐‘“|2 + . . . + |๐œ•๐‘‘ ๐‘“|2 )

1/2

.

If ๐›ผ = (๐›ผ1 , ๐›ผ2 , . . . , ๐›ผ๐‘‘ ) is a multi-index, where each ๐›ผ๐‘– โˆˆ โ„•, we define ๐‘ฅ๐›ผ as the monomial ๐›ผ ๐›ผ ๐›ผ ๐‘ฅ ๐›ผ = ๐‘ฅ 1 1 ๐‘ฅ2 2 โ‹ฏ ๐‘ฅ ๐‘‘ ๐‘‘ and ๐œ•๐›ผ ๐‘“ as the higher order derivative ๐›ผ

๐›ผ

๐›ผ

๐œ•๐›ผ ๐‘“ = ๐œ•1 1 ๐œ•2 2 โ‹ฏ ๐œ•๐‘‘ ๐‘‘ ๐‘“. The order of the multi-index ๐›ผ is given by |๐›ผ| = ๐›ผ1 + ๐›ผ2 + . . . + ๐›ผ๐‘‘ . A hypersurface in โ„๐‘‘ is a differentiable manifold ๐‘† of dimension ๐‘‘ โˆ’ 1. Locally, for each ๐‘ฅ0 โˆˆ ๐‘†, there exists an open set ๐‘ˆ that contains ๐‘ฅ0 such that ๐‘ˆ โˆฉ ๐‘† is the solution set to the equation (1.4)

๐œ™(๐‘ฅ) = 0,

for some continuously differentiable function ๐œ™ in ๐‘ˆ with โˆ‡๐œ™ โ‰  0. By the implicit function theorem, and relabeling the coordinates if needed, we can assume ๐œ™ is of the form ๐œ™(๐‘ฅ) = ๐‘ฅ๐‘‘ โˆ’ ๐œ“(๐‘ฅโ€ฒ ), and thus ๐‘ˆ โˆฉ ๐‘† is given by (1.5)

๐‘ฅ๐‘‘ = ๐œ“(๐‘ฅโ€ฒ ).

We say that ๐‘† is a ๐ถ ๐‘˜ -hypersurface if the function ๐œ™ above is in ๐ถ ๐‘˜ (๐‘ˆ) (and hence ๐œ“ is a ๐ถ ๐‘˜ function in its domain). A domain in โ„๐‘‘ is an open and connected subset ฮฉ โŠ‚ โ„๐‘‘ . The domain ฮฉ is a ๐ถ ๐‘˜ -domain if its boundary ๐œ•ฮฉ is a ๐ถ ๐‘˜ -hypersurface. If ฮฉ is a ๐ถ 1 -domain and ๐‘ฅ0 โˆˆ ๐œ•ฮฉ, then the normal vector at ๐‘ฅ0 is the unit vector ๐œˆ(๐‘ฅ0 ) orthogonal to the hypersuface ๐œ•ฮฉ pointing outwards of ฮฉ (Figure 1.2). Thus, โˆ‡๐œ™ ๐œˆ=ยฑ , |โˆ‡๐œ™| where ๐œ™ is a function that describes ๐œ•ฮฉ locally near ๐‘ฅ0 , as in (1.4).

1.3. Know thy calculus

7

Figure 1.2. The normal vector ๐œˆ at a point in the boundary of ฮฉ.

Example 1.6. The open ball ๐ต๐‘… (๐‘ฅ0 ) of radius ๐‘… and centered at ๐‘ฅ0 โˆˆ โ„๐‘‘ is a ๐ถ 1 -domain (in fact, a ๐ถ ๐‘˜ -domain for every ๐‘˜), with boundary equal to the sphere ๐‘† ๐‘… (๐‘ฅ0 ). Note that ๐•Š is the solution set to the equation ๐‘ฅ12 + ๐‘ฅ22 + . . . + ๐‘ฅ๐‘‘2 = 1. Hence, for ๐‘ฅ โˆˆ ๐•Š, ๐œˆ(๐‘ฅ) = ๐‘ฅ. The surface measure on a hypersurface ๐‘† is denoted by ๐‘‘๐œŽ. Locally, if ๐œ“ is as in (1.5), we have that ๐‘‘๐œŽ = โˆš1 + |โˆ‡๐œ“|2 ๐‘‘๐‘ฅโ€ฒ . 1.7. One can integrate over โ„๐‘‘ (or over a subset with rotational symmetry) by using spherical coordinates. If we write a point ๐‘ฅ โ‰  0 in โ„๐‘‘ as ๐‘ฅ = ๐‘Ÿ๐œ‰, where ๐‘Ÿ = |๐‘ฅ| > 0 and ๐œ‰ = ๐‘ฅ/|๐‘ฅ| โˆˆ ๐•Š, then โˆž

โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ โˆซ ๐‘“(๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰)๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ. โ„๐‘‘

0

๐•Š

1.8. The area of the unit sphere in โ„๐‘‘ is denoted by ๐œ”๐‘‘ . Thus, ๐œ”๐‘‘ = โˆซ ๐‘‘๐œŽ. ๐•Š

We leave it as an exercise (Exercise (7)) to prove that ๐œ”๐‘‘ =

2๐œ‹๐‘‘/2 , ฮ“(๐‘‘/2)

8

1. Motivation and preliminaries

where ฮ“(๐‘ ) is the gamma function given by โˆž

ฮ“(๐‘ ) = โˆซ ๐‘’โˆ’๐‘ก ๐‘ก๐‘ โˆ’1 ๐‘‘๐‘ก, 0

for every ๐‘  > 0. We will also make use of Theorem 1.9. Theorem 1.9 (Divergence theorem). Let ฮฉ โŠ‚ โ„๐‘‘ be a bounded ๐ถ 1 domain and ๐น a continuously differentiable vector field defined in a neighborhood of ฮฉ.ฬ„ Then โˆซ โˆ‡ โ‹… ๐น๐‘‘๐‘ฅ = โˆซ ๐น โ‹… ๐œˆ๐‘‘๐œŽ.

(1.10)

โ„ฆ

๐œ•โ„ฆ

If ๐น = (๐น 1 , ๐น 2 , . . . , ๐น ๐‘‘ ), then โˆ‡ โ‹… ๐น = ๐œ•1 ๐น 1 + ๐œ•2 ๐น 2 + . . . + ๐œ• ๐‘‘ ๐น ๐‘‘ is called the divergence of ๐น, and is also denoted by div ๐น. Note that the divergence theorem is a multi-dimensional version of the fundamental theorem of calculus. Indeed, in the line โ„, ฮฉ is just an open interval, say ฮฉ = (๐‘Ž, ๐‘), the normal vector at its boundary is given by ๐œˆ(๐‘Ž) = โˆ’1, ๐œˆ(๐‘) = 1, and โˆ‡ โ‹… ๐น is the derivative of ๐น, so (1.10) is ๐‘

โˆซ ๐น โ€ฒ (๐‘ฅ)๐‘‘๐‘ฅ = โˆ’๐น(๐‘Ž) + ๐น(๐‘). ๐‘Ž

1.11. Taking ๐น = (0, 0, . . . , ๐‘ข๐‘ฃ, . . . , 0), where the nonzero component is the ๐‘–th term, we obtain the formula for integration by parts: โˆซ โ„ฆ

๐œ•๐‘ข ๐œ•๐‘ฃ ๐‘ฃ๐‘‘๐‘ฅ = โˆซ ๐‘ข๐‘ฃ๐œˆ ๐‘– ๐‘‘๐œŽ โˆ’ โˆซ ๐‘ข ๐‘‘๐‘ฅ. ๐œ•๐‘ฅ๐‘– ๐œ•๐‘ฅ ๐‘– ๐œ•โ„ฆ โ„ฆ

Theorem 1.12 (Greenโ€™s identities). Let ฮฉ be a bounded ๐ถ 1 -domain in โ„๐‘‘ . (1) If ๐‘ข is continuously differentiable and ๐‘ฃ is twice continuously differentiable in a neighborhood of ฮฉ,ฬ„ then (1.13)

โˆซ (๐‘ขฮ”๐‘ฃ + โˆ‡๐‘ข โ‹… โˆ‡๐‘ฃ)๐‘‘๐‘ฅ = โˆซ ๐‘ข๐œ•๐œˆ ๐‘ฃ๐‘‘๐œŽ, โ„ฆ

๐œ•โ„ฆ

where ๐œ•๐œˆ ๐‘ฃ = โˆ‡๐‘ฃ โ‹… ๐œˆ is the normal derivative of ๐‘ฃ at the boundary of ฮฉ.

1.4. The Dirichlet principle

9

(2) If ๐‘ข and ๐‘ฃ are twice continuously differentiable in a neighborhood of ฮฉ,ฬ„ then (1.14)

โˆซ (๐‘ขฮ”๐‘ฃ โˆ’ ๐‘ฃฮ”๐‘ข)๐‘‘๐‘ฅ = โˆซ (๐‘ข๐œ•๐œˆ ๐‘ฃ โˆ’ ๐‘ฃ๐œ•๐œˆ ๐‘ข)๐‘‘๐œŽ. โ„ฆ

๐œ•โ„ฆ

Theorem 1.12 follows almost immediately from the divergence theorem (or 1.11), and we leave it as an exercise (Exercise (9)).

1.4. The Dirichlet principle The Green identities provide us with another motivation for the study of harmonic functions: they are minimizers of energy. Let ฮฉ be a ๐ถ 1 domain. We define the energy form on ฮฉ as the bilinear form (1.15)

โ„ฐ(๐‘ข, ๐‘ฃ) = โˆซ โˆ‡๐‘ข โ‹… โˆ‡๐‘ฃ๐‘‘๐‘ฅ, โ„ฆ

for smooth functions ๐‘ข and ๐‘ฃ in a neighborhood of ฮฉฬ„ (we denote the ฬ„ The energy of the function ๐‘ข, denoted space of such functions as ๐ถ โˆž (ฮฉ)). simply as โ„ฐ(๐‘ข), is then given by โ„ฐ(๐‘ข) = โ„ฐ(๐‘ข, ๐‘ข) = โˆซ |โˆ‡๐‘ข|2 ๐‘‘๐‘ฅ. โ„ฆ

We now consider the following question: can we find the function ๐‘ข that minimizes โ„ฐ(๐‘ข), given its values at the boundary ๐œ•ฮฉ of ฮฉ? That ฬ„ such that ๐‘ข|๐œ•โ„ฆ = ๐‘“ is, given a function ๐‘“ defined on ๐œ•ฮฉ, find ๐‘ข โˆˆ ๐ถ โˆž (ฮฉ) and ฬ„ and ๐‘ฃ|๐œ•โ„ฆ = ๐‘“}. โ„ฐ(๐‘ข) = min{โ„ฐ(๐‘ฃ) โˆถ ๐‘ฃ โˆˆ ๐ถ โˆž (ฮฉ) It is clear that we cannot expect the above problem to always have a ฬ„ ๐‘“ cannot be arbitrary because solution ๐‘ข. First, as we require ๐‘ข โˆˆ ๐ถ โˆž (ฮฉ), it must be the restriction of such a function to ๐œ•ฮฉ. Moreover, although it is true that the set ฬ„ and ๐‘ฃ|๐œ•โ„ฆ = ๐‘“} {โ„ฐ(๐‘ฃ) โˆถ ๐‘ฃ โˆˆ ๐ถ โˆž (ฮฉ) is bounded from below, because โ„ฐ(๐‘ฃ) โ‰ฅ 0 for any smooth function ๐‘ฃ, it is not clear whether it has a minimum or not. However, we have the following fact: in the case when โ„ฐ takes its minimum at ๐‘ข, then ๐‘ข is a harmonic function in ฮฉ, that is a function that satisfies the equation ฮ”๐‘ข = 0

10

1. Motivation and preliminaries

in ฮฉ. To prove this, suppose โ„ฐ takes its minimum at ๐‘ข for given values at ๐œ•ฮฉ. Now, for any function ๐‘ฃ โˆˆ ๐ถ๐‘โˆž (ฮฉ) and any ๐‘ก โˆˆ โ„, the function ๐‘ข + ๐‘ก๐‘ฃ is smooth and has the same values as ๐‘ข at the boundary. Since โ„ฐ(๐‘ข) is minimal, we have that โ„ฐ(๐‘ข) โ‰ค โ„ฐ(๐‘ข + ๐‘ก๐‘ฃ). Hence, as a function of ๐‘ก, the function ๐ผ(๐‘ก) = โ„ฐ(๐‘ข+๐‘ก๐‘ฃ) takes its minimum value at ๐‘ก = 0, and thus ๐ผ โ€ฒ (0) = 0.

(1.16) Now

๐ผ(๐‘ก) = โ„ฐ(๐‘ข + ๐‘ก๐‘ฃ) = โˆซ โˆ‡(๐‘ข + ๐‘ก๐‘ฃ) โ‹… โˆ‡(๐‘ข + ๐‘ก๐‘ฃ)๐‘‘๐‘ฅ โ„ฆ

= โˆซ |โˆ‡(๐‘ข)|2 ๐‘‘๐‘ฅ + 2๐‘ก โˆซ โˆ‡๐‘ข โ‹… โˆ‡๐‘ฃ๐‘‘๐‘ฅ + ๐‘ก2 โˆซ |โˆ‡๐‘ฃ|2 ๐‘‘๐‘ฅ, โ„ฆ

โ„ฆ

โ„ฆ

so ๐ผ โ€ฒ (๐‘ก) = 2 โˆซ โˆ‡๐‘ข โ‹… โˆ‡๐‘ฃ๐‘‘๐‘ฅ + 2๐‘ก โˆซ |โˆ‡๐‘ฃ|2 ๐‘‘๐‘ฅ โ„ฆ

โ„ฆ

and (1.16) implies โˆซ โˆ‡๐‘ข โ‹… โˆ‡๐‘ฃ๐‘‘๐‘ฅ = 0. โ„ฆ

By the Green identity (1.13), we thus have (1.17)

โˆซ ๐‘ฃฮ”๐‘ข๐‘‘๐‘ฅ = โˆซ ๐‘ฃ๐œ•๐œˆ ๐‘ข๐‘‘๐œŽ = 0, โ„ฆ

๐œ•โ„ฆ

because ๐‘ฃ is zero near the boundary. Moreover, since (1.17) holds for every ๐‘ฃ โˆˆ ๐ถ๐‘โˆž (ฮฉ), we conclude ฮ”๐‘ข = 0, and thus ๐‘ข is harmonic. We have left many open questions in the discussion above. We have mentioned that we cannot expect to have a minimizer ๐‘ข of the energy form satisfying that ๐‘ข|๐œ•โ„ฆ = ๐‘“ for any function ๐‘“. However, as a minimizer is a harmonic function, this leads to the following problem: given a domain ฮฉ and a function ๐‘“ defined on ๐œ•ฮฉ, find a harmonic function ๐‘ข in ฮฉ such that it is equal to ๐‘“ on the boundary. This is known as the Dirichlet problem, in honor of the french mathematician Lejeune Dirichlet. It opens a handful of questions, such as the following: โ€ข For which domains ฮฉ can we solve the Dirichlet problem? For which functions on its boundary does the solution exist?

Exercises

11

โ€ข If ๐‘ข is a harmonic function in ฮฉ, what can we say about its behavior at the boundary of ฮฉ? Does it extend continuously to ๐œ•ฮฉ? Throughout this text we will be discussing results related to the previous questions. In particular, we will focus our attention to harmonic functions in the domains ฮฉ = ๐”น, the unit ball, and ฮฉ = โ„๐‘‘+1 + , the upper half-space, and the behavior of such harmonic functions at the boundaries of their domains. The fact that the minimizers of the energy form โ„ฐ are harmonic functions is called the Dirichlet principle. The first to give it this name was Bernard Riemann in [Rie51], who used this fact to prove the result in complex analysis that we now know as the Riemann mapping theorem. See, for example, [Ull08] for a study of the Riemann mapping theorem and its relation to the Dirichlet problem.

Exercises (1) Let ๐‘… be a rotation in the plane. (a) Consider the change of variables (๐œ‰, ๐œ‚) = ๐‘…(๐‘ฅ, ๐‘ฆ). Then ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข ๐œ•2 ๐‘ข + = + . ๐œ•๐œ‚2 ๐œ•๐‘ฅ2 ๐œ•๐‘ฆ2 ๐œ•๐œ‰2 (b) If ๐‘ข is harmonic, then ๐‘ข โˆ˜ ๐‘… is also harmonic. (2) Let (๐‘Ÿ, ๐œƒ) be the polar coordinates of the plane. Then ฮ”๐‘ข =

๐œ•2 ๐‘ข 1 ๐œ•๐‘ข 1 ๐œ•2 ๐‘ข + + . ๐‘Ÿ ๐œ•๐‘Ÿ ๐‘Ÿ2 ๐œ•๐œƒ2 ๐œ•๐‘Ÿ2

(3) Let ๐‘ข be a harmonic function in โ„2 . Then there exists a conjugate harmonic function ๐‘ฃ to ๐‘ข. (Hint: Consider a line integral of the 1form ๐œ•๐‘ข ๐œ•๐‘ข โˆ’ ๐‘‘๐‘ฅ + ๐‘‘๐‘ฆ.) ๐œ•๐‘ฆ ๐œ•๐‘ฅ (4) If ๐‘ฃ 1 and ๐‘ฃ 2 are conjugate to ๐‘ข in the plane, then ๐‘ฃ 1 โˆ’ ๐‘ฃ 2 is constant. (5) (a) If 0 is conjugate to ๐‘ข in the plane, then ๐‘ข is constant. (b) If ๐‘“ is holomorphic in โ„‚ and real valued, then ๐‘“ is constant. (6) Let ฮ“(๐‘ ) be the gamma function.

12

1. Motivation and preliminaries (a) Integrate by parts to verify the identity ฮ“(๐‘  + 1) = ๐‘ ฮ“(๐‘ ). (b) For every ๐‘› โˆˆ โ„ค+ , ฮ“(๐‘›) = (๐‘› โˆ’ 1)!.

(7) (a) Use polar coordinates to verify the identity 2

โˆซ ๐‘’โˆ’๐œ‹|๐‘ฅ| ๐‘‘๐‘ฅ = 1. โ„2

(b) For every dimension ๐‘‘, 2

โˆซ ๐‘’โˆ’๐œ‹|๐‘ฅ| ๐‘‘๐‘ฅ = 1. โ„๐‘‘

(c) Use spherical coordinates to verify 2๐œ‹๐‘‘/2 . ฮ“(๐‘‘/2) (8) Use integration in spherical coordinates, fact 1.7, to prove that the volume of the unit ball ๐”น is given by ๐œ”๐‘‘ =

โˆซ ๐‘‘๐‘ฅ = ๐”น

๐œ”๐‘‘ . ๐‘‘

(9) Prove Theorem 1.12. (10) Consider the unit interval and, for smooth functions in [0, 1], define the form 1

โ„ฐ(๐‘“) = โˆซ ๐‘“โ€ฒ (๐‘ฅ)2 ๐‘‘๐‘ฅ. 0

(a) The minimizers of this form, given the values of ๐‘“ at ๐‘ฅ = 0 and ๐‘ฅ = 1, are the linear functions ๐‘“(๐‘ฅ) = ๐‘Ž๐‘ฅ + ๐‘. (b) If โ„ฐ(๐‘“) is a minimum, then โ„ฐ(๐‘“) = (๐‘“(1) โˆ’ ๐‘“(0))2 .

Chapter 2

Basic properties

2.1. The mean value property Let ฮฉ โŠ‚ โ„๐‘‘ be an open set. As discussed in Chapter 1, we say that a twice differentiable function ๐‘ข is harmonic in ฮฉ if it satisfies ฮ”๐‘ข = ๐œ•12 ๐‘ข + ๐œ•22 ๐‘ข + . . . + ๐œ•๐‘‘2 ๐‘ข = 0 in ฮฉ. Example 2.1. Any linear function ๐‘ข = ๐‘Ž1 ๐‘ฅ1 + ๐‘Ž2 ๐‘ฅ2 + . . . + ๐‘Ž๐‘‘ ๐‘ฅ๐‘‘ is harmonic in โ„๐‘‘ , as all of its second derivatives are zero. Observe that, in the case ๐‘‘ = 1, the linear functions ๐‘ข(๐‘ฅ) = ๐‘Ž๐‘ฅ + ๐‘ are precisely the functions that satisfy ๐‘ขโ€ณ (๐‘ฅ) = 0, so the only harmonic functions in โ„ (or in any interval in the real line) are the linear functions. Example 2.2. The quadratic polynomial ๐‘ข(๐‘ฅ, ๐‘ฆ) = ๐‘ฅ2 โˆ’ ๐‘ฆ2 is harmonic in โ„2 , as its second derivatives are equal to ๐œ•12 ๐‘ข = 2 and ๐œ•22 ๐‘ข = โˆ’2. Note that ๐‘ข is the real part of the holomorphic function ๐‘“(๐‘ง) = ๐‘ง2 . As observed in Section 1.2, the real and imaginary parts of an analytic function are harmoinc. The imaginary part of ๐‘ง2 , and thus a conjugate harmonic to ๐‘ข, is ๐‘ฃ(๐‘ฅ, ๐‘ฆ) = 2๐‘ฅ๐‘ฆ. Similarly, the functions ๐‘ข(๐‘ฅ, ๐‘ฆ) = โ„œ((๐‘ฅ + ๐‘–๐‘ฆ)๐‘› )

and

๐‘ฃ(๐‘ฅ, ๐‘ฆ) = โ„‘((๐‘ฅ + ๐‘–๐‘ฆ)๐‘› ),

the real and imaginary parts of ๐‘ง๐‘› , are harmonic in โ„2 for each ๐‘› โˆˆ โ„•. 13

14

2. Basic properties

Example 2.3. The function ๐‘ข(๐‘ฅ, ๐‘ฆ) = sin ๐‘ฅ sinh ๐‘ฆ is harmonic in โ„2 , because and ๐œ•22 ๐‘ข = ๐‘ข. ๐œ•12 ๐‘ข = โˆ’๐‘ข Note that ๐‘ข is the imaginary part of the holomorphic function โˆ’ cos ๐‘ง. It is easy to see that the harmonic functions in any open set in โ„๐‘‘ form a vector space because, if ๐‘ข and ๐‘ฃ are harmonic, then any linear combination ๐›ผ๐‘ข + ๐›ฝ๐‘ฃ of ๐‘ข and ๐‘ฃ is also a harmonic function. Moreover, the space of harmonic functions in โ„๐‘‘ is invariant under translations and orthogonal transformations (see Exercises (1) and (2)). We observed in Example 2.1 that the linear functions are harmonic functions and, in fact they are the only harmonic functions in โ„. We now make a rather immediate observation: if ๐‘ข is linear, say in the interval [๐‘Ž, ๐‘], then its value at the midpoint (๐‘Ž + ๐‘)/2 is the average of its values at ๐‘Ž and ๐‘, ๐‘Ž+๐‘ ๐‘ข(๐‘Ž) + ๐‘ข(๐‘) ๐‘ข( . )= 2 2 It turns out that this is true in every dimension. Theorem 2.4 (Mean value property). Let ๐‘ข be a harmonic function in a neighborhhod of the closed ball ๐ต๐‘Ÿฬ„ (๐‘ฅ0 ). Then (2.5)

๐‘ข(๐‘ฅ0 ) =

1 โˆซ ๐‘ข(๐œ‰)๐‘‘๐œŽ(๐œ‰). |๐‘†๐‘Ÿ (๐‘ฅ0 )| ๐‘† (๐‘ฅ ) ๐‘Ÿ

0

In other words, the average of the values of ๐‘ข over any sphere around ๐‘ฅ0 is equal to ๐‘ข(๐‘ฅ0 ). In the identity (2.5), ๐‘‘๐œŽ is the surface measure on the sphere ๐‘†๐‘Ÿ (๐‘ฅ0 ), and |๐‘†๐‘Ÿ (๐‘ฅ0 )| is its surface area, |๐‘†๐‘Ÿ (๐‘ฅ0 )| = ๐œ”๐‘‘ ๐‘Ÿ๐‘‘โˆ’1 . With an appropiate change of variables, we can also write (2.5) as (2.6)

๐‘ข(๐‘ฅ0 ) =

1 โˆซ ๐‘ข(๐‘ฅ0 + ๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰), ๐œ”๐‘‘ ๐•Š

where ๐•Š is the unit sphere around the origin.

2.1. The mean value property

15

2.7. Integrating over the ball ๐ต๐‘Ÿ (๐‘ฅ0 ) in spherical coordinates, it follows from (2.5) that, under the same assumptions of Theorem 2.4, ๐‘ข(๐‘ฅ0 ) =

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ0 )| ๐ต

๐‘ข(๐‘ฅ)๐‘‘๐‘ฅ =

๐‘Ÿ (๐‘ฅ0 )

๐‘‘ โˆซ ๐‘ข(๐‘ฅ0 + ๐‘Ÿ๐‘ฅ)๐‘‘๐‘ฅ, ๐œ”๐‘‘ ๐”น

where ๐”น is the unit ball centered at the origin. We leave this as an exercise (Exercise (3)). Again, this identity is immediate in โ„ (Exercise (4)). Proof of Theorem 2.4. We only need to prove the case ๐‘‘ โ‰ฅ 2, by the observations made before the statement of the theorem. By translating by ๐‘ฅ0 , we can assume ๐‘ฅ0 = 0 (see Exercise (1)). For 0 < ๐œ€ < ๐‘Ÿ, let ฮฉ = ๐ต๐‘Ÿ โงต ๐ต๐œ€ฬ„ , where ๐ต๐‘Ÿ = ๐ต๐‘Ÿ (0) and ๐ต๐œ€ = ๐ต๐œ€ (0), as in Figure 2.1. Define

Figure 2.1. The domain ฮฉ = ๐ต๐‘Ÿ โงต ๐ต๐œ€ฬ„ . On a point in ๐‘† ๐‘Ÿ , the normal vector ๐œˆ points away from the origin, while on a point in ๐‘† ๐œ€ points towards the origin.

the function ๐‘ฃ in โ„๐‘‘ โงต {0} by log |๐‘ฅ| ๐‘ฃ(๐‘ฅ) = { 2โˆ’๐‘‘ |๐‘ฅ|

๐‘‘=2

๐‘‘ โ‰ฅ 3. ๐‘ฅ Note that, for any ๐‘ฅ โˆˆ ฮฉ, โˆ‡๐‘ฃ(๐‘ฅ) = ๐‘ ๐‘‘ ๐‘‘ , where ๐‘ ๐‘‘ = 1 if ๐‘‘ = 2 and |๐‘ฅ| ๐‘ ๐‘‘ = 2 โˆ’ ๐‘‘ if ๐‘‘ โ‰ฅ 3. Also, ๐œ•ฮฉ = ๐‘†๐‘Ÿ โˆ’ ๐‘†๐œ€ , where we have written ๐‘†๐‘Ÿ and ๐‘†๐œ€ for ๐‘†๐‘Ÿ (0) and ๐‘†๐œ€ (0), respectively (as oriented manifolds,1 see Figure 2.1), 1 We are only interested, at this moment, in the fact that the normal vectors on โˆ’๐‘†๐‘’ point opposite to those at ๐‘†๐‘Ÿ .

16

2. Basic properties

and

๐‘ฅ ๐œˆ(๐‘ฅ) = { ๐‘Ÿ ๐‘ฅ โˆ’ ๐œ€

on ๐‘†๐‘Ÿ on โˆ’ ๐‘†๐‘’ .

Thus

๐‘๐‘‘ on ๐‘†๐‘Ÿ ๐‘‘โˆ’1 ๐‘Ÿ ๐œ•๐œˆ ๐‘ฃ = { ๐‘ ๐‘‘ โˆ’ ๐‘‘โˆ’1 on โˆ’ ๐‘†๐‘’ . ๐œ€ We can also verify explicitly that ฮ”๐‘ฃ = 0 (Exercise (5)), and hence โˆซ (๐‘ขฮ”๐‘ฃ โˆ’ ๐‘ฃฮ”๐‘ข)๐‘‘๐‘ฅ = 0. โ„ฆ

Applying Greenโ€™s identity (1.14), and the previous explicit calculations, we obtain 0 = โˆซ (๐‘ข๐œ•๐œˆ ๐‘ฃ โˆ’ ๐‘ฃ๐œ•๐œˆ ๐‘ข)๐‘‘๐œŽ ๐œ•โ„ฆ

= โˆซ (๐‘ข ๐‘†๐‘Ÿ

๐‘๐‘‘ ๐‘๐‘‘ โˆ’ ๐œ๐‘‘ (๐‘Ÿ)๐œ•๐œˆ ๐‘ข)๐‘‘๐œŽ โˆ’ โˆซ (๐‘ข ๐‘‘โˆ’1 โˆ’ ๐œ๐‘‘ (๐œ€)๐œ•๐œˆ ๐‘ข)๐‘‘๐œŽ, ๐‘Ÿ๐‘‘โˆ’1 ๐‘’ ๐‘†๐‘’

where log ๐‘  ๐‘‘ = 2 ๐œ๐‘‘ (๐‘ ) = { 1 ๐‘‘ โ‰ฅ 3, ๐‘ ๐‘‘โˆ’2 for ๐‘  = ๐‘‘ or ๐‘  = ๐œ€, which are constant over ๐‘†๐‘Ÿ and ๐‘†๐œ€ . Thus, since the surface integral of ๐œ•๐œˆ ๐‘ข over a sphere is zero (Exercise (6)), we obtain ๐‘๐‘‘ ๐‘๐‘‘ โˆซ ๐‘ข๐‘‘๐œŽ = ๐‘‘โˆ’1 โˆซ ๐‘ข๐‘‘๐œŽ ๐‘Ÿ๐‘‘โˆ’1 ๐‘†๐‘Ÿ ๐œ€ ๐‘†๐œ€ for any ๐œ€ > 0. Since ๐‘ข is continuous we obtain, taking ๐œ€ โ†’ 0 (see Exercise (7)), 1 โˆซ ๐‘ข๐‘‘๐œŽ = ๐‘ข(0). ๐œ”๐‘‘ ๐‘Ÿ๐‘‘โˆ’1 ๐‘†๐‘Ÿ โ–ก If a continuous function ๐‘ข in โ„ satisfies the mean value property, that is, ๐‘ข(๐‘ฅ) + ๐‘ข(๐‘ฆ) ๐‘ฅ+๐‘ฆ ๐‘ข( )= 2 2

2.1. The mean value property

17

for all ๐‘ฅ, ๐‘ฆ โˆˆ โ„, then ๐‘ข must be a linear function. Indeed, let ๐‘Ž = ๐‘ข(1) โˆ’ ๐‘ข(0) and ๐‘ = ๐‘ข(0), so we have ๐‘ข(1) = ๐‘Ž + ๐‘ and ๐‘ข(0) = ๐‘. Since ๐‘ข(1) = (๐‘ข(0) + ๐‘ข(2))/2, we see that ๐‘ข(2) = 2๐‘ข(1) โˆ’ ๐‘ข(0) = 2๐‘Ž + ๐‘, and we can verify, inductively, that (2.8)

๐‘ข(๐‘›) = ๐‘Ž๐‘› + ๐‘

for every ๐‘› โˆˆ โ„•. We can similarly prove that (2.8) holds for negative integers ๐‘›. Now, for every ๐‘› โˆˆ โ„ค, ๐‘ข(๐‘›) + ๐‘ข(๐‘› + 1) ๐‘Ž๐‘› + ๐‘ + ๐‘Ž(๐‘› + 1) + ๐‘ 2๐‘› + 1 = )= 2 2 2 2๐‘› + 1 = ๐‘Ž( ) + ๐‘, 2 and similarly for every number of the form ๐‘˜/2๐‘› , for every ๐‘˜ โˆˆ โ„ค and every ๐‘› โˆˆ โ„•. Since such numbers are dense in โ„ and ๐‘ข is continuous, we conclude that ๐‘ข(๐‘ฅ) = ๐‘Ž๐‘ฅ + ๐‘ ๐‘ข(

for every ๐‘ฅ โˆˆ โ„. As we have stated above, the linear functions are the harmonic functions in โ„ so, therefore, the continuous functions that satisfy the mean value property are precisely the harmonic functions. This is also true in higher dimensions. Theorem 2.9 (Converse to the mean value property). Let ฮฉ โŠ‚ โ„๐‘‘ be open and ๐‘ข a continuous function on ฮฉ that satisfies that, whenever ๐ต๐‘Ÿฬ„ (๐‘ฅ) โŠ‚ ฮฉ, (2.10)

๐‘ข(๐‘ฅ) =

1 โˆซ ๐‘ข(๐‘ฅ + ๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰). ๐œ”๐‘‘ ๐•Š

Then ๐‘ข โˆˆ ๐ถ โˆž (ฮฉ) and ๐‘ข is harmonic in ฮฉ. As harmonic functions must be twice differentiable, we must prove that a function ๐‘ข that satisfies (2.10) is at least twice differentiable before proving that ฮ”๐‘ข = 0 in ฮฉ. However, the conclusion of Theorem 2.9 is much stronger: ๐‘ข is actually an infinitely differentiable function. We thus conclude Corollary 2.11, which follows by applying Theorems 2.4 and 2.9. Corollary 2.11. If ๐‘ข is harmonic in an open set ฮฉ, then it is infinitely differentiable in ฮฉ.

18

2. Basic properties In order to prove Theorem 2.9, we will make use of Lemma 2.12.

Lemma 2.12. There exists a smooth radial function ๐œ™ on โ„๐‘‘ such that it is supported in ๐”น and โˆซ ๐œ™ = 1. Proof. Consider the function ๐œ“ on โ„ given by 1

๐‘’ (4๐‘กโˆ’1)(2๐‘กโˆ’1) ๐œ“(๐‘ฅ) = { 0

1/4 < ๐‘ก < 1/2 otherwise.

It is a standard calculus exercise to verify that ๐œ“ is a smooth function in โ„, supported in [1/4, 1/2]. Indeed, it is infinitely flat at the points 1/4 and 1/2 (see Figure 2.2). Now, we define on โ„๐‘‘ the function ๐œ™(๐‘ฅ) = ๐‘๐œ“(|๐‘ฅ|),

1

1

4

2

Figure 2.2. The cut-off function ๐œ“(๐‘ก). Note that it is supported in [1/4, 1/2], and infinitely flat at the points 1/4 and 1/2.

where ๐‘ is such that โˆซ ๐œ™(๐‘ฅ)๐‘‘๐‘ฅ = 1. โ„๐••

Such ๐‘ exists because ๐œ“ is nonnegative, and thus โˆซโ„๐•• ๐œ“(|๐‘ฅ|)๐‘‘๐‘ฅ > 0. Now, ๐œ“ is ๐ถ โˆž and supported away from zero, and hence ๐œ™ โˆˆ ๐ถ โˆž (โ„๐‘‘ ), because

2.1. The mean value property

19

๐‘ฅ โ†ฆ |๐‘ฅ| is smooth away from zero. Finally, as ๐œ“(๐‘ก) = 0 unless 1/4 < ๐‘ก < 1/2, then ๐œ™(๐‘ฅ) = 0 unless 1/4 < |๐‘ฅ| < 1/2, and thus supp ๐œ™ โŠ‚ ๐”น. โ–ก Proof of Theorem 2.9. Let ๐‘ข be a continuous function that satisfies ฬ„ (๐‘ฅ0 ) โŠ‚ ฮฉ. (2.10) for every ๐ต๐‘Ÿฬ„ (๐‘ฅ) โŠ‚ ฮฉ. Let ๐‘ฅ0 โˆˆ ฮฉ, and ๐œ€ > 0 such that ๐ต2๐œ€ ฬƒ Let ๐œ™(๐‘ฅ) = ๐œ™(|๐‘ฅ|) as in Lemma 2.12, and define the function ๐œ™๐œ€ (๐‘ฅ) = ๐œ€โˆ’๐‘‘ ๐œ™(๐œ€โˆ’1 ๐‘ฅ). Note that ๐œ™๐œ€ โˆˆ ๐ถ โˆž (โ„๐‘‘ ), it is supported in ๐ต๐œ€ (0), and โˆซ ๐œ™๐œ€ (๐‘ฅ)๐‘‘๐‘ฅ = 1. โ„๐••

In particular, for any ๐‘ฅ โˆˆ ๐ต๐œ€ (๐‘ฅ0 ), the function ๐‘ฆ โ†ฆ ๐œ™๐œ€ (๐‘ฅ โˆ’๐‘ฆ) is supported in ๐ต2๐œ€ (๐‘ฅ0 ) โŠ‚ ฮฉ. Hence, we observe that, for ๐‘ฅ โˆˆ ๐ต๐œ€ (๐‘ฅ0 ), โˆซ ๐‘ข(๐‘ฆ)๐œ™๐œ€ (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ = โˆซ ๐‘ข(๐‘ฅ โˆ’ ๐‘ฆ)๐œ™๐œ€ (๐‘ฆ)๐‘‘๐‘ฆ = โˆซ โ„๐••

โ„๐•• ๐œ€

๐‘ข(๐‘ฅ โˆ’ ๐‘ฆ)๐œ™๐œ€ (๐‘ฆ)๐‘‘๐‘ฆ

๐ต๐œ€ (0)

ฬƒ โˆ’1 ๐‘Ÿ)๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ. = โˆซ โˆซ ๐‘ข(๐‘ฅ โˆ’ ๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰) โ‹… ๐œ€โˆ’๐‘‘ ๐œ™(๐œ€ 0

๐•Š

Using (2.10) we obtain ๐œ€

ฬƒ โˆ’1 ๐‘Ÿ)๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ โˆซ ๐‘ข(๐‘ฆ)๐œ™๐œ€ (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ = ๐‘ข(๐‘ฅ) โ‹… ๐œ”๐‘‘ โˆซ ๐œ€โˆ’๐‘‘ ๐œ™(๐œ€ โ„๐••

0 ๐œ€

ฬƒ โˆ’1 ๐‘Ÿ)๐‘‘๐œŽ(๐œ‰)๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ = ๐‘ข(๐‘ฅ) โˆซ โˆซ ๐œ€โˆ’๐‘‘ ๐œ™(๐œ€ 0

๐•Š

= ๐‘ข(๐‘ฅ) โˆซ

๐œ™๐œ€ (๐‘ฆ)๐‘‘๐‘ฆ = ๐‘ข(๐‘ฅ).

๐ต๐œ€ (0)

Note that the function ๐‘ข(๐‘ฆ)๐œ™๐œ€ (๐‘ฅ โˆ’ ๐‘ฆ) is ๐ถ โˆž in ๐‘ฅ and continuous with compact support in ๐‘ฆ, so we can differentiate under the integral the function ๐‘ฅ โ†ฆ โˆซ ๐‘ข(๐‘ฆ)๐œ™๐œ€ (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ โ„๐••

as many times as we want, and thus we conclude ๐‘ข โˆˆ ๐ถ โˆž (๐ต๐œ€ (๐‘ฅ0 )). In particular, ฮ”๐‘ข is a continuous function in ๐ต๐œ€ (๐‘ฅ0 ). Now, for any ๐‘ฅ โˆˆ ๐ต๐œ€ (๐‘ฅ0 ) and 0 < ๐‘Ÿ < ๐œ€ such that ๐ต๐‘Ÿฬ„ (๐‘ฅ) โŠ‚ ๐ต๐œ€ (๐‘ฅ0 ), ๐‘ข(๐‘ฅ) =

1 โˆซ ๐‘ข(๐‘ฅ + ๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰), ๐œ”๐‘‘ ๐•Š

20

2. Basic properties

so if we differentiate with respect to ๐‘Ÿ we obtain 0= =

๐‘‘ โˆซ ๐‘ข(๐‘ฅ + ๐‘Ÿ๐œ‰)๐‘‘๐œŽ(๐œ‰) = โˆซ โˆ‡๐‘ข(๐‘ฅ + ๐‘Ÿ๐œ‰) โ‹… ๐œ‰๐‘‘๐œŽ(๐œ‰) ๐‘‘๐‘Ÿ ๐•Š ๐•Š 1 ๐‘Ÿ๐‘‘โˆ’1

โˆซ

๐œ•๐œˆ ๐‘ข๐‘‘๐œŽ =

๐‘†๐‘Ÿ (๐‘ฅ)

1 ๐‘Ÿ๐‘‘โˆ’1

โˆซ

ฮ”๐‘ข๐‘‘๐‘ฅ.

๐ต๐‘Ÿ (๐‘ฅ)

In the last equality we have used Greenโ€™s identity (1.14) with ๐‘ฃ = โˆ’1. Hence, the integral of ฮ”๐‘ข over any ball in ๐ต๐œ€ (๐‘ฅ0 ) is zero. By the continuity of ฮ”๐‘ข, ฮ”๐‘ข = 0 in ๐ต๐œ€ (๐‘ฅ0 ). Since ๐‘ฅ0 โˆˆ ฮฉ is arbitrary, we conclude ๐‘ข is harmonic in ฮฉ.

โ–ก

2.2. The maximum principle From the mean value property 2.4 we obtain another basic property of harmonic functions, the maximum principle. Corollary 2.13 (Maximum principle). If ฮฉ โŠ‚ โ„๐‘‘ is a domain and ๐‘ข is harmonic in ฮฉ, then ๐‘ข does not take a maximum nor a minumum in ฮฉ, unless ๐‘ข is constant. This is easy to see in the case ๐‘‘ = 1, where harmonic functions coincide with linear functions: if ๐‘ข is linear in the interval (๐‘Ž, ๐‘), then it clearly does not take neither a maximum or a minimum, because ๐‘ข is either strictly increasing or strictly decreasing, unless it is constant. Proof of Corollary 2.13. Suppose that ๐‘ข takes its maximum ๐‘€ at some ๐‘ฅ0 โˆˆ ฮฉ, so ๐‘ข(๐‘ฅ0 ) = ๐‘€. Let ๐‘ˆ = {๐‘ฅ โˆˆ ฮฉ โˆถ ๐‘ข(๐‘ฅ) = ๐‘€} be the set of points in ฮฉ where ๐‘ข takes the value ๐‘€. Note that ๐‘ˆ โ‰  โˆ… because ๐‘ฅ0 โˆˆ ๐‘ˆ. We prove that ๐‘ˆ = ฮฉ. First, ๐‘ˆ is closed in ฮฉ because ๐‘ˆ = ๐‘ขโˆ’1 ({๐‘€}) and ๐‘ข is continuous, so it is the pre-image of a closed set under a continuous function.2 Now let ๐‘ฅ โˆˆ ๐‘ˆ. Since ฮฉ is open, there exists ๐‘Ÿ > 0 such that ๐ต๐‘Ÿฬ„ (๐‘ฅ) โŠ‚ ฮฉ. By the mean value property, ๐‘ข(๐‘ฅ) =

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ.

๐‘Ÿ (๐‘ฅ)

2

See Sections A.2 and A.4 for a summary of results from topology in Euclidean spaces.

2.2. The maximum principle

21

As we are assuming ๐‘ฅ โˆˆ ๐‘ˆ, this integral must equal ๐‘€. Now ๐‘ข(๐‘ฆ) โ‰ค ๐‘€ for all ๐‘ฆ โˆˆ ฮฉ, because ๐‘€ is the maximum of ๐‘ข. Hence, if at some ๐‘ฆ โˆˆ ๐ต๐‘Ÿ (๐‘ฅ) we had ๐‘ข(๐‘ฆ) < ๐‘€, this integral would be smaller than ๐‘€, because ๐‘ข is continuous. Thus ๐‘ข(๐‘ฆ) = ๐‘€ for all ๐‘ฆ โˆˆ ๐ต๐‘Ÿ (๐‘ฅ). Therefore ๐ต๐‘Ÿ (๐‘ฅ) โŠ‚ ๐‘ˆ, and ๐‘ˆ is also open in ฮฉ. Since ฮฉ is connected, we conclude ๐‘ˆ = ฮฉ, and therefore ๐‘ข is the constant function ๐‘ข(๐‘ฅ) = ๐‘€. By taking โˆ’๐‘ข, we also see that ๐‘ข takes its minimum in ฮฉ only if it is constant. โ–ก The maximum principle implies that if ฮฉ is bounded, so ฮฉฬ„ is compact, and ๐‘ข is harmonic in ฮฉ and continuous on ฮฉ,ฬ„ then ๐‘ข takes its maximum (and its minimum) at the boundary of ฮฉ (Exercise (8)). Again, this is clear in the case of a linear function in a closed interval. The maximum principle also implies uniqueness of harmonic functions in a bounded domain ฮฉ, given their values on the boundary. See Exercise (9) for details. The maximum principle states that harmonic functions, unless they are constant, do not take their maxima nor minima in their domains. If the domain ฮฉ is bounded, a nonconstant harmonic function in ฮฉ may be bounded, of course, and in that case it may be possible to extend it to the boundary of ฮฉ, and hence its extrema would be achieved in ๐œ•ฮฉ. Even if the domain ฮฉ is unbounded, we may have bounded harmonic functions, as we will see later on. However, in the case where the domain of the harmonic function is all of the Euclidean space, we have the following result. Theorem 2.14 (Liouville). If ๐‘ข is harmonic and bounded in โ„๐‘‘ , then it is constant. Proof. Suppose ๐‘ข is harmonic and |๐‘ข(๐‘ฅ)| โ‰ค ๐‘€ for all ๐‘ฅ โˆˆ โ„๐‘‘ . We prove that ๐‘ข(๐‘ฅ) = ๐‘ข(0) for all ๐‘ฅ โˆˆ โ„๐‘‘ . Fix ๐‘ฅ โˆˆ โ„๐‘‘ and let ๐‘… > |๐‘ฅ|. By the mean value property in balls, 2.7, we have that ๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0) =

1 โˆซ |๐ต๐‘… (๐‘ฅ)| ๐ต

๐‘… (๐‘ฅ)

๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ โˆ’

1 โˆซ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ, |๐ต๐‘… | ๐ต ๐‘…

22

2. Basic properties

where ๐ต๐‘… = ๐ต๐‘… (0). Since |๐ต๐‘… (๐‘ฅ)| = |๐ต๐‘… | = ๐œ”๐‘‘ ๐‘…๐‘‘ /๐‘‘, we can write this difference as ๐‘‘ ๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0) = ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ โˆ’ โˆซ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ). (โˆซ ๐œ”๐‘‘ ๐‘…๐‘‘ ๐ต๐‘… (๐‘ฅ) ๐ต๐‘… Now, if ๐ด is the annulus ๐ด = {๐‘ฆ โˆˆ โ„๐‘‘ โˆถ ๐‘… โˆ’ |๐‘ฅ| โ‰ค |๐‘ฆ| โ‰ค ๐‘… + |๐‘ฅ|}, we see that the symmetric difference of the balls ๐ต๐‘… (๐‘ฅ) and ๐ต๐‘… satisfies ๐ต๐‘… (๐‘ฅ) โ–ณ ๐ต๐‘… โŠ‚ ๐ด. (See Figure 2.3) Indeed, if ๐‘ฆ โˆˆ ๐ต๐‘… (๐‘ฅ) โงต ๐ต๐‘… , then |๐‘ฆ โˆ’ ๐‘ฅ| < ๐‘… and |๐‘ฆ| โ‰ฅ

Figure 2.3. The annulus ๐ด containing the symmetric difference of the balls ๐ต๐‘… (๐‘ฅ) and ๐ต๐‘… .

๐‘… โ‰ฅ ๐‘… โˆ’ |๐‘ฅ|, and further |๐‘ฆ| โ‰ค |๐‘ฆ โˆ’ ๐‘ฅ| + |๐‘ฅ| < ๐‘… + |๐‘ฅ|; similarly, if ๐‘ฆ โˆˆ ๐ต๐‘… โงต ๐ต๐‘… (๐‘ฅ), then |๐‘ฆ| < ๐‘… โ‰ค ๐‘… + |๐‘ฅ| and |๐‘ฆ โˆ’ ๐‘ฅ| โ‰ฅ ๐‘…, so |๐‘ฆ| โ‰ฅ |๐‘ฆ โˆ’ ๐‘ฅ| โˆ’ |๐‘ฅ| โ‰ฅ ๐‘… โˆ’ |๐‘ฅ|. If we integrate using spherical coordinates, we obtain ๐‘…+|๐‘ฅ|

๐‘‘ ๐‘‘ โˆซ |๐‘ข(๐‘ฆ)|๐‘‘๐‘ฆ โ‰ค โˆซโˆซ ๐‘€๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ๐‘‘๐œŽ |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0)| โ‰ค ๐œ” ๐‘‘ ๐‘…๐‘‘ ๐ด ๐œ”๐‘‘ ๐‘…๐‘‘ ๐•Š ๐‘…โˆ’|๐‘ฅ| =

(๐‘… + |๐‘ฅ|)๐‘‘ โˆ’ (๐‘… โˆ’ |๐‘ฅ|)๐‘‘ ๐ถ ๐‘€ ๐‘‘ โ‹… ๐‘€๐œ” โ‹… โ‰ค ๐‘ฅ , ๐‘‘ ๐‘‘ ๐‘… ๐œ” ๐‘‘ ๐‘…๐‘‘

2.2. The maximum principle

23

where we have used the fact that |๐‘ข(๐‘ฆ)| โ‰ค ๐‘€ for all ๐‘ฆ โˆˆ โ„๐‘‘ , and the constant ๐ถ๐‘ฅ only depends on ๐‘‘ and |๐‘ฅ|. As ๐‘ฅ is fixed and ๐‘… is arbitrary, we conclude that |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0)| = 0, and therefore ๐‘ข(๐‘ฅ) = ๐‘ข(0). โ–ก We have seen above that, if ๐‘“(๐‘ง) is holomorphic, then it is real and imaginary parts are harmonic functions. Therefore, Theorem 2.14 implies that, if ๐‘“ is a holomorphic function in โ„‚ (such a function is called an entire function) and is bounded, then ๐‘“ must be constant (this is also know as Liouvilleโ€™s theorem). This fact provides a proof for the fundamental theorem of algebra: If ๐‘(๐‘ง) is polynomial over โ„‚ of degree at least 1, then it has a root in โ„‚. Indeed, if ๐‘(๐‘ง) is a polynomial over โ„‚ with no roots, then 1/๐‘(๐‘ง) is an entire bounded function, and thus constant, so ๐‘(๐‘ง) is a constant polynomial. See Exercise (10) for the details. We can refine Theorem 2.14 to obtain the same conclusion even when ๐‘ข is only bounded from below or from above. Theorem 2.15. If ๐‘ข is harmonic and nonnegative in โ„๐‘‘ , then it is constant. Proof. The proof of Theorem 2.15 follows similarly as the proof of Theorem 2.14, but we now have to be more careful when estimating the difference ๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0) =

๐‘‘ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ โˆ’ โˆซ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ), (โˆซ ๐œ”๐‘‘ ๐‘…๐‘‘ ๐ต๐‘… (๐‘ฅ) ๐ต๐‘…

for ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘… > |๐‘ฅ|. This time, we use the fact that ๐‘ข(๐‘ฆ) โ‰ฅ 0 to observe that, if again ๐ด is the annulus ๐ด = {๐‘ฆ โˆˆ โ„๐‘‘ โˆถ ๐‘… โˆ’ |๐‘ฅ| โ‰ค |๐‘ฆ| โ‰ค ๐‘… + |๐‘ฅ|}, then |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0)| โ‰ค =

๐‘‘ โˆซ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ ๐œ” ๐‘‘ ๐‘…๐‘‘ ๐ด ๐‘‘ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ โˆ’ โˆซ ๐‘ข(๐‘ฆ)๐‘‘๐‘ฆ) (โˆซ ๐œ”๐‘‘ ๐‘…๐‘‘ ๐ต๐‘…+|๐‘ฅ| ๐ต๐‘…โˆ’|๐‘ฅ|

where ๐ต๐‘…+|๐‘ฅ| and ๐ต๐‘…โˆ’|๐‘ฅ| are the balls of radii ๐‘… + |๐‘ฅ| and ๐‘… โˆ’ |๐‘ฅ| centered at 0, respectively. We use again the mean value property and, as before,

24

2. Basic properties

for some constant ๐ถ๐‘ฅ that depends only on ๐‘‘ and |๐‘ฅ|, |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(0)| โ‰ค

๐‘‘ |๐‘ข(0) โˆ’ |๐ต๐‘…โˆ’|๐‘ฅ| |๐‘ข(0)) (|๐ต ๐œ”๐‘‘ ๐‘…๐‘‘ ๐‘…+|๐‘ฅ|

(๐‘… + |๐‘ฅ|)๐‘‘ โˆ’ (๐‘… โˆ’ |๐‘ฅ|)๐‘‘ โ‹… ๐‘ข(0) ๐‘…๐‘‘ ๐ถ ๐‘ข(0) โ‰ค ๐‘ฅ . ๐‘… We can now conclude again that ๐‘ข(๐‘ฅ) = ๐‘ข(0). =

โ–ก

It is clear that we obtain the same conclusion of Theorem 2.15 whenever ๐‘ข is a harmonic function in โ„๐‘‘ and there exists a constant ๐›ผ โˆˆ โ„ such that either ๐‘ข(๐‘ฅ) โ‰ฅ ๐›ผ for all ๐‘ฅ โˆˆ โ„๐‘‘ , or ๐‘ข(๐‘ฅ) โ‰ค ๐›ผ for all ๐‘ฅ โˆˆ โ„๐‘‘ .

2.3. Poisson kernel and Poisson integrals in the ball We have seen that the value of a harmonic function at the center of a sphere is equal to the average of its values over the sphere. One can ask naturally if the value at any other point in the interior of the sphere is similarly determined by the values over the sphere, and if this value corresponds to a perhaps weighted average over such values. This is certainly the case for a linear function in a close interval [๐‘Ž, ๐‘]: if ๐‘ก โˆˆ (๐‘Ž, ๐‘) and ๐‘ข is linear, then ๐‘โˆ’๐‘ก ๐‘กโˆ’๐‘Ž ๐‘ข(๐‘Ž) + ๐‘ข(๐‘), ๐‘โˆ’๐‘Ž ๐‘โˆ’๐‘Ž which is a convex combination of ๐‘ข(๐‘Ž) and ๐‘ข(๐‘). ๐‘ข(๐‘ก) =

We will prove that this is true for harmonic functions in โ„๐‘‘ , for ๐‘‘ โ‰ฅ 2, as well. The weight function is called the Poisson kernel. For ๐‘ฅ โˆˆ ๐”น and ๐œ‰ โˆˆ ๐•Š, we define (2.16)

๐‘ƒ(๐‘ฅ, ๐œ‰) =

1 1 โˆ’ |๐‘ฅ|2 . ๐œ”๐‘‘ |๐‘ฅ โˆ’ ๐œ‰|๐‘‘

The Poisson kernel satisfies the following facts. 2.17. ๐‘ƒ(๐‘ฅ, ๐œ‰) > 0 for any ๐‘ฅ โˆˆ ๐”น and ๐œ‰ โˆˆ ๐•Š, which is easily seen from (2.16) because |๐‘ฅ| < 1. 2.18. For each fixed ๐œ‰ โˆˆ ๐•Š, the function ๐‘ฅ โ†ฆ ๐‘ƒ(๐‘ฅ, ๐œ‰) is harmonic in ๐”น. This is followed by explicit differentiation (Exercise (13)).

2.3. Poisson kernel and Poisson integrals in the ball

25

2.19. For each fixed ๐‘ฅ โˆˆ ๐”น, โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰) = 1.

(2.20)

๐•Š

This is clear if ๐‘ฅ = 0, since (2.21)

๐‘ƒ(0, ๐œ‰) =

1 ๐œ”๐‘‘

for any ๐œ‰ โˆˆ ๐•Š, and thus โˆซ ๐‘ƒ(0, ๐œ‰)๐‘‘๐œŽ(๐œ‰) = โˆซ ๐•Š

๐•Š

1 ๐‘‘๐œŽ(๐œ‰) = 1. ๐œ”๐‘‘

For ๐‘ฅ โ‰  0, ๐‘ฅ/|๐‘ฅ| โˆˆ ๐•Š and, since ๐‘ƒ(โ‹…, ๐‘ฅ/|๐‘ฅ|) is harmonic in ๐”น, the mean value property implies (2.22)

๐‘ƒ(0,

๐‘ฅ 1 ๐‘ฅ โˆซ ๐‘ƒ(|๐‘ฅ|๐œ‰, )= )๐‘‘๐œŽ(๐œ‰), |๐‘ฅ| ๐œ”๐‘‘ ๐•Š |๐‘ฅ|

because the integral on the right side is the average over the sphere centered at the origin of radius |๐‘ฅ| < 1, contained in the ball ๐”น. By the identity ||๐‘ฅ|๐œ‰ โˆ’ ๐‘ฅ | = |๐‘ฅ โˆ’ ๐œ‰|, | |๐‘ฅ| | known as the symmetry lemma (Exercise (14)), we have that 2

๐‘ƒ(|๐‘ฅ|๐œ‰,

1 1 โˆ’ ||๐‘ฅ|๐œ‰ | 1 1 โˆ’ |๐‘ฅ|2 ๐‘ฅ = = ๐‘ƒ(๐‘ฅ, ๐œ‰), )= |๐‘ฅ| ๐œ”๐‘‘ | ๐œ”๐‘‘ |๐‘ฅ โˆ’ ๐œ‰|๐‘‘ ๐‘ฅ |๐‘‘ ||๐‘ฅ|๐œ‰ โˆ’ |๐‘ฅ| |

and thus ๐‘ƒ(0,

๐‘ฅ 1 โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰). )= |๐‘ฅ| ๐œ”๐‘‘ ๐•Š

The identity (2.20) follows using (2.21). Observe that, if ๐‘‘ = 1, we simply have ๐œ”1 = 2 and hence ๐‘ƒ(๐‘ฅ, โˆ’1) =

1 1 โˆ’ |๐‘ฅ|2 1 = (1 โˆ’ ๐‘ฅ) 2 |๐‘ฅ โˆ’ (โˆ’1)| 2

and ๐‘ƒ(๐‘ฅ, 1) =

1 1 โˆ’ |๐‘ฅ|2 1 = (1 + ๐‘ฅ) 2 |๐‘ฅ โˆ’ 1| 2

26

2. Basic properties

for every ๐‘ฅ โˆˆ (โˆ’1, 1), since the boundary of the unit interval (โˆ’1, 1) is the set {โˆ’1, 1} of two points. Note that each function ๐‘ฅ โ†ฆ ๐‘ƒ(๐‘ฅ, ยฑ1) is linear, so it is harmonic in (โˆ’1, 1), and (2.20) is just ๐‘ƒ(๐‘ฅ, โˆ’1) + ๐‘ƒ(๐‘ฅ, 1) = 1. Example 2.23. In the case when ๐‘‘ = 2, we can write the Poisson kernel ๐‘ƒ(๐‘ฅ, ๐œ‰) in polar coordinates. Indeed, if ๐‘ฅ = ๐‘Ÿ๐‘’๐‘–๐œƒ for some 0 โ‰ค ๐‘Ÿ < 1 and ๐œ‰ = ๐‘’๐‘–๐œ , then ๐‘ƒ(๐‘Ÿ๐‘’๐‘–๐œƒ , ๐‘’๐‘–๐œ ) =

1 1 โˆ’ ๐‘Ÿ2 1 1 โˆ’ ๐‘Ÿ2 = โ‹… . 2๐œ‹ |๐‘Ÿ๐‘’๐‘–๐œƒ โˆ’ ๐‘’๐‘–๐œ |2 2๐œ‹ 1 + 2๐‘Ÿ cos(๐œ โˆ’ ๐œƒ) + ๐‘Ÿ2

Note that (2.20) is now the identity 2๐œ‹

1 โˆซ 2๐œ‹ 0

1 โˆ’ ๐‘Ÿ2 ๐‘‘๐œ = 1. 1 + 2๐‘Ÿ cos(๐œ โˆ’ ๐œƒ) + ๐‘Ÿ2

The following fact states that, as we approach a point in the boundary, the weight of the Poisson kernel concentrates on that point. 2.24. For any ๐œ โˆˆ ๐•Š and ๐œ‚ > 0, โˆซ

๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰) โ†’ 0

|๐œ‰โˆ’๐œ|โ‰ฅ๐œ‚

as ๐‘ฅ โ†’ ๐œ, where the integral is taken over the subset of ๐•Š of points ๐œ‰ โˆˆ ๐•Š that satisfy |๐œ‰ โˆ’ ๐œ| โ‰ฅ ๐œ‚. To verify this limit observe that, if |๐œ‰ โˆ’ ๐œ| โ‰ฅ ๐œ‚ and |๐‘ฅ โˆ’ ๐œ| < ๐œ‚/2, then ๐œ‚ ๐œ‚ |๐‘ฅ โˆ’ ๐œ‰| = |๐œ‰ โˆ’ ๐œ + ๐œ โˆ’ ๐‘ฅ| โ‰ฅ |๐œ‰ โˆ’ ๐œ| โˆ’ |๐œ โˆ’ ๐‘ฅ| > ๐œ‚ โˆ’ = , 2 2 so we have 1 1 โˆ’ |๐‘ฅ|2 1 1 โˆ’ |๐‘ฅ|2 ๐‘ƒ(๐‘ฅ, ๐œ‰) = โ‰ค . ๐œ”๐‘‘ |๐‘ฅ โˆ’ ๐œ‰|๐‘‘ ๐œ”๐‘‘ (๐œ‚/2)๐‘‘ Therefore, if |๐‘ฅ โˆ’ ๐œ| < ๐œ‚/2, โˆซ

๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰) โ‰ค โˆซ

|๐œ‰โˆ’๐œ|โ‰ฅ๐œ‚

|๐œ‰โˆ’๐œ|โ‰ฅ๐œ‚

1 1 โˆ’ |๐‘ฅ|2 2 ๐‘‘ ๐‘‘๐œŽ(๐œ‰) โ‰ค ( ) (1 โˆ’ |๐‘ฅ|2 ), ๐‘‘ ๐œ”๐‘‘ (๐œ‚/2) ๐œ‚

and hence โˆซ

๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰) โ†’ 0

|๐œ‰โˆ’๐œ|โ‰ฅ๐œ‚

as ๐‘ฅ โ†’ ๐œ, because |๐œ| = 1 and thus |๐‘ฅ| โ†’ 1.

2.3. Poisson kernel and Poisson integrals in the ball

27

The facts 2.17, 2.20 and 2.24 make the family {๐œ‰ โ†ฆ ๐‘ƒ(๐‘ฅ, ๐œ‰) โˆถ ๐‘ฅ โˆˆ ๐”น} of functions on ๐•Š resemble a family of good kernels as ๐‘ฅ โ†’ ๐œ โˆˆ ๐•Š, as defined in [SS03]. We will study such families later in this text. Let ๐‘“ โˆˆ ๐ถ(๐•Š). The Poisson integral of ๐‘“ is given by (2.25)

๐’ซ๐‘“(๐‘ฅ) = โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘“(๐œ‰)๐‘‘๐œŽ(๐œ‰), ๐•Š

for each ๐‘ฅ โˆˆ ๐”น. The function ๐‘ข(๐‘ฅ) = ๐’ซ๐‘“(๐‘ฅ) defined by the Poisson integral of ๐‘“ is well defined in ๐”น for any continuous function ๐‘“ on ๐•Š. This follows because ๐‘ƒ(๐‘ฅ, ๐œ‰) is continuous as well in ๐•Š. In fact, it is not required for ๐‘“ to be continuous on ๐•Š for the integral in (2.25) to be defined. It is sufficient for ๐‘“ to be Riemann-integrable on ๐•Š. 2.26. The Poisson integral ๐‘ข(๐‘ฅ) of ๐‘“ โˆˆ ๐ถ(๐•Š) is harmonic in ๐”น. This is followed by differentiating inside the integral (2.25), and using the fact that ๐‘ƒ(๐‘ฅ, ๐œ‰) is harmonic in ๐‘ฅ. (This is true for a Riemann-integrable function ๐‘“ on ๐•Š, as well; see Exercise (15).) In the case ๐‘‘ = 1, the Poisson integral of ๐‘“ โˆถ {โˆ’1, 1} โ†’ โ„ is the sum ๐‘ข(๐‘ฅ) = ๐‘ƒ(๐‘ฅ, โˆ’1)๐‘“(โˆ’1) + ๐‘ƒ(๐‘ฅ, 1)๐‘“(1) 1 1 = (1 โˆ’ ๐‘ฅ)๐‘“(โˆ’1) + (1 + ๐‘ฅ)๐‘“(1). 2 2 This is a linear combination of linear functions, so it is linear and clearly harmonic in (โˆ’1, 1). Note that ๐‘ข(๐‘ฅ) โ†’ ๐‘“(ยฑ1) as ๐‘ฅ โ†’ ยฑ1. The Poisson integral solves the Dirichlet problem for the ball: given ๐‘“ โˆˆ ๐ถ(๐•Š), find a function ๐‘ข on ๐”นฬ„ such that it is harmonic in the interior and coincides with ๐‘“ on the boundary, that is (2.27)

ฮ”๐‘ข = 0 { ๐‘ข=๐‘“

in ๐”น on ๐•Š.

We prove Theorem 2.28. Theorem 2.28. Let ๐‘“ โˆˆ ๐ถ(๐•Š) and ๐‘ข = ๐’ซ๐‘“ its Poisson integral. Then ๐‘ข is harmonic in ๐”น, extends continuously to ๐”นฬ„ and ๐‘ข|๐•Š = ๐‘“.

28

2. Basic properties

Proof. From fact 2.26, we know that ๐‘ข is harmonic in ๐”น. It is thus sufficient to prove that, for each ๐œ โˆˆ ๐•Š, ๐‘ข(๐‘ฅ) โ†’ ๐‘“(๐œ) as ๐‘ฅ โ†’ ๐œ. Since ๐‘“ is continuous on the compact set ๐•Š, it is bounded.3 Let ๐‘€ > 0 be such that |๐‘“(๐œ‰)| โ‰ค ๐‘€ for all ๐œ‰ โˆˆ ๐•Š. Given ๐œ€ > 0, we can choose ๐œ‚ > 0 such that, if |๐œ‰ โˆ’ ๐œ| < ๐œ‚, then ๐œ€ |๐‘“(๐œ‰) โˆ’ ๐‘“(๐œ)| < . 2 We write, using identity (2.20), |๐‘ข(๐‘ฅ) โˆ’ ๐‘“(๐œ)| = || โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘“(๐œ‰)๐‘‘๐œŽ(๐œ‰) โˆ’ ๐‘“(๐œ) โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)๐‘‘๐œŽ(๐œ‰)|| ๐•Š

๐•Š

โ‰ค โˆซ ๐‘ƒ(๐‘ฅ, ๐œ‰)|๐‘“(๐œ‰) โˆ’ ๐‘“(๐œ)|๐‘‘๐œŽ(๐œ‰) ๐•Š

=โˆซ

+โˆซ

|๐œ‰โˆ’๐œ| 0, we define the function in ๐”นฬ„ โˆ— ๐‘ฃ ๐œ€ (๐‘ฅ) = ๐‘ข(๐‘ฅ) โˆ’ ๐’ซ(๐‘ข|๐•Š )(๐‘ฅ) + ๐œ€(|๐‘ฅ|2โˆ’๐‘‘ โˆ’ 1).

Exercises

31

As |๐‘ฅ|2โˆ’๐‘‘ is harmonic in โ„๐‘‘ โงต {0}, ๐‘ฃ ๐œ€ is harmonic in ๐”นโˆ— and, if we set ๐’ซ(๐‘ข|๐•Š )(๐œ‰) = ๐‘ข(๐œ‰) if ๐œ‰ โˆˆ ๐•Š, ๐‘ฃ ๐œ€ is continuous on ๐”นฬ„ โˆ— . We observe that, for ๐œ‰ โˆˆ ๐•Š, ๐‘ฃ ๐œ€ (๐œ‰) = 0. Moreover, as ๐‘ฅ โ†’ 0, we have ๐‘ฃ ๐œ€ (๐‘ฅ) โ†’ โˆž because ๐‘ข is bounded near 0. Thus, by the maximum principle, ๐‘ฃ ๐œ€ (๐‘ฅ) > 0 for all ๐‘ฅ โˆˆ ๐”นโˆ— , because otherwise ๐‘ข would take a negative minimum in ๐”นโˆ— , and that is not possible by the maximum principle. Since ๐œ€ > 0 is arbitrary, we obtain that ๐‘ข(๐‘ฅ) โ‰ฅ ๐’ซ(๐‘ข|๐•Š )(๐‘ฅ) for all ๐‘ฅ โˆˆ ๐”นโˆ— . If we repeat the argument for โˆ’๐‘ข, we obtain ๐‘ข(๐‘ฅ) โ‰ค ๐’ซ(๐‘ข|๐•Š )(๐‘ฅ) for all ๐‘ฅ โˆˆ ๐”นโˆ— . Thus ๐‘ข(๐‘ฅ) = ๐’ซ(๐‘ข|๐•Š )(๐‘ฅ) in ๐”นโˆ— . Therefore, the Poisson integral ๐’ซ(๐‘ข|๐•Š )(๐‘ฅ) is the harmonic extension of ๐‘ข to all of ๐”น. โ–ก Note that, in the proof of Theorem 2.33, we are extending ๐‘ข to ๐‘ฅ = 0 by its average over ๐•Š. It is actually not necessary to assume that ๐‘ข is bounded near ๐‘ฅ0 to conclude that ๐‘ฅ0 is a removable singularity. See Exercise (21).

Exercises (1) A translation in โ„๐‘‘ is a map ๐‘‡ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ of the form ๐‘‡(๐‘ฅ) = ๐‘ฅ + โ„Ž, for some โ„Ž โˆˆ โ„๐‘‘ . (a) If ๐‘‡ is a translation, ฮ”(๐‘ข โˆ˜ ๐‘‡) = (ฮ”๐‘ข) โˆ˜ ๐‘‡. (b) If ๐‘ข is harmonic in โ„๐‘‘ and ๐‘‡ is a translation, then ๐‘ข โˆ˜ ๐‘‡ is also harmonic in โ„๐‘‘ . (2) An orthogonal transformation in โ„๐‘‘ is a map ๐‘ƒ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ of the form ๐‘ƒ(๐‘ฅ) = ๐ด๐‘ฅ, for some orthogonal ๐‘› ร— ๐‘› matrix ๐ด, that is, ๐ด satisfies that ๐ด๐ด๐‘ก = ๐ผ๐‘› , where ๐ด๐‘ก is the transpose of ๐ด and ๐ผ๐‘› is the ๐‘› ร— ๐‘› identity matrix. (a) If ๐‘ƒ is an orthogonal transformation, then ฮ”(๐‘ข โˆ˜ ๐‘ƒ) = (ฮ”๐‘ข) โˆ˜ ๐‘ƒ. (b) If ๐‘ข is harmonic in โ„๐‘‘ and ๐‘ƒ is orthogonal, then ๐‘ข โˆ˜ ๐‘ƒ is also harmonic in โ„๐‘‘ . (3) Prove fact 2.7.

32

2. Basic properties

(4) Let ๐‘ข(๐‘ฅ) = ๐‘Ž๐‘ฅ + ๐‘. Then ๐‘ฅ +๐‘Ÿ

0 1 โˆซ ๐‘ข(๐‘ฅ0 ) = ๐‘ข(๐‘ฅ)๐‘‘๐‘ฅ. 2๐‘Ÿ ๐‘ฅ โˆ’๐‘Ÿ 0

(5) Prove that the function ๐‘ฃ in the proof of Theorem 2.4 is harmonic. (6) Suppose ๐‘ข is harmonic in a neighborhood of ฮฉ,ฬ„ where ฮฉ is a ๐ถ 1 domain. Then โˆซ ๐œ•๐œˆ ๐‘ข ๐‘‘๐œŽ = 0. ๐œ•โ„ฆ

(7) Let ๐‘“ be Riemann-integrable on the rectangle ๐‘…, and continuous at the interior point ๐‘ฅ0 โˆˆ ๐‘…. As ๐œ€ โ†’ 0, 1 โˆซ ๐‘“ โ†’ ๐‘“(๐‘ฅ0 ); and (a) |๐ต๐œ€ (๐‘ฅ0 )| ๐ต (๐‘ฅ ) ๐œ€ 0 1 โˆซ ๐‘“ โ†’ ๐‘“(๐‘ฅ0 ). (b) |๐‘†๐œ€ (๐‘ฅ0 )| ๐‘† (๐‘ฅ ) ๐œ€

0

(8) If ฮฉ โŠ‚ โ„๐‘‘ is a bounded domain and ๐‘ข is harmonic in ฮฉ and continuous on ฮฉ,ฬ„ then ๐‘ข takes its maximum and its minimum on ๐œ•ฮฉ. (9) Let ฮฉ โŠ‚ โ„๐‘‘ be a bounded domain, ๐‘ข and ๐‘ฃ harmonic in ฮฉ and continuous on ฮฉ.ฬ„ If ๐‘ข = ๐‘ฃ on ๐œ•ฮฉ, then ๐‘ข = ๐‘ฃ in ฮฉ. (10) The following exercises provide the details of the proof of the fundamental theorem of algebra. (a) If ๐‘“ is a holomorphic function without zeroes in its domain ฮฉ, then 1/๐‘“ is holomorphic in ฮฉ. (b) If ๐‘(๐‘ง) is a polynomial over โ„‚, then either ๐‘(๐‘ง) is constant or |๐‘(๐‘ง)| โ†’ โˆž as |๐‘ง| โ†’ โˆž. (c) If ๐‘(๐‘ง) is polynomial over โ„‚ with no roots, then 1/๐‘(๐‘ง) is an entire bounded function. (11) If ๐‘“ is an entire function and its real part is nonnegative, then ๐‘“ is constant. (12) If ๐‘ข is a radial harmonic function in ๐”น, then it is constant. (13) ๐‘ฅ โ†ฆ ๐‘ƒ(๐‘ฅ, ๐œ‰) is harmonic in ๐”น, for each ๐œ‰ โˆˆ ๐•Š. (Hint: Write ๐‘ƒ(๐‘ฅ, ๐œ‰) = 2 โˆ’๐‘‘ ๐œ”โˆ’1 and use the identity ฮ”(๐‘ข๐‘ฃ) = (ฮ”๐‘ข)๐‘ฃ + 2โˆ‡๐‘ข โ‹… ๐‘‘ (1 โˆ’ |๐‘ฅ| )|๐‘ฅ โˆ’ ๐œ‰| โˆ‡๐‘ฃ + ๐‘ขฮ”๐‘ฃ.) (14) Symmetry Lemma: If ๐‘ฅ โˆˆ ๐”น and ๐œ‰ โˆˆ ๐•Š, then ||๐‘ฅ|๐œ‰ โˆ’ ๐‘ฅ | = |๐‘ฅ โˆ’ ๐œ‰|. | |๐‘ฅ| |

Exercises

33

See Figure 2.4.

Figure 2.4. If ๐‘ฅ โˆˆ ๐”น and ๐œ‰ โˆˆ ๐•Š, the distance between the points ๐‘ฅ/|๐‘ฅ| and |๐‘ฅ|๐œ‰ is the same as the distance between ๐‘ฅ and ๐œ‰, as stated by the symmetry lemma.

(15) Let ๐‘“ be Riemann-integrable on ๐•Š. Then its Poisson integral ๐‘ข is harmonic in ๐”น. (16) Hopf lemma: If ๐‘ข is a nonconstant harmonic function in ๐”น, is continuous on ๐”น,ฬ„ and attains its maximum at ๐œ โˆˆ ๐•Š, then there exists ๐‘ > 0 such that ๐‘ข(๐œ) โˆ’ ๐‘ข(๐‘Ÿ๐œ) > ๐‘(1 โˆ’ ๐‘Ÿ), for any 0 < ๐‘Ÿ < 1. (17) Harnack inequality: If ๐‘ข is a harmonic function in ๐”น, is continuous on ๐”น,ฬ„ and is positive, then 1 โˆ’ |๐‘ฅ| 1 + |๐‘ฅ| ๐‘ข(0) โ‰ค ๐‘ข(๐‘ฅ) โ‰ค ๐‘ข(0) ๐‘‘โˆ’1 (1 + |๐‘ฅ|) (1 โˆ’ |๐‘ฅ|)๐‘‘โˆ’1 for all ๐‘ฅ โˆˆ ๐”น. (18) If ๐‘ข is harmonic in ฮฉ and ๐ต๐‘Ÿฬ„ (๐‘ฅ0 ) โŠ‚ ฮฉ, then the values of ๐‘ข in ๐ต๐‘Ÿ (๐‘ฅ0 ) are determined by its values on ๐‘†๐‘Ÿ (๐‘ฅ0 ). (19) Let ๐‘ข๐‘› be a sequence of harmonic functions in ฮฉ such that ๐‘ข๐‘› โ‡‰ ๐‘ข on any compact ๐พ โŠ‚ ฮฉ. Then ๐‘ข is harmonic in ฮฉ. (20) Prove Theorem 2.33 for ๐‘‘ = 2.

34

2. Basic properties

(21) Let ๐‘ข be harmonic in a domain in โ„๐‘‘ with an isolated singularity at ๐‘ฅ0 . If ๐‘‘ = 2 and lim ๐‘ข(๐‘ฅ) log |๐‘ฅ โˆ’ ๐‘ฅ0 | = 0,

๐‘ฅโ†’๐‘ฅ0

or ๐‘‘ > 2 and lim ๐‘ข(๐‘ฅ)|๐‘ฅ โˆ’ ๐‘ฅ0 |๐‘‘โˆ’2 = 0,

๐‘ฅโ†’๐‘ฅ0

then ๐‘ฅ0 is a removable singularity.

Notes The results of this chapter are basic classical results, proven in every text in harmonic functions and partial differential equations. A classical reference for the theory of harmonic functions is [Kel67]. Theorem 2.4 is a result by Gauss [Gau40]. The proof presented here is the most popular and can be found, for instance, in [ABR01] or in [Fol95]. It can also be proven by differentiating the integral over a sphere with respect to its radius, as in [Eva10] or in [MS13]. The proof of Theorem 2.9 is also in [Fol95]. The proof of Theorem 2.14 is an elaboration of the proof by Nelson [Nel61]. Simรฉon Denis Poisson developed explicit expressions to solutions to the Laplace equation in terms of integrals over the sphere in [Poi20], and thus the Poisson kernel and integral are named after him. Theorem 2.33 is a result by Riemann [Rie51]. The proof presented here can be found in [ABR01].

Chapter 3

Fourier series

3.1. Separation of variables In Chapter 2 we solved the Dirichlet problem (2.27) using the Poisson integral, which provides an explicit integral form of the solution from its values on the boundary. We now attempt to solve the problem by decomposing the function ๐‘“ on ๐•Š in fundamental pieces, which is the original Fourier approach in [Fou55]. In this chapter we consider the Dirichlet problem in the unit disk ๐”ป in the plane โ„2 . In order to find such fundamental pieces, we first recall that the Laplacian in polar coordinates (๐‘Ÿ, ๐œƒ) is given by ฮ”๐‘ข =

๐œ•2 ๐‘ข 1 ๐œ•๐‘ข 1 ๐œ•2 ๐‘ข + + 2 2. 2 ๐‘Ÿ ๐œ•๐‘Ÿ ๐‘Ÿ ๐œ•๐œƒ ๐œ•๐‘Ÿ

(See Exercise (2) of Chapter 1.) We now search for solutions of the form ๐‘ข(๐‘Ÿ, ๐œƒ) = ๐‘ฃ(๐‘Ÿ)๐œ™(๐œƒ), where ๐‘ฃ(๐‘Ÿ) is a twice differentiable function defined for 0 โ‰ค ๐‘Ÿ < 1 and ๐œ™(๐œƒ) is a twice differentiable periodic function defined on โ„, with period 2๐œ‹, as the pair (๐‘Ÿ, ๐œƒ) denotes a point in the disk. Thus we want to solve the equation 1 1 ฮ”๐‘ข = ๐‘ฃโ€ณ (๐‘Ÿ)๐œ™(๐œƒ) + ๐‘ฃโ€ฒ (๐‘Ÿ)๐œ™(๐œƒ) + 2 ๐‘ฃ(๐‘Ÿ)๐œ™โ€ณ (๐œƒ) = 0, ๐‘Ÿ ๐‘Ÿ 35

36

3. Fourier series

which we can rewrite, when ๐‘ฃ(๐‘Ÿ) โ‰  0 and ๐œ™(๐œƒ) โ‰  0, as ๐œ™โ€ณ (๐œƒ) ๐‘Ÿ2 ๐‘ฃโ€ณ (๐‘Ÿ) + ๐‘Ÿ๐‘ฃโ€ฒ (๐‘Ÿ) =โˆ’ . ๐‘ฃ(๐‘Ÿ) ๐œ™(๐œƒ) The left hand side of this equation does not depend on ๐œƒ, and the right hand side does not depend on ๐‘Ÿ, so we conclude that both sides are equal to a constant, say, ๐œ† โˆˆ โ„. Hence we obtain the equations (3.1)

๐‘Ÿ2 ๐‘ฃโ€ณ (๐‘Ÿ) + ๐‘Ÿ๐‘ฃโ€ฒ (๐‘Ÿ) = ๐œ†๐‘ฃ(๐‘Ÿ)

and ๐œ™โ€ณ (๐œƒ) = โˆ’๐œ†๐œ™(๐œƒ),

(3.2)

subject to the constrains stated above. Equation (3.2) has periodic solutions ๐œ™(๐œƒ) = cos(๐‘›๐œƒ)

and

๐œ™(๐œƒ) = sin(๐‘›๐œƒ),

2

with period 2๐œ‹, when ๐œ† = ๐‘› and ๐‘› โˆˆ โ„•. We have a pair of linearly independent solutions for each natural number ๐‘› โ‰ฅ 1. For ๐‘› = 0, we have the linearly independent solutions ๐œ™(๐œƒ) = 1 and ๐œ™(๐œƒ) = ๐œƒ, but the latter is not periodic. Now, with ๐œ† = ๐‘›2 , equation (3.1) is ๐‘Ÿ2 ๐‘ฃโ€ณ (๐‘Ÿ) + ๐‘Ÿ๐‘ฃโ€ฒ (๐‘Ÿ) โˆ’ ๐‘›2 ๐‘ฃ(๐‘Ÿ) = 0 and has linearly independent solutions ๐‘ฃ(๐‘Ÿ) = ๐‘Ÿ๐‘›

and

๐‘ฃ(๐‘Ÿ) = ๐‘Ÿโˆ’๐‘›

for each ๐‘› โ‰ฅ 1, though only the former is well defined on [0, 1). Again, if ๐‘› = 0, we have the solutions ๐‘ฃ(๐‘Ÿ) = 1 and ๐‘ฃ(๐‘Ÿ) = log ๐‘Ÿ, but only the former is defined on [0, 1). We thus obtain, for each ๐‘› โˆˆ โ„•, the harmonic functions ๐‘ข๐‘› (๐‘Ÿ, ๐œƒ) = ๐‘Ÿ๐‘› (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)), where ๐‘Ž๐‘› , ๐‘๐‘› โˆˆ โ„ (note that ๐‘ข0 is the constant function ๐‘ข0 (๐‘Ÿ, ๐œƒ) = ๐‘Ž0 ). Any linear combination of such functions, ๐‘

(3.3)

๐‘ข(๐‘Ÿ, ๐œƒ) = โˆ‘ ๐‘Ÿ๐‘› (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)), ๐‘›=0

3.1. Separation of variables

37

is harmonic in the disk, and its limit at the boundary ๐‘Ÿ โ†’ 1 is the trigonometric polynomial ๐‘

(3.4)

๐‘(๐œƒ) = โˆ‘ (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)). ๐‘›=0

In fact, this limit is uniform as ๐‘Ÿ โ†’ 1 on ๐•Š. We have thus solved the problem ฮ”๐‘ข = 0 in ๐”ป { ๐‘ข=๐‘ on ๐•Š, where ๐‘ is the trigonometric polynomial (3.4). It is clear that the trigonometric polynomial (3.4) can be seen either as a function on ๐•Š or as a periodic function on โ„, through the map ๐œƒ โ†ฆ (cos ๐œƒ, sin ๐œƒ). In general, that is true for any 2๐œ‹-periodic function ๐‘“ on โ„: the function ๐น(cos ๐œƒ, sin ๐œƒ) = ๐‘“(๐œƒ) is a well-defined function on ๐•Š. We note that ๐‘“ is continuous on โ„ if and only if ๐น is continuous on ๐•Š, so we can identify the space ๐ถ(๐•Š) of continuous functions on ๐•Š with the subspace of ๐ถ(โ„) of periodic functions with period 2๐œ‹, or with the subspace of ๐ถ([0, 2๐œ‹]) of functions satisfying ๐‘“(0) = ๐‘“(2๐œ‹). Similarly, Riemann-integrable functions on ๐•Š can be identified with Riemann-integrable functions ๐‘“ on [0, 2๐œ‹] satisfying ๐‘“(0) = ๐‘“(2๐œ‹), as well as with 2๐œ‹-periodic functions on โ„ that are Riemann-integrable on each closed interval. We will interchangeably use the terms โ€œfunction on ๐•Šโ€, โ€œfunction on [0, 2๐œ‹]โ€ and โ€œ2๐œ‹-periodic function on โ„โ€ throughout this chapter. If we recall de Moivreโ€™s formula for complex numbers, (cos ๐œƒ + ๐‘– sin ๐œƒ)๐‘› = cos(๐‘›๐œƒ) + ๐‘– sin(๐‘›๐œƒ), we see that (3.4) is indeed a polynomial in cos ๐œƒ and sin ๐œƒ (that is why we call it a trigonometric polynomial), and thus a polynomial in the coordinate functions on ๐•Š. We can ask whether we can use the linear combinations (3.3) to solve the Dirichlet problem (3.5)

ฮ”๐‘ข = 0 { ๐‘ข=๐‘“

in ๐”ป on ๐•Š,

given any continuous function ๐‘“ on the boundary ๐•Š, and not only a trigonometric polynomial. More precisely, we have:

38

3. Fourier series Question 1: For ๐‘“ โˆˆ ๐ถ(๐•Š), is there a sequence of trigonometric polynomials ๐‘๐‘ that converge to ๐‘“ (pointwise or uniformly) such that the corresponding solutions ๐‘ข๐‘ as in (3.3) converge to a solution of the Dirichlet problem (3.5)? Question 2: For ๐‘“ โˆˆ ๐ถ(๐•Š), is there a series โˆž

โˆ‘ (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)) ๐‘›=0

such that the series โˆž

๐‘ข(๐‘Ÿ, ๐œƒ) = โˆ‘ ๐‘Ÿ๐‘› (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)) ๐‘›=0

converges to a solution of the Dirichlet problem (3.5)? Note that an affirmative answer to Question 2 would not necessarily give a positive answer to Question 1, as the partial sums of the series might not satisfy the requirement of the trigonometric polynomials ๐‘๐‘ and the functions ๐‘ข๐‘ . Note also that a sequence ๐‘๐‘ of trigonometric polynomials converging uniformly to ๐‘“ is by all means not unique, and thus neither the sequence ๐‘ข๐‘ . These questions also make sense if ๐‘“ is only Riemann-integrable, except for the fact that we cannot expect uniform convergence anymore because the uniform limit of continuous functions is continuous. In that case, we would restrict to convergence at certain points of ๐•Š.

3.2. Fourier series Consider the problem of writing a function ๐‘“ on ๐•Š as a series โˆž

(3.6)

โˆ‘ (๐‘Ž๐‘› cos ๐‘›๐œƒ + ๐‘๐‘› sin ๐‘›๐œƒ) ๐‘›=0

for ๐œƒ โˆˆ [0, 2๐œ‹], where we agree that ๐‘0 = 0. By Eulerโ€™s formula ๐‘’๐‘–๐‘ฅ = cos ๐‘ฅ + ๐‘– sin ๐‘ฅ, we can write, for ๐‘› โ‰ฅ 1, ๐‘Ž๐‘› cos ๐‘›๐œƒ + ๐‘๐‘› sin ๐‘›๐œƒ = ๐‘๐‘› ๐‘’๐‘–๐‘›๐œƒ + ๐‘โˆ’๐‘› ๐‘’โˆ’๐‘–๐‘›๐œƒ , where ๐‘ ยฑ๐‘› =

๐‘Ž๐‘› โˆ“ ๐‘–๐‘๐‘› . 2

3.2. Fourier series

39

If we set ๐‘ 0 = ๐‘Ž0 , the series (3.6) can be written as โˆž

(3.7)

โˆž

๐‘ 0 + โˆ‘ (๐‘๐‘› ๐‘’๐‘–๐‘›๐œƒ + ๐‘โˆ’๐‘› ๐‘’โˆ’๐‘–๐‘›๐œƒ ) = โˆ‘ ๐‘๐‘› ๐‘’๐‘–๐‘›๐œƒ , ๐‘›=1

๐‘›=โˆ’โˆž

where the double infinite series on the right of (3.7) is understood as the limit of the partial sums ๐‘

โˆ‘ ๐‘๐‘› ๐‘’๐‘–๐‘›๐œƒ ๐‘›=โˆ’๐‘

as ๐‘ โ†’ โˆž. Suppose the series (3.7) converges uniformly to ๐‘“, so ๐‘

๐‘ ๐‘ (๐œƒ) = โˆ‘ ๐‘๐‘› ๐‘’๐‘–๐‘›๐œƒ โ‡‰ ๐‘“(๐œƒ). ๐‘›=โˆ’๐‘

Thus, for any ๐‘› โˆˆ โ„ค (see Section A.3), 2๐œ‹

2๐œ‹

โˆซ

๐‘ ๐‘ (๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ โ†’ โˆซ

0

0

๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ.

Now for any ๐‘š โˆˆ โ„ค, 2๐œ‹

(3.8)

โˆซ

๐‘’๐‘–๐‘š๐œƒ ๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = {

0

2๐œ‹ ๐‘š = ๐‘› 0

๐‘šโ‰ 0

(Exercise (3)) and thus, for ๐‘ โ‰ฅ ๐‘›, 2๐œ‹

โˆซ

๐‘ ๐‘ (๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = 2๐œ‹๐‘๐‘› .

0

Therefore, if the series (3.7) converges uniformly to ๐‘“, the coefficients ๐‘๐‘› are given by ๐‘๐‘› =

1 โˆซ 2๐œ‹ 0

2๐œ‹

๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ

for each ๐‘› โˆˆ โ„ค. For any Riemann-integrable function ๐‘“ on ๐•Š, we define its Fourier coefficients, for ๐‘› โˆˆ โ„ค, by 2๐œ‹

(3.9)

ฬ‚ = 1 โˆซ ๐‘“(๐‘›) 2๐œ‹ 0

๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ.

40

3. Fourier series

The series ๐‘–๐‘›๐œƒ ฬ‚ โˆ‘ ๐‘“(๐‘›)๐‘’

(3.10)

๐‘›โˆˆโ„ค

is called the Fourier series of ๐‘“. Note that, since a Riemann-integrable function ๐‘“ is bounded, its Fourier coefficients (3.9) are bounded. Indeed, they are bounded by any bound for ๐‘“ because, if |๐‘“(๐œƒ)| โ‰ค ๐‘€ for all ๐œƒ, then 1 ฬ‚ โˆซ |๐‘“(๐‘›)| โ‰ค || 2๐œ‹ 0

(3.11)

2๐œ‹

2๐œ‹

๐‘“(๐œƒ)๐‘’

โˆ’๐‘–๐‘›๐œƒ

1 โˆซ ๐‘‘๐œƒ|| โ‰ค 2๐œ‹ 0

|๐‘“(๐œƒ)|๐‘‘๐œƒ โ‰ค ๐‘€.

We also observe that in the definition (3.9) of the Fourier coefficients of ๐‘“, as ๐‘“ is periodic, we can choose any interval of length 2๐œ‹ in the integration, as convenient (see Exercise (4)). Example 3.12 (The sawtooth function). Consider the periodic function ๐‘“, with period 2๐œ‹, given in [โˆ’๐œ‹, ๐œ‹) by ๐‘“(๐œƒ) = ๐œƒ (Figure 3.1). Its Fourier ฯ€

-2 ฯ€

-ฯ€

2ฯ€

ฯ€

3ฯ€

-ฯ€

Figure 3.1. The sawtooth function given by ๐‘“(๐œƒ) = ๐œƒ, โˆ’๐œ‹ โ‰ค ๐œƒ < ๐œ‹. Note that ๐‘“ is discontinuous at ๐œ‹.

coefficients are given, for ๐‘› = 0, by ๐œ‹

๐œ‹

ฬ‚ = 1 โˆซ ๐‘“(๐œƒ)๐‘‘๐œƒ = 1 โˆซ ๐œƒ๐‘‘๐œƒ = 0, ๐‘“(0) 2๐œ‹ โˆ’๐œ‹ 2๐œ‹ โˆ’๐œ‹ and, for ๐‘› โ‰  0, by ๐œ‹

๐œ‹

๐‘› ฬ‚ = 1 โˆซ ๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = 1 โˆซ ๐œƒ๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = ๐‘–(โˆ’1) , ๐‘“(๐‘›) 2๐œ‹ โˆ’๐œ‹ 2๐œ‹ โˆ’๐œ‹ ๐‘›

3.2. Fourier series

41

and thus the Fourier series of ๐‘“ is given by โˆž

(โˆ’1)๐‘› ๐‘–๐‘›๐œƒ (โˆ’1)๐‘› ๐‘’ = โˆ’2 โˆ‘ sin ๐‘›๐œƒ. ๐‘› ๐‘› ๐‘›โ‰ 0 ๐‘›=1

๐‘–โˆ‘

Using Dirichletโ€™s test,1 one can prove that this series converges for every ๐œƒ (Exercise (5)). At ๐œƒ = 0, the series clearly converges to 0, and at ๐œƒ = ๐œ‹/2, the series โˆž

โˆž

(โˆ’1)๐‘› ๐‘›๐œ‹ (โˆ’1)๐‘˜ 1 1 1 sin =2โˆ‘ = 2(1 โˆ’ + โˆ’ + โ‹ฏ ) ๐‘› 2 2๐‘˜ + 1 3 5 7 ๐‘›=1 ๐‘˜=0

โˆ’2 โˆ‘

converges to ๐œ‹/2, as seen in a calculus course. In both cases, the series at each ๐œƒ converges to ๐‘“(๐œƒ). However, at ๐œƒ = ๐œ‹, the series converges to 0, even though ๐‘“(๐œ‹) = โˆ’๐œ‹. Note that ๐‘“ is discontinuous at ๐œ‹. Example 3.13 (The sharkteeth function). We now consider the periodic function ๐‘” given in [โˆ’๐œ‹, ๐œ‹) by ๐‘”(๐œƒ) = |๐œƒ| (Figure 3.2). This time the ฯ€

-2 ฯ€

-ฯ€

ฯ€

2ฯ€

3ฯ€

Figure 3.2. The sharkteeth function given by ๐‘”(๐œƒ) = |๐œƒ|, โˆ’๐œ‹ โ‰ค ๐œƒ < ๐œ‹. Note that ๐‘” is continuous at every point.

function ๐‘” is continuous at every point. Its Fourier coefficients are given by ๐œ‹ ๐‘›=0 โŽง โŽช2 even ๐‘› โ‰  0 ๐‘”(๐‘›) ฬ‚ = 0 โŽจ 2 โŽชโˆ’ โŽฉ ๐œ‹๐‘›2 odd ๐‘› (Exercise (6)), and thus its Fourier series is given by โˆž

๐‘’๐‘–๐‘›๐œƒ cos(2๐‘˜ + 1)๐œƒ ๐œ‹ 2 ๐œ‹ 4 โˆ‘ 2 = โˆ’ โˆ‘ โˆ’ . 2 ๐œ‹ odd ๐‘› ๐‘› 2 ๐œ‹ ๐‘˜=0 (2๐‘˜ + 1)2 1

Theorem A.5 in Section A.1.

42

3. Fourier series

We see that, this time, the Fourier series of ๐‘” converges uniformly on [โˆ’๐œ‹, ๐œ‹), by the Weierstrass ๐‘€-test,2 as the coefficients decrease as 1/๐‘›2 . For example, for ๐œƒ = ๐œ‹/2, each ๐œ‹ cos(2๐‘˜ + 1) = 0, 2 so the series converges to ๐œ‹/2, the value of ๐‘” at ๐œ‹/2. We can ask then if the series converges to ๐‘”(๐œƒ) at any other ๐œƒ. Example 3.14 (The flounces function). Consider now the periodic function โ„Ž given in [โˆ’๐œ‹, ๐œ‹) by โ„Ž(๐œƒ) = (๐œ‹2 โˆ’๐œƒ2 )2 , shown in Figure 3.3. โ„Ž is not

-ฯ€

ฯ€

Figure 3.3. The flounces function given by โ„Ž(๐œƒ) = (๐œ‹2 โˆ’ ๐œƒ2 )2 , โˆ’๐œ‹ โ‰ค ๐œƒ < ๐œ‹. Note that โ„Ž is continuous and differentiable at every point.

only continuous, but also differentiable at everypoint, with โ„Žโ€ฒ (๐‘˜๐œ‹) = 0 for every ๐‘˜ โˆˆ โ„ค. Its Fourier coefficients (Exercise (7)) are given by 8 4 ๐œ‹ 15 ฬ‚ โ„Ž(๐‘›) = { (โˆ’1)๐‘› โˆ’24 4 ๐‘› so its Fourier series is then

๐‘›=0 ๐‘› โ‰  0, โˆž

8 4 (โˆ’1)๐‘› ๐‘–๐‘›๐œƒ 8 4 (โˆ’1)๐‘› ๐‘’ = cos ๐‘›๐œƒ. ๐œ‹ โˆ’ 24 โˆ‘ ๐œ‹ โˆ’ 48 โˆ‘ 4 15 15 ๐‘› ๐‘›4 ๐‘›โ‰ 0 ๐‘›=1 As in example 3.13 above, the Fourier series converges absolutely and uniformly, though this time it is not clear what the limit of the series is at any point. However, we observe that the series converges more rapidly than in the previous example, and thus one questions if the regularity of โ„Žโ€”the fact that it is not only continuous but also differentiableโ€”has any effect in the decrease of the coefficients. We have now another set of questions. 2

Theorem A.6 in Section A.1.

3.3. Abel means and Poisson integrals

43

Question 3: Does the Fourier series (3.10) converge? In what sense? Pointwise, uniformly? In any other sense? Does the regularity of the function have any effect on the convergence? Question 4: If the Fourier series (3.10) of ๐‘“ converges, does it converge to ๐‘“? It turns out that Question 4 can be answered thanks to our previous analysis of Poisson integrals.

3.3. Abel means and Poisson integrals โˆž

Consider a series โˆ‘๐‘›=0 ๐‘Ž๐‘› . The Abel means of โˆ‘ ๐‘Ž๐‘› are given by โˆž

๐ด๐‘Ÿ = โˆ‘ ๐‘Ž๐‘› ๐‘Ÿ๐‘› ,

(3.15)

๐‘›=0

for 0 < ๐‘Ÿ < 1. We say that the series โˆ‘ ๐‘Ž๐‘› is Abel-summable to ๐‘  if lim ๐ด๐‘Ÿ = ๐‘ . ๐‘Ÿโ†’1

All convergent series are also Abel-summable, a result know as Abelโ€™s theorem. Theorem 3.16. Suppose that the series โˆ‘ ๐‘Ž๐‘› converges to ๐‘ . Then โˆ‘ ๐‘Ž๐‘› is Abel-summable to ๐‘ . Proof. Let ๐‘ ๐‘› = ๐‘Ž0 + ๐‘Ž1 + . . . ๐‘Ž๐‘› be the ๐‘›th partial sum of the series, so we have ๐‘ ๐‘› โ†’ ๐‘ . In particular, ๐‘ ๐‘› is bounded, so there exists ๐‘€ > 0 such that |๐‘ ๐‘› | โ‰ค ๐‘€ for all ๐‘›. The boundedness of ๐‘ ๐‘› implies the convergence of โˆ‘ ๐‘ ๐‘› ๐‘Ÿ๐‘› , so for each 0 โ‰ค ๐‘Ÿ < 1, using the fact that ๐‘Ž๐‘› = ๐‘ ๐‘› โˆ’ ๐‘ ๐‘›โˆ’1 for each ๐‘› โ‰ฅ 1, โˆž

โˆž

๐ด๐‘Ÿ = โˆ‘ ๐‘Ž๐‘› ๐‘Ÿ๐‘› = ๐‘Ž0 + โˆ‘ (๐‘ ๐‘› โˆ’ ๐‘ ๐‘›โˆ’1 )๐‘Ÿ๐‘› ๐‘›=0

๐‘›=1 โˆž

โˆž

= ๐‘ 0 + โˆ‘ ๐‘ ๐‘› ๐‘Ÿ๐‘› โˆ’ โˆ‘ ๐‘ ๐‘›โˆ’1 ๐‘Ÿ๐‘› ๐‘›=1 โˆž

๐‘›=1 โˆž

โˆž

= โˆ‘ ๐‘ ๐‘› ๐‘Ÿ๐‘› โˆ’ โˆ‘ ๐‘ ๐‘› ๐‘Ÿ๐‘›+1 = (1 โˆ’ ๐‘Ÿ) โˆ‘ ๐‘ ๐‘› ๐‘Ÿ๐‘› . ๐‘›=0

๐‘›=0

๐‘›=0

44

3. Fourier series Given ๐œ€ > 0 there exists ๐‘ such that |๐‘ ๐‘› โˆ’ ๐‘ |
0 we have |๐‘“โ€ฒ (๐œ)| โ‰ค ๐‘€, so (3.38)

|๐‘”โ„Ž (๐œƒ)| โ‰ค 2๐‘€โ„Ž

52

3. Fourier series

for every ๐œƒ. Now, 2๐œ‹

๐‘” ห† โ„Ž (๐‘›) =

1 โˆซ (๐‘“(๐œƒ + โ„Ž) โˆ’ ๐‘“(๐œƒ โˆ’ โ„Ž))๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ 2๐œ‹ 0 2๐œ‹

2๐œ‹

=

1 โˆซ 2๐œ‹ 0

๐‘“(๐œƒ + โ„Ž)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ โˆ’

1 โˆซ 2๐œ‹ 0

=

1 โˆซ 2๐œ‹ 0

๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›(๐œƒโˆ’โ„Ž) ๐‘‘๐œƒ โˆ’

1 โˆซ 2๐œ‹ 0

=

๐‘’๐‘–๐‘›โ„Ž โˆ’ ๐‘’โˆ’๐‘–๐‘›โ„Ž โˆซ 2๐œ‹ 0

2๐œ‹

๐‘“(๐œƒ โˆ’ โ„Ž)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ

2๐œ‹

๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›(๐œƒ+โ„Ž) ๐‘‘๐œƒ

2๐œ‹

ฬ‚ ๐‘“(๐œƒ)๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = 2๐‘– sin ๐‘›โ„Ž๐‘“(๐‘›),

so we have ฬ‚ 2. โˆ‘ |ห† ๐‘”โ„Ž (๐‘›)|2 = 4 โˆ‘ | sin ๐‘›โ„Ž|2 |๐‘“(๐‘›)|

(3.39)

๐‘›โˆˆโ„ค

๐‘›โˆˆโ„ค

Lemma 3.36, together with equations (3.38) and (3.39), implies ฬ‚ 2โ‰ค โˆ‘ | sin ๐‘›โ„Ž|2 |๐‘“(๐‘›)| ๐‘›โˆˆโ„ค

1 1 โˆซ โ‹… 4 2๐œ‹ 0

2๐œ‹

|๐‘”โ„Ž (๐œƒ)|2 ๐‘‘๐œƒ โ‰ค

(2๐‘€โ„Ž)2 โ‹… 2๐œ‹ = ๐‘€ 2 โ„Ž2 . 8๐œ‹

In particular, for each ๐‘ โ‰ฅ 1, ฬ‚ 2 โ‰ค ๐‘€ 2 โ„Ž2 . | sin ๐‘›โ„Ž|2 |๐‘“(๐‘›)|

โˆ‘

(3.40)

2๐‘โˆ’1 โ‰ค|๐‘›| 0 such that, if |๐œ| < ๐œ‚, then ๐œ€ |๐‘“(๐œƒ โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ)| < . 2 Now, as ๐‘“ is Riemann-integrable, it is bounded, so there exists ๐‘€ > 0 such that |๐‘“(๐œ)| โ‰ค ๐‘€ for all ๐œ. By (3.48), there exists ๐พ such that, if ๐‘ โ‰ฅ ๐พ, ๐œ€ | 1 โˆซ ๐น๐‘ (๐œ)๐‘‘๐œ|| < . | 2๐œ‹ 4๐‘€ ๐œ‚โ‰ค|๐œ|โ‰ค๐œ‹ Therefore, for ๐‘ โ‰ฅ ๐พ, using (3.47), ๐œ‹

1 โˆซ ๐น๐‘ (๐œ)๐‘“(๐œƒ โˆ’ ๐œ)๐‘‘๐œ โˆ’ ๐‘“(๐œƒ)|| |๐œŽ๐‘ (๐œƒ) โˆ’ ๐‘“(๐œƒ)| = || 2๐œ‹ โˆ’๐œ‹

๐œ‹

๐œ‹

= ||

1 1 โˆซ ๐น (๐œ)๐‘“(๐œƒ โˆ’ ๐œ)๐‘‘๐œ โˆ’ ๐‘“(๐œƒ) โ‹… โˆซ ๐น (๐œ)๐‘‘๐œ|| 2๐œ‹ โˆ’๐œ‹ ๐‘ 2๐œ‹ โˆ’๐œ‹ ๐‘

= ||

1 โˆซ ๐น (๐œ)(๐‘“(๐œƒ โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ))๐‘‘๐œ|| 2๐œ‹ โˆ’๐œ‹ ๐‘

๐œ‹

โ‰ค

1 โˆซ ๐น (๐œ)|๐‘“(๐œƒ โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ)|๐‘‘๐œ 2๐œ‹ |๐œ| 0, โˆซ

|๐พ๐‘ (๐œƒ)|๐‘‘๐œƒ โ†’ 0

๐œ‚โ‰ค|๐œƒ|โ‰ค๐œ‹

as ๐‘ โ†’ โˆž. It is not hard to verify that the proof of Theorem 3.45 applies to any family of good kernels, so, if a function is continuous at ๐œƒ, then ๐œ‹

1 โˆซ ๐พ (๐œƒ โˆ’ ๐œ)๐‘“(๐œ)๐‘‘๐œ โ†’ ๐‘“(๐œƒ) 2๐œ‹ โˆ’๐œ‹ ๐‘ as ๐‘ โ†’ โˆž. The fact that the Cesร ro sums of a convergent series converge to the same limit implies the same results of Corollary 3.21. However, we additionally have the following stronger result. Corollary 3.49. The space of trigonometric polynomials is dense in ๐ถ(๐•Š). In other words, if ๐‘“ โˆˆ ๐ถ(๐•Š) and ๐œ€ > 0, there exists a trigonometric polynomial ๐‘ such that |๐‘“(๐œƒ) โˆ’ ๐‘(๐œƒ)| < ๐œ€ for every ๐œƒ. Proof. By Fejรฉrโ€™s theorem, the Cesร ro sums ๐œŽ๐‘ of the Fourier series of ๐‘“ converge uniformly to ๐‘“. In other words, given ๐œ€ > 0, there exists ๐‘ such that |๐‘“(๐œƒ) โˆ’ ๐œŽ๐‘ (๐œƒ)| < ๐œ€

3.6. Mean-square convergence

59

for all ๐œƒ. The results follows from the fact that ๐œŽ๐‘ is a trigonometric polynomial. โ–ก Corollary 3.49 gives an affirmative answer to Question 1 above.

3.6. Mean-square convergence We now come back to the orthogonality of the Fourier expansions to prove Theorem 3.50. Theorem 3.50. If ๐‘“ is a 2๐œ‹-periodic Riemann-integrable function and, for each ๐‘ โˆˆ โ„•, ๐‘ ๐‘–๐‘›๐œƒ ฬ‚ ๐‘ ๐‘ (๐œƒ) = โˆ‘ ๐‘“(๐‘›)๐‘’ , ๐‘›=โˆ’๐‘

then (3.51)

lim โ€–๐‘“ โˆ’ ๐‘ ๐‘ โ€– = 0.

๐‘โ†’โˆž

In other words, the Fourier series of ๐‘“ ๐‘–๐‘›๐œƒ ฬ‚ โˆ‘ ๐‘“(๐‘›)๐‘’ ๐‘›โˆˆโ„ค

converges to ๐‘“ in the sense, 2๐œ‹

๐‘

1 โˆซ 2๐œ‹ 0

2

๐‘–๐‘›๐œƒ | |๐‘“(๐œƒ) โˆ’ โˆ‘ ๐‘“(๐‘›)๐‘’ ฬ‚ | | ๐‘‘๐œƒ โ†’ 0 ๐‘›=โˆ’๐‘

as ๐‘ โ†’ โˆž. This is called mean-square convergence. Proof. As ๐‘“ is Riemann-integrable, it is bounded, so there exists ๐‘€ > 0 such that |๐‘“(๐œƒ)| โ‰ค ๐‘€ for all ๐œƒ. Now, given ๐œ€ > 0, by Theorem A.10 (Section A.3) we can choose a continuous function ๐‘” on ๐•Š such that |๐‘”(๐œƒ)| โ‰ค ๐‘€ and 2๐œ‹

โˆซ

|๐‘“(๐œƒ) โˆ’ ๐‘”(๐œƒ)|๐‘‘๐œƒ
0 small, let ๐‘”๐‘ be the continuous function on [โˆ’๐œ‹, ๐œ‹] that is equal to ๐œ™๐‘ (๐œƒ) in each interval [๐œƒ๐‘˜ + ๐›ฟ, ๐œƒ๐‘˜+1 โˆ’ ๐›ฟ], and is equal to the linear function from ๐œ™๐‘ (๐œƒ๐‘˜ โˆ’ ๐›ฟ) to ๐œ™๐‘ (๐œƒ๐‘˜ + ๐›ฟ) around each ๐œƒ๐‘˜ (see Figure 3.4).4 We now observe that ๐œƒ๐‘˜ =

Figure 3.4. The functions ๐ท๐‘ , ๐œ™๐‘ and ๐‘”๐‘ for ๐‘ = 5. ๐œ‹

โˆซ |๐œ™๐‘ (๐œƒ) โˆ’ ๐‘”๐‘ (๐œƒ)|๐‘‘๐œƒ = 2๐‘๐›ฟ, โˆ’๐œ‹

so, using the fact that |๐ท๐‘ (๐œƒ)| โ‰ค 2๐‘ + 1, we obtain ๐œ‹

1 1 1 โˆซ |๐ท (๐œƒ)||๐œ™๐‘ (๐œƒ) โˆ’ ๐‘”๐‘ (๐œƒ)|๐‘‘๐œƒ โ‰ค (2๐‘ + 1)2๐‘๐›ฟ < , 2๐œ‹ โˆ’๐œ‹ ๐‘ 2๐œ‹ 4 if we choose ๐›ฟ< 4

Cf. the proof of Theorem A.10.

๐œ‹ . 4๐‘(2๐‘ + 1)

3.7. Convergence for continuous functions

65

Hence ๐œ‹

1 โˆซ ๐ท (๐œƒ)๐‘”๐‘ (๐œƒ)๐‘‘๐œƒ 2๐œ‹ โˆ’๐œ‹ ๐‘ ๐œ‹

๐œ‹

1 1 โˆซ ๐ท (๐œƒ)๐œ™๐‘ (๐œƒ)๐‘‘๐œƒ โˆ’ โˆซ |๐ท (๐œƒ)||๐œ™๐‘ (๐œƒ) โˆ’ ๐‘”๐‘ (๐œƒ)|๐‘‘๐œƒ โ‰ฅ 2๐œ‹ โˆ’๐œ‹ ๐‘ 2๐œ‹ โˆ’๐œ‹ ๐‘ 4 โ‰ฅ 2 log ๐‘. ๐œ‹ Set ๐‘ large enough so that 4 log ๐‘ โ‰ฅ ๐‘€ + 1. ๐œ‹2 By Corollary 3.49, there exists a trigonometric polynomial ๐‘(๐œƒ) such that |๐‘”๐‘ (๐œƒ) โˆ’ ๐‘(๐œƒ)| < Hence

1 . 2๐‘ + 1

๐œ‹

1 โˆซ |๐ท (๐œƒ)||๐‘”๐‘ (๐œƒ) โˆ’ ๐‘(๐œƒ)|๐‘‘๐œƒ < 1 2๐œ‹ โˆ’๐œ‹ ๐‘ and, as above ๐œ‹

1 โˆซ ๐ท (๐œƒ)๐‘(๐œƒ)๐‘‘๐œƒ 2๐œ‹ โˆ’๐œ‹ ๐‘ ๐œ‹

โ‰ฅ

๐œ‹

1 1 โˆซ ๐ท (๐œƒ)๐‘”๐‘ (๐œƒ)๐‘‘๐œƒ โˆ’ โˆซ |๐ท (๐œƒ)||๐‘”๐‘ (๐œƒ) โˆ’ ๐‘(๐œƒ)|๐‘‘๐œƒ 2๐œ‹ โˆ’๐œ‹ ๐‘ 2๐œ‹ โˆ’๐œ‹ ๐‘

โ‰ฅ ๐‘€. Since |๐‘”๐‘ (๐œƒ)| โ‰ค 1, we clearly have |๐‘(๐œƒ)| โ‰ค 2.

โ–ก

Proof of Theorem 3.56. For each ๐‘˜ โ‰ฅ 0, by Lemma 3.57 we can find a trigonometric polynomial ๐‘ ๐‘˜ and an integer ๐‘ ๐‘˜ such that |๐‘ ๐‘˜ (๐œƒ)| โ‰ค 2 and ๐œ‹

(3.58)

1 โˆซ ๐ท (๐œƒ)๐‘ ๐‘˜ (๐œƒ)๐‘‘๐œƒ โ‰ฅ 2๐‘˜ . 2๐œ‹ โˆ’๐œ‹ ๐‘๐‘˜

Let ๐‘‘๐‘˜ be a sequence of integers such that ๐‘‘๐‘˜+1 > 3๐‘‘๐‘˜ and such that ๐‘‘๐‘˜ is at least as large as the degree of ๐‘ ๐‘˜ and as ๐‘ ๐‘˜ , so that we can write ๐‘‘๐‘˜

(3.59)

๐‘ ๐‘˜ (๐œƒ) = โˆ‘ ๐‘ ห†๐‘˜ (๐‘›)๐‘’๐‘–๐‘›๐œƒ , ๐‘›=โˆ’๐‘‘๐‘˜

66

3. Fourier series

and estimate (3.58) implies ๐‘๐‘˜

โˆ‘ ๐‘ ห†๐‘˜ (๐‘›) โ‰ฅ 2๐‘˜ .

(3.60)

๐‘›=โˆ’๐‘๐‘˜

Now define ๐‘ž๐‘˜ (๐œƒ) = ๐‘’2๐‘–๐‘‘๐‘˜ ๐œƒ ๐‘ ๐‘˜ (๐œƒ), so by (3.59) we have 3๐‘‘๐‘˜

๐‘‘๐‘˜

ห†๐‘˜ (๐‘›)๐‘’ ๐‘ž๐‘˜ (๐œƒ) = โˆ‘ ๐‘

๐‘–(๐‘›+2๐‘‘๐‘˜ )๐œƒ

๐‘›=โˆ’๐‘‘๐‘˜

= โˆ‘ ๐‘ ห†๐‘˜ (๐‘› โˆ’ 2๐‘‘๐‘˜ )๐‘’๐‘–๐‘›๐œƒ , ๐‘›=๐‘‘๐‘˜

so (3.60) is now 2๐‘‘๐‘˜ +๐‘๐‘˜

๐‘žห†๐‘˜ (๐‘›) โ‰ฅ 2๐‘˜ .

โˆ‘

(3.61)

๐‘›=2๐‘‘๐‘˜ โˆ’๐‘๐‘˜

Note that |๐‘ž๐‘˜ (๐œƒ)| โ‰ค 2 and, for each ๐‘›, ๐‘žห†๐‘˜ (๐‘›) โ‰  0 for at most one ๐‘˜ because ๐‘‘๐‘˜ > 3๐‘‘๐‘˜โˆ’1 . We can now define โˆž

๐‘“(๐œƒ) = โˆ‘ 2โˆ’๐‘˜ ๐‘ž๐‘˜ (๐œƒ). ๐‘˜=0

By the Weierstrass ๐‘€-test, the series above converges uniformly and, since each ๐‘ž๐‘˜ is continuous (itโ€™s a trigonometric polynomial), then ๐‘“ is continuous. Moreover, by the observations above and the uniform conฬ‚ = 2โˆ’๐‘˜ ๐‘žห†๐‘˜ (๐‘›) for ๐‘‘๐‘˜ โ‰ค ๐‘› โ‰ค vergence of the series, for each ๐‘› we have ๐‘“(๐‘›) 3๐‘‘๐‘˜ . Therefore, if ๐‘ ๐‘ (0) is the ๐‘th partial sum of the Fourier series of ๐‘“ at 0, 2๐‘‘๐‘˜ +๐‘๐‘˜

๐‘ 2๐‘‘๐‘˜ +๐‘๐‘˜ (0) โˆ’ ๐‘ 2๐‘‘๐‘˜ โˆ’๐‘๐‘˜ โˆ’1 (0) = 2โˆ’๐‘˜

โˆ‘

๐‘žห†๐‘˜ (๐‘›) โ‰ฅ 1,

๐‘›=2๐‘‘๐‘˜ โˆ’๐‘๐‘˜

by (3.61). Therefore, the sequence ๐‘ ๐‘ (0) is not a Cauchy sequence, and cannot converge. โ–ก By an appropiate translation of ๐‘“, we can verify the existence of a continuous function with divergent Fourier series at any ๐œƒ0 โˆˆ ๐•Š. By adding such functions, we conclude that there exist continuous functions with divergent Fourier series at any finite, or even countable infinite, number of points (Exercise (18)). Theorem 3.56 implies that being continuous is not enough for a function to have convergent Fourier series. However, if the function is โ€œregular enoughโ€, then we can guarantee its convergence, as in Theorem

3.7. Convergence for continuous functions

67

3.32, where we have ๐‘“ โˆˆ ๐ถ 1 (๐•Š). Moreover, it is enough for a function to be differentiable at a point to conclude that its Fourier series converges at that point. Theorem 3.62. Let ๐‘“ be Riemann-integrable on ๐•Š and differentiable at ๐œƒ0 . Then its Fourier series converges at ๐œƒ0 . Proof. Let ๐‘“ be differentiable at ๐œƒ0 . We want to prove that ๐œ‹

(3.63)

๐‘ ๐‘ (๐œƒ0 ) โˆ’ ๐‘“(๐œƒ0 ) =

1 โˆซ ๐ท (๐œ)(๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ))๐‘‘๐œ 2๐œ‹ โˆ’๐œ‹ ๐‘

converges to 0 as ๐‘ โ†’ โˆž. Since ๐ท๐‘ (๐œ) =

sin ๐‘๐œ cos ๐œ/2 + cos ๐‘๐œ sin ๐œ/2 sin(๐‘ + 1/2)๐œ = , sin ๐œ/2 sin ๐œ/2

we can write (3.63) as ๐œ‹

(3.64)

๐‘ ๐‘ (๐œƒ0 ) โˆ’ ๐‘“(๐œƒ0 ) =

1 sin ๐‘๐œ cos ๐œ/2 โˆซ (๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ))๐‘‘๐œ 2๐œ‹ โˆ’๐œ‹ sin ๐œ/2 ๐œ‹

+

1 โˆซ cos ๐‘๐œ(๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ))๐‘‘๐œ. 2๐œ‹ โˆ’๐œ‹

Note that the second integral in (3.64) is the Fourier coefficient ๐‘Ž๐‘ of the Riemann-integrable function ๐œ โ†ฆ ๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ), and thus converges to 0 as ๐‘ โ†’ โˆž, by the Riemannโ€“Lebesgue lemma 3.43. The first integral in (3.64) is the Fourier coefficient ๐‘๐‘ of the function ๐œ โ†ฆ cos ๐œ/2

๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ) , sin ๐œ/2

which is also Riemann-integrable as the limit lim

๐œโ†’0

๐‘“(๐œƒ0 โˆ’ ๐œ) โˆ’ ๐‘“(๐œƒ0 ) sin ๐œ/2

exists because ๐‘“ is differentiable at ๐œƒ0 . Thus, it also converges to 0 when ๐‘ โ†’ โˆž. โ–ก

68

3. Fourier series

Exercises (1) Calculate the solutions of equations (3.1) and (3.2) when ๐œ† < 0. (2) If the sequences ๐‘Ž๐‘› and ๐‘๐‘› are bounded, then โˆž

๐‘ข(๐‘Ÿ, ๐œƒ) = โˆ‘ ๐‘Ÿ๐‘› (๐‘Ž๐‘› cos(๐‘›๐œƒ) + ๐‘๐‘› sin(๐‘›๐œƒ)) ๐‘›=0

is harmonic in ๐”ป. (3) For ๐‘š, ๐‘› โˆˆ โ„ค, 2๐œ‹

โˆซ 0

2๐œ‹ ๐‘š = ๐‘› ๐‘’๐‘–๐‘š๐œƒ ๐‘’โˆ’๐‘–๐‘›๐œƒ ๐‘‘๐œƒ = { 0 ๐‘š โ‰  ๐‘›.

(4) If ๐‘“ is Riemann-integrable and periodic with period ๐‘‡, then ๐‘Ž+๐‘‡

โˆซ

๐‘‡

๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ

๐‘Ž

0

for any ๐‘Ž โˆˆ โ„. (5) Use Dirichletโ€™s test to prove that the Fourier series of Example 3.12 converges for every ๐œƒ. (6) Complete the details of Example 3.13. (7) Complete the details of Example 3.14. (8) If โˆ‘ ๐‘Ž๐‘› is Abel-summable to ๐‘  and ๐‘›๐‘Ž๐‘› โ†’ 0, then โˆ‘ ๐‘Ž๐‘› converges to ๐‘ . (9) Use Example 3.14 to obtain the identity โˆž

1 ๐œ‹4 = . 4 90 ๐‘› ๐‘›=1 โˆ‘

(10) Suppose ๐‘“ is a Riemann-integrable function that has left and right limits at ๐œƒ0 , say lim ๐‘“(๐œƒ) = ๐‘“(๐œƒ0 โˆ’)

๐œƒโ†’๐œƒ0โˆ’

and

lim ๐‘“(๐œƒ) = ๐‘“(๐œƒ0 +),

๐œƒโ†’๐œƒ0+

then ๐‘“(๐œƒ0 โˆ’) + ๐‘“(๐œƒ0 +) , 2 as ๐‘Ÿ โ†’ 1, where ๐’œ๐‘Ÿ ๐‘“(๐œƒ) are the Abel means of its Fourier series. ๐’œ๐‘Ÿ ๐‘“(๐œƒ0 ) โ†’

Exercises

69

(11) The following exercises prove inequality (3.35) for the inner product (3.33) and the partial sums (3.34). All functions are Riemannintegrable periodic functions with period 2๐œ‹. (a) For a function ๐‘“, if ๐‘ ๐‘ is given by (3.34) and ๐‘ โˆˆ ๐’ฏ๐‘ , then ๐‘“ โˆ’๐‘ ๐‘ and ๐‘ are orthogonal, โŸจ๐‘“ โˆ’ ๐‘ ๐‘ , ๐‘โŸฉ = 0. (b) (Pythagorasโ€™s theorem) If ๐‘“, ๐‘” are orthogonal, then โ€–๐‘“ + ๐‘”โ€–2 = โ€–๐‘“โ€–2 + โ€–๐‘”โ€–2 . (c) Conclude (3.35) by noting that ๐‘“ โˆ’ ๐‘ = ๐‘“ โˆ’ ๐‘ ๐‘ + ๐‘ ๐‘› โˆ’ ๐‘, and that ๐‘“ โˆ’ ๐‘ ๐‘ is orthogonal to ๐‘ ๐‘ โˆ’ ๐‘. (12) Let โ„›(๐•Š) be the space of Riemann-integrable functions on ๐•Š, and let ๐‘… = โ„›(๐•Š)/ โˆผ, where โˆผ is the equivalence relation so that ๐‘“ โˆผ ๐‘” if, and only if, ๐‘“(๐‘ฅ) = ๐‘”(๐‘ฅ) for all ๐‘ฅ except at a set of measure 0. (a) The subspace of continuous functions ๐ถ(๐•Š) identifies with itself in ๐‘…. (b) The bilinear form โŸจ๐‘“, ๐‘”โŸฉ =

1 โˆซ 2๐œ‹ 0

2๐œ‹

๐‘“(๐œƒ)๐‘”(๐œƒ)๐‘‘๐œƒ

induces an inner product on ๐‘…. (c) The quadratic form โ€–๐‘“โ€– = โˆšโŸจ๐‘“, ๐‘“โŸฉ induces a norm in ๐‘…. (13) We say that ๐‘“ is Hรถlder continuous with exponent ๐›ผ, and write ๐‘“ โˆˆ ๐ถ ๐›ผ (๐•Š), for some 0 < ๐›ผ โ‰ค 1, if there exists ๐‘€ > 0 such that |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฆ)| โ‰ค ๐‘€|๐‘ฅ โˆ’ ๐‘ฆ|๐›ผ . Prove Bernsteinโ€™s theorem: If ๐‘“ โˆˆ ๐ถ ๐›ผ (๐•Š) for some ๐›ผ > 1/2, then ฬ‚ < โˆž. (Hint: Proceed as in the proof of Theorem 3.32). โˆ‘ |๐‘“(๐‘›)| (14) If โˆ‘ ๐‘Ž๐‘› is Cesร ro-summable to ๐‘ , then is it Abel-summable to ๐‘ . (15) For any ๐œ โˆˆ โ„ and ๐‘ โ‰ฅ 1, ๐‘โˆ’1

๐‘›

1 1 sin2 (๐‘๐œ/2) โˆ‘ โˆ‘ ๐‘’๐‘–๐‘˜๐œ = . ๐‘ ๐‘›=0 ๐‘˜=โˆ’๐‘› ๐‘ sin2 (๐œ/2) (16) Suppose ๐‘“ is a Riemann-integrable function that has left and right limits at ๐œƒ0 , say lim ๐‘“(๐œƒ) = ๐‘“(๐œƒ0 โˆ’)

๐œƒโ†’๐œƒ0โˆ’

and

lim ๐‘“(๐œƒ) = ๐‘“(๐œƒ0 +),

๐œƒโ†’๐œƒ0+

70

3. Fourier series then

๐‘“(๐œƒ0 โˆ’) + ๐‘“(๐œƒ0 +) , 2 where ๐œŽ๐‘ (๐œƒ) are the Cesร ro sums of its Fourier series. ๐œŽ๐‘ (๐œƒ0 ) โ†’

(17) The Dirichlet kernel ๐‘

๐ท๐‘ (๐œƒ) = โˆ‘ ๐‘’๐‘–๐‘›๐œƒ ๐‘›=โˆ’๐‘

is given explicitly by sin(๐‘ + 1/2)๐œƒ . sin ๐œƒ/2 (18) (a) For each ๐œƒ โˆˆ ๐•Š, there exists ๐‘“ โˆˆ ๐ถ(๐•Š) with divergent Fourier series at ๐œƒ. (b) If ๐œƒ1 , ๐œƒ2 , . . . , ๐œƒ๐‘˜ โˆˆ ๐•Š, there exists ๐‘“ โˆˆ ๐ถ(๐•Š) with divergent Fourier series at ๐œƒ1 , ๐œƒ2 , . . . , ๐œƒ๐‘˜ . (c) If ๐ด โŠ‚ ๐•Š is countable, there exists ๐‘“ โˆˆ ๐ถ(๐•Š) with divergent Fourier series at each point in ๐ด. ๐ท๐‘ (๐œƒ) =

Notes The idea of solving Laplaceโ€™s equation, and the heat equation, by decomposing its solutions in trigonometric series was introduced by Fourier in [Fou55]. Parsevalโ€™s identity was stated by Marc-Antoine Parseval in [PdC06], and it was proven by Michel Plancherel in [Pla10]. The Riemannโ€“ Lebesgueโ€™s lemma was first proven by Riemann for the case of Riemannintegrable functions, and then proven for general measurable functions by Lebesgue in [Leb03]. Theorem 3.45 was proven by Lipรณt Fejรฉr in [Fej00]. An extensive discussion of summability methods can be found in [Zyg02]. The proof of Theorem 3.62 is taken from [SS03]. Bernsteinโ€™s theorem was proven in [Ber14]. The steps to prove it in Exercise (13) are also taken from [SS03], which are elaborated from its proof in [Zyg02]. The first to provide an explicit example of a continuous function with divergent Fourier series was Paul du Bois-Reymond in [du 76]. The construction discussed here follows [K8ฬˆ8]. Another construction can be found in [SS03].

Chapter 4

Poisson kernel in the half-space

4.1. The Poisson kernel in the half-space We now study harmonic functions in the upper half-space โ„๐‘‘+1 = {(๐‘ฅ, ๐‘ก) โˆถ ๐‘ฅ โˆˆ โ„๐‘‘ , ๐‘ก > 0}. + As in the case of the ball in โ„๐‘‘ , we will study explicit formulas for harmonic functions in โ„๐‘‘+1 + , given their values at its boundary โ„๐‘‘ ร— {0} = {(๐‘ฅ, 0) โˆถ ๐‘ฅ โˆˆ โ„๐‘‘ } Note that, while the ball is a bounded set, and hence its boundary is compact, the upper half-space is unbounded, and so is its boundary โ„๐‘‘ ร— {0}. This will force us to be more careful when defining objects analogous to the Poisson integrals studied before. The Poisson kernel for the upper half-space is given by the function (4.1)

๐‘ƒ๐‘ก (๐‘ฅ) =

๐œ”๐‘‘+1

(|๐‘ฅ|2

2๐‘ก , + ๐‘ก2 )(๐‘‘+1)/2

๐‘‘

defined for ๐‘ฅ โˆˆ โ„ and ๐‘ก > 0. Compare the function ๐‘ƒ๐‘ก (๐‘ฅ) with the Poisson kernel for the ball, given by 1 โˆ’ |๐‘ฅ|2 ๐‘ƒ(๐‘ฅ, ๐œ‰) = ๐œ”๐‘‘ |๐‘ฅ โˆ’ ๐œ‰|๐‘‘ 71

72

4. Poisson kernel in the half-space

for ๐‘ฅ โˆˆ ๐”น and ๐œ‰ โˆˆ ๐•Š. Both functions have a multiple of the distance to the boundary (2๐‘ก and (1 + |๐‘ฅ|)(1 โˆ’ |๐‘ฅ|), respectively), and both have the distance to a boundary point powered to the dimension: in the case of the upper half-space, this is the distance between (๐‘ฅ, ๐‘ก) to the origin, so it is |(๐‘ฅ, ๐‘ก)|๐‘‘+1 , and in the case of the ball, the distance between ๐‘ฅ and ๐œ‰, for each ๐œ‰ โˆˆ ๐•Š, so it is |๐‘ฅ โˆ’ ๐œ‰|๐‘‘ . Example 4.2. For ๐‘‘ = 1, since ๐œ”2 = 2๐œ‹, the circunference of the unit circle, the Poisson kernel for the upper half-plane is given by ๐‘ƒ๐‘ก (๐‘ฅ) =

1 ๐‘ก . 2 ๐œ‹ ๐‘ฅ + ๐‘ก2

๐‘ƒ๐‘ก (๐‘ฅ) is an even function, and since we can write ๐‘ƒ๐‘ก (๐‘ฅ) =

1 1 1 โ‹… , ๐‘ก ๐œ‹ (๐‘ฅ/๐‘ก)2 + 1

we see that it has a bump at the origin that increases as ๐‘ก โ†’ 0, as seen in Figure 4.1.

Figure 4.1. The Poisson kernel for the upper half-plane. Note that it is an even function, with a bump at the origin.

4.3. The function (๐‘ฅ, ๐‘ก) โ†ฆ ๐‘ƒ๐‘ก (๐‘ฅ) is harmonic in โ„๐‘‘+1 + . This can be verified explicitly by differentiating (Exercise (1)). 4.4. For each ๐‘ก > 0, ๐‘ƒ๐‘ก (๐‘ฅ) =

๐‘ฅ 1 ๐‘ƒ1 ( ). ๐‘‘ ๐‘ก ๐‘ก

4.1. The Poisson kernel in the half-space

73

Indeed, we have 2๐‘ก 2๐‘ก = ๐œ”๐‘‘+1 (|๐‘ฅ|2 + ๐‘ก2 )(๐‘‘+1)/2 ๐œ”๐‘‘+1 ๐‘ก๐‘‘+1 (|๐‘ฅ/๐‘ก|2 + 1)(๐‘‘+1)/2 1 2 ๐‘ฅ 1 = ๐‘‘ = ๐‘‘ ๐‘ƒ1 ( ). (๐‘‘+1)/2 2 ๐‘ก ๐‘ก ๐œ”๐‘‘+1 (|๐‘ฅ/๐‘ก| + 1) ๐‘ก

๐‘ƒ๐‘ก (๐‘ฅ) =

Thus, ๐‘ƒ๐‘ก is the dilation of ๐‘ƒ1 , which motivates the notation of ๐‘ก as a subindex parameter, rather than as another variable of ๐‘ƒ. For any ๐‘ก > 0, โˆซ ๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ = 1.

(4.5)

โ„๐‘‘

The integral in (4.5) must be understood in the improper sense, i.e. โˆซ ๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ = lim โˆซ ๐‘โ†’โˆž

โ„๐‘‘

๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ.

|๐‘ฅ|โ‰ค๐‘

The fact the this limit exists for ๐‘ก > 0 follows from the estimate 2๐‘ก 2๐‘ก โ‰ค , ๐‘ƒ๐‘ก (๐‘ฅ) = (๐‘‘+1)/2 2 2 ๐œ”๐‘‘+1 |๐‘ฅ|๐‘‘+1 ๐œ”๐‘‘+1 (|๐‘ฅ| + ๐‘ก ) so, for ๐‘ โ‰ฅ ๐‘€, using spherical coordinates we obtain |โˆซ |

๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ โˆ’ โˆซ

|๐‘ฅ|โ‰ค๐‘

|๐‘ฅ|โ‰ค๐‘€

๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ|| โ‰ค โ‰ค

๐‘‘๐‘ฅ 2๐‘ก โˆซ ๐œ”๐‘‘+1 ๐‘€ 0, โˆซ

๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ โ†’ 0

|๐‘ฅ|โ‰ฅ๐›ฟ

as ๐‘ก โ†’ 0. Thus, the weight of the Poisson kernel gets concentrated at the origin as ๐‘ก โ†’ 0. To verify this limit, observe first that, for any ๐‘ก > 0, ๐‘ƒ๐‘ก (๐‘ฅ) โ‰ค

2๐‘ก , ๐œ”๐‘‘+1 |๐‘ฅ|๐‘‘+1

74

4. Poisson kernel in the half-space

and thus, using spherical coordinates as above โˆž

โˆซ

๐‘ƒ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ โ‰ค ๐‘๐‘ก โˆซ

|๐‘ฅ|โ‰ฅ๐›ฟ

|๐‘ฅ|โ‰ฅ๐›ฟ

1 ๐‘ก ๐‘‘๐‘ฅ = ๐‘๐‘ก โˆซ โˆซ ๐‘‘+1 ๐‘‘๐œŽ๐‘Ÿ๐‘‘โˆ’1 ๐‘‘๐‘Ÿ = ๐‘โ€ฒ , ๐›ฟ |๐‘ฅ|๐‘‘+1 ๐‘Ÿ ๐›ฟ ๐•Š

โ€ฒ

where ๐‘, ๐‘ are the same constants as above, and all integrals at infinity are improper integrals. Therefore, the integral clearly goes to 0 as ๐‘ก โ†’ 0, for each ๐›ฟ > 0.

4.2. Poisson integrals in the half-space In this section we study the Poisson integrals in the half-space. Again, as in the previous section, we understand all integrals in โ„๐‘‘ in the improper sense. For this, let โ„›(โ„๐‘‘ ) be the space of bounded, locally Riemannintegrable functions ๐‘“ on โ„๐‘‘ (๐‘“ is Riemann-integrable on any rectangle in โ„๐‘‘ ) such that the sequence โˆซ

(4.7)

|๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ

|๐‘ฅ|โ‰ค๐‘

is bounded. For ๐‘“ โˆˆ โ„›(โ„๐‘‘ ), we define its improper integral as โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = lim โˆซ

(4.8)

๐‘โ†’โˆž

โ„๐‘‘

๐‘“(๐‘ฅ)๐‘‘๐‘ฅ.

|๐‘ฅ|โ‰ค๐‘

To see that this limit exists, note that, since the sequence (4.7) is increasing and bounded, it converges, so for any ๐œ€ > 0 there exists ๐พ such that, if ๐‘ โ‰ฅ ๐‘€ โ‰ฅ ๐พ, then โˆซ ๐‘€โ‰ค|๐‘ฅ|โ‰ค๐‘

|๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ = || โˆซ

|๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ โˆ’ โˆซ

|๐‘ฅ|โ‰ค๐‘€

|๐‘ฅ|โ‰ค๐‘

|๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ|| < ๐œ€.

Thus, for ๐‘ โ‰ฅ ๐‘€ โ‰ฅ ๐พ, |โˆซ |

๐‘“(๐‘ฅ)๐‘‘๐‘ฅ โˆ’ โˆซ

|๐‘ฅ|โ‰ค๐‘€

|๐‘ฅ|โ‰ค๐‘

๐‘“(๐‘ฅ)๐‘‘๐‘ฅ|| โ‰ค โˆซ

|๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ < ๐œ€,

๐‘€โ‰ค|๐‘ฅ|โ‰ค๐‘

and therefore the limit in (4.8) exists. Improper integrals on โ„๐‘‘ are translation invariant, that is, (4.9)

โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ โˆ’ โ„Ž)๐‘‘๐‘ฅ, โ„๐‘‘

โ„๐‘‘

4.2. Poisson integrals in the half-space

75

for any ๐‘“ โˆˆ โ„›(โ„๐‘‘ ) and any โ„Ž โˆˆ โ„๐‘‘ . They also satisfy, for any ๐‘Ÿ > 0, โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ =

(4.10)

โ„๐‘‘

๐‘ฅ 1 โˆซ ๐‘“( )๐‘‘๐‘ฅ. ๐‘Ÿ๐‘‘ โ„๐‘‘ ๐‘Ÿ

These two properties follow from the boundedness of the sequence (4.7), and are left as an exercise (Exercise (3)). For a bounded locally Riemann-integrable function ๐‘“ in โ„๐‘‘ , we define its Poisson integral to be the function ๐‘ข(๐‘ฅ, ๐‘ก) given by ๐‘ข(๐‘ฅ, ๐‘ก) = ๐’ซ๐‘ก ๐‘“(๐‘ฅ) = โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ,

(4.11)

โ„๐‘‘

for each (๐‘ฅ, ๐‘ก) โˆˆ โ„๐‘‘+1 + . The integral in (4.11) is well defined because ๐‘“ is bounded and thus ๐‘ฆ โ†ฆ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ) is in โ„›(โ„๐‘‘ ). Note that we can also write the Poisson integral as ๐‘ข(๐‘ฅ, ๐‘ก) = โˆซ ๐‘ƒ๐‘ก (๐‘ฆ)๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ, โ„๐‘‘ 1

using (4.9).

The Poisson integral ๐‘ข(๐‘ฅ, ๐‘ก) of a bounded function ๐‘“ is a bounded continuous function. Indeed, if |๐‘“(๐‘ฅ)| โ‰ค ๐‘€, |๐‘ข(๐‘ฅ, ๐‘ก)| โ‰ค โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)|๐‘“(๐‘ฆ)|๐‘‘๐‘ฆ โ‰ค ๐‘€ โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ = ๐‘€, โ„๐‘‘

โ„๐‘‘

by (4.5) and (4.9). The continuity follows from the continuity of ๐‘ƒ๐‘ก (๐‘ฅ), its integrability, and the boundedness of ๐‘“. Note that, for |๐‘ง| > 2|๐‘ฅ|, we have |๐‘ฅ โˆ’ ๐‘ง| > |๐‘ง|/2 and thus ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) โ‰ค

2๐‘ก 2๐‘‘+2 ๐‘ก < . ๐œ”๐‘‘+1 |๐‘ฅ โˆ’ ๐‘ง|๐‘‘+1 ๐œ”๐‘‘+1 |๐‘ง|๐‘‘+1

Hence, if ๐‘ > 2|๐‘ฅ|, 2๐‘‘+2 ๐œ”๐‘‘ ๐‘ก๐‘€ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง)๐‘“(๐‘ง)๐‘‘๐‘ง|| โ‰ค . ๐œ”๐‘‘+1 ๐‘ |๐‘ง|>๐‘

|โˆซ | 1

The operator ๐‘“ โ†ฆ ๐’ซ๐‘ก ๐‘“ is also called the Poisson semigroup operator. It satisfies ๐’ซ๐‘ก ๐’ซ๐‘  ๐‘“ = ๐’ซ๐‘ก+๐‘  ๐‘“ ๐‘‘

for any ๐‘“ โˆˆ โ„›(โ„ ) and ๐‘ก, ๐‘  > 0. This follows from the identity โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง)๐‘ƒ๐‘  (๐‘ง โˆ’ ๐‘ฆ)๐‘‘๐‘ง = ๐‘ƒ๐‘ก+๐‘  (๐‘ฅ โˆ’ ๐‘ฆ), โ„๐‘‘

which will be proved later using the Fourier transform (Chapter 10).

76

4. Poisson kernel in the half-space

Given (๐‘ฅ, ๐‘ก) โˆˆ โ„๐‘‘+1 and ๐œ€ > 0, choose ๐‘ > 2(|๐‘ฅ| + 1) such that + 2๐‘‘+2 ๐œ”๐‘‘ (๐‘ก + 1)๐‘€ ๐œ€ < . ๐œ”๐‘‘+1 ๐‘ 3 Now, choose ๐›ฟ > 0 such that ๐›ฟ < 1 and, if (|๐‘ฅ โˆ’ ๐‘ฆ|2 + (๐‘ก โˆ’ ๐‘ )2 )1/2 < ๐›ฟ, then ๐‘‘๐œ€ |๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) โˆ’ ๐‘ƒ๐‘  (๐‘ฆ โˆ’ ๐‘ง)| < 3๐œ”๐‘‘ ๐‘€๐‘ ๐‘‘ for any |๐‘ง| โ‰ค ๐‘. Such ๐›ฟ exists since the closed ball of radius ๐‘ + 1 is compact and, hence, ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) is uniformly continuous. Therefore, using the fact that |๐‘ฅ โˆ’ ๐‘ฆ| < 1 and |๐‘ก โˆ’ ๐‘ | < 1 if (|๐‘ฅ โˆ’ ๐‘ฆ|2 + (๐‘ก โˆ’ ๐‘ )2 )1/2 < ๐›ฟ, so that we have ๐‘ > 2|๐‘ฆ| and ๐‘  < ๐‘ก + 1, | | | โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง)๐‘“(๐‘ง)๐‘‘๐‘ง โˆ’ โˆซ ๐‘ƒ๐‘  (๐‘ฆ โˆ’ ๐‘ง)๐‘“(๐‘ง)๐‘‘๐‘ง| โ„๐‘‘

โ„๐‘‘

โ‰ค โˆซ |๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) โˆ’ ๐‘ƒ๐‘  (๐‘ฆ โˆ’ ๐‘ง)||๐‘“(๐‘ง)|๐‘‘๐‘ง โ„๐‘‘

2๐‘‘+2 ๐œ”๐‘‘ ๐‘ก๐‘€ 2๐‘‘+2 ๐œ”๐‘‘ ๐‘ ๐‘€ ๐‘‘๐œ€ ๐‘€๐‘‘๐‘ง + + ๐‘‘ ๐œ”๐‘‘+1 ๐‘ ๐œ”๐‘‘+1 ๐‘ |๐‘ง|โ‰ค๐‘ 3๐œ” ๐‘‘ ๐‘€๐‘ ๐œ€ ๐œ€ ๐œ€ < + + = ๐œ€. 3 3 3 โ‰คโˆซ

As above, we can also verify that ๐‘ข(๐‘ฅ, ๐‘ก) is differentiable, using the fact that ๐‘ƒ๐‘ก (๐‘ฅ) is continuously differentiable and in โ„›(โ„๐‘‘ ) for each ๐‘ก. Indeed, the partial derivatives of ๐‘ƒ๐‘ก (๐‘ฅ) are given by 2(๐‘‘ + 1)๐‘ก๐‘ฅ๐‘— ๐œ•๐‘ƒ๐‘ก =โˆ’ ๐œ•๐‘ฅ๐‘— ๐œ”๐‘‘+1 (|๐‘ฅ|2 + ๐‘ก2 )(๐‘‘+3)/2 for ๐‘— = 1, . . . , ๐‘‘, and 2(|๐‘ฅ|2 โˆ’ ๐‘‘๐‘ก2 ) ๐œ•๐‘ƒ๐‘ก , = ๐œ•๐‘ก ๐œ”๐‘‘+1 (|๐‘ฅ|2 + ๐‘ก2 )(๐‘‘+3)/2 and thus, as above, for |๐‘ง| > 2|๐‘ฅ|, | ๐œ•๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง)| โ‰ค ๐ถ๐‘ก | ๐œ•๐‘ฅ | |๐‘ง|๐‘‘+2 ๐‘—

and

| ๐œ•๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง)| โ‰ค ๐ถ , | ๐œ•๐‘ก | |๐‘ง|๐‘‘+1

4.3. Boundary limits

77

for some constant ๐ถ that depends only on ๐‘‘. Moreover, note that for โ„Ž โˆˆ โ„, ๐‘ข(๐‘ฅ + โ„Ž๐‘’๐‘— , ๐‘ก) โˆ’ ๐‘ข(๐‘ฅ, ๐‘ก) ๐œ•๐‘ƒ๐‘ก โˆ’โˆซ (๐‘ฅ โˆ’ ๐‘ง)๐‘“(๐‘ง)๐‘‘๐‘ง โ„Ž ๐œ•๐‘ฅ ๐‘— โ„๐•• =โˆซ ( โ„๐‘‘

๐‘ƒ๐‘ก (๐‘ฅ + โ„Ž๐‘’๐‘— โˆ’ ๐‘ง) โˆ’ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) ๐œ•๐‘ƒ โˆ’ ๐‘ก (๐‘ฅ โˆ’ ๐‘ง))๐‘“(๐‘ง)๐‘‘๐‘ง โ„Ž ๐œ•๐‘ฅ๐‘— =โˆซ ( โ„๐‘‘

๐œ•๐‘ƒ๐‘ก ๐œ•๐‘ƒ (๐‘ฆ โˆ’ ๐‘ง) โˆ’ ๐‘ก (๐‘ฅ โˆ’ ๐‘ง))๐‘“(๐‘ง)๐‘‘๐‘ง, ๐œ•๐‘ฅ๐‘— ๐œ•๐‘ฅ๐‘—

๐‘‘

where ๐‘ฆ โˆˆ โ„ is of the form ๐‘ฆ = ๐‘ฅ + ๐‘ ๐‘’๐‘— where ๐‘  is a number between 0 and โ„Ž that depends on ๐‘ฅ, ๐‘ก and ๐‘ง. Since ๐œ•๐‘ƒ๐‘ก ๐œ•๐‘ƒ (๐‘ฆ โˆ’ ๐‘ง) โ†’ ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) ๐œ•๐‘ฅ๐‘— ๐œ•๐‘ฅ๐‘— as โ„Ž โ†’ 0 for each ๐‘ง, we can proceed as above and conclude that ๐œ•๐‘ƒ๐‘ก ๐œ•๐‘ข =โˆซ (๐‘ฅ โˆ’ ๐‘ง)๐‘“(๐‘ง)๐‘‘๐‘ง, ๐œ•๐‘ฅ๐‘— ๐œ•๐‘ฅ ๐‘— โ„๐•• so we can โ€œdifferentiate under the integralโ€. Similarly for the partial derivative of ๐‘ข with respect to ๐‘ก. Hence, we have the following result. Proposition 4.12. If ๐‘“ is a bounded locally Riemann-integrable function in โ„๐‘‘ , its Poisson integral ๐‘ข(๐‘ฅ, ๐‘ก) is harmonic in โ„๐‘‘+1 + . Proof. Similarly, as in the above cases, we can differentiate inside the Poisson integral to verify that ๐‘‘

๐œ•2 ๐‘ƒ๐‘ก ๐œ•2 ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง) + (๐‘ฅ โˆ’ ๐‘ง))๐‘“(๐‘ง)๐‘‘๐‘ง. 2 ๐œ•๐‘ก2 ๐‘—=1 ๐œ•๐‘ฅ๐‘—

ฮ”๐‘ข(๐‘ฅ, ๐‘ก) = โˆซ ( โˆ‘ โ„๐••

The theorem then follows from the fact that (๐‘ฅ, ๐‘ก) โ†ฆ ๐‘ƒ๐‘ก (๐‘ฅ) is harmonic. We leave the details as an exercise (Exercise (4)). โ–ก

4.3. Boundary limits We now study the behavior of a Poisson integrals ๐‘ข(๐‘ฅ, ๐‘ก) as the point (๐‘ฅ, ๐‘ก) approaches a boundary point. We start with Theorem 4.13, analogous to Theorem 2.28.

78

4. Poisson kernel in the half-space

Theorem 4.13. Let ๐‘“ be a bounded locally Riemann-integrable function in โ„๐‘‘ , continuous at ๐‘ฅ0 โˆˆ โ„๐‘‘ , and let ๐‘ข(๐‘ฅ, ๐‘ก) be its Poisson integral. Then lim

(๐‘ฅ,๐‘ก)โ†’(๐‘ฅ0 ,0)

๐‘ข(๐‘ฅ, ๐‘ก) = ๐‘“(๐‘ฅ0 ).

In particular, if ๐‘“ is continuous at every point of โ„๐‘‘ , its Poisson integral ๐‘ข(๐‘ฅ, ๐‘ก) in โ„๐‘‘+1 extends continuously to the boundary โ„๐‘‘ ร— {0}, with + the value ๐‘“(๐‘ฅ) at each (๐‘ฅ, 0). Unsurprinsingly, the proof is very similar to the proofs of Theorems 2.28 and 3.45. Proof. Given ๐œ€ > 0, since ๐‘“ is continuous at ๐‘ฅ0 there exists ๐œ‚ > 0 such that, if |๐‘ฅ โˆ’ ๐‘ฅ0 | < ๐œ‚, then |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ0 )| < ๐œ€/2. Now, write ๐‘ข(๐‘ฅ, ๐‘ก) โˆ’ ๐‘“(๐‘ฅ0 ) = โˆซ ๐‘ƒ๐‘ก (๐‘ง)๐‘“(๐‘ฅ โˆ’ ๐‘ง)๐‘‘๐‘ง โˆ’ ๐‘“(๐‘ฅ0 ) โ„๐‘‘

= โˆซ ๐‘ƒ๐‘ก (๐‘ง)(๐‘“(๐‘ฅ โˆ’ ๐‘ง) โˆ’ ๐‘“(๐‘ฅ0 ))๐‘‘๐‘ง, โ„๐‘‘

where we have used (4.5). Now, if |๐‘ฅ โˆ’ ๐‘ฅ0 | < ๐œ‚/2 and |๐‘ง| < ๐œ‚/2, |๐‘ฅ โˆ’ ๐‘ง โˆ’ ๐‘ฅ0 | < ๐œ‚ and thus |โˆซ |

|๐‘ง| 0, โˆซ

|ฮฆ๐‘ก (๐‘ฅ)|๐‘‘๐‘ฅ โ†’ 0 as ๐‘ก โ†’ 0.

|๐‘ฅ|โ‰ฅ๐›ฟ

These operators are called convolution operators, and are denoted by ฮฆ๐‘ก โˆ— ๐‘“. Note that the Poisson integral of ๐‘“ is the convolution ๐‘ƒ๐‘ก โˆ— ๐‘“. The collection of functions {ฮฆ๐‘ก }๐‘ก>0 form a family of good kernels. See Exercise (9).

80

4. Poisson kernel in the half-space

Exercises (1) The function (๐‘ฅ, ๐‘ก) โ†ฆ ๐‘ƒ๐‘ก (๐‘ฅ) is harmonic in โ„๐‘‘+1 + . (2) For any dimension ๐‘‘ โ‰ฅ 1, โˆซ โ„๐‘‘

(|๐‘ฅ|2

๐‘‘๐‘ฅ ๐œ‹(๐‘‘+1)/2 = , (๐‘‘+1)/2 ฮ“((๐‘‘ + 1)/2) + 1)

and verify (4.5). (Hint: Use spherical coordinates and the identity โˆž

โˆซ ๐‘ก๐›ผ ๐‘’โˆ’๐‘ก๐‘  0

ฮ“(๐›ผ) ๐‘‘๐‘ก = ๐›ผ ๐‘ก ๐‘ 

for any ๐›ผ, ๐‘  > 0.) (3) Let ๐‘“ โˆˆ โ„›(โ„๐‘‘ ). (a) For any โ„Ž โˆˆ โ„๐‘‘ , โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ โˆ’ โ„Ž)๐‘‘๐‘ฅ. โ„๐‘‘

โ„๐‘‘

(b) For any ๐‘Ÿ > 0, โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โ„๐‘‘

1 ๐‘ฅ โˆซ ๐‘“( )๐‘‘๐‘ฅ. ๐‘Ÿ๐‘‘ โ„๐‘‘ ๐‘Ÿ

(Hint: The integrals โˆซ|๐‘ฅ|โ‰ค๐‘ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ form a Cauchy sequence.) (4) Complete the details of the proof of Proposition 4.12. (5) If ๐‘“ โˆˆ ๐ถ0 (โ„๐‘‘ ), then ๐‘ข(๐‘ฅ, ๐‘ก) โ†’ ๐‘“(๐‘ฅ) as ๐‘ก โ†’ 0, uniformly in ๐‘ฅ โˆˆ โ„๐‘‘ . (6) Let ๐‘“ โˆˆ ๐ถ๐‘ (โ„๐‘‘ ). Then โˆซ |๐‘“(๐‘ฅ โˆ’ โ„Ž) โˆ’ ๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ โ†’ 0 โ„๐‘‘ ๐‘‘

as โ„Ž โ†’ 0 in โ„ . (Hint: ๐‘“ is uniformly continuous on its compact support.) (7) Prove the following version of Fubiniโ€™s theorem: Let ๐‘“(๐‘ฅ, ๐‘ฆ) โˆˆ ๐ถ(โ„๐‘‘+๐‘‘ ) such that (a) there exists ๐ด > 0 such that, for all ๐‘ฆ โˆˆ โ„๐‘‘ , |๐‘“(๐‘ฅ, ๐‘ฆ)| โ‰ค

๐ด ; (|๐‘ฅ|๐‘‘+1 + 1)

Exercises

81

(b) there exists a compact ๐พ โŠ‚ โ„๐‘‘ such that, for all ๐‘ฅ โˆˆ โ„๐‘‘ , ๐‘“(๐‘ฅ, โ‹…) is supported in ๐พ. Then ๐‘“(๐‘ฅ, ๐‘ฆ) is integrable in โ„2๐‘‘ and โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฅ๐‘‘๐‘ฆ = โˆซ ( โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฅ)๐‘‘๐‘ฆ = โˆซ ( โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ)๐‘‘๐‘ฅ. โ„2๐‘‘

โ„๐‘‘

โ„๐‘‘

โ„๐‘‘

โ„๐‘‘

(8) If ๐‘“ โˆˆ ๐ถ๐‘ (โ„๐‘‘ ), โˆซ |๐‘ข(๐‘ฅ, ๐‘ก) โˆ’ ๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ โ†’ 0 โ„๐‘‘

as ๐‘ก โ†’ 0. (Hint: Use Exercises (6) and (7).) (9) Let {๐พ๐‘ก }๐‘ก>0 be a family of functions in โ„›(โ„๐‘‘ ). We say that it is a family of good kernels if โ€ข โˆซ ๐พ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ = 1 for all ๐‘ก > 0; โ„๐‘‘

โ€ข there exists ๐‘€ > 0 such that โˆซ |๐พ๐‘ก (๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค ๐‘€ for all ๐‘ก > 0; โ„๐‘‘

and โ€ข for ๐›ฟ > 0, โˆซ

|๐พ๐‘ก (๐‘ฅ)|๐‘‘๐‘ฅ โ†’ 0 as ๐‘ก โ†’ 0.

|๐‘ฅ|โ‰ฅ๐›ฟ ๐‘‘

(a) If ฮฆ โˆˆ โ„›(โ„ ) and โˆซ ฮฆ = 1, then its dilations {ฮฆ๐‘ก }๐‘ก>0 form a family of good kernels. (b) If {๐พ๐‘ก }๐‘ก>0 is a family of good kernels and ๐‘“ โˆˆ ๐ถ(โ„๐‘‘ ) is bounded, then, for each ๐‘ฅ โˆˆ โ„๐‘‘ , lim

(๐‘ฆ,๐‘ก)โ†’(๐‘ฅ,0)

๐พ๐‘ก โˆ— ๐‘“(๐‘ฆ) = ๐‘“(๐‘ฅ).

(c) If {๐พ๐‘ก }๐‘ก>0 is a family of good kernels and ๐‘“ โˆˆ ๐ถ0 (โ„๐‘‘ ), then ๐พ๐‘ก โˆ— ๐‘“ โ‡‰ ๐‘“ as ๐‘ก โ†’ 0. (10) (Principle of subordination) Let ๐ป๐‘ก (๐‘ฅ) be the heat kernel, ๐ป๐‘ก (๐‘ฅ) =

1 2 ๐‘’โˆ’|๐‘ฅ| /4๐‘ก . (4๐œ‹๐‘ก)๐‘‘/2

Then ๐‘ƒ๐‘ก (๐‘ฅ) =

๐‘ก 2โˆš๐œ‹

โˆž 2 /4๐‘ 

โˆซ ๐‘’โˆ’๐‘ก 0

๐ป๐‘  (๐‘ฅ)

๐‘‘๐‘  . ๐‘ 3/2

(Hint: use the identity of the gamma function as in Exercise (2).)

82

4. Poisson kernel in the half-space

Notes The results of this chapter are classical, and can be found in the references cited previously, as [ABR01]. An extended treatment on Poisson integrals can be found in [Ste70].

Chapter 5

Measure theory in Euclidean space

5.1. The need for an integration theory We have seen in Chapter 4 that, when dealing with Poisson integrals in the upper half-space, that improper integrals add a further difficulty when dealing with boundary limits. However, the main problem with Riemann integration is the fact that the Riemann integral is not compatible with pointwise limits. Indeed, consider the set ๐‘„ = [0, 1] โˆฉ โ„š of rational numbers in the interval [0, 1]. This set is countable, so we can write ๐‘„ = {๐‘ž๐‘› โˆถ ๐‘› โˆˆ โ„•}. Now, for each ๐‘›, let ๐‘“๐‘› be the function on [0, 1] given by 1 ๐‘ฅ = ๐‘ž๐‘— for some ๐‘— โ‰ค ๐‘› ๐‘“๐‘› (๐‘ฅ) = { 0 otherwise. Then, we see that ๐‘“๐‘› (๐‘ฅ) โ†’ ๐‘“(๐‘ฅ) for every ๐‘ฅ โˆˆ [0, 1], where ๐‘“ is the function 1 ๐‘“(๐‘ฅ) = { 0

๐‘ฅโˆˆ๐‘„ ๐‘ฅ โˆ‰ ๐‘„.

However, while each ๐‘“๐‘› is Riemann-integrable on [0, 1], ๐‘“ is not. Given any partition ๐’ซ = {๐‘ฅ0 = 0 < ๐‘ฅ1 < ๐‘ฅ2 < . . . < ๐‘ก๐‘› = 1} 83

84

5. Measure theory in Euclidean space

of [0, 1], we see that ๐‘€๐‘— = sup{๐‘“(๐‘ฅ) โˆถ ๐‘ฅ โˆˆ [๐‘ฅ๐‘—โˆ’1 , ๐‘ฅ๐‘— ]} = 1 for all ๐‘— = 1, . . . , ๐‘›, because every interval in โ„ contains a rational number, so the upper Riemann sum of ๐‘“ is ๐‘›

๐‘›

๐‘ˆ(๐‘“, ๐’ซ) = โˆ‘ ๐‘€๐‘— (๐‘ฅ๐‘— โˆ’ ๐‘ฅ๐‘—โˆ’1 ) = โˆ‘ (๐‘ฅ๐‘— โˆ’ ๐‘ฅ๐‘—โˆ’1 ) = 1. ๐‘—=1

๐‘—=1

On the other hand, ๐‘š๐‘— = inf{๐‘“(๐‘ฅ) โˆถ ๐‘ฅ โˆˆ [๐‘ฅ๐‘—โˆ’1 , ๐‘ฅ๐‘— ]} = 0 for all ๐‘— = 1, . . . , ๐‘›, because every interval in โ„ also contains an irrational number, so the lower Riemann sum is ๐‘›

๐ฟ(๐‘“, ๐’ซ) = โˆ‘ ๐‘š๐‘— (๐‘ฅ๐‘— โˆ’ ๐‘ฅ๐‘—โˆ’1 ) = 0. ๐‘—=1

As it is impossible to find a partition to get ๐‘ˆ(๐‘“, ๐’ซ) โˆ’ ๐ฟ(๐‘“, ๐’ซ) < 1, we conclude ๐‘“ is not Riemann integrable. We now ask if it is possible to extend the definition of the integral to satisfy the following requirements: (1) every Riemann-integrable function is also integrable in the new definition; (2) if each ๐‘“๐‘› is integrable and ๐‘“๐‘› โ†’ ๐‘“ pointwise, then ๐‘“ is also integrable; (3) under appropiate conditions, if ๐‘“๐‘› โ†’ ๐‘“ pointwise, then we have โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“ in the extended definition. Property (1) is necessary because we want to extend the definition of the integral, so that the results on integration we already have must still hold in the new definition. The โ€œappropiate conditionsโ€ above should not be too restrictive, of course. For example, the Riemann integral is consistent with uniform convergence, which is an hypothesis too strong to be useful in most situations. The purpose of this chapter is to give an introduction to measure and integration theory. Lebesgueโ€™s theory provides an integral that satisfies the above conditions, and will allow us to both extend the results of Chapter 4 and deal with further related problems.

5.2. Outer measure in Euclidean space

85

5.2. Outer measure in Euclidean space The first ingredient in Lebesgueโ€™s theory is the idea of measure of a set. Recall that if ๐‘… is a rectangle in โ„๐‘‘ , say ๐‘… = ๐ผ 1 ร— ๐ผ2 ร— โ‹ฏ ร— ๐ผ ๐‘‘ , where each ๐ผ๐‘— is a bounded interval, then its volume is given by vol(๐‘…) = |๐ผ1 | โ‹ฏ |๐ผ๐‘‘ |, the product of the lengths of the intervals ๐ผ1 , . . . , ๐ผ๐‘‘ . A cube ๐‘„ is a rectangle with all intervals ๐ผ๐‘— above of the same length, say ๐ฟ, and in that case vol(๐‘„) = ๐ฟ๐‘‘ . ๐‘„ is a dyadic cube if ๐ฟ = 2๐‘š for some ๐‘š โˆˆ โ„ค, and the limits of each ๐ผ๐‘— are of the form ๐‘˜ โ‹… 2๐‘š for some ๐‘˜ โˆˆ โ„ค. The intervals that define the rectangle ๐‘… may be open, closed, or neither. Note that ๐‘… is an open set if all the intervals ๐ผ๐‘— are open, and ๐‘… is closed if all of them are closed. For any rectangle ๐‘…, its interior ๐‘…0 and its closure ๐‘…ฬ„ satisfy vol(๐‘…0 ) = vol(๐‘…)ฬ„ = vol(๐‘…). For any ๐œ€ > 0, we can find an open rectangle ๐‘† โŠƒ ๐‘… such that vol(๐‘†) < vol(๐‘…) + ๐œ€, by widening (if necessary) each interval ๐ผ๐‘— , and a closed rectangle ๐‘‡ โŠ‚ ๐‘… such that vol(๐‘‡) > vol(๐‘…) โˆ’ ๐œ€, by shrinking (if necessary) each interval ๐ผ๐‘— (Exercise (1)). In order to extend the concept of volume to general sets ๐ด โŠ‚ โ„๐‘‘ , we first define the outer measure of ๐ด by |๐ด|โˆ— = inf { โˆ‘ vol(๐‘…๐‘— ) โˆถ each ๐‘…๐‘— is a rectangle and ๐ด โŠ‚ ๐‘—โ‰ฅ1

โ‹ƒ

๐‘…๐‘— }.

๐‘—โ‰ฅ1

Thus, |๐ด|โˆ— is the infimum of the set of all possible sums of volumes of rectangles covering ๐ด, where each cover may have a finite or a countable infinite number of rectangles. If no such sum is finite, then we say |๐ด|โˆ— = โˆž. Thus, for each ๐œ€ > 0, there exists a cover of rectangles ๐‘…1 , ๐‘…2 , . . . for ๐ด such that โˆ‘ vol(๐‘…๐‘— ) < |๐ด|โˆ— + ๐œ€, ๐‘—โ‰ฅ1

86

5. Measure theory in Euclidean space

and, by the observations above, we can choose all of these rectangles to be open, or all of them to be closed.1 The outer measure is also usually called exterior measure. 5.1. If ๐ด = โˆ…, then clearly |๐ด|โˆ— = 0. Also if ๐ด contains only one point, or if ๐ด is finite. In fact, if ๐ด is countable, |๐ด|โˆ— = 0 (Exercise (2)). As seen in Appendix A, if |๐ด|โˆ— = 0, we say that ๐ด has measure zero, or that itโ€™s a measure zero set. One can show that the countable union of measure zero sets is also of measure zero (Exercise (2)). The Cantor ternary set is also a set of measure zero (Exercise (3)). 5.2. If ๐‘… is a closed rectangle, then |๐‘…|โˆ— = vol(๐‘…). That is, its outer measure coincides with its volume. Indeed, first note that |๐‘…|โˆ— โ‰ค vol(๐‘…), because ๐‘… is a cover for itself. For the reverse inequality, suppose ๐œ€ > 0 is given and ๐‘…1 , ๐‘…2 , . . . are open rectangles that cover ๐‘… such that โˆ‘ vol(๐‘…๐‘— ) < |๐‘…|โˆ— + ๐œ€. ๐‘—โ‰ฅ1

Since ๐‘… is closed, it is compact, and thus we can assume we have a finite number of the ๐‘…๐‘— , say, we have ๐‘ of them. Now, the edges of these rectangles can be extended to form a grid of subrectangles ๐‘† ๐‘– of ๐‘…, as in

Figure 5.1. Given a cover of rectangles for a rectangle, the edges can be extended to form a grid of subrectangles.

Figure 5.1, each one of them also a subrectangle of the ๐‘…๐‘— , such that ๐‘…=

โ‹ƒ

๐‘†๐‘–

๐‘–

1

We use the convention that ๐‘ฅ < โˆž for all ๐‘ฅ โˆˆ โ„, and later on we will also use โˆž + โˆž = โˆž.

5.2. Outer measure in Euclidean space

87

and the ๐‘† ๐‘– are almost disjoint, that is, ๐‘† ๐‘– โˆฉ ๐‘† ๐‘˜ is either empty or just part of their boundary. Therefore ๐‘

vol(๐‘…) = โˆ‘ vol(๐‘† ๐‘– ) โ‰ค โˆ‘ vol(๐‘…๐‘— ) < |๐‘…|โˆ— + ๐œ€ ๐‘–

๐‘—=1

(Exercise (4)) and, since ๐œ€ > 0 is arbitrary, vol(๐‘…) โ‰ค |๐‘…|โˆ— . 5.3. If ๐‘… is any rectangle, then we also have that |๐‘…|โˆ— = vol(๐‘…). Indeed, |๐‘…|โˆ— โ‰ค vol(๐‘…) because again ๐‘… forms a cover for itself. Also, for any closed rectangle ๐‘† โŠ‚ ๐‘…, |๐‘†|โˆ— โ‰ค |๐‘…|โˆ— , because any cover of rectangles for ๐‘… also a cover for ๐‘†. Now, |๐‘†|โˆ— = vol(๐‘†) and, for any ๐œ€ > 0, we can choose ๐‘† such that vol(๐‘†) > vol(๐‘…) โˆ’ ๐œ€. Thus vol(๐‘…) โˆ’ ๐œ€ < |๐‘…|โˆ— , so we again have vol(๐‘…) โ‰ค |๐‘…|โˆ— because we can make ๐œ€ arbitrarily small. 5.4. The fact for rectangles used above is true for general subsets of โ„๐‘‘ : if ๐ด โŠ‚ ๐ต, then |๐ด|โˆ— โ‰ค |๐ต|โˆ— , as any cover of rectangles for ๐ต is also a cover for ๐ด. This property is called monotonicity. 5.5. If ๐ด = โ‹ƒ๐‘—โ‰ฅ1 ๐ด๐‘— , then |๐ด|โˆ— โ‰ค โˆ‘ |๐ด๐‘— |โˆ— . ๐‘—โ‰ฅ1 ๐‘—

Indeed, let ๐œ€ > 0 be given and, for each ๐ด๐‘— , let {๐‘…๐‘– } be a cover of rectangles for ๐ด๐‘— such that ๐œ€ ๐‘— โˆ‘ vol(๐‘…๐‘– ) < |๐ด๐‘— |โˆ— + ๐‘— . 2 ๐‘–โ‰ฅ1 ๐‘—

Thus โ‹ƒ๐‘–,๐‘—โ‰ฅ1 {๐‘…๐‘– } is a cover of rectangles for ๐ด, and ๐‘—

|๐ด|โˆ— โ‰ค โˆ‘ vol(๐‘…๐‘– ) < โˆ‘ (|๐ด๐‘— |โˆ— + ๐‘–,๐‘—โ‰ฅ1

๐‘—โ‰ฅ1

๐œ€ ) = โˆ‘ |๐ด๐‘— |โˆ— + ๐œ€. 2๐‘— ๐‘—โ‰ฅ1

We obtain the result because ๐œ€ > 0 is arbitrary. Property 5.5 is called countable subadditivity. A natural question to ask is whether we have equality in 5.5 when the ๐ด๐‘— are disjoint. In general we donโ€™t,2 but we have the following special cases. 2

This is beyond the goal of this text, but you can find an example in [Fol99, Section 1.1].

88

5. Measure theory in Euclidean space

5.6. If ๐ด, ๐ต โŠ‚ โ„๐‘‘ and dist(๐ด, ๐ต) > 0, then |๐ด โˆช ๐ต|โˆ— = |๐ด|โˆ— + |๐ต|โˆ— . Note that ๐ด and ๐ต are not only disjoint, but separated by a positive distance from each other, that is, there is some ๐›ฟ > 0 such that, for any ๐‘ฅ โˆˆ ๐ด, ๐‘ฆ โˆˆ ๐ต, then |๐‘ฅ โˆ’ ๐‘ฆ| > ๐›ฟ. By 5.5 we have |๐ด โˆช ๐ต|โˆ— โ‰ค |๐ด|โˆ— + |๐ต|โˆ— , so we have to prove the reverse inequality. Let ๐œ€ > 0 be given and {๐‘…๐‘— } be a cover of rectangles for ๐ด โˆช ๐ต such that โˆ‘ vol(๐‘…๐‘— ) < |๐ด โˆช ๐ต|โˆ— + ๐œ€ ๐‘—

and, by subdividing them if necessary, the diameter of each ๐‘…๐‘— is smaller than ๐›ฟ. Thus, if ๐‘…๐‘— โˆฉ ๐ด โ‰  โˆ…, then ๐‘…๐‘— โˆฉ ๐ต = โˆ…, and vice versa. Hence, if ๐”๐ด = {๐‘— โˆถ ๐‘…๐‘— โˆฉ ๐ด โ‰  โˆ…} and ๐”๐ต = {๐‘— โˆถ ๐‘…๐‘— โˆฉ ๐ต โ‰  โˆ…}, then ๐”๐ด โˆฉ ๐”๐ต = โˆ… and each of {๐‘…๐‘— }๐‘—โˆˆ๐”๐ด and {๐‘…๐‘— }๐‘—โˆˆ๐”๐ต is a cover for ๐ด and ๐ต, respectively. Thus |๐ด|โˆ— + |๐ต|โˆ— โ‰ค โˆ‘ vol(๐‘…๐‘— ) + โˆ‘ vol(๐‘…๐‘— ) โ‰ค โˆ‘ vol(๐‘…๐‘— ) < |๐ด โˆช ๐ต|โˆ— + ๐œ€, ๐‘—โˆˆ๐”๐ด

๐‘—โˆˆ๐”๐ต

๐‘—

and the result follows because ๐œ€ > 0 is arbitrary. Note that, inductively, we can extend 5.6 for any finite number of sets at positive distance. 5.7. If ๐ด = โ‹ƒ๐‘— ๐‘„๐‘— where the ๐‘„๐‘— are almost disjoint cubes, then |๐ด|โˆ— = โˆ‘ |๐‘„|โˆ— = โˆ‘ vol(๐‘„). ๐‘—

๐‘—

To verify this, let ๐œ€ > 0 be given and consider strictly thinner cubes ๐‘„ฬƒ๐‘— โŠ‚ ๐‘„๐‘— such that ๐œ€ vol(๐‘„ฬƒ๐‘— ) > vol(๐‘„๐‘— ) โˆ’ ๐‘— . 2 With strictly thinner we mean that the closure of each ๐‘„ฬƒ๐‘— is contained in the interior of ๐‘„๐‘— . Thus, for any ๐‘– โ‰  ๐‘—, dist(๐‘„ฬƒ ๐‘– , ๐‘„ฬƒ๐‘— ) > 0 and, by 5.6, for each ๐‘ we have ๐‘

๐‘

๐‘

๐‘

๐œ€ | | | โ‹ƒ ๐‘„ฬƒ๐‘— |โˆ— = โˆ‘ |๐‘„ฬƒ๐‘— |โˆ— > โˆ‘ (|๐‘„๐‘— |โˆ— โˆ’ 2๐‘— ) > โˆ‘ |๐‘„๐‘— |โˆ— โˆ’ ๐œ€. ๐‘—=1 ๐‘—=1 ๐‘—=1 ๐‘—=1

5.3. Measurable sets and measure

89

๐‘

Since ๐ด โŠƒ โ‹ƒ๐‘—=1 ๐‘„ฬƒ๐‘— , we have, by 5.4, ๐‘

|๐ด|โˆ— > โˆ‘ |๐‘„๐‘— |โˆ— โˆ’ ๐œ€. ๐‘—=1

As ๐‘ is arbitrary, we have |๐ด|โˆ— โ‰ฅ โˆ‘ |๐‘„๐‘— |โˆ— โˆ’ ๐œ€, ๐‘—

and we obtain the result because ๐œ€ > 0 is also arbitrary. Fact 5.7 is particularly useful when ๐ด is an open set, since every nonempty open set in โ„๐‘‘ is an almost disjoint union of closed cubes which can even be chosen to be dyadic (Exercise (5)). As we have mentioned above, we donโ€™t have equality in 5.5, even if we have a finite number of sets. Thus, we need to restrict ourselves to those sets on which the equality would be true.

5.3. Measurable sets and measure We say that ๐ด โŠ‚ โ„๐‘‘ is measurable if, for each ๐ต โŠ‚ โ„๐‘‘ , |๐ต|โˆ— = |๐ต โˆฉ ๐ด|โˆ— + |๐ต โงต ๐ด|โˆ— .

(5.8)

Note that ๐ต โˆฉ ๐ด and ๐ต โงต ๐ด are disjoint and ๐ต = (๐ต โˆฉ ๐ด) โˆช (๐ต โงต ๐ด), so ๐ด is measurable when it splits any other set ๐ต in parts whose outer measures add to the outer measure of ๐ต. Recall that we always have, by 5.5, the inequality |๐ต|โˆ— โ‰ค |๐ต โˆฉ ๐ด|โˆ— + |๐ต โงต ๐ด|โˆ— , so ๐ด is measurable if the reverse inequality |๐ต|โˆ— โ‰ฅ |๐ต โˆฉ ๐ด|โˆ— + |๐ต โงต ๐ด|โˆ— is true for all ๐ต โŠ‚ โ„๐‘‘ . 5.9. โˆ… and โ„๐‘‘ are measurable. This follows because ๐ตโˆฉโˆ…=โˆ…

and

๐ต โงต โˆ… = ๐ต,

๐ต โˆฉ โ„๐‘‘ = ๐ต

and

๐ต โงต โ„๐‘‘ = โˆ….

as well as 5.10. If ๐ด is measurable, then its complement โ„๐‘‘ โงต๐ด is measurable. This also follows immediately because ๐ต โˆฉ (โ„๐‘‘ โงต ๐ด) = ๐ต โงต ๐ด

and

๐ต โงต (โ„๐‘‘ โงต ๐ด) = ๐ต โˆฉ ๐ด.

90

5. Measure theory in Euclidean space

5.11. If ๐ด and ๐ต are measurable, then ๐ด โˆช ๐ต is measurable. To prove this note that, for any ๐ถ โŠ‚ โ„๐‘‘ , |๐ถ|โˆ— = |๐ถ โˆฉ ๐ด|โˆ— + |๐ถ โงต ๐ด|โˆ— = |๐ถ โˆฉ ๐ด โˆฉ ๐ต|โˆ— + |(๐ถ โˆฉ ๐ด) โงต ๐ต|โˆ— + |(๐ถ โงต ๐ด) โˆฉ ๐ต|โˆ— + |(๐ถ โงต ๐ด) โงต ๐ต|โˆ— , using (5.8) for each of ๐ด and ๐ต, which are both measurable. Now ๐ด โˆช ๐ต = (๐ด โˆฉ ๐ต) โˆช (๐ด โงต ๐ต) โˆช (๐ต โงต ๐ด), ๐ถ โˆฉ (๐ด โงต ๐ต) = (๐ถ โˆฉ ๐ด) โงต ๐ต,

and

๐ถ โˆฉ (๐ต โงต ๐ด) = (๐ถ โงต ๐ด) โˆฉ ๐ต,

so we have ๐ถ โˆฉ (๐ด โˆช ๐ต) = (๐ถ โˆฉ ๐ด โˆฉ ๐ต) โˆช (๐ถ โˆฉ (๐ด โงต ๐ต)) โˆช (๐ถ โˆฉ (๐ต โงต ๐ด)) = (๐ถ โˆฉ ๐ด โˆฉ ๐ต) โˆช ((๐ถ โˆฉ ๐ด) โงต ๐ต) โˆช ((๐ถ โงต ๐ด) โˆฉ ๐ต), and thus |๐ถ โˆฉ ๐ด โˆฉ ๐ต|โˆ— + |(๐ถ โˆฉ ๐ด) โงต ๐ต|โˆ— + |(๐ถ โงต ๐ด) โˆฉ ๐ต|โˆ— โ‰ฅ |๐ถ โˆฉ (๐ด โˆช ๐ต)|โˆ— . Also (๐ถ โงต ๐ด) โงต ๐ต = ๐ถ โงต (๐ด โˆช ๐ต), and hence |๐ถ|โˆ— โ‰ฅ |๐ถ โˆฉ (๐ด โˆช ๐ต)|โˆ— + |๐ถ โงต (๐ด โˆช ๐ต)|โˆ— , so ๐ด โˆช ๐ต is measurable. You can visualize the decomposition of ๐ถ used in the proof above with the Venn diagram shown in Figure 5.2. Inductively, for any ๐‘, the union of ๐‘ measurable sets ๐ด1 , . . . , ๐ด๐‘ is measurable. 5.12. If ๐ด1 , ๐ด2 , . . . are measurable, then their union ๐ด = โ‹ƒ๐‘— ๐ด๐‘— is also measurable. To prove this, let ๐ต1 = ๐ด1 and, for each ๐‘— โ‰ฅ 2, define ๐ต๐‘— = ๐‘—โˆ’1

๐‘

๐‘

๐ด๐‘— โงต โ‹ƒ๐‘–=1 ๐ด๐‘– . Thus the ๐ต๐‘— are disjoint, โ‹ƒ๐‘—=1 ๐ต๐‘— = โ‹ƒ๐‘—=1 ๐ด๐‘— for all ๐‘ and ๐‘

โ‹ƒ๐‘— ๐ต๐‘— = ๐ด. Also, each โ‹ƒ๐‘—=1 ๐ต๐‘— is measurable. Now, for ๐ถ โŠ‚ โ„๐‘‘ and any ๐‘, ๐‘

|๐ถ โˆฉ

โ‹ƒ ๐‘—=1

๐‘

๐ต๐‘— |โˆ— = |(๐ถ โˆฉ

โ‹ƒ

๐‘

๐ต๐‘— ) โˆฉ ๐ต๐‘ |โˆ— + |(๐ถ โˆฉ

๐‘—=1

๐‘—=1

๐‘โˆ’1

= |๐ถ โˆฉ ๐ต๐‘ |โˆ— + |๐ถ โˆฉ

โ‹ƒ

โ‹ƒ ๐‘—=1

๐ต๐‘— |โˆ— ,

๐ต๐‘— ) โงต ๐ต๐‘ |โˆ—

5.3. Measurable sets and measure

91

Figure 5.2. Venn diagram of the decomposition of ๐ถ โˆฉ (๐ด โˆช ๐ต) as (๐ถ โˆฉ ๐ด โˆฉ ๐ต) โˆช ((๐ถ โˆฉ ๐ด) โงต ๐ต) โˆช ((๐ถ โงต ๐ด) โˆฉ ๐ต).

and inductively ๐‘

|๐ถ โˆฉ

โ‹ƒ

๐‘

๐ต๐‘— |โˆ— = โˆ‘ |๐ถ โˆฉ ๐ต๐‘— |โˆ— . ๐‘—=1

๐‘—=1

Thus ๐‘

๐‘

|๐ถ|โˆ— = |๐ถ โˆฉ

โ‹ƒ ๐‘—=1

๐ต๐‘— |โˆ— + |๐ถ โงต

โ‹ƒ

๐‘

๐ต๐‘— |โˆ— = โˆ‘ |๐ถ โˆฉ ๐ต๐‘— |โˆ— + |๐ถ โงต

๐‘—=1

๐‘—=1

๐‘

โ‹ƒ

๐ต๐‘— |โˆ—

๐‘—=1

๐‘

โ‰ฅ โˆ‘ |๐ถ โˆฉ ๐ต๐‘— |โˆ— + |๐ถ โงต ๐ด|โˆ— , ๐‘—=1 ๐‘

because โ‹ƒ๐‘—=1 ๐ต๐‘— โŠ‚ ๐ด. As ๐‘ is arbitrary, we obtain |๐ถ|โˆ— โ‰ฅ โˆ‘ |๐ถ โˆฉ ๐ต๐‘— |โˆ— + |๐ถ โงต ๐ด|โˆ— โ‰ฅ || (๐ถ โˆฉ ๐ต๐‘— )|| + |๐ถ โงต ๐ด|โˆ— โ‹ƒ โˆ— ๐‘—โ‰ฅ1

= |๐ถ โˆฉ ๐ด| + |๐ถ โงต ๐ด|โˆ— , and we conclude ๐ด is measurable.

๐‘—โ‰ฅ1

92

5. Measure theory in Euclidean space

Note that we can conclude, by 5.10 and 5.12, that the intersection of a countable number of measurable sets is measurable, as we can write โ‹‚

๐ด๐‘— = โ„๐‘‘ โงต

๐‘—

โ‹ƒ

(โ„๐‘‘ โงต ๐ด๐‘— ).

๐‘—

5.13. A closed cube ๐‘„ is measurable. Let ๐ด โŠ‚ โ„๐‘‘ , and we want to prove |๐ด|โˆ— โ‰ฅ |๐ด โˆฉ ๐‘„|โˆ— + |๐ด โงต ๐‘„|โˆ— . This inequality is obvious if |๐ด|โˆ— = โˆž, so we assume |๐ด|โˆ— < โˆž. For each ๐‘—, define ๐‘„๐‘— = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ dist(๐‘ฅ, ๐‘„)
0 be given. (1) There exists an open set ๐‘ˆ โŠƒ ๐ด such that |๐‘ˆ โงต ๐ด| < ๐œ€. (2) There exists a closed set ๐ธ โŠ‚ ๐ด such that |๐ด โงต ๐ธ| < ๐œ€. (3) If |๐ด| < โˆž, there exists a compact set ๐พ โŠ‚ ๐ด such that |๐ดโงต๐พ| < ๐œ€. (4) If |๐ด| < โˆž, there exist finitely many closed cubes ๐‘„1 , ๐‘„2 , . . ., ๐‘„๐‘ ๐‘ such that, if ๐น = โ‹ƒ๐‘—=1 ๐‘„๐‘— , then |๐ดโ–ณ๐น| < ๐œ€. The symbol โ–ณ above denotes the symmetric difference of the sets, ๐ดโ–ณ๐น = (๐ด โงต ๐น) โˆช (๐น โงต ๐ด). 5.18. Moreover, statement (1) of Corollary 5.17 is equivalent to the statement that ๐ด is measurable. Indeed, suppose ๐ด โŠ‚ โ„๐‘‘ such that, for any ๐œ€ > 0, there exists an open ๐‘ˆ โŠƒ ๐ด such that |๐‘ˆ โงต ๐ด|โˆ— < ๐œ€.

96

5. Measure theory in Euclidean space

Let ๐ต โŠ‚ โ„๐‘‘ . Given ๐œ€ > 0, choose ๐‘ˆ โŠƒ ๐ด such that |๐‘ˆ โงต ๐ด|โˆ— < ๐œ€. Now ๐ต โˆฉ ๐ด โŠ‚ ๐ต โˆฉ ๐‘ˆ and ๐ต โงต ๐ด โŠ‚ (๐ต โงต ๐‘ˆ) โˆช (๐‘ˆ โงต ๐ด), so |๐ต โˆฉ ๐ด|โˆ— + |๐ต โงต ๐ด|โˆ— โ‰ค |๐ต โˆฉ ๐‘ˆ|โˆ— + |๐ต โงต ๐‘ˆ|โˆ— + |๐‘ˆ โงต ๐ด|โˆ— < |๐ต|โˆ— + ๐œ€, because ๐‘ˆ is measurable. Since ๐œ€ > 0 is arbitrary, |๐ต โˆฉ ๐ด|โˆ— + |๐ต โงต ๐ด|โˆ— โ‰ค |๐ต|โˆ— , so ๐ด is measurable.3 5.19. Lebesgue measure is translation and dilation invariant: For ๐ด โŠ‚ โ„๐‘‘ , ๐‘ฅ0 โˆˆ โ„๐‘‘ and ๐›ฟ > 0, we define ๐‘ฅ0 + ๐ด = {๐‘ฅ0 + ๐‘ฅ โˆถ ๐‘ฅ โˆˆ ๐ด}

and

๐›ฟ๐ด = {๐›ฟ๐‘ฅ โˆถ ๐‘ฅ โˆˆ ๐ด}.

These sets are measurable, and we have |๐‘ฅ0 + ๐ด| = |๐ด| and |๐›ฟ๐ด| = ๐›ฟ๐‘‘ |๐ด| (Exercises (12) and (13)).

5.4. Measurable functions In this section we discuss the functions that are to be integrated with respect to the Lebesgue measure discussed in Section 5.3. Such functions are called measurable. For convenience, we will allow functions to have infinite values at some points, so we consider the extended real line [โˆ’โˆž, โˆž] with the conventions ๐‘ฅ+โˆž=โˆž+๐‘ฅ=โˆž

for all ๐‘ฅ โˆˆ โ„,

๐‘ฅ โˆ’ โˆž = โˆ’โˆž + ๐‘ฅ = โˆ’โˆž

for all ๐‘ฅ โˆˆ โ„,

๐‘ฅโ‹…โˆž=โˆžโ‹…๐‘ฅ=โˆž

for all ๐‘ฅ โˆˆ โ„,

๐‘ฅ > 0,

๐‘ฅ โ‹… โˆž = โˆž โ‹… ๐‘ฅ = โˆ’โˆž

for all ๐‘ฅ โˆˆ โ„,

๐‘ฅ < 0,

and the corresponding products with โˆ’โˆž. We also agree that โˆž+โˆž=โˆž

and

โˆ’ โˆž โˆ’ โˆž = โˆ’โˆž,

or ยฑโˆž โ‹… โˆž = ยฑโˆž, but we do not define โˆž โˆ’ โˆž nor 0 โ‹… โˆž. We also say that โˆ’โˆž < ๐‘ฅ and ๐‘ฅ < โˆž for all ๐‘ฅ โˆˆ โ„. Let ๐‘“ โˆถ โ„๐‘‘ โ†’ [โˆ’โˆž, โˆž] be an extended real valued function. We say that ๐‘“ is measurable if, for all ๐‘Ž โˆˆ โ„, the set ๐‘“โˆ’1 ([โˆ’โˆž, ๐‘Ž)) = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘“(๐‘ฅ) < ๐‘Ž} 3

Statement (1) is indeed used as the definition of a measurable set in some texts, as [SS05].

5.4. Measurable functions

97

is measurable. The definition of a measurable function is equivalent to saying that ๐‘“โˆ’1 ([โˆ’โˆž, ๐‘Ž]) is measurable for all ๐‘Ž โˆˆ โ„. Indeed, since 1 1 [โˆ’โˆž, ๐‘Ž] = [ โˆ’ โˆž, ๐‘Ž + ) and [โˆ’โˆž, ๐‘Ž) = [ โˆ’ โˆž, ๐‘Ž โˆ’ ], โ‹ƒ โ‹‚ ๐‘— ๐‘— ๐‘—

๐‘—

โˆ’1

we see that the sets ๐‘“ ([โˆ’โˆž, ๐‘Ž)) are measurable for all ๐‘Ž โˆˆ โ„ if and only if the sets ๐‘“โˆ’1 ([โˆ’โˆž, ๐‘Ž]) are measurable for all ๐‘Ž โˆˆ โ„. We could also have used the sets ๐‘“โˆ’1 ((๐‘Ž, โˆž]) or ๐‘“โˆ’1 ((๐‘Ž, โˆž]), or even the pre-images of bounded intervals ๐‘“โˆ’1 ((๐‘Ž, ๐‘]), ๐‘“โˆ’1 ((๐‘Ž, ๐‘)), etc. together with ๐‘“โˆ’1 ({โˆž}) and ๐‘“โˆ’1 ({โˆ’โˆž}), to define measurability. We usually denote the set ๐‘“โˆ’1 ([โˆ’โˆž, ๐‘Ž)) by {๐‘“ < ๐‘Ž}, and the set ๐‘“ ([โˆ’โˆž, ๐‘Ž]) by {๐‘“ โ‰ค ๐‘Ž}. Similarly, ๐‘“โˆ’1 ((๐‘Ž, โˆž]) by {๐‘“ > ๐‘Ž}, ๐‘“โˆ’1 ([๐‘Ž, โˆž]) by {๐‘“ โ‰ฅ ๐‘Ž}, etc. โˆ’1

The fact that ๐‘“ is measurable is also equivalent to the fact that ๐‘“โˆ’1 (๐‘ˆ) is measurable for any open ๐‘ˆ โŠ‚ โ„, or ๐‘“โˆ’1 (๐ธ) is measurable for any closed ๐ธ โŠ‚ โ„ (Exercise (14)). The previous equivalence implies that all continuous functions ๐‘“ are measurable. Indeed, since every open set is measurable and, for any open ๐‘ˆ โŠ‚ โ„, ๐‘“โˆ’1 (๐‘ˆ) is open, then ๐‘“ is measurable, as stated above. The most important property of measurability of functions is its stability under pointwise limits. This is a consequence of the following result. Theorem 5.20. Let ๐‘“1 , ๐‘“2 , . . . be measurable extended real valued functions. Then the functions sup ๐‘“๐‘› ,

inf ๐‘“๐‘› ,

lim sup ๐‘“๐‘› ,

and

lim inf ๐‘“๐‘›

are measurable. In the theorem, sup ๐‘“๐‘› denotes the function sup ๐‘“๐‘› (๐‘ฅ) = sup{๐‘“๐‘› (๐‘ฅ) โˆถ ๐‘› โ‰ฅ 1}, which we define as โˆž in the case when the set {๐‘“๐‘› (๐‘ฅ) โˆถ ๐‘› โ‰ฅ 1} is not bounded. Similarly for inf ๐‘“๐‘› . lim sup ๐‘“๐‘› and lim inf ๐‘“๐‘› are defined by lim sup ๐‘“๐‘› (๐‘ฅ) = inf sup ๐‘“๐‘˜ (๐‘ฅ) = inf { sup{๐‘“๐‘˜ (๐‘ฅ) โˆถ ๐‘˜ โ‰ฅ ๐‘›} โˆถ ๐‘› โ‰ฅ 1}, ๐‘›โ‰ฅ1 ๐‘˜โ‰ฅ๐‘›

lim inf ๐‘“๐‘› (๐‘ฅ) = sup inf ๐‘“๐‘˜ (๐‘ฅ) = sup { inf{๐‘“๐‘˜ (๐‘ฅ) โˆถ ๐‘˜ โ‰ฅ ๐‘›} โˆถ ๐‘› โ‰ฅ 1}. ๐‘›โ‰ฅ1 ๐‘˜โ‰ฅ๐‘›

98

5. Measure theory in Euclidean space

Proof. To see that sup ๐‘“๐‘› is measurable, observe that {sup ๐‘“๐‘› > ๐‘Ž} =

โ‹ƒ

{๐‘“๐‘› > ๐‘Ž}

๐‘›

and each set {๐‘“๐‘› > ๐‘Ž} is measurable. Similarly, inf ๐‘“๐‘› is measurable because {inf ๐‘“๐‘› < ๐‘Ž} = {๐‘“ < ๐‘Ž} โ‹ƒ ๐‘› ๐‘›

and also each {๐‘“๐‘› < ๐‘Ž} is measurable. Now, lim sup ๐‘“๐‘› is measurable because, for each ๐‘›, sup๐‘˜โ‰ฅ๐‘› ๐‘“๐‘˜ is measurable, and therefore inf๐‘›โ‰ฅ1 sup๐‘˜โ‰ฅ๐‘› ๐‘“๐‘˜ is measurable. Similarly for lim inf ๐‘“๐‘› . โ–ก Corollary 5.21. If ๐‘“1 , ๐‘“2 , . . . is a sequence of extended real valued measurable functions and ๐‘“๐‘› โ†’ ๐‘“ pointwise, then ๐‘“ is measurable. โ–ก

Proof. If ๐‘“๐‘› โ†’ ๐‘“, then lim sup ๐‘“๐‘› = lim inf ๐‘“๐‘› = ๐‘“. ๐‘‘

5.22. If ๐‘“ is measurable and ๐‘”(๐‘ฅ) = ๐‘“(๐‘ฅ) for all ๐‘ฅ โˆˆ โ„ except at a set of measure zero, then ๐‘” is also measurable. Indeed, for each ๐‘Ž โˆˆ โ„, whenever ๐‘”(๐‘ฅ) < ๐‘Ž we have either ๐‘“(๐‘ฅ) < ๐‘Ž or ๐‘“(๐‘ฅ) โ‰ฅ ๐‘Ž, so {๐‘” < ๐‘Ž} is contained in the union of {๐‘“ < ๐‘Ž} and {๐‘“ โ‰ฅ ๐‘Ž} โˆฉ {๐‘” < ๐‘Ž}. However, if ๐‘“(๐‘ฅ) < ๐‘Ž, it could happen that ๐‘”(๐‘ฅ) โ‰ฅ ๐‘Ž, so we need to remove the set {๐‘“ < ๐‘Ž} โˆฉ {๐‘” โ‰ฅ ๐‘Ž}. Thus, we have {๐‘” < ๐‘Ž} = {๐‘“ < ๐‘Ž} โˆช ({๐‘“ โ‰ฅ ๐‘Ž} โˆฉ {๐‘” < ๐‘Ž}) โงต ({๐‘“ < ๐‘Ž} โˆฉ {๐‘” โ‰ฅ ๐‘Ž}). As the two sets {๐‘“ โ‰ฅ ๐‘Ž} โˆฉ {๐‘” < ๐‘Ž} and {๐‘“ < ๐‘Ž} โˆฉ {๐‘” โ‰ฅ ๐‘Ž} are contained in the set where ๐‘“(๐‘ฅ) โ‰  ๐‘”(๐‘ฅ), they are sets of measure zero and thus measurable, so we conclude {๐‘” < ๐‘Ž} is measurable. If ๐‘”(๐‘ฅ) = ๐‘“(๐‘ฅ) for all ๐‘ฅ โˆˆ โ„๐‘‘ except at a set of measure zero, we say that ๐‘” = ๐‘“ almost everywhere, or at almost every point, and we denote it by a.e. In general, we say that a property ๐‘ƒ(๐‘ฅ) holds a.e. if the set where ๐‘ƒ(๐‘ฅ) is false is of measure zero. We can then refine Corollary 5.21 as the following statement. Corollary 5.23. If ๐‘“1 , ๐‘“2 , . . . is a sequence of extended real valued measurable functions and ๐‘“๐‘› โ†’ ๐‘“ a.e., then ๐‘“ is measurable. 5.24. If ๐‘“ is measurable and ๐‘˜ โˆˆ โ„ค+ , then (๐‘“)๐‘˜ is measurable. This follows because, if ๐‘˜ is odd, {(๐‘“)๐‘˜ < ๐‘Ž} = {๐‘“ < ๐‘Ž1/๐‘˜ }

5.4. Measurable functions

99

and, if ๐‘˜ is even, {(๐‘“)๐‘˜ < ๐‘Ž} = {โˆ’๐‘Ž1/๐‘˜ < ๐‘“ < ๐‘Ž1/๐‘˜ } if ๐‘Ž > 0, or empty if ๐‘Ž โ‰ค 0, and both of the sets on right side are measurable because ๐‘“ is measurable. 5.25. If ๐‘“ and ๐‘” are measurable and real valued, then ๐‘“ + ๐‘” and ๐‘“๐‘” are measurable. For the sum, observe that (5.26)

{๐‘“ + ๐‘” < ๐‘Ž} =

โ‹ƒ

({๐‘“ < ๐‘Ž โˆ’ ๐‘Ÿ} โˆฉ {๐‘” < ๐‘Ÿ})

๐‘Ÿโˆˆโ„š

and that each set {๐‘“ < ๐‘Ž โˆ’ ๐‘Ÿ} โˆฉ {๐‘” < ๐‘Ÿ} is measurable. To prove the identity (5.26), first note that, if ๐‘“(๐‘ฅ) < ๐‘Ž โˆ’ ๐‘Ÿ and ๐‘”(๐‘ฅ) < ๐‘Ÿ, we clearly have ๐‘“(๐‘ฅ) + ๐‘”(๐‘ฅ) < ๐‘Ž, so the union of the right side of (5.26) is contained in the left side. For the reverse inclusion, assume ๐‘“(๐‘ฅ)+๐‘”(๐‘ฅ) < ๐‘Ž, so ๐‘”(๐‘ฅ) < ๐‘Žโˆ’๐‘“(๐‘ฅ). Let ๐‘Ÿ โˆˆ โ„š such that ๐‘”(๐‘ฅ) < ๐‘Ÿ < ๐‘Žโˆ’๐‘“(๐‘ฅ). Hence ๐‘”(๐‘ฅ) < ๐‘Ÿ and ๐‘“(๐‘ฅ) < ๐‘Žโˆ’๐‘Ÿ, so ๐‘ฅ โˆˆ {๐‘“ < ๐‘Ž โˆ’ ๐‘Ÿ} โˆฉ {๐‘” < ๐‘Ÿ}, which implies (5.26). Now, the multiplication ๐‘“๐‘” is a measurable function because we can write 1 ๐‘“๐‘” = ((๐‘“ + ๐‘”)2 โˆ’ (๐‘“ โˆ’ ๐‘”)2 ), 4 and each of the functions (๐‘“ ยฑ ๐‘”)2 is measurable by the previous results. 5.27. If ๐ด is measurable set, its characteristic function 1 ๐œ’๐ด (๐‘ฅ) = { 0

๐‘ฅโˆˆ๐ด ๐‘ฅ โˆ‰ ๐ด,

is a measurable function. Indeed, for each ๐‘Ž โˆˆ โ„, ๐‘‘

โŽงโ„ {๐œ’๐ด < ๐‘Ž} = โ„๐‘‘ โงต ๐ด โŽจ โŽฉโˆ…

๐‘Ž>1 0 0}.

๐‘˜=0

We have thus partitioned the set where ๐‘“ is positive in the 22๐‘› + 1 sets 2๐‘› ๐ด0๐‘› , ๐ด1๐‘› , . . . , ๐ด2๐‘› โˆ’1 and ๐ต๐‘› . Now define the simple function 22๐‘› โˆ’1

๐œ™๐‘› = โˆ‘ ๐‘˜=0

๐‘˜ ๐œ’ ๐‘˜ + 2๐‘› ๐œ’๐ต๐‘› . 2๐‘› ๐ด๐‘›

Note that 0 โ‰ค ๐œ™๐‘› โ‰ค 2๐‘› , and ๐œ™๐‘› splits the values of ๐‘“, up to 2๐‘› , in small jumps of size 2โˆ’๐‘› (see Figure 5.4). The ๐œ™๐‘› clearly satisfy ๐œ™๐‘› โ‰ค ๐œ™๐‘›+1 and

2n

Figure 5.4. The approximation of a function with simple functions.

๐œ™๐‘› โ†— ๐‘“. Moreover, given ๐‘€ > 0, if ๐ด = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘“(๐‘ฅ) โ‰ค ๐‘€}, then, for ๐‘› such that 2๐‘› > ๐‘€, |๐‘“(๐‘ฅ) โˆ’ ๐œ™๐‘› (๐‘ฅ)| < 2โˆ’๐‘› โ–ก

for all ๐‘ฅ โˆˆ ๐ด. Therefore ๐œ™๐‘› โ‡‰ ๐‘“ on ๐ด. f (x)

f + (x)

f - (x)

Figure 5.5. The positive and negative parts of a function ๐‘“(๐‘ฅ).

5.30. We can apply Theorem 5.29 to approximate a general measurable function ๐‘“ with with simple functions ๐œ™๐‘› such that |๐œ™๐‘› | โ‰ค |๐œ™๐‘›+1 |, |๐œ™๐‘› | โ‰ค |๐‘“| and ๐œ™๐‘› โ†’ ๐‘“, and the convergence to be uniform on any set where ๐‘“

102

5. Measure theory in Euclidean space

is bounded. We just need to write ๐‘“ = ๐‘“+ โˆ’ ๐‘“โˆ’ , where ๐‘“+ and ๐‘“โˆ’ are the positive and negative parts of ๐‘“, respectively, given by ๐‘“+ = ๐‘“ โ‹… ๐œ’{๐‘“โ‰ฅ0}

๐‘“โˆ’ = โˆ’๐‘“ โ‹… ๐œ’{๐‘“โ‰ค0} .

and

(See Figure 5.5.) We leave the details as an exercise (Exercise (16)). Note that we also have |๐‘“| = ๐‘“+ + ๐‘“โˆ’ .

Exercises (1) Let ๐‘… be a rectangle and ๐œ€ > 0. Then there exist an open rectangle ๐‘† โŠƒ ๐‘… and a closed rectangle ๐‘‡ โŠ‚ ๐‘… such that vol(๐‘†) < vol(๐‘…) + ๐œ€

and

vol(๐‘‡) > vol(๐‘…) โˆ’ ๐œ€.

(2) (a) If ๐ด is countable, then |๐ด|โˆ— = 0. (b) If ๐ด1 , ๐ด2 , . . . are sets of measure zero, then their union has measure zero. (3) Let ๐‘‹ be the Cantor ternary set ๐ถ, constructed by the removal of middle third intervals starting from ๐ถ0 = [0, 1]. Then ๐ถ is of measure zero. (4) Let ๐‘… = [๐‘Ž1 , ๐‘1 ] ร— [๐‘Ž2 , ๐‘2 ] ร— โ‹ฏ ร— [๐‘Ž๐‘‘ , ๐‘๐‘‘ ] be a rectangle and, for each ๐‘– = 1, 2, . . . , ๐‘‘, ๐‘—

๐‘—

๐‘—

๐‘—

๐’ซ๐‘— = {๐‘ฅ0 = ๐‘Ž๐‘— < ๐‘ฅ1 < ๐‘ฅ2 < . . . < ๐‘ฅ๐‘๐‘— = ๐‘๐‘— } a partition of [๐‘Ž๐‘— , ๐‘๐‘— ]. If ๐’ฎ is the grid of subrectangles ๐‘… of the form ๐‘—

๐‘—

๐‘† = ๐ผ1 ร— ๐ผ2 ร— โ‹ฏ ร— ๐ผ๐‘‘ , where each ๐ผ๐‘— = [๐‘ฅ๐‘–โˆ’1 , ๐‘ฅ๐‘– ], then vol(๐‘…) = โˆ‘ vol(๐‘†). ๐‘†โˆˆ๐’ฎ

(5) If ๐‘ˆ โŠ‚ โ„๐‘‘ is open and nonempty, then ๐‘ˆ = โ‹ƒ๐‘— ๐‘„๐‘— where ๐‘„๐‘— are dyadic almost disjoint closed cubes. For the proof, follow the next steps: (a) For ๐‘š โ‰ฅ 0, let ๐””๐‘š be the collection of closed dyadic cubes ๐‘„ with sides of length 2โˆ’๐‘š such that ๐‘„ โŠ‚ ๐‘ˆ. (b) Show that, for ๐‘š โ‰ค ๐‘›, ๐‘„ โˆˆ ๐””๐‘š and ๐‘„โ€ฒ โˆˆ ๐””๐‘› , then either ๐‘„ โŠƒ ๐‘„โ€ฒ or they are almost disjoint.

Notes

103

(c) Define ๐”0 = ๐””0 and, for each ๐‘š > 0, let ๐”๐‘š be collection of ๐‘„ โˆˆ ๐””๐‘š such that ๐‘„ is almost disjoint to every ๐‘„โ€ฒ โˆˆ ๐””๐‘˜ , for ๐‘˜ = 0, . . . , ๐‘š โˆ’ 1. (d) Let ๐” = โˆช๐‘šโ‰ฅ0 ๐”๐‘š . (e) Prove that ๐‘ˆ= ๐‘„. โ‹ƒ ๐‘„โˆˆ๐”

๐‘‘

(6) For any ๐ด โŠ‚ โ„ , |๐ด|โˆ— = inf{|๐‘ˆ|โˆ— โˆถ ๐‘ˆ is open and ๐ด โŠ‚ ๐‘ˆ}. (7) If ๐ด1 , ๐ด2 , . . . are measurable, then โ‹‚๐‘— ๐ด๐‘— is measurable. (8) A set of measure zero is measurable. (9) Let ๐ด โŠ‚ โ„๐‘‘ . The following are equivalent. (a) ๐ด is measurable. (b) ๐ด = ๐‘ƒ โงต ๐‘€, where ๐‘ƒ is a ๐บ ๐›ฟ set and |๐‘€| = 0. (c) ๐ด = ๐‘„ โˆช ๐‘, where ๐‘„ is an ๐น๐œ set and |๐‘| = 0. (10) Part (2) of Corollary 5.16 is false if all ๐ด๐‘— have infinite measure. (11) Prove Corollary 5.17. (12) For ๐ด โŠ‚ โ„๐‘‘ measurable and ๐‘ฅ0 โˆˆ โ„๐‘‘ , |๐‘ฅ0 + ๐ด| = |๐ด|. (13) Let ๐ด โŠ‚ โ„๐‘‘ be a measurable set. (a) For ๐›ฟ > 0, |๐›ฟ๐ด| = ๐›ฟ๐‘‘ |๐ด|. (b) For a ๐‘‘-tuple ๐›ฟ ฬ„ = (๐›ฟ1 , . . . , ๐›ฟ ๐‘‘ ) with each ๐›ฟ๐‘— > 0, ๐‘— = 1, . . . , ๐‘‘, define ฬ„ = {(๐›ฟ1 ๐‘ฅ1 , . . . , ๐›ฟ ๐‘‘ ๐‘ฅ๐‘‘ ) โˆถ (๐‘ฅ1 , . . . , ๐‘ฅ๐‘‘ ) โˆˆ ๐ด}. ๐›ฟ๐ด ฬ„ = ๐›ฟ1 โ‹ฏ ๐›ฟ ๐‘‘ |๐ด|. Then |๐›ฟ๐ด| (14) Let ๐‘“ โˆถ โ„๐‘‘ โ†’ [โˆ’โˆž, โˆž] be an extended valued function. Then the following are equivalent. (a) ๐‘“ is measurable. (b) For any open ๐‘ˆ โŠ‚ โ„, ๐‘“โˆ’1 (๐‘ˆ) is measurable. (c) For any closed ๐ธ โŠ‚ โ„, ๐‘“โˆ’1 (๐ธ) is measurable. (15) If ๐‘“ โˆถ โ„ โ†’ [โˆ’โˆž, โˆž] is monotone, then it is measurable. (16) Write the details for Fact 5.30.

Notes Measure theory has a long history with a long list of motivations, as discussed in detail in the texts [Bre07] and [Bre08] by David M. Bressoud.

104

5. Measure theory in Euclidean space

Henri Lebesgueโ€™s contribution was to consider countable covers for sets, as in the definition of outer measure presented here, and the approximation of a function by decomposition of its image, as it was done in 5.29. Lebesgueโ€™s theory is discussed in his papers [Leb98], [Leb99a], and [Leb99b], as well as his text [Leb04]. More detailed introductions to measure theory can be found in [Fol99] and [SS05].

Chapter 6

Lebesgue integral and Lebesgue spaces

6.1. Integration of measurable functions We are now ready to define the Lebesgue integral of a measurable function, and we first consider nonnegative functions. By Theorem 5.29, we can approximate any nonnegative measurable function by simple functions. Thus, we start by defining the integral of a nonnegative simple function. 6.1.1. Nonnegative simple functions. Let ๐œ™ be a nonnegative simple function on โ„๐‘‘ and โˆ‘๐‘— ๐‘Ž๐‘— ๐œ’๐ด๐‘— its reduced form, so each ๐‘Ž๐‘— > 0 and ๐ด๐‘– โˆฉ ๐ด๐‘— = โˆ… if ๐‘– โ‰  ๐‘—. We define the integral of ๐œ™ by (6.1)

โˆซ ๐œ™ = โˆ‘ ๐‘Ž๐‘— |๐ด๐‘— |. ๐‘—

The integral of ๐œ™ is also written โˆซ ๐œ™(๐‘ฅ)๐‘‘๐‘ฅ if we need to explicitly make reference to its argument ๐‘ฅ. We require no special conditions on the sets ๐ด๐‘— that define the simple function ๐œ™, except to be measurable. Thus, the integral (6.1) may 105

106

6. Lebesgue integral and Lebesgue spaces

be infinite, for example, if some |๐ด๐‘— | = โˆž. Note that assuming ๐‘Ž๐‘— > 0 avoids the conflicting operations โˆž โˆ’ โˆž or 0 โ‹… โˆž in (6.1). 6.2. For any ๐‘ > 0, โˆซ ๐‘๐œ™ = ๐‘ โˆซ ๐œ™. This follows clearly from the fact that ๐‘ โˆ‘ ๐‘Ž๐‘— ๐œ’๐ด๐‘— = โˆ‘ ๐‘๐‘Ž๐‘— ๐œ’๐ด๐‘— , ๐‘—

๐‘—

if ๐‘ > 0. 6.3. For nonnegative simple functions ๐œ™ and ๐œ“, โˆซ(๐œ™ + ๐œ“) = โˆซ ๐œ™ + โˆซ ๐œ“. Indeed, if ๐œ™ = โˆ‘๐‘— ๐‘Ž๐‘— ๐œ’๐ด๐‘— and ๐œ“ = โˆ‘๐‘˜ ๐‘๐‘˜ ๐œ’๐ต๐‘˜ , then we can write the reduced form of ๐œ™ + ๐œ“ as ๐œ™ + ๐œ“ = โˆ‘ ๐‘ ๐‘– ๐œ’๐ถ๐‘– = โˆ‘(๐‘โ€ฒ๐‘– + ๐‘โ€ณ๐‘– )๐œ’๐ถ๐‘– , ๐‘–

๐‘–

where each ๐ด๐‘— =

๐ถ๐‘– ,

โ‹ƒ

๐ต๐‘˜ =

๐‘Ž ๐‘โ€ฒ๐‘– = { ๐‘— 0

if ๐ถ๐‘– โŠ‚ ๐ด๐‘— otherwise

โ‹ƒ

๐ถ๐‘– ,

๐‘๐‘˜

if ๐ถ๐‘– โŠ‚ ๐ต๐‘˜

0

otherwise.

๐ถ๐‘– โˆฉ๐ต๐‘˜ โ‰ โˆ…

๐ถ๐‘– โˆฉ๐ด๐‘— โ‰ โˆ…

and

๐‘โ€ณ๐‘– = {

Note that each ๐ถ๐‘– is either of the form ๐ด๐‘— โˆฉ ๐ต๐‘˜ for some ๐‘—, ๐‘˜, of the form ๐ด๐‘— โงต โ‹ƒ๐‘˜ ๐ต๐‘˜ for some ๐‘—, or ๐ต๐‘˜ โงต โ‹ƒ๐‘— ๐ด๐‘— for some ๐‘˜. Thus โˆซ(๐œ™ + ๐œ“) = โˆ‘ ๐‘ ๐‘– |๐ถ๐‘– | = โˆ‘(๐‘โ€ฒ๐‘– + ๐‘โ€ณ๐‘– )|๐ถ๐‘– | = โˆ‘ ๐‘โ€ฒ๐‘– |๐ถ๐‘– | + โˆ‘ ๐‘โ€ณ๐‘– |๐ถ๐‘– | ๐‘–

=โˆ‘

๐‘–;๐‘โ€ฒ๐‘– โ‰ 0

๐‘–

โˆ‘

๐‘โ€ฒ๐‘– |๐ถ๐‘– | + โˆ‘

๐‘— ๐ถ๐‘– โˆฉ๐ด๐‘— โ‰ โˆ…

๐‘โ€ณ๐‘– |๐ถ๐‘– |

โˆ‘

๐‘˜ ๐ถ๐‘– โˆฉ๐ต๐‘˜ โ‰ โˆ…

= โˆ‘ ๐‘Ž๐‘—

โˆ‘

๐‘—

๐ถ๐‘– โˆฉ๐ด๐‘— โ‰ โˆ…

|๐ถ๐‘– | + โˆ‘ ๐‘๐‘˜ ๐‘˜

โˆ‘

|๐ถ๐‘– |

๐ถ๐‘– โˆฉ๐ต๐‘˜ โ‰ โˆ…

= โˆ‘ ๐‘Ž๐‘— |๐ด๐‘— | + โˆ‘ ๐‘๐‘˜ |๐ต๐‘˜ | = โˆซ ๐œ™ + โˆซ ๐œ“, ๐‘—

๐‘˜

because the sets ๐ถ๐‘– are pairwise disjoint and measurable.

๐‘–;๐‘โ€ณ ๐‘– โ‰ 0

6.1. Integration of measurable functions

107

6.4. If ๐œ™ โ‰ฅ 0 is simple, ๐‘ฅ0 โˆˆ โ„๐‘‘ and ๐œ“ is the translation of ๐œ™ by ๐‘ฅ0 , ๐œ“(๐‘ฅ) = ๐œ™(๐‘ฅ โˆ’ ๐‘ฅ0 ), then โˆซ ๐œ“ = โˆซ ๐œ™. This follows from the fact that |๐‘ฅ0 + ๐ด| = |๐ด| for any measurable set ๐ด โŠ‚ โ„๐‘‘ (Exercise (12), Chapter 5), because this implies that โˆซ ๐œ’๐ด (๐‘ฅ โˆ’ ๐‘ฅ0 )๐‘‘๐‘ฅ = โˆซ ๐œ’๐‘ฅ0 +๐ด = |๐‘ฅ0 + ๐ด| = |๐ด| = โˆซ ๐œ’๐ด , and thus the result follows for any linear combination of ๐œ’๐ด , by 6.2 and 6.3. 6.5. If ๐œ™ โ‰ฅ 0 is simple, ๐›ฟ > 0 and ๐œ“ is the dilation of ๐œ™ by ๐›ฟ, ๐œ“(๐‘ฅ) = ๐›ฟโˆ’๐‘‘ ๐œ™(๐‘ฅ/๐›ฟ), then โˆซ ๐œ“ = โˆซ ๐œ™. This follows from the fact that |๐›ฟ๐ด| = ๐›ฟ๐‘‘ |๐ด| for any measurable set ๐ด โŠ‚ โ„๐‘‘ (Exercise (13), Chapter 5), because this implies that, using 6.2, โˆซ ๐›ฟโˆ’๐‘‘ ๐œ’๐ด (๐‘ฅ/๐›ฟ)๐‘‘๐‘ฅ = ๐›ฟโˆ’๐‘‘ โˆซ ๐œ’๐›ฟ๐ด = ๐›ฟโˆ’๐‘‘ |๐›ฟ๐ด| = ๐›ฟโˆ’๐‘‘ โ‹… ๐›ฟ๐‘‘ |๐ด| = โˆซ ๐œ’๐ด , and thus, as above, the result follows for any linear combination of ๐œ’๐ด . 6.6. If ๐œ™ โ‰ค ๐œ“, โˆซ ๐œ™ โ‰ค โˆซ ๐œ“. Using the same decomposition as in 6.3, we have โˆซ ๐œ™ = โˆ‘ ๐‘Ž๐‘— |๐ด๐‘— | = โˆ‘ ๐‘โ€ฒ๐‘– |๐ถ๐‘– | โ‰ค โˆ‘ ๐‘โ€ณ๐‘– |๐ถ๐‘– | = โˆ‘ ๐‘๐‘˜ |๐ต๐‘˜ | = โˆซ ๐œ“, ๐‘—

since each

๐‘โ€ฒ๐‘–

โ‰ค

๐‘–

๐‘–

๐‘˜

๐‘โ€ณ๐‘– .

If ๐ด โŠ‚ โ„๐‘‘ is measurable, we define the integral of ๐œ™ over ๐ด as โˆซ ๐œ™ = โˆซ ๐œ™๐œ’๐ด .

(6.7)

๐ด

Note that, if ๐œ™ = โˆ‘๐‘— ๐‘Ž๐‘— ๐œ’๐ด๐‘— , then โˆซ ๐œ™ = โˆ‘ ๐‘Ž๐‘— |๐ด๐‘— โˆฉ ๐ด|. ๐ด

๐‘—

108

6. Lebesgue integral and Lebesgue spaces

6.8. If ๐ด, ๐ต โŠ‚ โ„๐‘‘ are measurable and ๐ด โˆฉ ๐ต = โˆ…, then โˆซ

๐œ™ = โˆซ ๐œ™ + โˆซ ๐œ™.

๐ดโˆช๐ต

๐ด

๐ต

Follows from the fact that, since ๐ด โˆฉ ๐ต = โˆ…, ๐œ’๐ถโˆฉ(๐ดโˆช๐ต) = ๐œ’๐ถโˆฉ๐ด + ๐œ’๐ถโˆฉ๐ต for any set ๐ถ. 6.1.2. Nonnegative measurable functions. For any measurable extended real valued function ๐‘“ โ‰ฅ 0, we can approximate ๐‘“ pointwise with simple functions ๐œ™, by Theorem 5.29. This allows us to define the integral of ๐‘“ by (6.9)

โˆซ ๐‘“ = sup { โˆซ ๐œ™ โˆถ ๐œ™ is simple and 0 โ‰ค ๐œ™ โ‰ค ๐‘“}.

Again, it is possible to have โˆซ ๐‘“ = โˆž. The integral of ๐‘“ is also usually denoted by โˆซ ๐‘“, โ„๐••

if one wants to make explicit the fact that ๐‘“ is a function on โ„๐‘‘ , or โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ, โ„๐••

making the argument ๐‘ฅ explicit. As before, โˆซ๐ด ๐‘“ means โˆซ ๐‘“ = โˆซ ๐‘“๐œ’๐ด , ๐ด ๐‘‘

for any measurable ๐ด โŠ‚ โ„ . The definition of the integral implies the following properties, which follow from their versions for simple functions. 6.10. For any measurable function ๐‘“ โ‰ฅ 0 and any ๐‘ > 0, โˆซ ๐‘๐‘“ = ๐‘ โˆซ ๐‘“. This follows directly from 6.2 and the definition (6.9).

6.1. Integration of measurable functions

109

6.11. The translation and dilation invariance of the integral for simple functions, 6.4 and 6.5, also imply the invariance of the integral for measurable ๐‘“ โ‰ฅ 0, by (6.9). That is, if ๐‘“ โ‰ฅ 0 is measurable, ๐‘ฅ0 โˆˆ โ„๐‘‘ and ๐›ฟ > 0, then โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฅ0 )๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ and โˆซ ๐›ฟโˆ’๐‘‘ ๐‘“(๐‘ฅ/๐›ฟ)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ. 6.12. If ๐‘“, ๐‘” are measurable and 0 โ‰ค ๐‘“ โ‰ค ๐‘”, then โˆซ ๐‘“ โ‰ค โˆซ ๐‘”. Again, this follows from the definition (6.9), and the fact that, if ๐œ™ is a simple function that satisfies 0 โ‰ค ๐œ™ โ‰ค ๐‘“, then 0 โ‰ค ๐œ™ โ‰ค ๐‘” and thus { โˆซ ๐œ™ โˆถ ๐œ™ is simple and 0 โ‰ค ๐œ™ โ‰ค ๐‘“} โŠ‚ { โˆซ ๐œ™ โˆถ ๐œ™ is simple and 0 โ‰ค ๐œ™ โ‰ค ๐‘”}. The supremum of the set on the left side cannot be larger than that of the set on the right side. We are ready to state and prove the first of the Lebesgue integration theorems, called the monotone convergence theorem. Theorem 6.13 (Monotone convergence). If ๐‘“๐‘› โ‰ฅ 0 are measurable and ๐‘“๐‘› โ†— ๐‘“, then โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“. Proof. The limit ๐‘“ is measurable by Theorem 5.20. By 6.12, the sequence โˆซ ๐‘“๐‘› is increasing, so the limit exists (possibly โˆž). Also, by 6.12, โˆซ ๐‘“๐‘› โ‰ค โˆซ ๐‘“ for all ๐‘›, so lim โˆซ ๐‘“๐‘› โ‰ค โˆซ ๐‘“. To prove the reverse inequality, we verify that the limit of the integrals is at least as large as the integral of any simple function ๐œ™ that

110

6. Lebesgue integral and Lebesgue spaces

satisfies 0 โ‰ค ๐œ™ โ‰ค ๐‘“, so this limit is an upper bound for the set of all such integrals. By the definition (6.9) of the integral, this will imply lim โˆซ ๐‘“๐‘› โ‰ฅ โˆซ ๐‘“. So let ๐œ™ be a simple function such that 0 โ‰ค ๐œ™ โ‰ค ๐‘“, and let 0 < ๐œ€ < 1. Define ๐ธ๐‘› = {๐‘ฅ โˆถ ๐‘“๐‘› (๐‘ฅ) โ‰ฅ (1 โˆ’ ๐œ€)๐œ™(๐‘ฅ)}. Since ๐‘“๐‘› is increasing, the sequence ๐ธ๐‘› of sets is increasing because, if ๐‘ฅ โˆˆ ๐ธ๐‘› , then ๐‘“๐‘›+1 (๐‘ฅ) โ‰ฅ ๐‘“๐‘› (๐‘ฅ) โ‰ฅ (1 โˆ’ ๐œ€)๐œ™(๐‘ฅ) and thus ๐‘ฅ โˆˆ ๐ธ๐‘›+1 . We also have โ‹ƒ๐‘› ๐ธ๐‘› = โ„๐‘‘ because ๐‘“๐‘› โ†— ๐‘“ and ๐‘“ โ‰ฅ ๐œ™. Suppose ๐œ™ = โˆ‘ ๐‘Ž๐‘— ๐œ’๐ด๐‘— . Then, for each ๐‘›, โˆซ ๐‘“๐‘› โ‰ฅ โˆซ ๐‘“๐‘› ๐œ’๐ธ๐‘› โ‰ฅ (1 โˆ’ ๐œ€) โˆซ ๐œ™๐œ’๐ธ๐‘› = (1 โˆ’ ๐œ€) โˆ‘ ๐‘Ž๐‘— |๐ด๐‘— โˆฉ ๐ธ๐‘› |. ๐‘—

For each ๐‘—, the sequence of sets ๐ด๐‘— โˆฉ ๐ธ๐‘› is increasing in ๐‘› and โ‹ƒ

๐ด๐‘— โˆฉ ๐ธ๐‘› = ๐ด๐‘— ,

๐‘›

so we have, by the monotone continuity of the measure (Corollary 5.16), |๐ด๐‘— โˆฉ ๐ธ๐‘› | โ†’ |๐ด๐‘— | and thus lim โˆซ ๐‘“๐‘› โ‰ฅ (1 โˆ’ ๐œ€) โˆซ ๐œ™. Since 0 < ๐œ€ < 1 is arbitrary, we obtain lim โˆซ ๐‘“๐‘› โ‰ฅ โˆซ ๐œ™, โ–ก

as desired.

The monotone convergence theorem can be used to prove the linear and analytic properties of the Lebesgue integral, as in the following two facts. 6.14. If ๐‘“, ๐‘” โ‰ฅ are measurable, then โˆซ(๐‘“ + ๐‘”) = โˆซ ๐‘“ + โˆซ ๐‘”.

6.1. Integration of measurable functions

111

This is the analogue of fact 6.3 for nonnegative functions. For the proof, let ๐œ™๐‘› and ๐œ“๐‘› be sequences of simple functions with 0 โ‰ค ๐œ™๐‘› โ‰ค ๐‘“ and 0 โ‰ค ๐œ“๐‘› โ‰ค ๐‘” such that ๐œ™๐‘› โ†— ๐‘“ and ๐œ“๐‘› โ†— ๐‘”, which exist by Theorem 5.29. Thus ๐œ™๐‘› + ๐œ“๐‘› โ†— ๐‘“ + ๐‘” and, by the monotone convergence theorem, โˆซ(๐‘“ + ๐‘”) = lim โˆซ(๐œ™๐‘› + ๐œ“๐‘› ) = lim โˆซ ๐œ™๐‘› + lim โˆซ ๐œ“๐‘› = โˆซ ๐‘“ + โˆซ ๐‘”. Inductively, we can extend 6.14 to any finite number of nonnegative measurable functions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ , so ๐‘

(6.15)

๐‘

โˆซ โˆ‘ ๐‘“๐‘— = โˆ‘ โˆซ ๐‘“๐‘— . ๐‘—=1

๐‘—=1

The monotone convergence theorem also implies, in fact, that the integral of a series of nonnegative functions is equal to the series of the integrals. 6.16. If ๐‘“๐‘› is a sequence of nonnegative measurable functions and ๐‘“ = โˆ‘๐‘› ๐‘“๐‘› , then โˆซ ๐‘“ = โˆ‘ โˆซ ๐‘“๐‘› . ๐‘›

Indeed, the partial sums ๐‘ ๐‘ =

๐‘ โˆ‘๐‘›=1 ๐‘“๐‘›

satisfy

๐‘

โˆซ ๐‘ ๐‘ = โˆ‘ โˆซ ๐‘“๐‘› , ๐‘›=1

by (6.15). Since each ๐‘“๐‘› is nonnegative, ๐‘ ๐‘ โ†— ๐‘“, so by the monotone convergence theorem we obtain ๐‘

โˆ‘ โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“. ๐‘›=1

The monotone convergence theorem is true even if we only assume ๐‘“๐‘› โ†— ๐‘“ a.e. (Exercise (4)). If the sequence ๐‘“๐‘› is not monotone, nor converges to ๐‘“, we can still say something about the sequence of integrals. The following result, our second of the Lebesgue theorems, is known as Fatouโ€™s lemma.

112

6. Lebesgue integral and Lebesgue spaces

Theorem 6.17 (Fatouโ€™s lemma). For any sequence ๐‘“๐‘› of nonnegative measurable functions, โˆซ lim inf ๐‘“๐‘› โ‰ค lim inf โˆซ ๐‘“๐‘› . Proof. For each ๐‘›, consider the function inf๐‘˜โ‰ฅ๐‘› ๐‘“๐‘˜ given by inf ๐‘“๐‘˜ (๐‘ฅ) = inf{๐‘“๐‘˜ (๐‘ฅ) โˆถ ๐‘˜ โ‰ฅ ๐‘›},

๐‘˜โ‰ฅ๐‘›

the largest lower bound of the set of values ๐‘“๐‘˜ (๐‘ฅ) for ๐‘˜ โ‰ฅ ๐‘›. Thus, for each ๐‘— โ‰ฅ ๐‘›, we have inf๐‘˜โ‰ฅ๐‘› ๐‘“๐‘˜ โ‰ค ๐‘“๐‘— and thus โˆซ inf ๐‘“๐‘˜ โ‰ค โˆซ ๐‘“๐‘— ,

(6.18)

๐‘˜โ‰ฅ๐‘›

by 6.12. Now, since ๐‘— โ‰ฅ ๐‘› is arbitrary, the number on the left of (6.18) is then a lower bound for the set { โˆซ ๐‘“๐‘˜ โˆถ ๐‘˜ โ‰ฅ ๐‘›}, so we have that โˆซ inf ๐‘“๐‘˜ โ‰ค inf โˆซ ๐‘“๐‘˜ .

(6.19)

๐‘˜โ‰ฅ๐‘›

๐‘˜โ‰ฅ๐‘›

Since inf๐‘˜โ‰ฅ๐‘› ๐‘“๐‘˜ โ†— lim inf ๐‘“๐‘› as ๐‘› โ†’ โˆž, the monotone convergence theorem implies โˆซ lim inf ๐‘“๐‘› = lim โˆซ inf ๐‘“๐‘˜ . ๐‘›

๐‘˜โ‰ฅ๐‘›

Therefore, by (6.19), โˆซ lim inf ๐‘“๐‘› โ‰ค lim inf โˆซ ๐‘“๐‘˜ = lim inf โˆซ ๐‘“๐‘› . ๐‘› ๐‘˜โ‰ฅ๐‘›

โ–ก 6.20. In the case when ๐‘“๐‘› โ†’ ๐‘“, Fatouโ€™s lemma implies that โˆซ ๐‘“ โ‰ค lim inf โˆซ ๐‘“๐‘› . Thus, the integral of the limit of a sequence of functions can never be larger that the limit of the integrals, if it exists.

6.1. Integration of measurable functions

113

Example 6.21. In general, we donโ€™t have equality in 6.20. Consider the sequence ๐‘“๐‘› = ๐œ’(๐‘›,๐‘›+1] . Clearly ๐‘“๐‘› โ†’ 0, but each โˆซ ๐‘“๐‘› = 1, so โˆซ ๐‘“๐‘› โ†’ 1 > 0. 6.1.3. Extended real valued functions. Now we consider the integral for general measurable extended real valued functions ๐‘“. We say the ๐‘“ is integrable if โˆซ ๐‘“+ < โˆž

โˆซ ๐‘“โˆ’ < โˆž,

and

where ๐‘“+ and ๐‘“โˆ’ are the positive and negative parts of ๐‘“, defined above by ๐‘“+ = ๐‘“ โ‹… ๐œ’{๐‘“โ‰ฅ0} and ๐‘“โˆ’ = โˆ’๐‘“ โ‹… ๐œ’{๐‘“โ‰ค0} . + โˆ’ See Figure 5.5. As |๐‘“| = ๐‘“ + ๐‘“ , the integrability of ๐‘“ is equivalent to โˆซ |๐‘“| < โˆž. Note that ๐‘“ = ๐‘“+ โˆ’ ๐‘“โˆ’ . Thus, if ๐‘“ is integrable, its integral is defined by โˆซ ๐‘“ = โˆซ ๐‘“+ โˆ’ โˆซ ๐‘“โˆ’ . Example 6.22. Let ๐‘“ โˆถ โ„ โ†’ โ„ be given by sin ๐‘ฅ ๐‘ฅ>0 ๐‘“(๐‘ฅ) = { ๐‘ฅ 0 ๐‘ฅ โ‰ค 0. ๐‘“ is not integrable. We have that (2๐‘˜+1)๐œ‹

โˆž

โˆซ ๐‘“+ = โˆ‘ โˆซ ๐‘˜=0 2๐‘˜๐œ‹

sin ๐‘ฅ ๐‘‘๐‘ฅ ๐‘ฅ (2๐‘˜+1)๐œ‹

โˆž

1 โˆซ (2๐‘˜ + 1)๐œ‹ 2๐‘˜๐œ‹ ๐‘˜=0

โ‰ฅ โˆ‘

sin ๐‘ฅ๐‘‘๐‘ฅ = โˆž,

because each integral in the series is equal to 2.1 1

Recall, however, that the improper integral of ๐‘“ indeed exists: โˆž

โˆซ 0

๐‘…

sin ๐‘ฅ ๐œ‹ sin ๐‘ฅ ๐‘‘๐‘ฅ = lim โˆซ ๐‘‘๐‘ฅ = . ๐‘ฅ ๐‘ฅ 2 ๐‘…โ†’โˆž 0

114

6. Lebesgue integral and Lebesgue spaces

It is not hard to verify that the integral of extended real valued functions, by the definition above, satisfies the linear properties โˆซ ๐‘๐‘“ = ๐‘ โˆซ ๐‘“

(6.23) and (6.24)

โˆซ ๐‘“ + ๐‘” = โˆซ ๐‘“ + โˆซ ๐‘”.

Indeed, for the proof of (6.23), consider first the case ๐‘ > 0. We have ๐‘๐‘“ = (๐‘๐‘“)+ โˆ’ (๐‘๐‘“)โˆ’ = ๐‘๐‘“+ โˆ’ ๐‘๐‘“โˆ’ , so โˆซ ๐‘๐‘“ = โˆซ ๐‘๐‘“+ โˆ’ โˆซ ๐‘๐‘“โˆ’ = ๐‘ โˆซ ๐‘“+ โˆ’ ๐‘ โˆซ ๐‘“โˆ’ = ๐‘ inf ๐‘“, where we have used 6.10. The case ๐‘ = 0 is trivial, as we have both sides of (6.23) equal to 0. If ๐‘ < 0, we write ๐‘๐‘“ = (โˆ’๐‘)(โˆ’๐‘“) and thus โˆซ ๐‘๐‘“ = โˆซ(โˆ’๐‘)(โˆ’๐‘“) = (โˆ’๐‘) โˆซ(โˆ’๐‘“) = โˆ’๐‘( โˆซ ๐‘“โˆ’ โˆ’ โˆซ ๐‘“+ ) = ๐‘ โˆซ ๐‘“, because (โˆ’๐‘“)+ = ๐‘“โˆ’ and (โˆ’๐‘“)โˆ’ = ๐‘“+ . Since we are allowing the functions to be extended real valued, the undefined operation 0โ‹…โˆž will occur in the case ๐‘ = 0 at the points where ๐‘“ is equal to โˆž. However this may only occur in a set of measure zero (Exercise (3)) for an integrable function ๐‘“, so the scalar multiplication ๐‘๐‘“ is zero a.e. and โˆซ ๐‘๐‘“ = 0 (Exercise (1)). To prove (6.24), first note that, in the case when the measurable functions ๐‘“, ๐‘” satisfy ๐‘“, ๐‘” โ‰ฅ 0, ๐‘“ โˆ’ ๐‘” โ‰ฅ 0 and โˆซ ๐‘” < โˆž, then โˆซ(๐‘“ โˆ’ ๐‘”) = โˆซ ๐‘“ โˆ’ โˆซ ๐‘”. Indeed, by 6.14, we have โˆซ ๐‘“ = โˆซ(๐‘“ โˆ’ ๐‘” + ๐‘”) = โˆซ(๐‘“ โˆ’ ๐‘”) + โˆซ ๐‘”, and we can substract โˆซ ๐‘” from both sides because โˆซ ๐‘” < โˆž. Now, to verify (6.24), for extended real valued functions ๐‘“, ๐‘” we have (๐‘“ + ๐‘”)+ = ๐‘“+ ๐œ’๐ด โˆ’ ๐‘“โˆ’ ๐œ’๐ด + ๐‘”+ ๐œ’๐ด โˆ’ ๐‘”โˆ’ ๐œ’๐ด ,

6.1. Integration of measurable functions

115

where ๐ด = {๐‘ฅ โˆถ ๐‘“(๐‘ฅ) + ๐‘”(๐‘ฅ) โ‰ฅ 0}, and similarly for (๐‘“ + ๐‘”)โˆ’ (with opposite signs). Thus, by the previous observation and 6.14, โˆซ(๐‘“ + ๐‘”)+ = โˆซ(๐‘“+ ๐œ’๐ด โˆ’ ๐‘“โˆ’ ๐œ’๐ด + ๐‘”+ ๐œ’๐ด โˆ’ ๐‘”โˆ’ ๐œ’๐ด ) = โˆซ(๐‘“+ ๐œ’๐ด + ๐‘”+ ๐œ’๐ด ) โˆ’ โˆซ(๐‘“โˆ’ ๐œ’๐ด + ๐‘”โˆ’ ๐œ’๐ด ) = โˆซ ๐‘“+ ๐œ’๐ด + โˆซ ๐‘”+ ๐œ’๐ด โˆ’ โˆซ ๐‘“โˆ’ ๐œ’๐ด โˆ’ โˆซ ๐‘”โˆ’ ๐œ’๐ด = โˆซ ๐‘“๐œ’๐ด + โˆซ ๐‘”๐œ’๐ด . Combining with the corresponding expresion for โˆซ(๐‘“ + ๐‘”)โˆ’ we obtain (6.24). We leave the rest of the details as an exercise (Exercise (7)). Again, as we are allowing the functions to be extended valued, we may have the operation โˆž โˆ’ โˆž. As before, this may only occur in a set of measure zero, which doesnโ€™t modify the integrals (Exercise (2)). 6.1.4. Complex valued functions. Now, let ๐‘“ โˆถ โ„๐‘‘ โ†’ โ„‚. We say that ๐‘“ is integrable if both its real and imaginary parts are measurable and โˆซ |๐‘“| < โˆž. In this case, the integral of ๐‘“ is defined by โˆซ ๐‘“ = โˆซ โ„œ๐‘“ + ๐‘– โˆซ โ„‘๐‘“, where โ„œ๐‘“ and โ„‘๐‘“ are the real and imaginary parts of ๐‘“, respectively. Note that the complex valued function ๐‘“ is integrable if and only if its real and imaginary parts are integrable. As above, the integral of ๐‘“ is denoted by โˆซ ๐‘“ โ„๐••

when we want to make the Euclidean space โ„๐‘‘ explicit, or by โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ, โ„๐••

if we need to show explicitly the variable ๐‘ฅ of ๐‘“. We denote the set of complex valued integrable functions by ๐ฟ1 (โ„๐‘‘ ). We sometimes denote it simply by ๐ฟ1 , if there is no confusion. It is not hard to see that ๐ฟ1 is a complex vector space and ๐‘“ โ†ฆ โˆซ ๐‘“ is a linear

116

6. Lebesgue integral and Lebesgue spaces

functional on ๐ฟ1 (Exercise (8)). This time we donโ€™t have to worry about operations with infinity, but one still needs to verify carefully the identity โˆซ ๐›ผ๐‘“ = ๐›ผ โˆซ ๐‘“ for complex scalars ๐›ผ. 6.25. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), ๐‘ฅ0 โˆˆ โ„๐‘‘ and ๐›ฟ > 0, let ๐‘”(๐‘ฅ) = ๐‘“(๐‘ฅ โˆ’ ๐‘ฅ0 ) be the translation of ๐‘“ by ๐‘ฅ0 , and โ„Ž(๐‘ฅ) = ๐›ฟโˆ’๐‘‘ ๐‘“(๐‘ฅ/๐›ฟ) the dilation of ๐‘“ by ๐›ฟ. Then, by applying 6.11 to (โ„œ๐‘“)ยฑ , (โ„‘๐‘“)ยฑ we obtain โˆซ ๐‘” = โˆซ โ„Ž = โˆซ ๐‘“. 6.26. For ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), | โˆซ ๐‘“| โ‰ค โˆซ |๐‘“|. | | This inequality is obvious if โˆซ ๐‘“ = 0. If ๐‘“ is real valued, | โˆซ ๐‘“| = | โˆซ ๐‘“+ โˆ’ โˆซ ๐‘“โˆ’ | โ‰ค โˆซ ๐‘“+ + โˆซ ๐‘“โˆ’ = โˆซ |๐‘“|. | | | | If ๐‘“ is complex valued and โˆซ ๐‘“ โ‰  0, write โˆซ ๐‘“ = ๐‘Ÿ๐‘’๐‘–๐œƒ , its polar form. Then | โˆซ ๐‘“| = ๐‘Ÿ = ๐‘’โˆ’๐‘–๐œƒ โˆซ ๐‘“ = โˆซ ๐‘’โˆ’๐‘–๐œƒ ๐‘“, | | so โˆซ ๐‘’โˆ’๐‘–๐œƒ ๐‘“ is a real number, and hence โˆซ ๐‘’โˆ’๐‘–๐œƒ ๐‘“ = โˆซ โ„œ(๐‘’โˆ’๐‘–๐œƒ ๐‘“) + ๐‘– โˆซ โ„‘(๐‘’โˆ’๐‘–๐œƒ ๐‘“) = โˆซ โ„œ(๐‘’โˆ’๐‘–๐œƒ ๐‘“). Therefore, by 6.12, | โˆซ ๐‘“| = โˆซ ๐‘’โˆ’๐‘–๐œƒ ๐‘“ = โˆซ โ„œ(๐‘’โˆ’๐‘–๐œƒ ๐‘“) | | โ‰ค โˆซ |โ„œ(๐‘’โˆ’๐‘–๐œƒ ๐‘“)| โ‰ค โˆซ |๐‘’โˆ’๐‘–๐œƒ ๐‘“| = โˆซ |๐‘“|. For a measurable ๐ด โŠ‚ โ„๐‘‘ , we say that ๐‘“ is integrable on ๐ด if โˆซ |๐‘“| = โˆซ |๐‘“|๐œ’๐ด < โˆž. ๐ด

If ๐‘“ is integrable on ๐ด, we have โˆซ ๐‘“ = โˆซ ๐‘“๐œ’๐ด . ๐ด

6.1. Integration of measurable functions

117

We denote the set of integrable functions on ๐ด by ๐ฟ1 (๐ด). Recall that we say that ๐‘“ = ๐‘” almost everywhere (and we write a.e.) if the set {๐‘ฅ โˆถ ๐‘“(๐‘ฅ) โ‰  ๐‘”(๐‘ฅ)} has measure zero. Proposition 6.27. Let ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ). The following are equivalent. (1) ๐‘“ = ๐‘” a.e. (2) โˆซ |๐‘“ โˆ’ ๐‘”| = 0. (3) For all measurable ๐ด โŠ‚ โ„๐‘‘ , โˆซ ๐‘“ = โˆซ ๐‘”. ๐ด

๐ด

Proof. We prove the implications (1) โ‡’ (2) โ‡’ (3) โ‡’ (1). (1) โ‡’ (2): If ๐‘“ = ๐‘” a.e., then |๐‘“ โˆ’ ๐‘”| = 0 a.e. Hence โˆซ |๐‘“ โˆ’ ๐‘”| = 0. (2) โ‡’ (3): For any measurable ๐ด โŠ‚ โ„๐‘‘ , by 6.26, | โˆซ ๐‘“ โˆ’ โˆซ ๐‘”| = | โˆซ(๐‘“ โˆ’ ๐‘”)๐œ’ | โ‰ค โˆซ |๐‘“ โˆ’ ๐‘”|๐œ’ โ‰ค โˆซ |๐‘“ โˆ’ ๐‘”| = 0. ๐ด| ๐ด | | | ๐ด

๐ด

(3) โ‡’ (1): We prove the contrapositive, so assume ๐‘“ โ‰  ๐‘” in a set of positive measure. Then at least one of the functions โ„œ(๐‘“ โˆ’ ๐‘”)ยฑ or โ„‘(๐‘“ โˆ’ ๐‘”)ยฑ is positive in a set of positive measure. Assume ๐ด = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ โ„œ(๐‘“ โˆ’ ๐‘”)+ > 0} has positive measure. Note that, if ๐‘ฅ โˆˆ ๐ด, โ„œ(๐‘“ โˆ’ ๐‘”)โˆ’ (๐‘ฅ) = 0. Thus โˆซ ๐‘“ โˆ’ โˆซ ๐‘” = โˆซ (๐‘“ โˆ’ ๐‘”) = โˆซ โ„œ(๐‘“ โˆ’ ๐‘”) + ๐‘– โˆซ โ„‘(๐‘“ โˆ’ ๐‘”) ๐ด

๐ด

๐ด

๐ด

๐ด

and โˆซ โ„œ(๐‘“ โˆ’ ๐‘”) = โˆซ โ„œ(๐‘“ โˆ’ ๐‘”)+ > 0, ๐ด

so โˆซ๐ด ๐‘“ โ‰  โˆซ๐ด ๐‘”.

๐ด

โ–ก

We are ready for our third Lebesgue convergence theorem for integrals, known as the dominated convergence theorem.

118

6. Lebesgue integral and Lebesgue spaces

Theorem 6.28 (Dominated convergence). Let ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that ๐‘“๐‘› โ†’ ๐‘“ and there exists ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ) with |๐‘“๐‘› | โ‰ค ๐‘” for all ๐‘›. Then ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“. Proof. ๐‘“ is measurable by Corollary 5.21 and is integrable by 6.12, because |๐‘“| โ‰ค ๐‘”. By taking real and imaginary parts, we can assume all functions are real valued. Since |๐‘“๐‘› | โ‰ค ๐‘”, we have ๐‘” ยฑ ๐‘“๐‘› โ‰ฅ 0 for all ๐‘›, and we can apply Fatouโ€™s lemma (Theorem 6.17) to the sequences of nonnegative functions ๐‘” ยฑ ๐‘“๐‘› . Hence we have โˆซ(๐‘” + ๐‘“) โ‰ค lim inf โˆซ(๐‘” + ๐‘“๐‘› ) = โˆซ ๐‘” + lim inf โˆซ ๐‘“๐‘› and โˆซ(๐‘” โˆ’ ๐‘“) โ‰ค lim inf โˆซ(๐‘” โˆ’ ๐‘“๐‘› ) = โˆซ ๐‘” โˆ’ lim sup โˆซ ๐‘“๐‘› . As โˆซ ๐‘” < โˆž and โˆซ(๐‘” ยฑ ๐‘“) = โˆซ ๐‘” ยฑ โˆซ ๐‘“, we have lim sup โˆซ ๐‘“๐‘› โ‰ค โˆซ ๐‘“ โ‰ค lim inf โˆซ ๐‘“๐‘› , โ–ก

and therefore โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“.

As in the case of the monotone convergence theorem, Theorem 6.28 also holds if we only assume ๐‘“๐‘› โ†’ ๐‘“ a.e. As a corollary, we can extend 6.16 to complex valued integrable functions. Corollary 6.29. Let ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that โˆ‘๐‘› โˆซ |๐‘“๐‘› | < โˆž. Then the series โˆ‘๐‘› ๐‘“๐‘› converges almost everywhere to an integrable function ๐‘“ and โˆซ ๐‘“ = โˆ‘ โˆซ ๐‘“๐‘› . ๐‘›

Proof. Consider the function ๐‘” = โˆ‘๐‘› |๐‘“๐‘› |. Then ๐‘” is the limit of an increasing sequence, because each |๐‘“๐‘› | โ‰ฅ 0. By the monotone convergence

6.1. Integration of measurable functions

119

theorem we have โˆซ ๐‘” = โˆ‘ โˆซ |๐‘“๐‘› | < โˆž, ๐‘› 1

๐‘‘

so ๐‘” โˆˆ ๐ฟ (โ„ ). In particular, ๐‘” is finite almost everywhere (Exercise (3)) and hence โˆ‘๐‘› ๐‘“๐‘› converges almost everywhere. Also, for every ๐‘, ๐‘

๐‘

๐‘›=1

๐‘›=1

| โˆ‘ ๐‘“ | โ‰ค โˆ‘ |๐‘“ | โ‰ค ๐‘”. ๐‘›| ๐‘› | By the dominated convergence theorem, if โˆ‘๐‘› ๐‘“๐‘› โ†’ ๐‘“, we have that ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and โˆซ ๐‘“ = โˆ‘ โˆซ ๐‘“๐‘› . ๐‘›

โ–ก 6.30. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and, for each ๐‘› โ‰ฅ 1, ๐‘“๐‘› = ๐‘“๐œ’๐ต๐‘› , where ๐ต๐‘› is the ball around the origin of radius ๐‘›, then clearly |๐‘“๐‘› | โ‰ค |๐‘“|, ๐‘“๐‘› โ†’ ๐‘“ and thus, by the dominated convergence theorem, โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“. This can be written as โˆซ ๐‘“ โ†’ โˆซ ๐‘“, ๐ต๐‘›

โ„๐••

or, explicitly, lim โˆซ

๐‘›โ†’โˆž

|๐‘ฅ|โ‰ค๐‘›

๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ, โ„๐••

a fact that will later be useful to calculate integrals explicitly. There is nothing special about the balls ๐ต๐‘› : we can use any increasing sequence of measurable sets ๐ด๐‘› such that โ‹ƒ๐‘› ๐ด๐‘› = โ„๐‘‘ and obtain the same result. Corollary 6.31 states the conditions for the continuity and differentiability of integrals, also implied by the dominated convergence theorem. We leave its proof as an exercise (Exercise (9)). Corollary 6.31. Let ๐‘“(๐‘ฅ, ๐‘ก) be a function on โ„๐‘‘ ร— [๐‘Ž, ๐‘] such that ๐‘“(โ‹…, ๐‘ก) โˆˆ ๐ฟ1 (โ„๐‘‘ ) for each ๐‘ก โˆˆ [๐‘Ž, ๐‘]. (1) If ๐‘“(๐‘ฅ, โ‹…) is continuous on [๐‘Ž, ๐‘] for each ๐‘ฅ โˆˆ โ„๐‘‘ and there exists ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that |๐‘“(๐‘ฅ, ๐‘ก)| โ‰ค ๐‘”(๐‘ฅ) for all ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘ก โˆˆ [๐‘Ž, ๐‘], then ๐‘ก โ†ฆ โˆซโ„๐•• ๐‘“(๐‘ฅ, ๐‘ก)๐‘‘๐‘ฅ is continuous on [๐‘Ž, ๐‘].

120

6. Lebesgue integral and Lebesgue spaces (2) If ๐‘“(๐‘ฅ, โ‹…) is differentiable in (๐‘Ž, ๐‘) for each ๐‘ฅ โˆˆ โ„๐‘‘ and there exists โ„Ž โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that | ๐‘“(๐‘ฅ, ๐‘ก) โˆ’ ๐‘“(๐‘ฅ, ๐‘ ) | โ‰ค โ„Ž(๐‘ฅ) | | ๐‘กโˆ’๐‘  for all ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘ก, ๐‘  โˆˆ (๐‘Ž, ๐‘), ๐‘ก โ‰  ๐‘ , then ๐‘ก โ†ฆ โˆซโ„๐•• ๐‘“(๐‘ฅ, ๐‘ก)๐‘‘๐‘ฅ is differentiable in (๐‘Ž, ๐‘) and ๐‘‘ ๐œ• โˆซ ๐‘“(๐‘ฅ, ๐‘ก)๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ, ๐‘ก)๐‘‘๐‘ฅ. ๐‘‘๐‘ก โ„๐•• ๐œ•๐‘ก โ„๐••

6.2. Fubiniโ€™s theorem In this section we prove Fubiniโ€™s theorem on iterated integrals. For ๐‘‘1 , ๐‘‘2 โˆˆ โ„ค+ , we write the elements of the space โ„๐‘‘1 +๐‘‘2 as (๐‘ฅ, ๐‘ฆ), where ๐‘ฅ โˆˆ โ„๐‘‘1 and ๐‘ฆ โˆˆ โ„๐‘‘2 . Now consider a measurable set ๐ด โŠ‚ โ„๐‘‘1 +๐‘‘2 . For each ๐‘ฅ โˆˆ โ„๐‘‘1 , the ๐‘ฅ-section of ๐ด is defined as ๐ด๐‘ฅ = {๐‘ฆ โˆˆ โ„๐‘‘2 โˆถ (๐‘ฅ, ๐‘ฆ) โˆˆ ๐ด}, while the ๐‘ฆ-section of ๐ด is defined by ๐ด ๐‘ฆ = {๐‘ฅ โˆˆ โ„๐‘‘1 โˆถ (๐‘ฅ, ๐‘ฆ) โˆˆ ๐ด}. ๐ด๐‘ฅ and ๐ด ๐‘ฆ are the projections of the cross sections of ๐ด onto โ„๐‘‘2 and โ„๐‘‘1 , respectively (see Figure 6.1). It may happen that ๐ด๐‘ฅ or ๐ด ๐‘ฆ are nonmeasurable sets. However, we have the following result. Lemma 6.32. The sets ๐ด๐‘ฅ and ๐ด ๐‘ฆ are measurable for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and a.e. ๐‘ฆ โˆˆ โ„๐‘‘2 , respectively, the functions ๐‘ฅ โ†ฆ |๐ด๐‘ฅ | and ๐‘ฆ โ†ฆ |๐ด ๐‘ฆ | are measurable, and |๐ด| = โˆซ |๐ด๐‘ฅ |๐‘‘๐‘ฅ = โˆซ |๐ด ๐‘ฆ |๐‘‘๐‘ฆ.

(6.33)

โ„๐‘‘ 2

โ„๐‘‘ 1

Proof. The lemma follows immediately if ๐ด = ๐‘… ร— ๐‘†, where ๐‘… โŠ‚ โ„๐‘‘1 and ๐‘† โŠ‚ โ„๐‘‘2 are rectangles, because then ๐‘ฅโˆˆ๐‘… ๐‘ฅโˆ‰๐‘…

and

๐‘… ๐ด๐‘ฆ = { โˆ…

|๐ด๐‘ฅ | = |๐‘†|๐œ’๐‘… (๐‘ฅ)

and

|๐ด ๐‘ฆ | = |๐‘…|๐œ’๐‘† (๐‘ฆ),

๐ด๐‘ฅ = {

๐‘† โˆ…

๐‘ฆโˆˆ๐‘† ๐‘ฆ โˆ‰ ๐‘†,

so

6.2. Fubiniโ€™s theorem

121

Figure 6.1. The ๐‘ฅ-section and the ๐‘ฆ-section of ๐ด. Note that they are the projections of the cross sections of ๐ด onto โ„๐‘‘2 and โ„๐‘‘1 .

and |๐ด| = |๐‘…| โ‹… |๐‘†| = โˆซ |๐‘†|๐œ’๐‘… (๐‘ฅ)๐‘‘๐‘ฅ = โˆซ |๐‘…|๐œ’๐‘† (๐‘ฆ)๐‘‘๐‘ฆ. โ„๐‘‘2

โ„๐‘‘ 1

If ๐ด is a finite almost disjoint union of closed rectangles, then each ๐ด๐‘ฅ and ๐ด ๐‘ฆ is a finite disjoint union of closed rectangles (except for a finite number of points ๐‘ฅ and ๐‘ฆ corresponding to intersecting boundaries), so they are measurable, and (6.33) follows by the linearity of the integral. ๐‘ Indeed, if ๐ด = โ‹ƒ๐‘—=1 ๐‘…๐‘— where the ๐‘…๐‘— are almost disjoint, then ๐ด๐‘ฅ = ๐‘

๐‘

โ‹ƒ๐‘—=1 (๐‘…๐‘— )๐‘ฅ and ๐ด ๐‘ฆ = โ‹ƒ๐‘—=1 (๐‘…๐‘— )๐‘ฆ are also almost disjoint unions, so ๐‘

|๐ด๐‘ฅ | = โˆ‘ |(๐‘…๐‘— )๐‘ฅ | ๐‘—=1

๐‘

and

|๐ด ๐‘ฆ | = โˆ‘ |(๐‘…๐‘— )๐‘ฆ | ๐‘—=1

122

6. Lebesgue integral and Lebesgue spaces

except at most a finite number of ๐‘ฅ and ๐‘ฆ. Hence, ๐‘

๐‘

|๐ด| = โˆ‘ |๐‘…๐‘— | = โˆ‘ โˆซ |(๐‘…๐‘— )๐‘ฅ |๐‘‘๐‘ฅ ๐‘—=1 โ„๐‘‘1

๐‘—=1 ๐‘

=โˆซ

โˆ‘ |(๐‘…๐‘— )๐‘ฅ |๐‘‘๐‘ฅ = โˆซ |๐ด๐‘ฅ |๐‘‘๐‘ฅ,

โ„๐‘‘1 ๐‘—=1

โ„๐‘‘1

and similarly for the integrals of the ๐‘ฆ-sections. By Exercise (5) of Chapter 5, if ๐ด is open then it is the countable union of almost disjoint closed cubes, so the lemma follows by the monotone convergence theorem. Indeed, if we write ๐ด = โ‹ƒ๐‘— ๐‘…๐‘— , where the ๐‘…๐‘— are almost disjoint closed rectangles, then we also have ๐ด = โ‹ƒ๐‘› ๐ด๐‘› , where the ๐ด๐‘› is an increasing sequence of sets which are a finite union of almost disjoint rectangles, so by monotone continuity |๐ด| = lim |๐ด๐‘› | = lim โˆซ |(๐ด๐‘› )๐‘ฅ |๐‘‘๐‘ฅ = โˆซ |๐ด๐‘ฅ |๐‘‘๐‘ฅ ๐‘›

๐‘›

โ„๐‘‘ 1

โ„๐‘‘1

as |(๐ด๐‘› )๐‘ฅ | โ†— |๐ด๐‘ฅ | for each ๐‘ฅ. The result for the ๐‘ฆ-sections follows in the same way. If ๐ด is a bounded ๐บ ๐›ฟ set, then ๐ด = โ‹‚๐‘› ๐‘ˆ๐‘› , where ๐‘ˆ๐‘› is a decreasing sequence of bounded open sets. Then ๐ด๐‘ฅ =

โ‹‚

(๐‘ˆ๐‘› )๐‘ฅ

and

๐‘›

๐ด๐‘ฆ =

โ‹‚

(๐‘ˆ๐‘› )๐‘ฆ

๐‘›

๐‘ฆ

are measurable. Since |(๐‘ˆ1 )๐‘ฅ |, |(๐‘ˆ1 ) | < โˆž because ๐‘ˆ1 is bounded, the sequences |(๐‘ˆ๐‘› )๐‘ฅ | and |(๐‘ˆ๐‘› )๐‘ฆ | converge to |๐ด๐‘ฅ | and |๐ด ๐‘ฆ |, respectively, satisfy that |(๐‘ˆ๐‘› )๐‘ฅ | โ‰ค |(๐‘ˆ1 )๐‘ฅ |, |(๐‘ˆ๐‘› )๐‘ฆ | โ‰ค |(๐‘ˆ1 )๐‘ฆ |, and โˆซ |(๐‘ˆ1 )๐‘ฅ |๐‘‘๐‘ฅ = โˆซ |(๐‘ˆ1 )๐‘ฆ |๐‘‘๐‘ฆ = |๐‘ˆ1 | < โˆž, โ„๐‘‘ 2

โ„๐‘‘1

๐‘ฆ

so |(๐‘ˆ1 )๐‘ฅ | and |(๐‘ˆ1 ) | are integrable. Thus (6.33) follows by the dominated convergence theorem, following similar lines as in the previous cases. For a nonbounded ๐บ ๐›ฟ set ๐ด, we can write ๐ด = โ‹ƒ๐‘› (๐ด โˆฉ ๐ต๐‘› ), where each ๐ต๐‘› is the open ball of radius ๐‘› around the origin, and thus ๐ด is the increasing union of bounded ๐บ ๐›ฟ sets, so the lemma again follows by the monotone convergence theorem.

6.2. Fubiniโ€™s theorem

123

Let ๐ด be a measure zero set. By Corollary 5.17(1), for each ๐‘› there exists an open set ๐‘ˆ๐‘› โŠƒ ๐ด such that |๐‘ˆ๐‘› | < 1/๐‘›. If ๐‘ˆ = โ‹‚๐‘› ๐‘ˆ๐‘› , then ๐‘ˆ is a ๐บ ๐›ฟ set of measure zero, ๐‘ˆ๐‘ฅ and ๐‘ˆ ๐‘ฆ are measurable, and โˆซ |๐‘ˆ๐‘ฅ |๐‘‘๐‘ฅ = โˆซ |๐‘ˆ ๐‘ฆ |๐‘‘๐‘ฆ = 0. โ„๐‘‘ 2

โ„๐‘‘1

Thus |๐‘ˆ๐‘ฅ | = 0 for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and |๐‘ˆ ๐‘ฆ | = 0 for a.e. ๐‘ฆ โˆˆ โ„๐‘‘2 . Since ๐ด โŠ‚ ๐‘ˆ, each ๐ด๐‘ฅ โŠ‚ ๐‘ˆ๐‘ฅ and ๐ด ๐‘ฆ โŠ‚ ๐‘ˆ ๐‘ฆ , so ๐ด๐‘ฅ is a measure zero set for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and ๐ด ๐‘ฆ is a measure zero set for a.e. ๐‘ฆ โˆˆ โ„๐‘‘2 , and thus measurable. Therefore (6.33) is true for the measure zero set ๐ด. For a general measurable set ๐ด, we can use Corollary 5.17(1) as above to write ๐ด = ๐‘ˆ โงต ๐‘, where ๐‘ˆ is a ๐บ ๐›ฟ set and ๐‘ is a measure zero set. Since ๐ด๐‘ฅ = ๐‘ˆ๐‘ฅ โงต ๐‘๐‘ฅ and ๐ด ๐‘ฆ = ๐‘ˆ ๐‘ฆ โงต ๐‘ ๐‘ฆ , ๐ด๐‘ฅ and ๐ด ๐‘ฆ are measurable for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and for a.e. ๐‘ฆ โˆˆ โ„๐‘‘2 , respectively, and (6.33) follows by the previous cases. โ–ก Indeed, the monotone convergence theorem is the protagonist in the proof of Lemma 6.32. Do not be surprised, as it will be in many of the results further in this book. It is, of course, in the proof of Theorem 6.34. Given a function ๐‘“ โˆถ โ„๐‘‘1 +๐‘‘2 โ†’ โ„‚ we define, for each ๐‘ฅ โˆˆ โ„๐‘‘1 , the function ๐‘“๐‘ฅ on โ„๐‘‘2 by ๐‘“๐‘ฅ (๐‘ฆ) = ๐‘“(๐‘ฅ, ๐‘ฆ) and, for each ๐‘ฆ โˆˆ โ„๐‘‘2 , the function ๐‘“๐‘ฆ on โ„๐‘‘1 by ๐‘“๐‘ฆ (๐‘ฅ) = ๐‘“(๐‘ฅ, ๐‘ฆ). Theorem 6.34 (Fubini). If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘1 +๐‘‘2 ), for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and ๐‘ฆ โˆˆ โ„๐‘‘2 we have ๐‘“๐‘ฅ โˆˆ ๐ฟ1 (โ„๐‘‘2 ) and ๐‘“๐‘ฆ โˆˆ ๐ฟ1 (โ„๐‘‘1 ) and โˆซ

(6.35)

โ„๐‘‘1 +๐‘‘2

๐‘“ = โˆซ ( โˆซ ๐‘“๐‘ฅ )๐‘‘๐‘ฅ = โˆซ ( โˆซ ๐‘“๐‘ฆ )๐‘‘๐‘ฆ. โ„๐‘‘1

โ„๐‘‘ 2

โ„๐‘‘2

โ„๐‘‘ 1

Proof. By taking (โ„œ๐‘“)ยฑ and (โ„‘๐‘“)ยฑ , we can assume ๐‘“ โ‰ฅ 0. If ๐‘“ = ๐œ’๐ด for some measurable set ๐ด โŠ‚ โ„๐‘‘1 +๐‘‘2 , the identity (6.35) is the same as (6.33). By linearity of the integral, (6.35) follows if ๐‘“ is a simple function. For general ๐‘“ โ‰ฅ 0, take a sequence of nonnegative simple functions ๐œ™๐‘› โ†— ๐‘“, as in Theorem 5.29. Then (๐œ™๐‘› )๐‘ฅ โ†— ๐‘“๐‘ฅ and (๐œ™๐‘› )๐‘ฆ โ†— ๐‘“๐‘ฆ as well, for any ๐‘ฅ โˆˆ โ„๐‘‘1 and ๐‘ฆ โˆˆ โ„๐‘‘2 . Thus ๐‘“๐‘ฅ and ๐‘“๐‘ฆ are measurable for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 , ๐‘ฆ โˆˆ โ„๐‘‘2 and by the monotone convergence theorem we have the two limits โˆซ (๐œ™๐‘› )๐‘ฅ โ†— โˆซ ๐‘“๐‘ฅ โ„๐‘‘ 2

โ„๐‘‘2

and

โˆซ (๐œ™๐‘› )๐‘ฆ โ†— โˆซ ๐‘“๐‘ฆ , โ„๐‘‘1

โ„๐‘‘1

124

6. Lebesgue integral and Lebesgue spaces

and a second application of the monotone convergence theorem gives us โˆซ ( โˆซ (๐œ™๐‘› )๐‘ฅ )๐‘‘๐‘ฅ โ†’ โˆซ ( โˆซ ๐‘“๐‘ฅ )๐‘‘๐‘ฅ โ„๐‘‘2

โ„๐‘‘ 1

โ„๐‘‘ 1

and

โ„๐‘‘2

โˆซ ( โˆซ (๐œ™๐‘› )๐‘ฆ )๐‘‘๐‘ฆ โ†’ โˆซ ( โˆซ ๐‘“๐‘ฆ )๐‘‘๐‘ฆ. โ„๐‘‘ 2

โ„๐‘‘2

โ„๐‘‘ 1

โ„๐‘‘1

Equation (6.35) follows because, once more, the monotone convergence theorem implies โˆซ ๐œ™๐‘› โ†’ โˆซ ๐‘“ on โ„๐‘‘1 +๐‘‘2 . Since we are assuming โˆซ ๐‘“ < โˆž, we see that all integrals in (6.35) are finite, so the functions ๐‘ฅ โ†ฆ โˆซ ๐‘“๐‘ฅ and ๐‘ฆ โ†ฆ โˆซ ๐‘“๐‘ฆ are finite almost everywhere, so ๐‘“๐‘ฅ and ๐‘“๐‘ฆ are integrable for a.e. ๐‘ฅ โˆˆ โ„๐‘‘1 and ๐‘ฆ โˆˆ โ„๐‘‘2 . โ–ก We usually write (6.35) as ๐‘“ = โˆซ โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ๐‘‘๐‘ฅ = โˆซ โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฅ๐‘‘๐‘ฆ,

โˆซ โ„๐‘‘1 +๐‘‘2

โ„๐‘‘1

โ„๐‘‘ 2

โ„๐‘‘2

โ„๐‘‘1

and we call the second and third integrals as the iterated integrals of โˆซ ๐‘“. From the proof of Fubiniโ€™s theorem, we can see that (6.35) is true if we only assume ๐‘“ โ‰ฅ 0, as the integrability of ๐‘“ was only used to conclude the integrability of ๐‘“๐‘ฅ and ๐‘“๐‘ฆ , needed to extend the result to general integrable functions. The result for the case of nonnegative functions is known as Tonelliโ€™s theorem. We usually use Tonelliโ€™s theorem to verify the integrability of ๐‘“, as we can estimate โˆซ |๐‘“| using its iterated integrals. 6.36. An immediate consequence of Fubiniโ€™s theorem is following inequality, known as Minkowskiโ€™s inequality, โˆซ || โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ||๐‘‘๐‘ฅ โ‰ค โˆซ โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ. โ„๐‘‘ 1

โ„๐‘‘ 2

โ„๐‘‘2

โ„๐‘‘ 1

6.3. The Lebesgue space ๐ฟ1 We observed above that ๐ฟ1 (โ„๐‘‘ ) is a vector space under the usual operations (๐‘“ + ๐‘”)(๐‘ฅ) = ๐‘“(๐‘ฅ) + ๐‘”(๐‘ฅ)

and

(๐›ผ๐‘“)(๐‘ฅ) = ๐›ผ ๐‘“(๐‘ฅ),

which follow from the basic properties of the Lebesgue integral. Now we can define โ€–๐‘“โ€–๐ฟ1 = โˆซ |๐‘“|,

6.3. The Lebesgue space ๐ฟ1

125

and see that it satisfies: (1) โ€–๐‘“โ€–๐ฟ1 โ‰ฅ 0 for all ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ); (2) โ€–๐›ผ๐‘“โ€–๐ฟ1 = |๐›ผ| โ€–๐‘“โ€–๐ฟ1 for all ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐›ผ โˆˆ โ„‚; and (3) โ€–๐‘“ + ๐‘”โ€–๐ฟ1 โ‰ค โ€–๐‘“โ€–๐ฟ1 + โ€–๐‘”โ€–๐ฟ1 , which also follow from the basic properties of the integral. For instance, inequality (3), called the triangle inequality, follows from 6.12, โ€–๐‘“ + ๐‘”โ€–๐ฟ1 = โˆซ |๐‘“ + ๐‘”| โ‰ค โˆซ(|๐‘“| + |๐‘”|) = โˆซ |๐‘“| + โˆซ |๐‘”| = โ€–๐‘“โ€–๐ฟ1 + โ€–๐‘”โ€–๐ฟ1 , because |๐‘“(๐‘ฅ) + ๐‘”(๐‘ฅ)| โ‰ค |๐‘“(๐‘ฅ)| + |๐‘”(๐‘ฅ)| for all ๐‘ฅ. So we can almost say that โ€– โ‹… โ€–๐ฟ1 defines a norm, except for the fact that the condition โ€œโ€–๐‘“โ€–๐ฟ1 = 0 if and only if ๐‘“ = 0โ€ is not true. For instance, take ๐ด = โ„š (or any other subset of โ„ of measure zero) and ๐‘“ = ๐œ’๐ด . Then ๐‘“ โ‰  0, but โ€–๐‘“โ€–๐ฟ1 = 0. However, by Proposition 6.27, ๐‘“ = ๐‘” a.e. if and only if โˆซ |๐‘“โˆ’๐‘”| = 0, so ๐‘“ = ๐‘” a.e. if and only if โ€–๐‘“ โˆ’ ๐‘”โ€–๐ฟ1 = 0. Thus, if we define the equivalence relation ๐‘“ โˆผ ๐‘” if and only if ๐‘“ = ๐‘” a.e., we see that โ€– โ‹… โ€–๐ฟ1 defines a norm on the space โ„’ = ๐ฟ1 (โ„๐‘‘ )/ โˆผ of equivalence classes. Indeed, first note that the sum and scalar multiplication are well defined on โ„’, because if ๐‘“1 โˆผ ๐‘“2 and ๐‘”1 โˆผ ๐‘”2 , then {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘“1 (๐‘ฅ) + ๐‘”1 (๐‘ฅ) โ‰  ๐‘“2 (๐‘ฅ) + ๐‘”2 (๐‘ฅ)} โŠ‚ {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘“1 (๐‘ฅ) โ‰  ๐‘“2 (๐‘ฅ)} โˆช {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘”1 (๐‘ฅ) โ‰  ๐‘”2 (๐‘ฅ)}, so ๐‘“1 + ๐‘”1 = ๐‘“2 + ๐‘”2 a.e., and similarly for the scalar multiplication. For โ€– โ‹… โ€–๐ฟ1 , note that if ๐‘“ โˆผ ๐‘” then |๐‘“| = |๐‘”| a.e., so โˆซ |๐‘“| = โˆซ |๐‘”| and hence โ€–๐‘“โ€–๐ฟ1 = โ€–๐‘”โ€–๐ฟ1 . Thus, โ€– โ‹… โ€–๐ฟ1 is well defined on โ„’, and it is a norm, by (1), (2) and (3). Therefore, โ„’ is a normed space. As there is no reason for confusion, we will also denote the space โ„’ of equivalence classes of integrable functions by ๐ฟ1 (โ„๐‘‘ ). 6.37. The norm โ€– โ‹… โ€–๐ฟ1 is invariant under translations and dilations, which follows immediately by 6.25. Theorem 6.38. ๐ฟ1 (โ„๐‘‘ ) is a complete normed space.

126

6. Lebesgue integral and Lebesgue spaces Any norm โ€– โ‹… โ€– on a vector space ๐‘‹ induces the metric ๐‘‘(๐‘“, ๐‘”) = โ€–๐‘“ โˆ’ ๐‘”โ€–,

so it makes ๐‘‹ a metric space. Recall that a complete metric space is a metric space where its Cauchy sequences converge. If ๐‘‹ is a complete normed vector space, we call it a Banach space. See Appendix A.5. Proof of Theorem 6.38. We use Theorem A.13: assume that we have a sequence ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that โˆ‘๐‘› โ€–๐‘“๐‘› โ€–๐ฟ1 < โˆž, and we have to prove that the series โˆ‘๐‘› ๐‘“๐‘› converges in ๐ฟ1 (โ„๐‘‘ ), that is, there exists ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that ๐‘

|| โˆ‘ ๐‘“ โˆ’ ๐‘“|| โ†’ 0 ๐‘› || ||๐ฟ1 ๐‘›=1

as ๐‘ โ†’ โˆž. By Corollary 6.29, there exists ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) such that โˆ‘๐‘› ๐‘“๐‘› = ๐‘“ a.e. If we set ๐‘” = โˆ‘๐‘› |๐‘“๐‘› | as in the proof of 6.29, then ๐‘

| โˆ‘ ๐‘“ โˆ’ ๐‘“| โ‰ค 2๐‘” ๐‘› | | ๐‘›=1

for all ๐‘. Since ๐‘

| โˆ‘ ๐‘“ โˆ’ ๐‘“| โ†’ 0 ๐‘› | |

a.e.,

๐‘›=1

the dominated convergence theorem implies that ๐‘

โˆซ || โˆ‘ ๐‘“๐‘› โˆ’ ๐‘“|| โ†’ 0. ๐‘›=1

โ–ก 6.39. Simple functions are dense in ๐ฟ1 (โ„๐‘‘ ). This follows by 5.30, which states that we can find simple functions ๐œ™๐‘› such that |๐œ™๐‘› | โ‰ค |๐œ™๐‘›+1 |, |๐œ™๐‘› | โ‰ค |๐‘“| and ๐œ™๐‘› โ†’ ๐‘“, and the dominated convergence theorem. Note that, since each ๐œ™๐‘› is integrable, if ๐œ™๐‘› = โˆ‘ ๐‘Ž๐‘— ๐œ’๐ด๐‘— , ๐‘—

is the reduced form of ๐œ™๐‘› , then the sets ๐ด๐‘— have finite measure.

6.3. The Lebesgue space ๐ฟ1

127

6.40. The space ๐ถ๐‘ (โ„๐‘‘ ) of continuous functions of compact support is also dense in ๐ฟ1 (โ„๐‘‘ ). By 6.30, we can approximate any ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) with integrable functions with compact support, and thus we can assume that the simple functions in 6.39 also have compact support, so each ๐ด๐‘— in their reduced form is bounded. Now, given ๐œ€ > 0, we can find a compact set ๐น๐‘— โŠ‚ ๐ด๐‘— and a bounded open set ๐‘ˆ ๐‘— โŠƒ ๐ด๐‘— such that |๐ด๐‘— โงต๐น๐‘— |, |๐‘ˆ ๐‘— โงต๐ด๐‘— | < ๐œ€, and by Theorem A.11 in Appendix A there exists a continuous function ๐‘“๐‘— , with 0 โ‰ค ๐‘“๐‘— โ‰ค 1, such that ๐‘“๐‘— is supported in ๐‘ˆ ๐‘— and ๐‘“ = 1 on ๐น๐‘— . Thus โˆซ |๐œ’๐ด๐‘— โˆ’ ๐‘“๐‘— | โ‰ค |๐‘ˆ ๐‘— โงต ๐ด๐‘— | < 2๐œ€, so โˆซ || โˆ‘ ๐‘Ž๐‘— ๐œ’๐ด๐‘— โˆ’ โˆ‘ ๐‘Ž๐‘— ๐‘“๐‘— || โ‰ค 2๐œ€ โˆ‘ |๐‘Ž๐‘— |, ๐‘—

๐‘—

๐‘—

which implies the result because ๐œ€ > 0 is arbitrary. Note that, explicitly, 6.39 and 6.40 imply that, for every ๐œ€ > 0, we can find a simple ๐œ™ and ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ) such that โ€–๐‘“ โˆ’ ๐œ™โ€–๐ฟ1 < ๐œ€

and

โ€–๐‘“ โˆ’ ๐‘”โ€–๐ฟ1 < ๐œ€.

6.41. In particular, 6.40 implies that translations are continuous under the norm in ๐ฟ1 , in the following sense. Let ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and, for each ๐‘ฆ โˆˆ โ„๐‘‘ , let ๐‘“๐‘ฆ (๐‘ฅ) = ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) be the translation of ๐‘“ by ๐‘ฆ. Then lim โ€–๐‘“๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ1 = 0,

๐‘ฆโ†’0

so ๐‘“๐‘ฆ โ†’ ๐‘“ in ๐ฟ1 as ๐‘ฆ โ†’ 0. Note first that the result is true for ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ), because every continuous function of compact support is uniformly continuous. Indeed, asume ๐‘” is supported in the ball ๐ต๐‘Ÿ . Given ๐œ€ > 0, there exists ๐›ฟ > 0 such that ๐›ฟ < 1 and, if |๐‘ฆ| < ๐›ฟ then ๐œ€ |๐‘”(๐‘ฅ โˆ’ ๐‘ฆ) โˆ’ ๐‘”(๐‘ฅ)| < . |๐ต๐‘Ÿ+1 | Hence, if |๐‘ฆ| < ๐›ฟ, โ€–๐‘”๐‘ฆ โˆ’ ๐‘”โ€–๐ฟ1 = โˆซ ๐ต๐‘Ÿ+1 ๐‘ฆ

|๐‘”(๐‘ฅ โˆ’ ๐‘ฆ) โˆ’ ๐‘”(๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค โˆซ ๐ต๐‘Ÿ+1

๐œ€ ๐‘‘๐‘ฅ = ๐œ€. |๐ต๐‘Ÿ+1 |

1

Thus ๐‘” โ†’ ๐‘” in ๐ฟ as ๐‘ฆ โ†’ 0. For any other ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and given ๐œ€ > 0, choose ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ) such that โ€–๐‘“ โˆ’ ๐‘”โ€–๐ฟ1 < ๐œ€. Thus, by the triangle inequality and the invariance of the norm under translations, โ€–๐‘“๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ1 โ‰ค โ€–๐‘“๐‘ฆ โˆ’ ๐‘”๐‘ฆ โ€–๐ฟ1 + โ€–๐‘”๐‘ฆ โˆ’ ๐‘”โ€–๐ฟ1 + โ€–๐‘” โˆ’ ๐‘“โ€–๐ฟ1 < 2๐œ€ + โ€–๐‘”๐‘ฆ โˆ’ ๐‘”โ€–๐ฟ1 .

128

6. Lebesgue integral and Lebesgue spaces

Since โ€–๐‘”๐‘ฆ โˆ’ ๐‘”โ€–๐ฟ1 โ†’ 0 and ๐œ€ > 0 is arbitrary, we conclude that ๐‘“๐‘ฆ โ†’ ๐‘“ in ๐ฟ1 as ๐‘ฆ โ†’ 0. 6.42. For ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), their convolution is defined by ๐‘“ โˆ— ๐‘”(๐‘ฅ) = โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘”(๐‘ฆ)๐‘‘๐‘ฆ. โ„๐••

By Minkowskiโ€™s inequality and the translation invariance of the integral, โˆซ |๐‘“ โˆ— ๐‘”(๐‘ฅ)|๐‘‘๐‘ฅ = โˆซ || โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘”(๐‘ฆ)๐‘‘๐‘ฆ||๐‘‘๐‘ฅ โ„๐••

โ„๐••

โ„๐••

โ‰ค โˆซ โˆซ |๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘”(๐‘ฆ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ โ„๐••

โ„๐••

= โˆซ |๐‘”(๐‘ฆ)| โˆซ |๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ โ„๐••

โ„๐••

= โˆซ |๐‘”(๐‘ฆ)| โˆซ |๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ = โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”โ€–๐ฟ1 . โ„๐••

โ„๐••

Therefore ๐‘“ โˆ— ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ) and its norm satisfies (6.43)

โ€–๐‘“ โˆ— ๐‘”โ€–๐ฟ1 โ‰ค โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”โ€–๐ฟ1 .

Note that we have ๐‘“ โˆ— ๐‘” = ๐‘” โˆ— ๐‘“ for any ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ) (Exercise (10)). 6.44. Recall, as an example, the convolution operators 4.14 defined by ๐‘“ โ†ฆ ฮฆ๐‘ก โˆ— ๐‘“, where the dilations ฮฆ๐‘ก (๐‘ฅ) =

1 ๐‘ฅ ฮฆ( ), ๐‘ก ๐‘ก๐‘‘

form a family of good kernels. Note that this is true for any ฮฆ โˆˆ ๐ฟ1 (โ„๐‘‘ ) with โˆซ ฮฆ = 1, because we donโ€™t use the continuity of ฮฆ in Exercise (9) of Chapter 4. Also (6.43) and 6.37 imply โ€–ฮฆ๐‘ก โˆ— ๐‘“โ€–๐ฟ1 โ‰ค โ€–ฮฆ๐‘ก โ€–๐ฟ1 โ€–๐‘“โ€–๐ฟ1 = โ€–ฮฆโ€–๐ฟ1 โ€–๐‘“โ€–๐ฟ1 , so the convolutions ฮฆ๐‘ก โˆ— ๐‘“ are uniformly bounded in ๐ฟ1 .

6.3. The Lebesgue space ๐ฟ1

129

Now, by Minkwoskiโ€™s inequality and invariance under dilations, โ€–ฮฆ๐‘ก โˆ— ๐‘“ โˆ’ ๐‘“โ€–๐ฟ1 = โˆซ ||๐‘“ โˆ— ฮฆ๐‘ก (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ) โˆซ ฮฆ๐‘ก ||๐‘‘๐‘ฅ โ„๐••

โ„๐••

= โˆซ || โˆซ (๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) โˆ’ ๐‘“(๐‘ฅ))ฮฆ๐‘ก (๐‘ฆ)๐‘‘๐‘ฆ||๐‘‘๐‘ฅ โ„๐••

โ„๐••

โ‰ค โˆซ โˆซ |๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) โˆ’ ๐‘“(๐‘ฅ)| |ฮฆ๐‘ก (๐‘ฆ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ โ„๐••

โ„๐••

= โˆซ โˆซ |๐‘“(๐‘ฅ โˆ’ ๐‘ก๐‘ฆ) โˆ’ ๐‘“(๐‘ฅ)| |ฮฆ(๐‘ฆ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ โ„๐••

โ„๐••

= โˆซ โ€–๐‘“๐‘ก๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ1 |ฮฆ(๐‘ฆ)|๐‘‘๐‘ฆ. โ„๐••

Now, by 6.41, โ€–๐‘“๐‘ก๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ1 โ†’ 0 for each ๐‘ฆ โˆˆ โ„๐‘‘ as ๐‘ก โ†’ 0. Since โ€–๐‘“๐‘ก๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ1 โ‰ค 2โ€–๐‘“โ€–๐ฟ1 and ฮฆ is integrable, the dominated convergence theorem implies that โ€–ฮฆ๐‘ก โˆ— ๐‘“ โˆ’ ๐‘“โ€–๐ฟ1 โ†’ 0

as ๐‘ก โ†’ 0.

6.45. As a consequence of 6.44, we also see that the space ๐ถ๐‘โˆž (โ„๐‘‘ ) of smooth functions of compact support are dense in ๐ฟ1 (โ„๐‘‘ ). Indeed, by Corollary 6.31, if ๐œ™ โˆˆ ๐ถ๐‘โˆž (โ„๐‘‘ ) and ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), ๐œ™โˆ—๐‘“ is a smooth function because we can differentiate ๐œ™ โˆ— ๐‘“(๐‘ฅ) inside the integral. Now, if ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ), then ๐œ™๐‘ก โˆ— ๐‘” has compact support for any ๐‘ก > 0, and ๐œ™๐‘ก โˆ— ๐‘” โ†’ ๐‘” in ๐ฟ1 as ๐‘ก โ†’ 0, if โˆซ ๐œ™ = 1. Therefore, given ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐œ€ > 0, we can choose ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ) such that โ€–๐‘”โˆ’๐‘“โ€–๐ฟ1 < ๐œ€/2, and for a fixed ๐œ™ โˆˆ ๐ถ๐‘โˆž (โ„๐‘‘ ) with โˆซ ๐œ™ = 1 we can choose ๐‘ก > 0 such that โ€–๐œ™๐‘ก โˆ— ๐‘” โˆ’ ๐‘”โ€–๐ฟ1 < ๐œ€/2, and hence โ€–๐œ™๐‘ก โˆ— ๐‘” โˆ’ ๐‘“โ€–๐ฟ1 โ‰ค โ€–๐œ™๐‘ก โˆ— ๐‘” โˆ’ ๐‘”โ€–๐ฟ1 + โ€–๐‘” โˆ’ ๐‘“โ€–๐ฟ1 < ๐œ€. 6.46. Observe that, if {๐พ๐‘ก } is a family of good kernels, ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) is bounded and continuous at ๐‘ฅ, then ๐พ๐‘ก โˆ— ๐‘“(๐‘ฅ) โ†’ ๐‘“(๐‘ฅ) as ๐‘ก โ†’ 0, as the same proof as in the ones the case of the Poisson integral in Chapter 4 will apply to this case.

130

6. Lebesgue integral and Lebesgue spaces

6.4. The Lebesgue space ๐ฟ2 We now define ๐ฟ2 (โ„๐‘‘ ) as the set of square integrable measurable functions on โ„๐‘‘ , that is, functions that satisfy โˆซ |๐‘“|2 < โˆž. As in the case of ๐ฟ1 (โ„๐‘‘ ), we in fact identify ๐ฟ2 (โ„๐‘‘ ) to be the set of equivalence classes with respect to the relation ๐‘“ โˆผ ๐‘” if and only if ๐‘“(๐‘ฅ) = ๐‘”(๐‘ฅ) for almost every ๐‘ฅ. ๐ฟ2 (โ„๐‘‘ ) is a vector space (Exercise (11)) with inner product (6.47)

โŸจ๐‘“, ๐‘”โŸฉ = โˆซ ๐‘“(๐‘ฅ)๐‘”(๐‘ฅ)๐‘‘๐‘ฅ. โ„๐••

The integral in (6.47) converges, as Cauchyโ€™s inequality |๐‘“(๐‘ฅ)๐‘”(๐‘ฅ)| โ‰ค

1 (|๐‘“(๐‘ฅ)|2 + |๐‘”(๐‘ฅ)|2 ) 2

implies ๐‘“๐‘”ฬ„ is integrable whenever |๐‘“|2 and |๐‘”|2 are integrable. The inner product (6.47) induces the ๐ฟ2 norm โ€–๐‘“โ€–๐ฟ2 =

โˆซ |๐‘“|2 . โˆš

Recall the Cauchyโ€“Schwarz inequality | โˆซ ๐‘“(๐‘ฅ)๐‘”(๐‘ฅ)๐‘‘๐‘ฅ| โ‰ค โ€–๐‘“โ€– 2 โ€–๐‘”โ€– 2 , ๐ฟ ๐ฟ | | which is the main ingredient when verifying that โ€– โ‹… โ€–๐ฟ2 satisfies the triangle inequality. The ๐ฟ2 norm is also invariant by translations, as the ๐ฟ1 norm above, although it is not invariant by dilations. Theorem 6.48. ๐ฟ2 (โ„๐‘‘ ) is a complete inner product space. As in the proof of Theorem 6.38, the proof of Theorem 6.48 consists on verifying that every absolutely convergent series of functions in ๐ฟ2 (โ„๐‘‘ ) converges in ๐ฟ2 , and we leave the proof as an exercise (Exercise (12)). Moreover, we can also prove that the simple functions, as well as the space ๐ถ๐‘ (โ„๐‘‘ ) of continuous functions of compact support, are dense in

6.4. The Lebesgue space ๐ฟ2

131

๐ฟ2 (โ„๐‘‘ ), with similar proofs to the ๐ฟ1 (โ„๐‘‘ ) case. The density of ๐ถ๐‘ (โ„๐‘‘ ) in ๐ฟ2 (โ„๐‘‘ ) also implies the continuity of translations in the ๐ฟ2 norm, that is (6.49)

โ€–๐‘“๐‘ฆ โˆ’ ๐‘“โ€–๐ฟ2 โ†’ 0

as ๐‘ฆ โ†’ 0.

To discuss the analogous results to convolutions in ๐ฟ2 (โ„๐‘‘ ), we need the following result. Theorem 6.50. Let ๐‘“ be a measurable function such that, for any ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ), ๐‘“๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘€ = sup {|| โˆซ ๐‘“๐‘”|| โˆถ ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘”โ€–๐ฟ2 = 1} < โˆž. Then ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘“โ€–๐ฟ2 = ๐‘€. Theorem 6.50 can be seen as the converse to the Cauchyโ€“Schwarz inequality. Proof. The result is trivial if ๐‘“ = 0 a.e., so we assume it is not. Let ๐œ™๐‘› be a sequence of simple funtions such that |๐œ™๐‘› | โ‰ค |๐œ™๐‘›+1 |, |๐œ™๐‘› | โ‰ค |๐‘“| and ๐œ™๐‘› โ†’ ๐‘“, as in 5.30. If we set ๐œ“๐‘› = ๐œ™๐‘› ๐œ’๐ต๐‘› , then ๐œ“๐‘› โ†’ ๐‘“ with |๐œ“๐‘› | โ†— |๐‘“| and each ๐œ“๐‘› is supported in the ball ๐ต๐‘› , so ๐œ“๐‘› โˆˆ ๐ฟ2 (โ„๐‘‘ ). Since we are assuming ๐‘“ is not zero almost everywhere, we can also assume each โ€–๐œ“๐‘› โ€–๐ฟ2 > 0. We write ๐‘“ in polar coordinates as ๐‘“(๐‘ฅ) = |๐‘“(๐‘ฅ)|๐‘’๐‘–๐œƒ(๐‘ฅ) , and define ๐‘”๐‘› =

|๐œ“๐‘› |๐‘’โˆ’๐‘–๐œƒ . โ€–๐œ“๐‘› โ€–๐ฟ2

The sequence ๐‘”๐‘› has the following properties. โ€ข ๐‘”๐‘› โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘”๐‘› โ€–๐ฟ2 = 1. Indeed, โˆซ |๐‘”๐‘› |2 =

1 โˆซ |๐œ“๐‘› |2 = 1. โ€–๐œ“๐‘› โ€–2๐ฟ2

โ€ข โˆซ |๐œ“๐‘› | |๐‘”๐‘› | = โ€–๐œ“๐‘› โ€–๐ฟ2 . This follows directly by the calculation โˆซ |๐œ“๐‘› | |๐‘”๐‘› | = โˆซ |๐œ“๐‘› |

|๐œ“๐‘› | 1 โˆซ |๐œ“๐‘› |2 = โ€–๐œ“๐‘› โ€–๐ฟ2 . = โ€–๐œ“๐‘› โ€–๐ฟ2 โ€–๐œ“๐‘› โ€–๐ฟ2

132

6. Lebesgue integral and Lebesgue spaces โ€ข โˆซ ๐‘“๐‘”๐‘› = โˆซ |๐‘“| |๐‘”๐‘› |. It follows from the fact that ๐‘“๐‘’โˆ’๐‘–๐œƒ = |๐‘“|, because ๐‘“๐‘”๐‘› = ๐‘“

|๐œ“๐‘› |๐‘’โˆ’๐‘–๐œƒ |๐œ“๐‘› | = |๐‘“| = |๐‘“| |๐‘”๐‘› |. โ€–๐œ“โ€–๐ฟ2 โ€–๐œ“โ€–๐ฟ2

Thus, by Fatouโ€™s lemma and the above properties โ€–๐‘“โ€–๐ฟ2 โ‰ค lim inf โ€–๐œ“๐‘› โ€–๐ฟ2 = lim inf โˆซ |๐œ“๐‘› | |๐‘”๐‘› | โ‰ค lim inf โˆซ |๐‘“| |๐‘”๐‘› | = lim inf โˆซ ๐‘“๐‘”๐‘› โ‰ค ๐‘€.

Hence ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘“โ€–๐ฟ2 โ‰ค ๐‘€. By the Cauchyโ€“Schwarz inequality, for any ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ) such that โ€–๐‘”โ€–๐ฟ2 = 1, | โˆซ ๐‘“๐‘”| โ‰ค โ€–๐‘“โ€– 2 โ€–๐‘”โ€– 2 = โ€–๐‘“โ€– 2 . ๐ฟ ๐ฟ ๐ฟ | | โ–ก

Therefore ๐‘€ โ‰ค โ€–๐‘“โ€–๐ฟ2 , as required.

6.51. As a consequence of Theorem 6.50, we have the Minkowskiโ€™s inequality for ๐ฟ2 , 2

โˆš

โˆซ || โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ|| ๐‘‘๐‘ฅ โ‰ค โˆซ โ„๐‘‘ 1

โ„๐‘‘2

โ„๐‘‘2

โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|2 ๐‘‘๐‘ฅ๐‘‘๐‘ฆ. โˆš โ„๐‘‘1

Note that, if we write ๐‘“๐‘ฆ = ๐‘“(๐‘ฅ, ๐‘ฆ), this can be written as || โˆซ ๐‘“ ๐‘‘๐‘ฆ|| โ‰ค โˆซ โ€–๐‘“ โ€– 2 ๐‘‘๐‘ฆ, ๐‘ฆ ๐‘ฆ ๐ฟ || ||๐ฟ2 so it can be seen as an extension of the triangle inequality for integrals of functions. This follows by Theorem 6.50 because, for any ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘1 ) such that โ€–๐‘”โ€–๐ฟ2 = 1, using Tonelliโ€™s theorem and the Cauchyโ€“Schwarz

6.4. The Lebesgue space ๐ฟ2

133

inequality, โˆซ || โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ|| |๐‘”(๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค โˆซ โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|๐‘‘๐‘ฆ |๐‘”(๐‘ฅ)|๐‘‘๐‘ฅ โ„๐‘‘ 1

โ„๐‘‘ 2

โ„๐‘‘ 1

โ„๐‘‘ 2

= โˆซ โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)| |๐‘”(๐‘ฅ)|๐‘‘๐‘ฅ๐‘‘๐‘ฆ โ„๐‘‘2

โ‰คโˆซ โ„๐‘‘ 2

=โˆซ โ„๐‘‘2

โ„๐‘‘ 1

โˆš

โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|2 ๐‘‘๐‘ฅ โ‹… โ€–๐‘”โ€–๐ฟ2 ๐‘‘๐‘ฆ โ„๐‘‘ 1

โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|2 ๐‘‘๐‘ฅ๐‘‘๐‘ฆ. โˆš โ„๐‘‘1

Therefore, if the last integral is finite, ๐‘ฅ โ†ฆ ( โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ)๐‘”(๐‘ฅ) โ„๐‘‘2

is integrable and satisfies | โˆซ ( โˆซ ๐‘“(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ)๐‘”(๐‘ฅ)๐‘‘๐‘ฅ| โ‰ค โˆซ | | โ„๐‘‘ 1

โ„๐‘‘ 2

โ„๐‘‘2

โˆซ |๐‘“(๐‘ฅ, ๐‘ฆ)|2 ๐‘‘๐‘ฅ๐‘‘๐‘ฆ, โˆš โ„๐‘‘ 1

so we obtain the result. 6.52. We can obtain the following inequality for convolutions. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ), then ๐‘“ โˆ— ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘“ โˆ— ๐‘”โ€–๐ฟ2 โ‰ค โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”โ€–๐ฟ2 . This follows by Minkowskiโ€™s inequality because 2

โˆซ |๐‘“ โˆ— ๐‘”(๐‘ฅ)|2 ๐‘‘๐‘ฅ = โˆซ || โˆซ ๐‘“(๐‘ฆ)๐‘”(๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ|| ๐‘‘๐‘ฅ โˆš โˆš โ‰คโˆซ

โˆซ |๐‘“(๐‘ฆ)|2 |๐‘”(๐‘ฅ โˆ’ ๐‘ฆ)|2 ๐‘‘๐‘ฅ๐‘‘๐‘ฆ โˆš

โˆซ |๐‘”(๐‘ฅ โˆ’ ๐‘ฆ)|2 ๐‘‘๐‘ฅ๐‘‘๐‘ฆ โˆš = โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”โ€–๐ฟ2 , = โˆซ |๐‘“(๐‘ฆ)|

where have used the invariance of the ๐ฟ2 norm under translations.

134

6. Lebesgue integral and Lebesgue spaces

We can also prove analogous result in ๐ฟ2 to the convergence of convolution operators with good kernels discussed in 6.44, in ๐ฟ1 . We leave it as an exercise.

Exercises (1) Let ๐‘“ โ‰ฅ 0 be measurable. Then โˆซ ๐‘“ = 0 if and only if ๐‘“ = 0 a.e. (2) Let ๐‘“ โ‰ฅ 0 be measurable and ๐ด a set of measure 0. Then โˆซ๐ด ๐‘“ = 0. (3) Let ๐‘“ โ‰ฅ 0 be measurable. If โˆซ ๐‘“ < โˆž, then ๐‘“ < โˆž a.e. (4) If ๐‘“๐‘› โ‰ฅ 0 are measurable and ๐‘“๐‘› โ†— ๐‘“ a.e., then โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“. (5) If ๐‘“๐‘› โ‰ฅ 0 are measurable, ๐‘“๐‘› โ†’ ๐‘“ and โˆซ ๐‘“ = lim โˆซ ๐‘“๐‘› < โˆž, then โˆซ ๐‘“๐‘› โ†’ โˆซ ๐‘“ ๐ด

๐ด

for all measurable ๐ด โŠ‚ โ„๐‘‘ . The statement is false if โˆซ ๐‘“ = lim โˆซ ๐‘“๐‘› = โˆž. (6) Fatouโ€™s lemma implies the monotone convergence theorem. (7) Complete the details of the proof of (6.24). (8) The set ๐ฟ1 (โ„๐‘‘ ) of integrable functions is a vector space with the usual pointwise operations, and ๐‘“ โ†ฆ โˆซ ๐‘“ defines a linear functional on ๐ฟ1 (โ„๐‘‘ ). (9) Prove Corollary 6.31. (10) If ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ๐‘“ โˆ— ๐‘” = ๐‘” โˆ— ๐‘“. (11) The set of equivalence clases ๐ฟ2 (โ„๐‘‘ ) of square integrable functions forms a vector space, and (6.47) defines an inner product. (12) Prove Theorem 6.48. (13) Let ฮฆ โˆˆ ๐ฟ1 (โ„๐‘‘ ) with โˆซ ฮฆ = 1. If {ฮฆ๐‘ก }๐‘ก>0 are the dilations of ฮฆ, then lim โ€–ฮฆ๐‘ก โˆ— ๐‘“ โˆ’ ๐‘“โ€–๐ฟ2 = 0 ๐‘กโ†’0

for any ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ).

Notes

135

Notes Lebesgue developed what is now known as the Lebesgue integral in his papers cited in Chapter 5, as well as [Leb01a], [Leb01b], [Leb01c], his thesis [Leb02] and his text [Leb04], where he also proved the dominated convergence theorem for the special case when the functions are dominated on [0, 1] by a constant. The general case was proved in [Leb10]. Fatouโ€™s lemma was stated and proved by Pierre Fatou in [Fat06]. Beppo Levi stated and proved the monotone convergence theorem in [Lev06]. The spaces ๐ฟ๐‘ (for 1 โ‰ค ๐‘ โ‰ค โˆž, although we only discussed ๐‘ = 1, 2) were introduced by Frigyes Riesz in [Rie10], where he used the letter โ€œ๐ฟโ€ to denote them in honor to Lebesgue. Hermann Minkowski proved the inequality now known as Minkowskiโ€™s inequality for sums and series in [Min96], and the integral version is due to Riesz [Rie13].

Chapter 7

Maximal functions

7.1. Indefinite integrals and averages A well known result from calculus states that, if ๐‘“ is Riemann-integrable on the interval [๐‘Ž, ๐‘], then 1 โˆซ โ„Žโ†’0 โ„Ž ๐‘ฅ

๐‘ฅ+โ„Ž

lim

๐‘“ = ๐‘“(๐‘ฅ)

at each point ๐‘ฅ โˆˆ (๐‘Ž, ๐‘) where ๐‘“ is continuous. This says that, if ๐‘ฅ

๐น(๐‘ฅ) = โˆซ ๐‘“ ๐‘Ž

is the indefinite integral of ๐‘“, then ๐น is differentiable at each point ๐‘ฅ where ๐‘“ is continuous, and ๐น โ€ฒ (๐‘ฅ) = ๐‘“(๐‘ฅ). This result is true if ๐‘“ is Lebesgue integrable, of course, and, in fact, can be extended to functions on โ„๐‘‘ if we replace difference quotients by averages. Proposition 7.1. Let ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and continuous at ๐‘ฅ โˆˆ โ„๐‘‘ . Then 1 โˆซ |๐ต ๐‘Ÿโ†’0 ๐‘Ÿ (๐‘ฅ)| ๐ต

lim

๐‘“ = ๐‘“(๐‘ฅ).

๐‘Ÿ (๐‘ฅ)

Proof. Given ๐œ€ > 0, let ๐›ฟ > 0 such that, if |๐‘ฆ โˆ’ ๐‘ฅ| < ๐›ฟ, then |๐‘“(๐‘ฆ) โˆ’ ๐‘“(๐‘ฅ)| < ๐œ€. 137

138

7. Maximal functions

Then, if ๐‘Ÿ < ๐›ฟ, | 1 โˆซ | |๐ต (๐‘ฅ)| ๐‘Ÿ

๐ต๐‘Ÿ (๐‘ฅ)

๐‘“ โˆ’ ๐‘“(๐‘ฅ)|| โ‰ค

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

|๐‘“(๐‘ฆ) โˆ’ ๐‘“(๐‘ฅ)|๐‘‘๐‘ฆ


0

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

๐‘Ÿ (๐‘ฅ)

|๐‘“|,

7.2. The Hardyโ€“Littlewood maximal function

139

for each ๐‘ฅ โˆˆ โ„๐‘‘ . Note that we are taking the supremum of the averages of |๐‘“| over all balls centered at ๐‘ฅ, of all radii. If the set of averages is not bounded, then ๐‘€๐‘“(๐‘ฅ) = โˆž. If ๐‘“ is bounded, say, |๐‘“(๐‘ฅ)| โ‰ค ๐ด for all ๐‘ฅ โˆˆ โ„๐‘‘ (or almost everywhere), then ๐‘€๐‘“(๐‘ฅ) is finite at every ๐‘ฅ, of course, and |๐‘€๐‘“(๐‘ฅ)| โ‰ค ๐ด. Example 7.3. A priori, nothing guarantees that ๐‘€๐‘“ is finite in a positive measure set, or anywhere (see Exercises (2) and (3)). It is easy to see that, in general, ๐‘€๐‘“ โˆ‰ ๐ฟ1 (โ„๐‘‘ ), even if ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ). For example, consider ๐‘“ = ๐œ’[0,1] on โ„. For ๐‘ฅ > 1, ๐‘€๐‘“(๐‘ฅ) โ‰ฅ

1 โˆซ 2๐‘ฅ 0

2๐‘ฅ

๐œ’[0,1] =

1 , 2๐‘ฅ

and thus ๐‘€๐‘“ โˆ‰ ๐ฟ1 (โ„). However, Hardy and Littlewood proved that, indeed, ๐‘€๐‘“ is finite almost everywhere if ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and, even though it might not be integrable, Example 7.3 is essentially the worst case scenario. Theorem 7.4 (Hardyโ€“Littlewood). There exists ๐ด > 0 such that, for any ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐›ผ > 0, |{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘€๐‘“(๐‘ฅ) > ๐›ผ}| โ‰ค

๐ด โ€–๐‘“โ€–๐ฟ1 . ๐›ผ

We say then that the maximal function is an operator of weak type (1, 1) . We first observe that Theorem 7.4 states that the situation of Example 7.3 is pretty much the โ€œworst caseโ€ we would get for ๐‘€๐‘“, as the function 1/๐‘ฅ is the natural example of a measurable function ๐‘“ that satisfies ๐ด |{๐‘ฅ โˆˆ โ„ โˆถ |๐‘“(๐‘ฅ)| > ๐›ผ}| = ๐›ผ for some constant ๐ด. Indeed, we clearly have 1 2 |{๐‘ฅ โˆˆ โ„ โˆถ | | > ๐›ผ}| = ๐‘ฅ ๐›ผ for any ๐›ผ > 0, as the set on the left hand side is the interval (โˆ’๐›ผ, ๐›ผ).

140

7. Maximal functions In fact, any integrable function satisfies the above inequality, as |{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“(๐‘ฅ)| > ๐›ผ}| = โˆซ

1๐‘‘๐‘ฅ

{๐‘ฅโˆˆโ„๐‘‘ โˆถ|๐‘“(๐‘ฅ)|>๐›ผ}

โ‰คโˆซ {๐‘ฅโˆˆโ„๐‘‘ โˆถ|๐‘“(๐‘ฅ)|>๐›ผ}

|๐‘“(๐‘ฅ)| ๐‘‘๐‘ฅ ๐›ผ

1 (7.5) โ‰ค โ€–๐‘“โ€–๐ฟ1 . ๐›ผ The inequality (7.5) is called Chebyshevโ€™s inequality. In the proof of Theorem 7.4, we will use Lemma 7.6, due to Vitali. Lemma 7.6 (Vitali). Let ๐ต1 , ๐ต2 , . . . , ๐ต๐‘ be a finite collection of balls in โ„๐‘‘ . Then there exist disjoint ๐ต๐‘–1 , ๐ต๐‘–2 , . . . , ๐ต๐‘–๐‘˜ among them such that ๐‘

๐‘˜

๐‘‘ | | | โ‹ƒ ๐ต๐‘— | โ‰ค 3 โˆ‘ |๐ต๐‘–๐‘— |. ๐‘—=1

๐‘—=1

In other words, from any finite collection of balls we can obtain a disjoint subcollection such that, even as its union might be a smaller set than the union of the original balls, the volume of the new union is at least as large as a fixed proportion of the original total volume. Proof. Given ๐ต1 , ๐ต2 , . . . , ๐ต๐‘ , choose a ball of maximal radius, say ๐ต๐‘–1 . Once ๐ต๐‘–1 , ๐ต๐‘–2 , . . . , ๐ต๐‘–๐‘— are chosen, choose ๐ต๐‘–๐‘—+1 as a ball of maximal radius among all remaining balls disjoint to ๐ต๐‘–1 , ๐ต๐‘–2 , . . . , ๐ต๐‘–๐‘— , until exhausting the collection. For each ๐ต๐‘— in the collection, there exists ๐‘–๐‘™ such that โ€ข ๐ต๐‘–๐‘™ intersects ๐ต๐‘— ; and โ€ข the radius of ๐ต๐‘–๐‘™ is at least as large as the radius of ๐ต๐‘— . Otherwise, ๐ต๐‘— would have been chosen in the construction. Thus, if 3๐ต๐‘–๐‘™ is the ball with the same center as ๐ต๐‘–๐‘™ and three times its radius, then ๐ต๐‘— โŠ‚ 3๐ต๐‘–๐‘™ (see Figure 7.1) and, thus, ๐‘

โ‹ƒ ๐‘—=1

๐‘˜

๐ต๐‘— โŠ‚

โ‹ƒ

3๐ต๐‘–๐‘— .

๐‘—=1

The lemma follows from the fact that |3๐ต๐‘–๐‘— | = 3๐‘‘ |๐ต๐‘–๐‘— |.

โ–ก

7.2. The Hardyโ€“Littlewood maximal function

141

Figure 7.1. The ball 3๐ต๐‘–๐‘™ with the same center as ๐ต๐‘–๐‘™ and three times its radius contains all balls intersecting it with smaller or equal radii.

Proof of Theorem 7.4. As the measure of any measurable set is the supremum of the measures of its compact subsets, we let ๐พ โŠ‚ {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ ๐‘€๐‘“(๐‘ฅ) > ๐›ผ} be compact. Hence, it is enough to estimate the measure of ๐พ. Now, by definition, for each ๐‘ฅ โˆˆ ๐พ there exists a ball ๐ต๐‘ฅ centered at ๐‘ฅ such that 1 โˆซ |๐‘“| > ๐›ผ. |๐ต๐‘ฅ | ๐ต ๐‘ฅ

The balls ๐ต๐‘ฅ , for ๐‘ฅ โˆˆ ๐พ, cover the compact set ๐พ, so there exists a finite collection ๐ต1 , ๐ต2 , . . . , ๐ต๐‘ of them such that ๐‘

๐พโŠ‚

โ‹ƒ

๐ต๐‘— .

๐‘—=1

By Lemma 7.6, there exist disjoint ๐ต๐‘–1 , ๐ต๐‘–2 , . . . , ๐ต๐‘–๐‘˜ among them such that ๐‘

๐‘˜

๐‘‘ | | | โ‹ƒ ๐ต๐‘— | โ‰ค 3 โˆ‘ |๐ต๐‘–๐‘— |. ๐‘—=1

๐‘—=1

142

7. Maximal functions

Thus ๐‘

๐‘˜

|๐พ| โ‰ค || ๐ต | โ‰ค 3๐‘‘ โˆ‘ |๐ต๐‘–๐‘— |. โ‹ƒ ๐‘—| ๐‘—=1

๐‘—=1

Now, for each of those balls, 1 โˆซ |๐‘“| > ๐›ผ, |๐ต๐‘–๐‘— | ๐ต ๐‘–๐‘—

so |๐ต๐‘–๐‘— |
๐›ผ, for any fixed positive ๐›ผ, is finite. 7.7. We can also consider, for ๐‘“ โˆˆ ๐ฟ1loc (โ„๐‘‘ ), the maximal function 1 ฬƒ โˆซ |๐‘“| โˆถ ๐ต is a ball with ๐‘ฅ โˆˆ ๐ต}. ๐‘€๐‘“(๐‘ฅ) = sup { |๐ต| ๐ต ฬƒ is the supremum of the averages of |๐‘“| over all balls that That is, ๐‘€๐‘“ contain ๐‘ฅ, and not only those centered at ๐‘ฅ. If ๐‘ฅ โˆˆ ๐ต and ๐ต has radius ๐‘Ÿ, then ๐ต โŠ‚ ๐ต2๐‘Ÿ (๐‘ฅ) and |๐ต2๐‘Ÿ (๐‘ฅ)| = 2๐‘‘ |๐ต|, and thus 1 2๐‘‘ โˆซ |๐‘“| โ‰ค โˆซ |๐ต| ๐ต |๐ต2๐‘Ÿ (๐‘ฅ)| ๐ต

|๐‘“|,

2๐‘Ÿ (๐‘ฅ)

ฬƒ ฬƒ also satisfies the concluand hence ๐‘€๐‘“(๐‘ฅ) โ‰ค 2๐‘‘ ๐‘€๐‘“(๐‘ฅ). Therefore, ๐‘€๐‘“ sion of the Hardyโ€“Littlewood maximal theorem.

7.3. The Lebesgue differentiation theorem

143

7.3. The Lebesgue differentiation theorem We are now ready to discuss the question posed in Section 7.1, on whether lim ๐‘Ÿโ†’0

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

๐‘“

๐‘Ÿ (๐‘ฅ)

exists for any ๐‘ฅ โˆˆ โ„๐‘‘ , if ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ). In fact, it turns out that the limit exists almost everywhere, as stated by Lebesgueโ€™s theorem. Theorem 7.8 (Lebesgue). If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then, for almost every ๐‘ฅ โˆˆ โ„๐‘‘ , lim ๐‘Ÿโ†’0

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

๐‘“ = ๐‘“(๐‘ฅ).

๐‘Ÿ (๐‘ฅ)

Lebesgueโ€™s theorem 7.8 states that the limit not only exists, but that it is actually equal to ๐‘“ almost everywhere. This might seem remarkable when compared to Proposition 7.1, because ๐‘“ could be discontinuous everywhere.1 Proof of Theorem 7.8. Let ๐น โŠ‚ โ„๐‘‘ be the set where either the limit does not exist or is not equal to ๐‘“. Thus, ๐น = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ lim sup |๐ผ๐‘Ÿ ๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| > 0}. ๐‘Ÿโ†’0

where we have written ๐ผ๐‘Ÿ ๐‘“(๐‘ฅ) for the average of ๐‘“ on the ball of radius ๐‘Ÿ around ๐‘ฅ, 1 โˆซ ๐ผ๐‘Ÿ ๐‘“(๐‘ฅ) = ๐‘“. |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต (๐‘ฅ) ๐‘Ÿ

We want to prove that |๐น| = 0. As ๐น = โ‹ƒ๐‘› ๐น1/๐‘› , where ๐น๐›ผ = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ lim sup |๐ผ๐‘Ÿ ๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| > ๐›ผ}, ๐‘Ÿโ†’0

it is sufficient to prove that |๐น๐›ผ | = 0 for any ๐›ผ > 0. Given ๐œ€ > 0, let ๐‘” โˆˆ ๐ถ๐‘ (โ„๐‘‘ ) such that โ€–๐‘“ โˆ’ ๐‘”โ€–๐ฟ1 < ๐œ€. By Proposition 7.1, lim ๐ผ๐‘Ÿ ๐‘”(๐‘ฅ) = ๐‘”(๐‘ฅ) ๐‘Ÿโ†’0

1 Recall, however, that Riemann-integrable functions are indeed continuous almost everywhere (Theorem A.9).

144

7. Maximal functions

for every ๐‘ฅ โˆˆ โ„๐‘‘ , and thus lim sup|๐ผ๐‘Ÿ ๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| ๐‘Ÿโ†’0

= lim sup |๐ผ๐‘Ÿ (๐‘“ โˆ’ ๐‘”)(๐‘ฅ) + ๐ผ๐‘Ÿ ๐‘”(๐‘ฅ) โˆ’ ๐‘”(๐‘ฅ) + ๐‘”(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| ๐‘Ÿโ†’0

= lim sup |๐ผ๐‘Ÿ (๐‘“ โˆ’ ๐‘”)(๐‘ฅ)| + |๐‘”(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| ๐‘Ÿโ†’0

โ‰ค ๐‘€(๐‘“ โˆ’ ๐‘”)(๐‘ฅ) + |๐‘”(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)|, where, in the last inequality, we have used the fact |๐ผ๐‘Ÿ (๐‘“ โˆ’ ๐‘”)(๐‘ฅ)| โ‰ค

1 โˆซ |๐ต๐‘Ÿ (๐‘ฅ)| ๐ต

|๐‘“ โˆ’ ๐‘”| โ‰ค ๐‘€(๐‘“ โˆ’ ๐‘”)(๐‘ฅ).

๐‘Ÿ (๐‘ฅ)

Then |๐น๐›ผ | โ‰ค |{๐‘ฅ โˆถ ๐‘€(๐‘“ โˆ’ ๐‘”)(๐‘ฅ) >

๐›ผ ๐›ผ }| + |{๐‘ฅ โˆถ |๐‘”(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| > }|. 2 2

The first term is estimated by the Hardyโ€“Littlewood theorem, |{๐‘ฅ โˆถ ๐‘€(๐‘“ โˆ’ ๐‘”)(๐‘ฅ) >

๐›ผ ๐ด 2๐ด }| โ‰ค โ€–๐‘“ โˆ’ ๐‘”โ€–๐ฟ1 < ๐œ€, 2 ๐›ผ/2 ๐›ผ

and the second term by Chebyshevโ€™s inequality (7.5), |{๐‘ฅ โˆถ |๐‘”(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| >

๐›ผ 2 1 }| โ‰ค โ€–๐‘” โˆ’ ๐‘“โ€–๐ฟ1 < ๐œ€ 2 ๐›ผ/2 ๐›ผ

Therefore 2(๐ด + 1) ๐œ€, ๐›ผ and the theorem follows because ๐œ€ > 0 is arbitrary. |๐น๐›ผ |
0 such that, for every ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and (๐‘ฅ, ๐‘ก) โˆˆ โ„๐‘‘+1 + , |๐‘ข(๐‘ฅ, ๐‘ก)| โ‰ค ๐ด๐‘€๐‘“(๐‘ฅ). Hence Poisson integrals are uniformly estimated from above by the maximal function. Proof. We estimate the Poisson integral ๐‘ข(๐‘ฅ, ๐‘ก) by integrating over dyadic annuli around the point ๐‘ฅ. Indeed, since ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), for each (๐‘ฅ, ๐‘ก) โˆˆ โ„๐‘‘+1 we can write + โˆž

๐‘ข(๐‘ฅ, ๐‘ก) = โˆซ

๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ + โˆ‘ โˆซ

๐ต๐‘ก (๐‘ฅ)

๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ,

๐‘˜=0 ๐ด๐‘˜ (๐‘ฅ)

where each ๐ด๐‘˜ (๐‘ฅ) is the annulus ๐ด๐‘˜ (๐‘ฅ) = {๐‘ฆ โˆˆ โ„๐‘‘ โˆถ 2๐‘˜ ๐‘ก โ‰ค |๐‘ฅ โˆ’ ๐‘ฆ| < 2๐‘˜+1 ๐‘ก}. Now ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ) =

2 2๐‘ก โ‰ค , (๐‘‘+1)/2 2 2 ๐œ”๐‘‘+1 ๐‘ก๐‘‘ ๐œ”๐‘‘+1 (|๐‘ฅ โˆ’ ๐‘ฆ| + ๐‘ก )

so the first integral in the sum above is estimated by |โˆซ |

๐ต๐‘ก (๐‘ฅ)

๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ|| โ‰ค

2 ๐œ”๐‘‘+1 ๐‘ก๐‘‘

โˆซ

|๐‘“(๐‘ฆ)|๐‘‘๐‘ฆ โ‰ค ๐‘๐‘€๐‘“(๐‘ฅ),

๐ต๐‘ก (๐‘ฅ)

where ๐‘ > 0 is a constant that depends only on the dimension ๐‘‘. Similarly, if |๐‘ฅ โˆ’ ๐‘ฆ| โ‰ฅ 2๐‘˜ ๐‘ก, we have ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ) โ‰ค

2๐‘ก 2 โ‰ค , ๐œ”๐‘‘+1 |๐‘ฅ โˆ’ ๐‘ฆ|๐‘‘+1 ๐œ”๐‘‘+1 2๐‘˜(๐‘‘+1) ๐‘ก๐‘‘

146

7. Maximal functions

and thus |โˆซ |

๐ด๐‘˜ (๐‘ฅ)

๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ|| โ‰ค

2 โˆซ ๐œ”๐‘‘+1 2๐‘˜(๐‘‘+1) ๐‘ก๐‘‘ ๐ต ๐‘˜+1 2

|๐‘“(๐‘ฆ)|๐‘‘๐‘ฆ (๐‘ฅ) ๐‘ก

๐‘โ€ฒ โ‰ค ๐‘˜ ๐‘€๐‘“(๐‘ฅ), 2 โ€ฒ where ๐‘ > 0 is another constant that depends only on ๐‘‘. Therefore โˆž

๐‘โ€ฒ ๐‘€๐‘“(๐‘ฅ) = ๐ด๐‘€๐‘“(๐‘ฅ), 2๐‘˜ ๐‘˜=0

|๐‘ข(๐‘ฅ, ๐‘ก)| โ‰ค ๐‘๐‘€๐‘“(๐‘ฅ) + โˆ‘

โ–ก

where ๐ด = ๐‘ + 2๐‘โ€ฒ .

We can now state the following result on pointwise boundary limits of Poisson integrals. Theorem 7.12. Lef ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘ข(๐‘ฅ, ๐‘ก) its Poisson integral. Then lim ๐‘ข(๐‘ฅ, ๐‘ก) = ๐‘“(๐‘ฅ) ๐‘กโ†’0

๐‘‘

for almost every ๐‘ฅ โˆˆ โ„ . The proof of Theorem 7.12 follows as the proof of Theorem 7.8, by showing that the set {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ lim sup |๐‘ข(๐‘ฅ, ๐‘ก) โˆ’ ๐‘“(๐‘ฅ)| > ๐›ผ} ๐‘กโ†’0

has measure 0 for any ๐›ผ > 0, by first approximating with a compactly supported continuous function and then comparing with the maximal function, using Proposition 7.11. We leave the details as an exercise (Exercise (5)). If we compare Theorem 7.12 with Theorem 4.13, we see that this time we are only approaching the boundary point (๐‘ฅ, 0) vertically, as we only consider ๐‘ข(๐‘ฅ, ๐‘ก) as ๐‘ก > 0, while in Theorem 4.13 we approach (๐‘ฅ, 0) from any direction in the upper half-space. We can extend Theorem 7.12 if we consider nontangential limits. For ๐‘ฅ โˆˆ โ„๐‘‘ , the cone of aperture ๐œƒ > 0 over ๐‘ฅ is defined as the set ฮ“๐œƒ (๐‘ฅ) = {(๐‘ฆ, ๐‘ก) โˆˆ โ„๐‘‘+1 โˆถ |๐‘ฅ โˆ’ ๐‘ฆ| < ๐œƒ๐‘ก}. + See Figure 7.2. Thus, we consider the limit of ๐‘ข(๐‘ฆ, ๐‘ก) when we approach (๐‘ฅ, 0) within this cone. We first prove Theorem 7.13.

7.4. Boundary limits of harmonic functions

147

Figure 7.2. The cone ฮ“๐œƒ (๐‘ฅ) over the point ๐‘ฅ โˆˆ โ„๐‘‘ .

Theorem 7.13. For any ๐œƒ > 0, there exists ๐ด๐œƒ > 0 such that, for any ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘ฅ โˆˆ โ„๐‘‘ , |๐‘ข(๐‘ฆ, ๐‘ก)| โ‰ค ๐ด๐œƒ ๐‘€๐‘“(๐‘ฅ) for all (๐‘ฆ, ๐‘ก) โˆˆ ฮ“๐œƒ (๐‘ฅ). Proof. By Proposition 7.11, the theorem follows once we prove that there exists ๐‘ ๐œƒ > 0 such that, for all ๐‘ง โˆˆ โ„๐‘‘ , (7.14)

๐‘ƒ๐‘ก (๐‘ฆ โˆ’ ๐‘ง) โ‰ค ๐‘ ๐œƒ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง),

whenever (๐‘ฆ, ๐‘ก) โˆˆ ฮ“๐œƒ (๐‘ฅ). For this, fix ๐‘ฅ, ๐‘ง โˆˆ โ„๐‘‘ and (๐‘ฆ, ๐‘ก) โˆˆ ฮ“๐œƒ (๐‘ฅ). If |๐‘ฅ โˆ’ ๐‘ง| โ‰ฅ 2|๐‘ฅ โˆ’ ๐‘ฆ|, then 1 1 |๐‘ฆ โˆ’ ๐‘ง| = |๐‘ฅ โˆ’ ๐‘ง + ๐‘ฆ โˆ’ ๐‘ฅ| โ‰ฅ |๐‘ฅ โˆ’ ๐‘ง| โˆ’ |๐‘ฆ โˆ’ ๐‘ฅ| โ‰ฅ |๐‘ฅ โˆ’ ๐‘ง| โˆ’ |๐‘ฅ โˆ’ ๐‘ง| = |๐‘ฅ โˆ’ ๐‘ง|, 2 2 and hence 1 1 |๐‘ฆ โˆ’ ๐‘ง|2 + ๐‘ก2 โ‰ฅ |๐‘ฅ โˆ’ ๐‘ง|2 + ๐‘ก2 โ‰ฅ (|๐‘ฅ โˆ’ ๐‘ง|2 + ๐‘ก2 ), 4 4 so 2๐‘ก ๐‘ƒ๐‘ก (๐‘ฆ โˆ’ ๐‘ง) = ๐œ”๐‘‘+1 (|๐‘ฆ โˆ’ ๐‘ง|2 + ๐‘ก2 )(๐‘‘+1)/2 2๐‘ก = 2๐‘‘+1 ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ง). โ‰ค (๐‘‘+1)/2 2 2 ๐œ”๐‘‘+1 ((|๐‘ฅ โˆ’ ๐‘ง| + ๐‘ก )/4)

148

7. Maximal functions

If |๐‘ฅ โˆ’ ๐‘ง| < 2|๐‘ฅ โˆ’ ๐‘ฆ|, we have |๐‘ฅ โˆ’ ๐‘ง| < 2๐œƒ๐‘ก beacuse (๐‘ฆ, ๐‘ก) โˆˆ ฮ“๐œƒ (๐‘ฅ), and thus 1 1 1 1 |๐‘ฆ โˆ’ ๐‘ง|2 + ๐‘ก2 โ‰ฅ ๐‘ก2 = ๐‘ก2 + ๐‘ก2 โ‰ฅ 2 |๐‘ฅ โˆ’ ๐‘ง|2 + ๐‘ก2 โ‰ฅ ๐‘(|๐‘ฅ โˆ’ ๐‘ง|2 + ๐‘ก2 ), 2 2 2 8๐œƒ where ๐‘ = min{1/8๐œƒ2 , 1/2}, and thus, as above, ๐‘ƒ๐‘ก (๐‘ฆ โˆ’ ๐‘ง) โ‰ค

1

๐‘ƒ๐‘ก (๐‘ฅ ๐‘(๐‘‘+1)/2

โˆ’ ๐‘ง).

We obtain the inequality (7.14) with ๐‘ ๐œƒ = max{2๐‘‘+1 , 1/๐‘(๐‘‘+1)/2 }.

โ–ก

We then have the following result on the existence of nontangential limits. Corollary 7.15. Lef ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘ข(๐‘ฅ, ๐‘ก) its Poisson integral. Then, for any ๐œƒ > 0, lim ๐‘ข(๐‘ฅ, ๐‘ก) = ๐‘“(๐‘ฅ) (๐‘ฆ,๐‘ก)โ†’(๐‘ฅ,0) (๐‘ฆ,๐‘ก)โˆˆฮ“๐œƒ (๐‘ฅ)

for almost every ๐‘ฅ โˆˆ โ„๐‘‘ . We leave its proof as an exercise (Exercise (10)).

Exercises (1) The set ๐ฟ1loc (โ„๐‘‘ ) of locally integrable functions is a vector space. (2) If ๐‘“ โˆˆ ๐ฟ1loc (โ„๐‘‘ ), then ๐‘€๐‘“ might be infinite at every point. (3) If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ๐‘€๐‘“ might be infinite at some points. (4) Define on โ„ the function 1 ๐‘“(๐‘ฅ) = { |๐‘ฅ|(log |๐‘ฅ|)2 0

if |๐‘ฅ| โ‰ค 1/2 otherwise.

Then ๐‘“ โˆˆ ๐ฟ1 (โ„), but ๐‘€๐‘“ โˆ‰ ๐ฟ1loc (โ„). (5) Prove Theorem 7.12. (6) We say that {๐พ๐‘ก }๐‘ก>0 is a family of better kernels if it satisfies โ€ข โˆซ ๐พ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ = 1 for all ๐‘ก > 0; โ„๐‘‘

Exercises

149

โ€ข there exists ๐ด > 0 such that |๐พ๐‘ก (๐‘ฅ)| โ‰ค ๐‘ก > 0; and โ€ข there exists ๐ดโ€ฒ > 0 such that |๐พ๐‘ก (๐‘ฅ)| โ‰ค

๐ด for all ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘ก๐‘‘ ๐ดโ€ฒ ๐‘ก for all ๐‘ฅ โˆˆ โ„๐‘‘ |๐‘ฅ|๐‘‘+1

and ๐‘ก > 0. (a) If ฮฆ โˆˆ ๐ฟ1 (โ„๐‘‘ ), โˆซ ฮฆ = 1 and |ฮฆ(๐‘ฅ)| โ‰ค ๐ด/(1 + |๐‘ฅ|)๐‘‘+1 , then its dilations {ฮฆ๐‘ก }๐‘ก>0 form a family of better kernels. (b) If {๐พ๐‘ก }๐‘ก>0 is a family of better kernels, then it is a family of good kernels. (c) If {๐พ๐‘ก }๐‘ก>0 is a family of better kernels, then there exists a constant ๐‘ > 0 such that, if ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), |๐พ๐‘ก โˆ— ๐‘“(๐‘ฅ)| โ‰ค ๐‘๐‘€๐‘“(๐‘ฅ) for all ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘ก > 0. (d) If {๐พ๐‘ก }๐‘ก>0 is a family of better kernels and ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then lim ๐พ๐‘ก โˆ— ๐‘“(๐‘ฅ) = ๐‘“(๐‘ฅ) ๐‘กโ†’0

for almost every ๐‘ฅ โˆˆ โ„๐‘‘ . (7) The results of the previous exercise are still true if we change the third hypothesis by |๐พ๐‘ก (๐‘ฅ)| โ‰ค

๐ดโ€ฒ ๐‘ก๐œ€ , |๐‘ฅ|๐‘‘+๐œ€

for some ๐œ€ > 0. (8) If {๐พ๐‘ก }๐‘ก>0 is a family of better kernels and ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) is continuous at ๐‘ฅ, then lim ๐พ๐‘ก โˆ— ๐‘“(๐‘ฅ) = ๐‘“(๐‘ฅ). ๐‘กโ†’0

(Hint: Write ๐‘“ = ๐‘“ โ‹… ๐œ’๐ต๐›ฟ (๐‘ฅ) + ๐‘“ โ‹… (1 โˆ’ ๐œ’๐ต๐›ฟ (๐‘ฅ) ), where ๐ต๐›ฟ (๐‘ฅ) is a ball on which ๐‘“ is bounded.) (9) Let {๐พ๐‘ก }๐‘ก>0 be a family that satisfies the hypothesis of a family of better kernels except that, instead of the first hypothesis in Exercise (6), it satisfies, for some ๐œ† โˆˆ โ„‚, โˆซ ๐พ๐‘ก (๐‘ฅ)๐‘‘๐‘ฅ = ๐œ† โ„๐‘‘

for all ๐‘ก > 0. Then lim ๐พ๐‘ก โˆ— ๐‘“(๐‘ฅ) = ๐œ†๐‘“(๐‘ฅ) ๐‘กโ†’0

150

7. Maximal functions for almost every ๐‘ฅ โˆˆ โ„๐‘‘ .

(10) Prove Corollary 7.15. (11) Let {๐พ๐‘ก }๐‘ก>0 be a family of better kernels and ๐œƒ > 0. (a) There exists a constant ๐‘ ๐œƒ > 0 such that, for ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘ฅ โˆˆ โ„๐‘‘ , |๐พ๐‘ก โˆ— ๐‘“(๐‘ฆ)| โ‰ค ๐‘ ๐œƒ ๐‘€๐‘“(๐‘ฅ) for all (๐‘ฆ, ๐‘ก) โˆˆ ฮ“๐œƒ (๐‘ฅ). (b) If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then lim

(๐‘ฆ,๐‘ก)โ†’(๐‘ฅ,0) (๐‘ฆ,๐‘ก)โˆˆฮ“๐œƒ (๐‘ฅ)

๐พ๐‘ก โˆ— ๐‘“(๐‘ฆ) = ๐‘“(๐‘ฅ)

for almost every ๐‘ฅ โˆˆ โ„๐‘‘ .

Notes The maximal function was introduced by Hardy and Littlewood in [HL30], where they proved Theorem 7.4 for the 1-dimensional case. They also proved Theorem 7.11 in the same paper, for Poisson integrals in the circle. Lemma 7.6 appeared in the paper [Vit08] by Giuseppe Vitali, in the 1-dimensional case. The ๐‘‘-dimensional case, and the proof presented here, is due to Stephan Banach [Ban24]. Lebesgueโ€™s differentiation theorem 7.8 appeared in the one variable case in his book [Leb04]. A proof using Vitaliโ€™s lemma appeared in [Leb10], and the proof using the Hardyโ€“Littlewood maximal theorem is due to Riesz [Rie32]. Our discussion on nontangential limits can be found in [Ste70].

Chapter 8

Fourier transform

8.1. Integrable functions In this chapter we discuss the representation of a function on โ„๐‘‘ in terms of its Fourier transform, analogously to the Fourier series representation of a function in a circle discussed in previous chapters. The Fourier transform of a function ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) is defined as ฬ‚ = โˆซ ๐‘“(๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ, ๐‘“(๐œ‰)

(8.1)

โ„๐‘‘

for each ๐œ‰ โˆˆ โ„๐‘‘ . The integral above converges since ๐‘“, and thus the function ๐‘ฅ โ†ฆ ๐‘“(๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ , is integrable for each ๐œ‰. 8.2. For any ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), ๐‘“ ฬ‚ is a bounded continuous function. Indeed, ฬ‚ |๐‘“(๐œ‰)| โ‰ค โˆซ |๐‘“(๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ |๐‘‘๐‘ฅ = โ€–๐‘“โ€–๐ฟ1 , โ„๐‘‘

and, if ๐œ‰ โ†’ ๐œ‰0 , then ฬ‚ โˆ’ ๐‘“(๐œ‰ ฬ‚ 0 )| โ‰ค โˆซ |๐‘“(๐‘ฅ)| |๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ โˆ’ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰0 |๐‘‘๐‘ฅ โ†’ 0, |๐‘“(๐œ‰) โ„๐‘‘

by the dominated convergence theorem. 8.3. The Fourier transform defines on ๐ฟ1 (โ„๐‘‘ ) a linear operator, i.e. 151

152

8. Fourier transform ห† ฬ‚ โ€ข for each ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐›ผ โˆˆ โ„‚, ๐›ผ๐‘“(๐œ‰) = ๐›ผ๐‘“(๐œ‰); and 1 ๐‘‘ ฬ‚ + ๐‘”(๐œ‰). โ€ข for ๐‘“, ๐‘” โˆˆ ๐ฟ (โ„ ), ๐‘“ห† + ๐‘”(๐œ‰) = ๐‘“(๐œ‰) ฬ‚

By 8.2, this operator continuously maps ๐ฟ1 (โ„๐‘‘ ) into the space ๐ถ๐ต (โ„๐‘‘ ) of bounded continuous functions with the uniform norm. Example 8.4. If ๐‘“ = ๐œ’[โˆ’1,1] in โ„, then, for ๐œ‰ โ‰  0, 1

2๐œ‹๐‘–๐œ‰ sin(2๐œ‹๐œ‰) โˆ’ ๐‘’โˆ’2๐œ‹๐‘–๐œ‰ ฬ‚ = โˆซ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ = ๐‘’ ๐‘“(๐œ‰) = . 2๐œ‹๐‘–๐œ‰ ๐œ‹๐œ‰ โˆ’1

In particular, ๐‘“ ฬ‚ โˆ‰ ๐ฟ1 (โ„๐‘‘ ). Although the Fourier transform of a function ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) is not necessarily an integrable function, as in Example 8.4, we observe that, in this ฬ‚ example, ๐‘“(๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž. This property is true for all functions ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ). ฬ‚ Proposition 8.5 (Riemannโ€“Lebesgue lemma). If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ๐‘“(๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž. Proof. One can check explicitly that, if ๐‘… is a rectangle in โ„๐‘‘ , then ๐œ’ ห†๐‘… (๐œ‰) โ†’ 0

as |๐œ‰| โ†’ โˆž.

(Exercise (2)). If ๐ด โŠ‚ โ„๐‘‘ is a finite measure set, by Corollary 5.17(4), for any ๐œ€ > 0 there exist cubes ๐‘„1 , . . . , ๐‘„๐‘› such that |๐ดโ–ณ โ‹ƒ ๐‘„๐‘— | < ๐œ€, so โ€–๐œ’๐ด โˆ’ ๐œ’โ‹ƒ ๐‘„๐‘— โ€–๐ฟ1 < ๐œ€ and thus |ห† ๐œ’๐ด (๐œ‰) โˆ’ ๐œ’ห† โ‹ƒ ๐‘„๐‘— (๐œ‰)| < ๐œ€. Since we can write ๐‘

๐œ’โ‹ƒ ๐‘„๐‘— = โˆ‘ ๐œ’๐‘…๐‘— , ๐‘—=1

where each ๐‘…๐‘— is a rectangle, we have that ๐œ’ห† โ‹ƒ ๐‘„๐‘— (๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž, and thus ๐œ’ ห† ๐ด (๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž beacuse ๐œ€ > 0 above is arbitrary. We ฬ‚ thus have that, for a simple ๐œ™ โˆˆ ๐ฟ1 (โ„๐‘‘ ), ๐œ™(๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž, because an integrable simple function is the linear combination of characteristic functions of finite measure sets. We thus get the proposition for every ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), because the simple functions are dense in ๐ฟ1 (โ„๐‘‘ ). โ–ก The Riemannโ€“Lebesgue lemma implies that the Fourier transform defines a linear operator on ๐ฟ1 (โ„๐‘‘ ) into ๐ถ0 (โ„๐‘‘ ).

8.1. Integrable functions

153

8.6. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), the Fourier transform of its dilation ๐‘“๐‘ก (๐‘ฅ) = ๐‘กโˆ’๐‘‘ ๐‘“(๐‘ฅ/๐‘ก), for ๐‘ก > 0, is given by ฬ‚ ๐‘“ห†๐‘ก (๐œ‰) = ๐‘“(๐‘ก๐œ‰). This follows from the dilation property of Lebesgue measure. Indeed, 1 ๐‘“ห†๐‘ก (๐œ‰) = โˆซ ๐‘“๐‘ก (๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = ๐‘‘ โˆซ ๐‘“(๐‘ฅ/๐‘ก)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ ๐‘ก โ„๐‘‘ โ„๐‘‘ ฬ‚ = โˆซ ๐‘“(๐‘ฆ)๐‘’โˆ’2๐œ‹๐‘–๐‘ก๐‘ฆโ‹…๐œ‰ ๐‘‘๐‘ฆ = ๐‘“(๐‘ก๐œ‰), โ„๐‘‘

where we have applied the dilation ๐‘ฅ โ†ฆ ๐‘ก๐‘ฆ. 8.7. The translation invariance of Lebesgue measure on โ„๐‘‘ further implies, for ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), โ€ข if ๐‘“โ„Ž (๐‘ฅ) = ๐‘“(๐‘ฅ โˆ’ โ„Ž) for some โ„Ž โˆˆ โ„๐‘‘ , then ห†โ„Ž (๐œ‰) = ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…โ„Ž ๐‘“(๐œ‰); ฬ‚ ๐‘“ and โ€ข if ๐‘”(๐‘ฅ) = ๐‘’2๐œ‹๐‘–๐‘ฅโ‹…โ„Ž ๐‘“(๐‘ฅ) for some โ„Ž โˆˆ โ„๐‘‘ , then ฬ‚ โˆ’ โ„Ž). ๐‘”(๐œ‰) ฬ‚ = ๐‘“(๐œ‰ We leave the proof of these properties as an exercise (Exercise (3)). 2

Example 8.8. Consider the function ๐บ(๐‘ฅ) = ๐‘’โˆ’๐œ‹|๐‘ฅ| in โ„๐‘‘ . Then 2 ฬ‚ ๐บ(๐œ‰) = ๐‘’โˆ’๐œ‹|๐œ‰| = ๐บ(๐œ‰),

so ๐บ is equal to its own Fourier transform. To prove this by Fubiniโ€™s theorem it is enough to show that โˆž 2

โˆซ ๐‘’โˆ’๐œ‹๐‘ฅ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ = ๐‘’โˆ’๐œ‹๐œ‰

2

โˆ’โˆž 2

2

2

2

for each ๐œ‰ โˆˆ โ„, because ๐‘’โˆ’๐œ‹|๐‘ฅ| = ๐‘’โˆ’๐œ‹๐‘ฅ1 ๐‘’โˆ’๐œ‹๐‘ฅ2 โ‹ฏ ๐‘’โˆ’๐œ‹๐‘ฅ๐‘‘ . Since โˆž

๐‘ 2

2

โˆซ ๐‘’โˆ’๐œ‹๐‘ฅ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ = lim โˆซ ๐‘’โˆ’๐œ‹๐‘ฅ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ ๐‘โ†’โˆž

โˆ’โˆž

โˆ’๐‘ ๐‘

2

2

= ๐‘’โˆ’๐œ‹๐œ‰ lim โˆซ ๐‘’โˆ’๐œ‹(๐‘ฅ+๐‘–๐œ‰) ๐‘‘๐‘ฅ, ๐‘โ†’โˆž

โˆ’๐‘

it remains to verify that ๐‘

๐‘ 2

2

lim โˆซ ๐‘’โˆ’๐œ‹(๐‘ฅ+๐‘–๐œ‰) ๐‘‘๐‘ฅ = lim โˆซ ๐‘’โˆ’๐œ‹๐‘ฅ ๐‘‘๐‘ฅ,

๐‘โ†’โˆž

โˆ’๐‘

๐‘โ†’โˆž

โˆ’๐‘

154

8. Fourier transform

as we know that the integral on the right hand side is equal to 1. The above identity is achieved by considering the contour integral 2

โˆซ ๐‘’โˆ’๐œ‹๐‘ง ๐‘‘๐‘ง = 0 ๐›พ

over the rectangle ๐›พ with vertices ๐‘, ๐‘ + ๐‘–๐œ‰, โˆ’๐‘ + ๐‘–๐œ‰ and โˆ’๐‘. See Figure 8.1. We leave the details as an exercise (Exercise (4)).

Figure 8.1. The contour ๐›พ to obtain the Fourier transform of the function ๐บ(๐‘ฅ).

8.9. The function ๐บ(๐‘ฅ) of Example 8.8 is called the Gaussian kernel, and is related to the heat kernel introduced in Exercise (10) of Chapter 4, ๐ป๐‘ก (๐‘ฅ) =

1 2 ๐‘’โˆ’|๐‘ฅ| /4๐‘ก . ๐‘‘/2 (4๐œ‹๐‘ก)

Note that ๐ป๐‘ก correspond to the dilation ๐บโˆš4๐œ‹๐‘ก (๐‘ฅ) of the Gaussian. By the dilation property 8.6 of the Fourier transform we have that ห†๐‘ก (๐œ‰) = ๐‘’โˆ’4๐œ‹2 ๐‘ก|๐œ‰|2 . ๐ป Example 8.10. Consider now, for a fixed ๐‘ก > 0, the Poisson kernel ๐‘ƒ๐‘ก (๐‘ฅ). Then ห†๐‘ก (๐œ‰) = ๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| . ๐‘ƒ To show this, recall that ๐‘ƒ๐‘ก is the dilation of ๐‘ƒ1 , so it is sufficient to prove that ๐‘ƒห†1 (๐œ‰) = ๐‘’โˆ’2๐œ‹|๐œ‰| . We calculate ฮ“((๐‘‘ + 1)/2) ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ โˆซ ๐‘ƒห†1 (๐œ‰) = โˆซ ๐‘ƒ1 (๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = ๐‘‘๐‘ฅ, (๐‘‘+1)/2 (๐‘‘+1)/2 2 ๐œ‹ โ„๐‘‘ โ„๐‘‘ (|๐‘ฅ| + 1)

8.1. Integrable functions

155

where we have used the fact 2๐œ‹(๐‘‘+1)/2 . ฮ“((๐‘‘ + 1)/2)

๐œ”๐‘‘+1 = Now, by the identity

โˆž

1 1 ๐‘‘๐‘  2 โˆซ ๐‘ (๐‘‘+1)/2 ๐‘’โˆ’(|๐‘ฅ| +1)๐‘  , = (๐‘‘+1)/2 2 ๐‘  ฮ“((๐‘‘ + 1)/2) 0 (|๐‘ฅ| + 1) (see Exercise (2) of Chapter 4) and Fubiniโ€™s theorem we obtain ๐‘ƒห†1 (๐‘ฅ) = =

โˆž

1 ๐œ‹(๐‘‘+1)/2

โˆซ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ( โˆซ ๐‘ (๐‘‘+1)/2 ๐‘’โˆ’(|๐‘ฅ| โ„๐‘‘ โˆž

2 +1)๐‘ 

0

๐‘‘๐‘  )๐‘‘๐‘ฅ ๐‘ 

1 2 โˆซ ๐‘ (๐‘‘โˆ’1)/2 ๐‘’โˆ’๐‘  ( โˆซ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘’โˆ’|๐‘ฅ| ๐‘  ๐‘‘๐‘ฅ)๐‘‘๐‘ . ๐œ‹(๐‘‘+1)/2 0 ๐‘‘ โ„

From 8.8 and the dilation ๐‘ฅ โ†ฆ (โˆš๐œ‹/๐‘ )๐‘ฆ, the integral inside is 2

โˆซ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘’โˆ’|๐‘ฅ| ๐‘  ๐‘‘๐‘ฅ = โ„๐‘‘

and hence

โˆž

1 โˆซ ๐‘ƒห†1 (๐‘ฅ) = โˆš๐œ‹ 0

1 โˆš๐‘ 

๐œ‹๐‘‘/2 โˆ’๐œ‹2 |๐œ‰|2 /๐‘  ๐‘’ , ๐‘ ๐‘‘/2

๐‘’โˆ’๐‘  ๐‘’โˆ’๐œ‹

2 |๐œ‰|2 /๐‘ 

๐‘‘๐‘ .

We calculate the last integral in two steps: (I) For any ๐‘ข > 0, โˆž

โˆž

๐‘’โˆ’2๐œ‹๐‘–แต†๐‘ฃ 1 1 โˆซ โˆซ ๐‘‘๐‘ฃ = ๐œ‹ โˆ’โˆž 1 + ๐‘ฃ2 โˆš๐œ‹ 0

1 โˆš๐‘ 

๐‘’โˆ’๐‘  ๐‘’โˆ’๐œ‹

2 แต†2 /๐‘ 

(II) For any ๐‘ข > 0, โˆž

1 ๐‘’โˆ’2๐œ‹๐‘–แต†๐‘ฃ โˆซ ๐‘‘๐‘ฃ = ๐‘’โˆ’2๐œ‹แต† . ๐œ‹ โˆ’โˆž 1 + ๐‘ฃ2 Step (I) follows from the identities โˆž

1 2 = โˆซ ๐‘’โˆ’(1+๐‘ฃ )๐‘  ๐‘‘๐‘  1 + ๐‘ฃ2 0 and

โˆž 2

โˆซ ๐‘’โˆ’2๐œ‹๐‘–แต†๐‘ฃ ๐‘’โˆ’๐‘ฃ ๐‘  ๐‘‘๐‘ฃ = โˆ’โˆž

๐œ‹ โˆ’๐œ‹2 แต†2 /๐‘  ๐‘’ . โˆš๐‘ 

๐‘‘๐‘ ;

156

8. Fourier transform

The first is a straightforward calculation, while the second follows from Example 8.8 in the 1-dimensional case and the dilation property 8.6 of the Fourier transform. For Step (II), we calculate the contour integral โˆซ ๐›พ

๐‘’โˆ’2๐œ‹๐‘–แต†๐‘ง ๐‘‘๐‘ง 1 + ๐‘ง2

over the lower semicircle around the origin of radius ๐‘, and we let ๐‘ โ†’ โˆž. See Figure 8.2. We leave the details of Steps (I) and (II) as an exercise (Exercise (5)).

Figure 8.2. The contour ๐›พ to obtain the Fourier transform of the Poisson kernel ๐‘ƒ1 (๐‘ฅ). Note that the function ๐‘ง โ†ฆ ๐‘’โˆ’2๐œ‹๐‘–แต†๐‘ง /(1 + ๐‘ง2 ) has a pole at โˆ’๐‘–.

8.2. The Fourier inversion formula A natural question to ask is whether one can recover a function from its Fourier transform. Recall that the expansion formula ๐‘–๐‘›๐œƒ ฬ‚ ๐‘“(๐œƒ) = โˆ‘ ๐‘“(๐‘›)๐‘’ ๐‘›โˆˆโ„ค

of a function on the circle in terms of its Fourier series holds for appropiate continuous functions. In fact, it holds whenever the series ฬ‚ โˆ‘ |๐‘“(๐‘›)| ๐‘›โˆˆโ„ค

converges, so the Fourier series of ๐‘“ is absolutely convergent.

8.2. The Fourier inversion formula

157

Example 8.8, as well as the convergence of the Fourier series above, suggests that, if ๐‘“ ฬ‚ is integrable, then we should have the following Fourier inversion formula, 2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ ๐‘“(๐‘ฅ) = โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰.

(8.11)

โ„๐‘‘

It is indeed the case. Theorem 8.12. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) as well as ๐‘“ ฬ‚ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then the Fourier inversion formula (8.11) holds for almost every ๐‘ฅ โˆˆ โ„๐‘‘ . In the proof of Theorem 8.12 we will use Lemma 8.13. Lemma 8.13. If ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), then โˆซ ๐‘“๐‘”ฬ‚ = โˆซ ๐‘“๐‘”.ฬ‚ โ„๐‘‘

โ„๐‘‘

Proof. We first observe that both integrals converge because ๐‘“ and ๐‘” are integrable, and both ๐‘“ ฬ‚ and ๐‘”ฬ‚ are continuous and bounded. Also, if we define ๐น(๐‘ฅ, ๐‘ฆ) = ๐‘“(๐‘ฅ)๐‘”(๐‘ฆ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐‘ฆ ๐‘‘ ๐‘‘ on โ„ ร— โ„ , then ๐น is integrable, because โˆซ

|๐น| โ‰ค โˆซ |๐‘“(๐‘ฅ)|๐‘‘๐‘ฅ โˆซ |๐‘”(๐‘ฆ)|๐‘‘๐‘ฆ < โˆž.

โ„๐‘‘ ร—โ„๐‘‘

โ„๐‘‘

โ„๐‘‘

Thus, by Fubiniโ€™s theorem, โˆซ โ„๐‘‘ ร—โ„๐‘‘

๐น = โˆซ ( โˆซ ๐น(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฅ)๐‘‘๐‘ฆ โ„๐‘‘

โ„๐‘‘

= โˆซ ( โˆซ ๐‘“(๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐‘ฆ ๐‘‘๐‘ฅ)๐‘”(๐‘ฆ)๐‘‘๐‘ฆ โ„๐‘‘

โ„๐‘‘

ฬ‚ = โˆซ ๐‘“(๐‘ฆ)๐‘”(๐‘ฆ)๐‘‘๐‘ฆ, โ„๐‘‘

and โˆซ โ„๐‘‘ ร—โ„๐‘‘

๐น = โˆซ ( โˆซ ๐น(๐‘ฅ, ๐‘ฆ)๐‘‘๐‘ฆ)๐‘‘๐‘ฅ โ„๐‘‘

โ„๐‘‘

= โˆซ ๐‘“(๐‘ฅ)( โˆซ ๐‘”(๐‘ฆ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐‘ฆ ๐‘‘๐‘ฅ)๐‘‘๐‘ฅ โ„๐‘‘

โ„๐‘‘

= โˆซ ๐‘“(๐‘ฅ)๐‘”(๐‘ฅ)๐‘‘๐‘ฅ. ฬ‚ โ„๐‘‘

158

8. Fourier transform โ–ก

Proof of Theorem 8.12. For each ๐‘ฅ โˆˆ โ„๐‘‘ and ๐‘ก > 0, consider the function 2 ๐‘”(๐œ‰) = ๐‘’โˆ’๐œ‹๐‘ก|๐œ‰| ๐‘’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ . By 8.6, 8.7 and 8.8, its Fourier transform is the Gaussian kernel ๐‘”(๐‘ฆ) ฬ‚ =

1 ๐‘ก๐‘‘/2

2

๐‘’โˆ’๐œ‹|๐‘ฅโˆ’๐‘ฆ| /๐‘ก .

The collection {๐พ๐‘ก }๐‘ก>0 given by 1 โˆ’๐œ‹|๐‘ฅ|2 /๐‘ก ๐‘’ ๐‘ก๐‘‘/2 is a collection of better kernels (see Exercise (6) of Chapter 7; see also Exercise (6) of this chapter), because it is the collection of dilations of the Gaussian kernel seen in 8.8. By Lemma 8.13 with ๐‘“ and ๐‘”, we obtain ๐พ๐‘ก (๐‘ฅ) =

(8.14)

โˆ’๐œ‹๐‘ก|๐œ‰|2 2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘’ ๐‘‘๐œ‰ = โˆซ ๐‘“(๐‘ฆ)๐พ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ, โ„๐‘‘

โ„๐‘‘

where we have again used 8.7. Since ๐‘“ ฬ‚ โˆˆ ๐ฟ1 (โ„๐‘‘ ), the left side of (8.14) converges to 2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰ โ„๐‘‘

as ๐‘ก โ†’ 0, by the dominated convergence theorem. Since the collection {๐พ๐‘ก }๐‘ก>0 is a collection of better kernels, โˆซ ๐‘“(๐‘ฆ)๐พ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘‘๐‘ฆ = ๐‘“ โˆ— ๐พ๐‘ก (๐‘ฅ) โ†’ ๐‘“(๐‘ฅ) โ„๐‘‘

for almost every ๐‘ฅ โˆˆ โ„๐‘‘ , by Exercise (6) of Chapter 7.

โ–ก

8.15. Note that, in particular, if ๐‘“ and ๐‘“ ฬ‚ are integrable and ๐‘“ is continuous at ๐‘ฅ โˆˆ โ„๐‘‘ , then 2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ ๐‘“(๐‘ฅ) = โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰. โ„๐‘‘

(Exercise (8) of Chapter 7.) Example 8.16. Since the Fourier transform of the Poisson kernel ๐‘ƒ๐‘ก (๐‘ฅ) is given by ห†๐‘ก (๐œ‰) = ๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| , ๐‘ƒ

8.3. Mean-square convergence

159

ห†๐‘ก is integrable, we have that by 8.10, and clearly ๐‘ƒ โˆซ ๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| ๐‘’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐œ‰ = ๐‘ƒ๐‘ก (๐‘ฅ), โ„๐‘‘

for every ๐‘ฅ โˆˆ โ„๐‘‘ . Moreover, since both ๐‘ƒ๐‘ก (๐‘ฅ) = ๐‘ƒ๐‘ก (โˆ’๐‘ฅ) and ๐‘’โˆ’2๐œ‹๐‘ก|๐‘ฅ| = ๐‘’โˆ’2๐œ‹๐‘ก|โˆ’๐‘ฅ| , we see that the Fourier transform of ๐‘’โˆ’2๐œ‹๐‘ก|๐‘ฅ| is ๐‘ƒ๐‘ก (๐œ‰), as well. Theorem 8.12 leads to the question of whether the Fourier transform of a given function is integrable. In general, we have seen that the Fourier transform is continuous and has limit 0 at infinity but, in general, itโ€™s not a function in ๐ฟ1 (โ„๐‘‘ ), as the simple example of a characteristic function shows (Example 8.4). However, an application of the limit results above and the monotone convergence theorem give us the following test. ฬ‚ Proposition 8.17. Let ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) be continuous at 0. If ๐‘“(๐œ‰) โ‰ฅ 0 for ๐‘‘ 1 ๐‘‘ ฬ‚ every ๐œ‰ โˆˆ โ„ , then ๐‘“ โˆˆ ๐ฟ (โ„ ). Proof. From equation (8.14) (which is true for all ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ )) and the hypothesis that ๐‘“ is continuous at 0, we have that โˆ’๐œ‹๐‘ก|๐œ‰|2 ฬ‚ lim โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰ = ๐‘“(0). ๐‘กโ†’0

โ„๐‘‘

ฬ‚ โ‰ฅ 0, we clearly have that Also, since ๐‘“(๐œ‰) โˆ’๐œ‹๐‘ก|๐œ‰|2 ฬ‚ ฬ‚ ๐‘“(๐œ‰)๐‘’ โ†— ๐‘“(๐œ‰)

as ๐‘ก โ†’ 0 so, by the monotone convergence theorem, โˆ’๐œ‹๐‘ก|๐œ‰|2 ฬ‚ ฬ‚ โˆซ ๐‘“(๐œ‰)๐‘‘๐œ‰ = lim โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰. โ„๐‘‘

๐‘กโ†’0

โ„๐‘‘

Therefore, ๐‘“ ฬ‚ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and, in fact, โ€–๐‘“โ€–ฬ‚ ๐ฟ1 = ๐‘“(0).

โ–ก

8.3. Mean-square convergence The Fourier transform is only defined for ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), as we require that the integral (8.1) converges. However, it is possible to extend the definition of the Fourier transform to functions in ๐ฟ2 (โ„๐‘‘ ), and even obtain its inverse.

160

8. Fourier transform

For this, we will use the fact that the set ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) is dense in ๐ฟ (โ„๐‘‘ ): for any ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), there exists a sequence ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) that converges to ๐‘“ in ๐ฟ2 (โ„๐‘‘ ), that is 2

โˆซ |๐‘“ โˆ’ ๐‘“๐‘› |2 โ†’ 0. โ„๐‘‘

This follows from the fact that both ๐ฟ1 (โ„๐‘‘ ) and ๐ฟ2 (โ„๐‘‘ ) contain the space ๐ถ๐‘ (โ„๐‘‘ ) of continuous functions of compact support, and that ๐ถ๐‘ (โ„๐‘‘ ) is dense in both ๐ฟ1 (โ„๐‘‘ ) and ๐ฟ2 (โ„๐‘‘ ), as we discussed in Chapter 6. We have Theorem 8.18. Theorem 8.18. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ), then โˆซ |๐‘“|2 = โˆซ |๐‘“|ฬ‚ 2 . โ„๐‘‘

โ„๐‘‘

That is, if โŸจโ‹…, โ‹…โŸฉ is the inner product in ๐ฟ2 (โ„๐‘‘ ), then the operator ๐‘“ โ†ฆ ๐‘“ฬ‚ is an isometry on the subset ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) of ๐ฟ2 (โ„๐‘‘ ), so โ€–๐‘“โ€–๐ฟ2 = โ€–๐‘“โ€–ฬ‚ ๐ฟ2 . By the polarization identity, we also have โŸจ๐‘“, ๐‘”โŸฉ = โŸจ๐‘“,ฬ‚ ๐‘”โŸฉฬ‚

(8.19)

for any ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ )โˆฉ๐ฟ2 (โ„๐‘‘ ) (Exercise (7)). See Section A.5 for a review of results of inner product spaces. Proof. For the proof of Theorem 8.18, we will use the identity ห† ฬ‚ ๐‘”(๐œ‰), ๐‘“ โˆ— ๐‘”(๐œ‰) = ๐‘“(๐œ‰) ฬ‚ for any ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), which follows from Fubiniโ€™s theorem applied to the function ๐น(๐‘ฅ, ๐‘ฆ) = ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘”(๐‘ฆ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ (Exercise (8)). In particular, if we take ๐‘”(๐‘ฅ) = ๐‘“(โˆ’๐‘ฅ) and โ„Ž = ๐‘“ โˆ— ๐‘”, then ฬ‚ ๐‘”(๐œ‰) ฬ‚ 2, ฬ‚ = ๐‘“(๐œ‰) โ„Ž(๐œ‰) ฬ‚ = |๐‘“(๐œ‰)| because ฬ‚ ๐‘”(๐œ‰) ฬ‚ = โˆซ ๐‘“(โˆ’๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = โˆซ ๐‘“(๐‘ฅ)๐‘’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = ๐‘“(๐œ‰). โ„๐‘‘

โ„๐‘‘

8.3. Mean-square convergence

161

Now, โ„Ž is a continuous function because, by the Cauchyโ€“Schwarz inequality, |โ„Ž(๐‘ฅ) โˆ’ โ„Ž(๐‘ฆ)| = |๐‘“ โˆ— ๐‘”(๐‘ฅ) โˆ’ ๐‘“ โˆ— ๐‘”(๐‘ฆ)| = || โˆซ (๐‘“(๐‘ฅ โˆ’ ๐‘ง) โˆ’ ๐‘“(๐‘ฆ โˆ’ ๐‘ง))๐‘”(๐‘ง)๐‘‘๐‘ง|| โ„๐‘‘

โ‰ค โ€–๐‘“(๐‘ฅ โˆ’ โ‹…) โˆ’ ๐‘“(๐‘ฆ โˆ’ โ‹…)โ€–๐ฟ2 โ€–๐‘”โ€–๐ฟ2 goes to 0 as ๐‘ฆ โ†’ ๐‘ฅ, by the continuity of translations in the ๐ฟ2 norm (6.49). Then, โ„Ž is continuous at 0 and โ„Ž ฬ‚ โ‰ฅ 0, so by Proposition 8.17 we have that โ„Ž ฬ‚ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and โ„Ž(0) = โˆซ โ„Ž.ฬ‚ โ„๐‘‘

Therefore โˆซ |๐‘“|ฬ‚ 2 = โˆซ โ„Ž ฬ‚ = โ„Ž(0) = ๐‘“ โˆ— ๐‘”(0) = โˆซ ๐‘“(โˆ’๐‘ฅ)๐‘”(๐‘ฅ)๐‘‘๐‘ฅ โ„๐••

โ„๐‘‘

โ„๐••

= โˆซ ๐‘“(โˆ’๐‘ฅ)๐‘“(โˆ’๐‘ฅ)๐‘‘๐‘ฅ = โˆซ |๐‘“|2 . โ„๐••

โ„๐••

โ–ก Theorem 8.18 allows us to extend the definition of the Fourier transform to any function in ๐ฟ2 (โ„๐‘‘ ). Indeed, if ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), as we have seen above, there exists a sequence ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) such that ๐‘“๐‘› โ†’ ๐‘“ in ๐ฟ2 (โ„๐‘‘ ). Now the sequence of Fourier transforms ๐‘“๐‘›ฬ‚ is a Cauchy sequence in ๐ฟ2 (โ„๐‘‘ ), as โ€–๐‘“๐‘›ฬ‚ โˆ’ ๐‘“๐‘šฬ‚ โ€–๐ฟ2 = โ€–๐‘“๐‘› โˆ’ ๐‘“๐‘š โ€–๐ฟ2 and the fact that ๐‘“๐‘› converges in ๐ฟ2 (โ„๐‘‘ ). ๐ฟ2 (โ„๐‘‘ ) is complete, so ๐‘“๐‘›ฬ‚ converges in ๐ฟ2 (โ„๐‘‘ ) and we can define โ„ฑ๐‘“ = lim ๐‘“๐‘›ฬ‚ .

(8.20)

The operator โ„ฑ โˆถ ๐ฟ2 (โ„๐‘‘ ) โ†’ ๐ฟ2 (โ„๐‘‘ ) is well defined because, if ๐‘“๐‘› โ†’ ๐‘“ and ๐‘”๐‘› โ†’ ๐‘“ in ๐ฟ2 (โ„๐‘‘ ), then โ€–๐‘“๐‘›ฬ‚ โˆ’ ๐‘”๐‘›ฬ‚ โ€–๐ฟ2 = โ€–๐‘“๐‘› โˆ’ ๐‘”๐‘› โ€–๐ฟ2 โ†’ 0, so lim ๐‘“๐‘›ฬ‚ = lim ๐‘”๐‘›ฬ‚ 2

๐‘‘

in ๐ฟ (โ„ ). It is also an isometry, as clearly (8.21)

โ€–โ„ฑ๐‘“โ€–๐ฟ2 = โ€–๐‘“โ€–๐ฟ2

162

8. Fourier transform

for all ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ). Theorem 8.22. The operator โ„ฑ โˆถ ๐ฟ2 (โ„๐‘‘ ) โ†’ ๐ฟ2 (โ„๐‘‘ ) is a unitary operator in ๐ฟ2 (โ„๐‘‘ ) with inverse โ„ฑ โˆ’1 ๐‘”(๐‘ฅ) = โ„ฑ๐‘”(โˆ’๐‘ฅ). A unitary operator is a surjective isometry. Theorem 8.22 is commonly known as Plancherelโ€™s theorem. Proof. We first observe that โ„ฑ has closed range. If โ„ฑ๐‘“๐‘› โ†’ ๐‘” in ๐ฟ2 (โ„๐‘‘ ), then โ€–๐‘“๐‘› โˆ’ ๐‘“๐‘š โ€–๐ฟ2 = โ€–โ„ฑ๐‘“๐‘› โˆ’ โ„ฑ๐‘“๐‘š โ€–๐ฟ2 because โ„ฑ is an isometry. Thus ๐‘“๐‘› is Cauchy in ๐ฟ2 (โ„๐‘‘ ), so it converges, say ๐‘“๐‘› โ†’ ๐‘“ in ๐ฟ2 (โ„๐‘‘ ). Hence โ„ฑ๐‘“๐‘› โ†’ โ„ฑ๐‘“, and ๐‘” = โ„ฑ๐‘“. Let โ„ณ be the range of โ„ฑ. If โ„ณ โ‰  ๐ฟ2 (โ„๐‘‘ ), the orthogonal complement to โ„ณ would be nontrivial (see Appendix A.5), so there would be a nonzero ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ) so that โŸจโ„ฑ๐‘“, ๐‘”โŸฉ = 0 2

๐‘‘

for every ๐‘“ โˆˆ ๐ฟ (โ„ ), which implies โˆซ(โ„ฑ๐‘“)๐‘” = 0 for all ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ). Now, the density of ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) and Lemma 8.13 imply that โˆซ โ„ฑ๐‘“ ๐‘” = โˆซ ๐‘“ โ„ฑ๐‘” โ„๐‘‘

โ„๐‘‘

(Exercise (10)), so โˆซ ๐‘“ โ„ฑ๐‘” = 0 โ„๐‘‘ 2

๐‘‘

for every ๐‘“ โˆˆ ๐ฟ (โ„ ), and thus, choosing ๐‘“ = โ„ฑ๐‘”, we obtain โˆซ |โ„ฑ๐‘”|2 = 0. โ„๐‘‘

But then โ€–๐‘”โ€–๐ฟ2 = โ€–โ„ฑ๐‘”โ€–๐ฟ2 = 0, contradicting the fact that ๐‘” โ‰  0 in ๐ฟ2 (โ„๐‘‘ ). Thus โ„ณ = ๐ฟ2 (โ„๐‘‘ ) and โ„ฑ is surjective. In order to show that โ„ฑ โˆ’1 ๐‘”(๐‘ฅ) = โ„ฑ๐‘”(โˆ’๐‘ฅ), we first note that, for any ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), ฬ„ โ„ฑ๐‘“(๐œ‰) = โ„ฑ ๐‘“(โˆ’๐œ‰)

8.3. Mean-square convergence

163

ฬ„ = ๐‘“(๐‘ฅ). Indeed, if ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) and where ๐‘“ ฬ„ is the function ๐‘“(๐‘ฅ) 2 ๐‘‘ ๐‘“๐‘› โ†’ ๐‘“ in ๐ฟ (โ„ ), then ๐‘“๐‘›ฬ‚ (๐œ‰) = โˆซ ๐‘“๐‘› (๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = โˆซ ๐‘“๐‘› (๐‘ฅ)๐‘’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ = ๐‘“๐‘›ฬ‚ฬ„ (โˆ’๐œ‰), โ„๐••

โ„๐••

ฬ„ so ๐‘“๐‘›ฬ‚ (๐œ‰) โ†’ โ„ฑ ๐‘“(โˆ’๐œ‰) in ๐ฟ2 (โ„๐‘‘ ). Then, for any ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), if we set โ„Ž(๐‘ฅ) = โ„ฑ๐‘”(โˆ’๐‘ฅ), ฬ„ โŸจโ„ฑโ„Ž, ๐‘“โŸฉ = โˆซ โ„ฑโ„Ž(๐‘ฅ)๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โˆซ โ„Ž(๐‘ฅ)โ„ฑ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ โ„๐••

โ„๐••

ฬ„ ฬ„ = โˆซ โ„ฑ๐‘”(๐‘ฅ)โ„ฑ ๐‘“(โˆ’๐‘ฅ)๐‘‘๐‘ฅ = โˆซ โ„ฑ๐‘”(โˆ’๐‘ฅ)โ„ฑ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ โ„๐••

โ„๐••

= โˆซ โ„ฑ๐‘”(๐‘ฅ)โ„ฑ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ = โŸจโ„ฑ๐‘”, โ„ฑ๐‘“โŸฉ = โŸจ๐‘”, ๐‘“โŸฉ, โ„๐••

where in the last identiy we have used the fact that โ„ฑ is an isometry. Since ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ) is arbitrary, we have shown that โ„Ž = โ„ฑ โˆ’1 ๐‘”. โ–ก For a function ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), we will write its Fourier transform โ„ฑ๐‘“ simply as ๐‘“.ฬ‚ 8.23. If ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ), then ๐‘“๐‘ = ๐‘“๐œ’๐ต๐‘ โˆˆ ๐ฟ1 (โ„๐‘‘ ), where ๐œ’๐ต๐‘ is the characteristic function of the ball ๐ต๐‘ of radius ๐‘ around the origin. Indeed, โˆซ |๐‘“๐‘ | = โˆซ |๐‘“|๐œ’๐ต๐‘ โ‰ค โ€–๐‘“โ€–๐ฟ2 โ‹… โˆš|๐ต๐‘ | < โˆž โ„๐••

โ„๐••

for each ๐‘. Moreover, ๐‘“๐‘ โ†’ ๐‘“ in ๐ฟ2 (โ„๐‘‘ ) as ๐‘ โ†’ โˆž, because โ€–๐‘“ โˆ’ ๐‘“๐‘ โ€–2๐ฟ2 = โˆซ

|๐‘“(๐‘ฅ)|2 ๐‘‘๐‘ฅ โ†’ 0.

|๐‘ฅ|โ‰ฅ๐‘

Thus ฬ‚ = lim ๐‘“๐‘ฬ‚ (๐œ‰) = lim โˆซ ๐‘“(๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ๐‘‘๐‘ฅ ๐‘“(๐œ‰) ๐‘โ†’โˆž

2

๐‘โ†’โˆž

๐ต๐‘

๐‘‘

in ๐ฟ (โ„ ) (i.e. the limit is taken in the ๐ฟ2 sense). By Theorem 8.22, 2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ ๐‘‘๐œ‰ ๐‘“(๐‘ฅ) = lim โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘โ†’โˆž

2

๐ต๐‘

๐‘‘

in ๐ฟ (โ„ ), that is 2

2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ ฬ‚ โˆซ ||๐‘“(๐‘ฅ) โˆ’ โˆซ ๐‘“(๐œ‰)๐‘’ ๐‘‘๐œ‰ || ๐‘‘๐‘ฅ โ†’ 0 โ„๐••

๐ต๐‘

164

8. Fourier transform

as ๐‘ โ†’ โˆž. Hence we have an analog of Theorem 3.50, on the meansquare convergence of Fourier series, for the Fourier transform. 8.24. If ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ), we have seen above that their convolution ๐‘“ โˆ— ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ) and โ€–๐‘“ โˆ— ๐‘”โ€–๐ฟ2 โ‰ค โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”โ€–๐ฟ2 . Thus, if ๐‘”๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ) โˆฉ ๐ฟ2 (โ„๐‘‘ ) and ๐‘”๐‘› โ†’ ๐‘” in ๐ฟ2 (โ„๐‘‘ ), then โ€–๐‘“ โˆ— ๐‘”๐‘› โˆ’ ๐‘“ โˆ— ๐‘”โ€–๐ฟ2 โ‰ค โ€–๐‘“โ€–๐ฟ1 โ€–๐‘”๐‘› โˆ’ ๐‘”โ€–๐ฟ2 , so ๐‘“ โˆ— ๐‘”๐‘› โ†’ ๐‘“ โˆ— ๐‘” in ๐ฟ2 (โ„๐‘‘ ). Note that we also have ๐‘“ โˆ— ๐‘”๐‘› โˆˆ ๐ฟ1 (โ„๐‘‘ ), and thus ๐‘“ห† โˆ— ๐‘”๐‘› = ๐‘“๐‘”ฬ‚ ๐‘›ฬ‚ for each ๐‘›. Therefore we obtain ห† ๐‘“ โˆ— ๐‘” = ๐‘“๐‘”ฬ‚ ฬ‚ for any ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ). Example 8.25. Let ๐‘“ โˆˆ ๐ฟ2 (โ„๐‘‘ ) and let ๐’ซ๐‘ก ๐‘“(๐‘ฅ) be its Poisson integral. Then, since ๐’ซ๐‘ก ๐‘“ = ๐‘ƒ๐‘ก โˆ— ๐‘“, we have โˆ’2๐œ‹๐‘ก|๐œ‰| ฬ‚ ฬ‚ ห† ๐‘“(๐œ‰). ๐’ซห† ๐‘ก ๐‘“(๐œ‰) = ๐‘ƒ๐‘ก (๐œ‰)๐‘“(๐œ‰) = ๐‘’

Exercises (1) The Fourier transform is a linear operator: (a) If ๐›ผ โˆˆ โ„‚ and ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ห† ฬ‚ ๐›ผ๐‘“(๐œ‰) = ๐›ผ๐‘“(๐œ‰). (b) If ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ฬ‚ + ๐‘”(๐œ‰). ๐‘“ห† + ๐‘”(๐œ‰) = ๐‘“(๐œ‰) ฬ‚ (2) If ๐‘… is a rectangle in โ„๐‘‘ , then ๐œ’ ห†๐‘… (๐œ‰) โ†’ 0 as |๐œ‰| โ†’ โˆž. (3) Let ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ). (a) If ๐‘“โ„Ž (๐‘ฅ) = ๐‘“(๐‘ฅ + โ„Ž) is the translation of ๐‘“ with โ„Ž โˆˆ โ„๐‘‘ , then ห†โ„Ž (๐œ‰) = ๐‘’2๐œ‹๐‘–โ„Žโ‹…๐œ‰ ๐‘“(๐œ‰). ฬ‚ ๐‘“ (b) If ๐‘”(๐‘ฅ) = ๐‘’2๐œ‹๐‘–๐‘ฅโ‹…โ„Ž with โ„Ž โˆˆ โ„๐‘‘ , then ฬ‚ โˆ’ โ„Ž). ๐‘”(๐œ‰) ฬ‚ = ๐‘“(๐œ‰

Notes

165

(4) For any ๐œ‰ โˆˆ โ„, ๐‘

๐‘ 2

2

lim โˆซ ๐‘’โˆ’๐œ‹(๐‘ฅ+๐‘–๐œ‰) ๐‘‘๐‘ฅ = lim โˆซ ๐‘’โˆ’๐œ‹๐‘ฅ ๐‘‘๐‘ฅ = 1.

๐‘โ†’โˆž

๐‘โ†’โˆž

โˆ’๐‘

โˆ’๐‘ 2

(Hint: Consider the contour integral โˆซ๐›พ ๐‘’โˆ’๐œ‹๐‘ง ๐‘‘๐‘ง = 0 over the rectangle ๐›พ with vertices ๐‘, ๐‘ + ๐‘–๐œ‰, โˆ’๐‘ + ๐‘–๐œ‰ and โˆ’๐‘.) (5) Let ๐‘ข > 0. Then โˆž โˆ’2๐œ‹๐‘–แต†๐‘ฃ โˆž 1 ๐‘’ 1 1 โˆ’๐‘  โˆ’๐œ‹2 แต†2 /๐‘  โˆซ โˆซ (a) ๐‘‘๐‘ฃ = ๐‘’ ๐‘’ ๐‘‘๐‘ ; ๐œ‹ โˆ’โˆž 1 + ๐‘ฃ2 โˆš๐œ‹ 0 โˆš๐‘  โˆž โˆ’2๐œ‹๐‘–แต†๐‘ฃ 1 ๐‘’ โˆซ (b) ๐‘‘๐‘ฃ = ๐‘’โˆ’2๐œ‹แต† . ๐œ‹ โˆ’โˆž 1 + ๐‘ฃ2 2

(6) If ฮฆ(๐‘ฅ) = ๐‘’โˆ’๐œ‹|๐‘ฅ| , the collection {ฮฆ๐‘ก (๐‘ฅ)}๐‘ก>0 of its dilations is a collection of better kernels (see Exercise (6) of Chapter 9). (7) Use the polarization identity for complex inner products spaces to show 8.19. (8) If ๐‘“, ๐‘” โˆˆ ๐ฟ1 (โ„๐‘‘ ), then ห† ฬ‚ ๐‘”(๐œ‰). ๐‘“ โˆ— ๐‘”(๐œ‰) = ๐‘“(๐œ‰) ฬ‚ (Hint: Apply Fubiniโ€™s theorem to the function ๐น(๐‘ฅ, ๐‘ฆ) = ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ)๐‘”(๐‘ฆ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅโ‹…๐œ‰ on โ„๐‘‘ ร— โ„๐‘‘ .) (9) Let ๐‘“ โˆˆ ๐ฟ1 (โ„๐‘‘ ) and ๐‘… a rotation on โ„๐‘‘ . (a) If ๐‘” = ๐‘“ โˆ˜ ๐‘…, ฬ‚ ๐‘”(๐œ‰) ฬ‚ = ๐‘“(๐‘…๐œ‰). (b) If ๐‘“ is a radial function, then ๐‘“ ฬ‚ is also radial. (10) For every ๐‘“, ๐‘” โˆˆ ๐ฟ2 (โ„๐‘‘ ), โˆซ โ„ฑ๐‘“๐‘” = โˆซ ๐‘“โ„ฑ๐‘”. โ„๐••

โ„๐••

Notes The results in this chapter, and a deeper study of the Fourier transform, can be found in [SW71]. As we mentioned above, Theorem 8.22 is usually known as Plancherelโ€™s theorem, and the collective results of Section 8.3 are commonly described as Plancherelโ€™s theory due to Michel Plancherelโ€™s work in [Pla10].

Chapter 9

Hilbert transform

9.1. The conjugate function Let ๐‘“ โˆˆ ๐ฟ1 (โ„) and ๐‘ข(๐‘ฅ, ๐‘ก) be its Poisson integral, ๐‘ข(๐‘ฅ, ๐‘ก) = ๐’ซ๐‘ก ๐‘“(๐‘ฅ) = โˆซ ๐‘ƒ๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ, โ„

where ๐‘ƒ๐‘ก is the Poisson kernel on โ„, ๐‘ƒ๐‘ก (๐‘ฅ) =

1 ๐‘ก . 2 ๐œ‹ ๐‘ฅ + ๐‘ก2

We have seen above, in 4.3, that ๐‘ข(๐‘ฅ, ๐‘ก) is harmonic in the upper half plane โ„2+ . Consider now the function (9.1)

๐‘„๐‘ก (๐‘ฅ) =

1 ๐‘ฅ . ๐œ‹ ๐‘ฅ2 + ๐‘ก 2

The function ๐‘„๐‘ก is not integrable, as it only decays as 1/|๐‘ฅ| when |๐‘ฅ| โ†’ โˆž. However, it is bounded for each ๐‘ก > 0, so we can calculate (9.2)

๐‘ฃ(๐‘ฅ, ๐‘ก) = โˆซ ๐‘„๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ โ„

for any ๐‘“ โˆˆ ๐ฟ1 (โ„). ๐‘ฃ(๐‘ฅ, ๐‘ก) is a conjugate harmonic function to ๐‘ข(๐‘ฅ, ๐‘ก) in โ„2+ , so the function ๐‘“(๐‘ฅ, ๐‘ก) = ๐‘ข(๐‘ฅ, ๐‘ก) + ๐‘–๐‘ฃ(๐‘ฅ, ๐‘ก) 167

168

9. Hilbert transform

is holomorphic in โ„2+ . To prove this, it is sufficient to verify explicitly that (๐‘ฅ, ๐‘ก) โ†ฆ ๐‘„๐‘ก (๐‘ฅ) is harmonic, and so is ๐‘ฃ(๐‘ฅ, ๐‘ก), in โ„2+ , and that ๐‘ข and ๐‘ฃ satisfy the Cauchy-Riemann equations (Exercise (1)). In fact, we note that ๐‘– ๐‘ƒ๐‘ก (๐‘ฅ) + ๐‘–๐‘„๐‘ก (๐‘ฅ) = , ๐œ‹(๐‘ฅ + ๐‘–๐‘ก) so ๐‘ƒ๐‘ฆ (๐‘ฅ) and ๐‘„๐‘ฆ (๐‘ฅ), for ๐‘ฆ > 0, are the real and imaginary parts of ๐‘–/๐œ‹๐‘ง, where ๐‘ง = ๐‘ฅ + ๐‘–๐‘ฆ. We also observe that ๐‘„๐‘ก (๐‘ฅ) =

1 1 ๐‘ฅ/๐‘ก 1 ๐‘ฅ ( ) = ๐‘„1 ( ), ๐‘ก ๐œ‹ (๐‘ฅ/๐‘ก)2 + 1 ๐‘ก ๐‘ก

so {๐‘„๐‘ก }๐‘ก>0 is the family of dilations of ๐‘„1 . However, since ๐‘„1 โˆ‰ ๐ฟ1 (โ„), {๐‘„๐‘ก }๐‘ก>0 is not a family of good kernels, so we cannot yet determine the existence of (9.3)

lim โˆซ ๐‘„๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ. ๐‘กโ†’0

โ„

Note that the limit (9.3) represents the limit at the boundary of the conjugate harmonic function ๐‘ฃ to the Poisson integral of ๐‘“. We have the following questions. Question 1: If ๐‘“ โˆˆ ๐ฟ1 (โ„), does the limit (9.3) exist for any ๐‘ฅ โˆˆ โ„? Question 2: Under what conditions on ๐‘“ can we guarantee, and in what sense, that the limit (9.3) exists?

9.2. Mean-square convergence We first consider Question 2 by asking if the limit (9.3) exists in the mean-square sense, that is, in ๐ฟ2 (โ„). The first problem we encounter is that, since ๐‘„๐‘ก is not integrable, we cannot guarantee that the convolution ๐‘„๐‘ก โˆ— ๐‘“ exists if ๐‘“ โˆˆ ๐ฟ2 (โ„). However, since indeed ๐‘„๐‘ก โˆˆ ๐ฟ2 (โ„), the convolution ๐‘„๐‘ก โˆ— ๐‘“ โˆˆ ๐ฟ2 (โ„) if ๐‘“ โˆˆ ๐ฟ1 (โ„) and, in particular, if ๐‘“ โˆˆ ๐ฟ1 (โ„) โˆฉ ๐ฟ2 (โ„). Thus, before we consider the limit as ๐‘ก โ†’ 0, we have to ask Question 3. Question 3: Can we extend the operator ๐‘“ โ†ฆ ๐‘„๐‘ก โˆ— ๐‘“ from ๐ฟ1 (โ„) โˆฉ ๐ฟ2 (โ„) to ๐ฟ2 (โ„), for any ๐‘ก > 0? In that case, does the limit (9.3) exist in the ๐ฟ2 sense?

9.2. Mean-square convergence

169

To answer Question 3 we need to find out if the family of operators ๐‘“ โ†ฆ ๐‘„๐‘ก โˆ— ๐‘“ is uniformly bounded in ๐ฟ2 (โ„), that is, if there exists a constant ๐ด > 0, independent of ๐‘ก, such that โ€–๐‘„๐‘ก โˆ— ๐‘“โ€–๐ฟ2 โ‰ค ๐ดโ€–๐‘“โ€–๐ฟ2 1

2

for every ๐‘“ โˆˆ ๐ฟ (โ„) โˆฉ ๐ฟ (โ„) and all ๐‘ก > 0. This will allow us to extend ๐‘„๐‘ก โˆ— ๐‘“ to all of ๐ฟ2 (โ„), as we did in the case of the Fourier transform in the previous chapter, for each ๐‘ก > 0, and then take the limit as ๐‘ก โ†’ 0. Lemma 9.4. For any ๐‘ก > 0, the Fourier transform of ๐‘„๐‘ก in ๐ฟ2 (โ„) is given by ห†๐‘ก (๐œ‰) = โˆ’๐‘– sgn(๐œ‰)๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| , where sgn(๐œ‰) is the sign function the function ๐‘„ of ๐œ‰. Proof. As we noted in 8.23, the Fourier transform of the ๐ฟ2 function ๐‘„๐‘ก is given by ๐‘

๐‘

๐‘ฅ ห†๐‘ก (๐œ‰) = lim โˆซ ๐‘„๐‘ก (๐‘ฅ)๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ = 1 lim โˆซ ๐‘„ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ. 2 + ๐‘ก2 ๐œ‹ ๐‘ฅ ๐‘โ†’โˆž ๐‘โ†’โˆž โˆ’๐‘ โˆ’๐‘ As ๐‘„๐‘ก is odd, the above limit is 0 if ๐œ‰ = 0. For ๐œ‰ โ‰  0, the limit on the right can be obtained by using the residue theorem [Gam01]. Indeed, if ๐œ‰ < 0, consider the contour ๐›พ given by the upper semicircle of radius ๐‘

Figure 9.1. The contour ๐›พ in the proof of Lemma 9.4. For ๐‘ sufficiently large, the pole ๐‘–๐‘ก of the function ๐‘“(๐‘ง) is inside ๐›พ.

with center at the origin (as in Figure 9.1), and 1 ๐‘ง ๐‘“(๐‘ง) = ๐‘’โˆ’2๐œ‹๐‘–๐‘ง๐œ‰ . ๐œ‹ ๐‘ง 2 + ๐‘ก2

170

9. Hilbert transform

Then, by the residue theorem, โˆซ ๐‘“(๐‘ง)๐‘‘๐‘ง = 2๐œ‹๐‘– Res๐‘ง=๐‘–๐‘ก ๐‘“(๐‘ง) = ๐‘–๐‘’2๐œ‹๐‘ก๐œ‰ . ๐›พ

We also have โˆซ ๐‘“(๐‘ง)๐‘‘๐‘ง = ๐›พ ๐‘

โˆซ โˆ’๐‘

๐œ‹

1 ๐‘ฅ 1 ๐‘๐‘’๐‘–๐œƒ ๐‘–๐œƒ ๐‘’โˆ’2๐œ‹๐‘–๐‘ฅ๐œ‰ ๐‘‘๐‘ฅ + โˆซ ๐‘’โˆ’2๐œ‹๐‘–๐‘๐‘’ ๐œ‰ ๐‘–๐‘๐‘’๐‘–๐œƒ ๐‘‘๐œƒ, 2 2 ๐‘–๐œƒ )2 + ๐‘ก2 ๐œ‹๐‘ฅ +๐‘ก ๐œ‹ (๐‘๐‘’ 0

where the integrals on the right hand side correspond to the integrals over the segment [โˆ’๐‘, ๐‘] and the semicircle ๐‘๐‘’๐‘–๐œƒ , 0 โ‰ค ๐œƒ โ‰ค ๐œ‹, of ๐›พ, respectively. The first integral, as ๐‘ โ†’ โˆž, gives our desired limit, while the second satisfies ๐œ‹

๐œ‹

๐‘๐‘’๐‘–๐œƒ ๐‘–๐œƒ ๐‘–๐œƒ |โˆซ 1 ๐‘’โˆ’2๐œ‹๐‘–๐‘๐‘’ ๐œ‰ ๐‘–๐‘๐‘’๐‘–๐œƒ ๐‘‘๐œƒ|| โ‰ค ๐ถ โˆซ |๐‘’โˆ’2๐œ‹๐‘–๐‘๐‘’ ๐œ‰ |๐‘‘๐œƒ | ๐‘–๐œƒ )2 + ๐‘ก2 ๐œ‹ (๐‘๐‘’ 0 0 ๐œ‹

= ๐ถ โˆซ ๐‘’2๐œ‹๐‘๐œ‰ sin ๐œƒ ๐‘‘๐œƒ 0 ๐œ‹/2

โ‰ค 2๐ถ โˆซ

๐‘’โˆ’4๐‘|๐œ‰|๐œƒ ๐‘‘๐œƒ โ†’ 0,

0

where we have used the fact that ๐œ‰ < 0 and the estimate 2 sin ๐œƒ โ‰ฅ ๐œƒ ๐œ‹ on [0, ๐œ‹/2], along with the symmetry of sin ๐œƒ around ๐œƒ = ๐œ‹/2. Thus, we obtain ห†๐‘ก (๐œ‰) = ๐‘–๐‘’2๐œ‹๐‘ก๐œ‰ ๐‘„ for ๐œ‰ < 0. We leave as an exercise the details for ๐œ‰ > 0 (Exercise (2)), where we obtain ห†๐‘ก (๐œ‰) = โˆ’๐‘–๐‘’โˆ’2๐œ‹๐‘ก๐œ‰ . ๐‘„ Combining the above we conclude ห†๐‘ก (๐œ‰) = โˆ’๐‘– sgn(๐œ‰)๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| . ๐‘„ โ–ก 1

2

Corollary 9.5. If ๐‘“ โˆˆ ๐ฟ (โ„) โˆฉ ๐ฟ (โ„), then, for any ๐‘ก > 0, โ€–๐‘„๐‘ก โˆ— ๐‘“โ€–๐ฟ2 โ‰ค โ€–๐‘“โ€–๐ฟ2 .

9.2. Mean-square convergence

171

Proof. The Fourier transform is an isometry in ๐ฟ2 (โ„), so, using Lemma 9.4, โˆ’2๐œ‹๐‘ก|๐œ‰| ฬ‚ ห† โ€–๐‘„๐‘ก โˆ— ๐‘“โ€–๐ฟ2 = โ€–๐‘„ ๐‘“โ€–๐ฟ2 โ‰ค โ€–๐‘“โ€–ฬ‚ ๐ฟ2 = โ€–๐‘“โ€–๐ฟ2 . ๐‘ก โˆ— ๐‘“โ€–๐ฟ2 = โ€– โˆ’ ๐‘– sgn(๐œ‰)๐‘’

โ–ก Corollary 9.5 allows us to define the convolution ๐‘„๐‘ก โˆ— ๐‘“ for any ๐‘“ โˆˆ ๐ฟ2 (โ„), and we can do it in two different ways. The first way is analogous to the way we extended the Fourier transform to ๐ฟ2 (โ„๐‘‘ ): using the fact that ๐ฟ1 (โ„) โˆฉ ๐ฟ2 (โ„) is dense in ๐ฟ2 (โ„), we can find, for any ๐‘“ โˆˆ ๐ฟ2 (โ„), a sequence ๐‘“๐‘› โˆˆ ๐ฟ1 (โ„) โˆฉ ๐ฟ2 (โ„) such that ๐‘“๐‘› โ†’ ๐‘“ in ๐ฟ2 (โ„), and then define ๐‘„๐‘ก โˆ— ๐‘“ as the limit in ๐ฟ2 (โ„) of the sequence ๐‘„๐‘ก โˆ— ๐‘“๐‘› . This limit exists because ๐‘„๐‘ก โˆ— ๐‘“๐‘› is a Cauchy sequence in ๐ฟ2 (โ„), by Corollary 9.5, and ๐ฟ2 (โ„) is complete. The other way is through the Fourier transform of ๐‘„๐‘ก โˆ— ๐‘“. Since the Fourier transform is a unitary operator, by Theorem 8.18, we can define, for ๐‘“ โˆˆ ๐ฟ2 (โ„), ๐‘„๐‘ก โˆ— ๐‘“ โˆˆ ๐ฟ2 (โ„) as the function whose Fourier transform is equal to (9.6)

โˆ’2๐œ‹๐‘ก|๐œ‰| ฬ‚ ห† ๐‘„ ๐‘“(๐œ‰). ๐‘ก โˆ— ๐‘“(๐œ‰) = โˆ’๐‘– sgn(๐œ‰)๐‘’

That is, we define, for ๐‘“ โˆˆ ๐ฟ2 (โ„), ๐‘„๐‘ก โˆ— ๐‘“ as the inverse Fourier transform of (9.6). Moreover, it is clear that, as ๐‘ก โ†’ 0, the limit ห† lim ๐‘„ ๐‘ก โˆ— ๐‘“(๐œ‰) ๐‘กโ†’0

ฬ‚ exists for every ๐œ‰ โˆˆ โ„, and is equal to โˆ’๐‘– sgn(๐œ‰)๐‘“(๐œ‰). This limit also converges in ๐ฟ2 (โ„) because 2

ฬ‚ โˆ’ ( โˆ’ ๐‘– sgn(๐œ‰)๐‘“(๐œ‰)) ฬ‚ | ๐‘‘๐œ‰ โˆซ | โˆ’ ๐‘– sgn(๐œ‰)๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| ๐‘“(๐œ‰) โ„ 2

ฬ‚ | ๐‘‘๐œ‰ โ†’ 0, = โˆซ |(๐‘’โˆ’2๐œ‹๐‘ก|๐œ‰| โˆ’ 1)๐‘“(๐œ‰) โ„

by the dominated convergence theorem. We can therefore make Definition 9.7. Definition 9.7. Let ๐‘“ โˆˆ ๐ฟ2 (โ„). The Hilbert transform of ๐‘“ is the function ๐ป๐‘“ โˆˆ ๐ฟ2 (โ„) equal to the inverse Fourier transform of ห† ฬ‚ ๐ป๐‘“(๐œ‰) = โˆ’๐‘– sgn(๐œ‰)๐‘“(๐œ‰).

172

9. Hilbert transform

9.8. The Hilbert transform ๐ป is an isometry in ๐ฟ2 (โ„), because ฬ‚ ฬ‚ | โˆ’ ๐‘– sgn(๐œ‰)๐‘“(๐œ‰)| = |๐‘“(๐œ‰)| and hence ห† ๐ฟ2 (โ„) = โ€–๐‘“โ€–ฬ‚ ๐ฟ2 (โ„) = โ€–๐‘“โ€–๐ฟ2 (โ„) . โ€–๐ป๐‘“โ€–๐ฟ2 (โ„) = โ€–๐ป๐‘“โ€– With the observations above, and Definition 9.7, we have answered Questions 2 and 3 of this and Section 9.1. However, to answer Question 1, we need to understand the local nature of the convolution ๐‘„๐‘ก โˆ— ๐‘“ as ๐‘ก โ†’ 0, which we discuss in Section 9.3.

9.3. The Hilbert transform of integrable functions In this section we study the Hilbert transform of integrable functions. As noted above, the kernels ๐‘„๐‘ก are not integrable, and thus we cannot discuss the behavior of ๐‘„๐‘ก โˆ— ๐‘“ as ๐‘ก โ†’ 0 pointwise, nor in ๐ฟ1 (โ„), directly. However, we have Lemma 9.9. Lemma 9.9. For ๐‘“ โˆˆ ๐ฟ1 (โ„), lim ( โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘กโ†’0

โ„

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘ฆ ๐‘‘๐‘ฆ โˆ’ โˆซ ๐‘‘๐‘ฆ) = 0 ๐‘ฆ ๐‘ฆ2 + ๐‘ก 2 |๐‘ฆ|โ‰ฅ๐‘ก

for almost every ๐‘ฅ โˆˆ โ„. Proof. Define

๐‘ฆ 1 โˆ’ 2+1 ๐‘ฆ ๐‘ฆ ฮฆ(๐‘ฆ) = { ๐‘ฆ ๐‘ฆ2 + 1

Since

|๐‘ฆ| โ‰ฅ 1 |๐‘ฆ| < 1.

๐‘ฆ 1 1 โˆ’ =โˆ’ 2 +1 ๐‘ฆ ๐‘ฆ(๐‘ฆ + 1) for |๐‘ฆ| > 1, ฮฆ is integrable. Moreover, the collection of its dilations ๐‘ฆ 1 โˆ’ |๐‘ฆ| โ‰ฅ ๐‘ก 2 + ๐‘ก2 1 ๐‘ฆ ๐‘ฆ ๐‘ฆ ฮฆ๐‘ก (๐‘ฆ) = ฮฆ( ) = { ๐‘ฆ ๐‘ก ๐‘ก |๐‘ฆ| < ๐‘ก ๐‘ฆ2 + ๐‘ก 2 ฮฆ(๐‘ฆ) =

๐‘ฆ2

satisfies (1) โˆซ ฮฆ๐‘ก = 0 for all ๐‘ก > 0; โ„

9.3. The Hilbert transform of integrable functions

173

1 for all ๐‘ก > 0 and ๐‘ฆ โˆˆ โ„; and ๐‘ก ๐‘ก (3) |ฮฆ๐‘ก (๐‘ฆ)| โ‰ค 2 for all ๐‘ก > 0 and ๐‘ฆ โˆˆ โ„, ๐‘ฆ โ‰  0. ๐‘ฆ (2) |ฮฆ๐‘ก (๐‘ฆ)| โ‰ค

The first one follows because ฮฆ is odd, and the other two from the explicit form of ฮฆ๐‘ก (๐‘ฆ) (Exercise (4)). Then โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) โ„

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘ฆ ๐‘‘๐‘ฆ โˆ’ โˆซ ๐‘‘๐‘ฆ = ฮฆ๐‘ก โˆ— ๐‘“(๐‘ฅ) ๐‘ฆ ๐‘ฆ2 + ๐‘ก 2 |๐‘ฆ|โ‰ฅ๐‘ก

and {ฮฆ๐‘ก }๐‘ก>0 is a collection of better kernels, as introduced in Exercises (6) and (9) of Chapter 7, with constant ๐œ† = 0 in the latter. Therefore, lim ฮฆ๐‘ก โˆ— ๐‘“(๐‘ฅ) = 0 ๐‘กโ†’0

โ–ก

for almost every ๐‘ฅ โˆˆ โ„. Lemma 9.9 implies that, for almost every ๐‘ฅ โˆˆ โ„, the limit lim โˆซ ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘กโ†’0

โ„

๐‘ฆ ๐‘‘๐‘ฆ ๐‘ฆ2 + ๐‘ก 2

exists if and only if lim โˆซ ๐‘กโ†’0

|๐‘ฆ|โ‰ฅ๐‘ก

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘‘๐‘ฆ ๐‘ฆ

exists. Note that the latter limit exists at ๐‘ฅ โˆˆ โ„ if ๐‘“ โˆˆ ๐ฟ1 (โ„) and ๐‘ฅ is not in the support of ๐‘“, because lim ๐‘กโ†’0

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘“(๐‘ฆ) ๐‘“(๐‘ฆ) 1 1 1 โˆซ ๐‘‘๐‘ฆ = lim โˆซ ๐‘‘๐‘ฆ = โˆซ ๐‘‘๐‘ฆ, ๐œ‹ |๐‘ฆ|โ‰ฅ๐‘ก ๐‘ฆ ๐‘ฅ โˆ’ ๐‘ฆ ๐œ‹ ๐‘ฅ โˆ’๐‘ฆ ๐‘กโ†’0 ๐œ‹ |๐‘ฅโˆ’๐‘ฆ|โ‰ฅ๐‘ก โ„

as there exist some ๐›ฟ > 0 such that, if |๐‘ฅ โˆ’ ๐‘ฆ| < ๐›ฟ, then ๐‘ฆ โˆ‰ supp ๐‘“, so ๐‘“(๐‘ฆ) = 0. The limit also exists if ๐‘“ is differentiable at ๐‘ฅ, because in that case the integrals ๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) โˆซ ๐‘‘๐‘ฆ ๐‘ฆ ๐‘กโ‰ค|๐‘ฆ| 0, โ€–๐ป๐‘“โ€–๐ฟ1 โ‰ค ๐ดโ€–๐‘“โ€–๐ฟ1 ๐ถ๐‘โˆž (โ„).

for every ๐‘“ โˆˆ However, this is far from true, because ๐ป๐‘“ might not even be integrable. Example 9.11. Let ๐‘“ โˆˆ ๐ถ๐‘โˆž (โ„) such that ๐‘“ โ‰ฅ 0, ๐‘“(๐‘ฅ) = 1 if 0 โ‰ค ๐‘ฅ โ‰ค 1, and ๐‘“(๐‘ฅ) = 0 if ๐‘ฅ โ‰ฅ 2 or ๐‘ฅ โ‰ค โˆ’1 (as in Figure 9.2). Then, for ๐‘ฅ > 2,

Figure 9.2. A function ๐‘“ โˆˆ ๐ถ๐‘โˆž (โ„) such that ๐‘“ โ‰ฅ 0, ๐‘“(๐‘ฅ) = 1 if 0 โ‰ค ๐‘ฅ โ‰ค 1, and ๐‘“(๐‘ฅ) = 0 if ๐‘ฅ โ‰ฅ 2 or ๐‘ฅ โ‰ค โˆ’1. 2

๐ป๐‘“(๐‘ฅ) =

1

๐‘“(๐‘ฆ) 1 1 1 1 โˆซ ๐‘‘๐‘ฆ โ‰ฅ โˆซ ๐‘‘๐‘ฆ โ‰ฅ , ๐œ‹ โˆ’1 ๐‘ฅ โˆ’ ๐‘ฆ ๐œ‹ 0 ๐‘ฅโˆ’๐‘ฆ ๐œ‹๐‘ฅ

so ๐ป๐‘“ is not an integrable function. We see then that the Hilbert transform cannot be bounded in ๐ฟ1 (โ„). However we have Theorem 9.12. Theorem 9.12. There exists a constant ๐ด > 0 such that, for any ๐‘“ โˆˆ ๐ถ๐‘โˆž (โ„) and ๐›ผ > 0, ๐ด |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป๐‘“(๐‘ฅ)| > ๐›ผ}| โ‰ค โ€–๐‘“โ€–๐ฟ1 . ๐›ผ As in the case of the maximal function, the Hilbert transform is of weak type (1, 1) (although, so far, it is only defined on ๐ถ๐‘โˆž (โ„)).

9.3. The Hilbert transform of integrable functions

175

Proof. Fix ๐›ผ > 0. We will write the function ๐‘“ as ๐‘” + ๐‘, the sum of a โ€œgoodโ€ and a โ€œbadโ€ part. For this, we construct a collection โ„ of dyadic intervals (intervals of the form [๐‘˜ โ‹… 2๐‘› , (๐‘˜ + 1) โ‹… 2๐‘› ], with ๐‘˜, ๐‘› โˆˆ โ„ค) in the following way. Let ๐‘ be large enough so that 1 โˆซ|๐‘“| โ‰ค ๐›ผ |๐ผ| ๐ผ for every dyadic interval ๐ผ of length |๐ผ| = 2๐‘ . Such ๐‘ exists because ๐‘“ is integrable. Now, subdivide each ๐ผ in two subintervals ๐ผ โ€ฒ of half the length of ๐ผ. For each one, we have either 1 โˆซ |๐‘“| โ‰ค ๐›ผ |๐ผ โ€ฒ | ๐ผ โ€ฒ

1 โˆซ |๐‘“| > ๐›ผ. |๐ผ โ€ฒ | ๐ผ โ€ฒ

or

In the second case, we add ๐ผ โ€ฒ to the collection โ„. Note that we have ๐›ผ
๐›ผ/2}| โ‰ค โˆซ โ„

|๐ป๐‘”|2 4 20 โ‰ค 2 โ‹… 5๐›ผโ€–๐‘“โ€–๐ฟ1 = โ€–๐‘“โ€–๐ฟ1 . ๐›ผ ๐›ผ (๐›ผ/2)2

Define ๐‘ = ๐‘“ โˆ’ ๐‘”. Since |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป๐‘“(๐‘ฅ)| > ๐›ผ}| โ‰ค |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป๐‘”(๐‘ฅ)| > ๐›ผ/2}| + |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป๐‘(๐‘ฅ)| > ๐›ผ/2}|, it remains to estimate |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป๐‘(๐‘ฅ)| > ๐›ผ/2}|. ๐‘ is what we refer as the bad part of ๐‘“, because we have no control over the size of |๐‘(๐‘ฅ)|. However, ๐‘ = 0 outside of ฮฉ and, for each ๐ผ โˆˆ โ„, โˆซ๐‘ = โˆซ๐‘“ โˆ’ โˆซ๐‘” = โˆซ๐‘“ โˆ’ โˆซ( ๐ผ

๐ผ

๐ผ

๐ผ

๐ผ

1 โˆซ๐‘“) = 0. |๐ผ| ๐ผ

Write ๐‘ = โˆ‘ ๐‘๐ผ , ๐ผโˆˆโ„

where each ๐‘๐ผ = ๐‘ โ‹… ๐œ’๐ผ , for ๐ผ โˆˆ โ„. Then ๐ป๐‘ = โˆ‘ ๐ป๐‘๐ผ . ๐ผโˆˆโ„

Note that each ๐ป๐‘๐ผ is defined almost everywhere, since ๐‘๐ผ is either 0, outside of ๐ผ, or 1 โˆซ๐‘“ ๐‘๐ผ (๐‘ฅ) = ๐‘“(๐‘ฅ) โˆ’ |๐ผ| ๐ผ in the interior of ๐ผ, so it is differentiable, as we are assuming that ๐‘“ is differentiable.

9.3. The Hilbert transform of integrable functions

177

Write 3๐ผ for the interval with the same center as ๐ผ, but 3 times its length (see Figure 9.3).

Figure 9.3. The interval 3๐ผ has the same center as ๐ผ and 3 times its length.

If ๐‘ฅ โˆ‰ 3๐ผ, and ๐‘ฆ0 is the center of ๐ผ,

๐ป๐‘๐ผ (๐‘ฅ) =

๐‘ (๐‘ฆ) ๐‘ (๐‘ฆ) 1 1 โˆซ ๐ผ ๐‘‘๐‘ฆ โˆ’ โˆซ ๐ผ ๐‘‘๐‘ฆ ๐œ‹ ๐ผ ๐‘ฅโˆ’๐‘ฆ ๐œ‹ ๐ผ ๐‘ฅ โˆ’ ๐‘ฆ0

=

1 1 1 โˆซ๐‘ (๐‘ฆ)( โˆ’ )๐‘‘๐‘ฆ ๐œ‹ ๐ผ ๐ผ ๐‘ฅ โˆ’ ๐‘ฆ ๐‘ฅ โˆ’ ๐‘ฆ0

=

๐‘ฆ โˆ’ ๐‘ฆ0 1 โˆซ๐‘ (๐‘ฆ) ๐‘‘๐‘ฆ, ๐œ‹ ๐ผ ๐ผ (๐‘ฅ โˆ’ ๐‘ฆ)(๐‘ฅ โˆ’ ๐‘ฆ0 )

where, in the first equality, we have used the fact that โˆซ๐ผ ๐‘๐ผ = 0. As ๐‘ฅ โˆ‰ 3๐ผ and ๐‘ฆ โˆˆ ๐ผ in the integral above, we see that

|๐‘ฆ โˆ’ ๐‘ฆ0 | โ‰ค

1 1 |๐ผ| โ‰ค |๐‘ฅ โˆ’ ๐‘ฆ|, 2 2

so |๐‘ฅ โˆ’ ๐‘ฆ0 | โ‰ฅ |๐‘ฅ โˆ’ ๐‘ฆ| โˆ’ |๐‘ฆ โˆ’ ๐‘ฆ0 | โ‰ฅ

1 |๐‘ฅ โˆ’ ๐‘ฆ|. 2

Thus

|๐ป๐‘๐ผ (๐‘ฅ)| โ‰ค

|๐‘ฆ โˆ’ ๐‘ฆ0 | |๐‘ (๐‘ฆ)| 2 1 โˆซ|๐‘ (๐‘ฆ)| ๐‘‘๐‘ฆ โ‰ค |๐ผ| โˆซ ๐ผ 2 ๐‘‘๐‘ฆ, ๐œ‹ ๐ผ ๐ผ ๐œ‹ |๐‘ฅ โˆ’ ๐‘ฆ|2 |๐‘ฅ โˆ’ ๐‘ฆ| ๐ผ

178

9. Hilbert transform

and therefore โˆซ

|๐ป๐‘๐ผ (๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค

โ„โงต3๐ผ

|๐‘ (๐‘ฆ)| 1 |๐ผ| โˆซ โˆซ ๐ผ 2 ๐‘‘๐‘ฆ๐‘‘๐‘ฅ ๐œ‹ |๐‘ฅ โˆ’ ๐‘ฆ| โ„โงต3๐ผ ๐ผ

โ‰ค

1 1 |๐ผ| โˆซ|๐‘๐ผ (๐‘ฆ)| โˆซ ๐‘‘๐‘ฅ๐‘‘๐‘ฆ ๐œ‹ |๐‘ฅ โˆ’ ๐‘ฆ|2 ๐ผ |๐‘ฅโˆ’๐‘ฆ|โ‰ฅ|๐ผ|

โ‰ค

2 โˆซ|๐‘ (๐‘ฆ)|๐‘‘๐‘ฆ ๐œ‹ ๐ผ ๐ผ

โ‰ค

1 2 | โˆซ ๐‘“(๐‘ฆ) โˆ’ โˆซ๐‘“|๐‘‘๐‘ฆ ๐œ‹ ๐ผ| |๐ผ| ๐ผ |

โ‰ค

2 1 โˆซ( โˆซ|๐‘“|)) ( โˆซ|๐‘“| + ๐œ‹ ๐ผ |๐ผ| ๐ผ ๐ผ

=

4 8 โˆซ|๐‘“| โ‰ค ๐›ผ|๐ผ|, ๐œ‹ ๐ผ ๐œ‹

because ๐ผ โˆˆ โ„. Hence, if we define ฮฉโˆ— =

โ‹ƒ

3๐ผ,

๐ผโˆˆโ„

we obtain โˆซ

|๐ป๐‘(๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค โˆ‘ โˆซ

โ„โงตโ„ฆโˆ—

|๐ป๐‘๐ผ (๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค

๐ผโˆˆโ„ โ„โงต3๐ผ

=

8๐›ผ โˆ‘ |๐ผ| ๐œ‹ ๐ผโˆˆโ„

8๐›ผ 8 |ฮฉ| โ‰ค โ€–๐‘“โ€–๐ฟ1 . ๐œ‹ ๐œ‹

Thus, by Chebyshevโ€™s inequality, |{๐‘ฅ โˆˆ โ„ โงต ฮฉโˆ— โˆถ |๐ป๐‘(๐‘ฅ)| > ๐›ผ/2}| โ‰ค

16/๐œ‹ 2 โˆซ |๐ป๐‘(๐‘ฅ)|๐‘‘๐‘ฅ โ‰ค โ€–๐‘“โ€–๐ฟ1 . ๐›ผ โ„โงตโ„ฆโˆ— ๐›ผ

It just remains to estimate |{๐‘ฅ โˆˆ ฮฉโˆ— โˆถ |๐ป๐‘(๐‘ฅ)| > ๐›ผ/2}| โ‰ค |ฮฉโˆ— | โ‰ค โˆ‘ |3๐ผ| = 3|ฮฉ| โ‰ค ๐ผโˆˆโ„

3 โ€–๐‘“โ€–๐ฟ1 . ๐›ผ โ–ก

Note that, from the proof above, we can take the constant ๐ด in Theorem 9.12 as 16 ๐ด = 20 + + 3 < 29. ๐œ‹

9.4. Convergence in measure

179

9.4. Convergence in measure Theorem 9.12 doesnโ€™t guarantee that the limit lim ๐‘กโ†’0

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) 1 โˆซ ๐‘‘๐‘ฆ ๐œ‹ |๐‘ฆ|โ‰ฅ๐‘ก ๐‘ฆ

converges for a function in ๐ฟ1 . However, the Hilbert transform can be defined โ€œin measureโ€. We say that the sequence ๐‘“๐‘› converges in measure to ๐‘“ if, for every ๐œ€ > 0, |{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘› (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| โ‰ฅ ๐œ€}| โ†’ 0 as ๐‘› โ†’ โˆž. The function ๐‘“ is unique almost everywhere (Exercise (6)). If a sequence ๐‘“๐‘› โ†’ ๐‘“ in ๐ฟ1 , then it converges in measure to ๐‘“, but almost everywhere convergence does not imply convergence in measure (Exercise (7)). A sequence ๐‘“๐‘› of measurable functions on โ„๐‘‘ is Cauchy in measure if, for every ๐œ€ > 0, |{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘› (๐‘ฅ) โˆ’ ๐‘“๐‘š (๐‘ฅ)| โ‰ฅ ๐œ€}| โ†’ 0 as ๐‘›, ๐‘š โ†’ โˆž. If ๐‘“๐‘› converges in measure to ๐‘“, then ๐‘“๐‘› is Cauchy in measure because {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘› (๐‘ฅ) โˆ’ ๐‘“๐‘š (๐‘ฅ)| โ‰ฅ ๐œ€} โŠ‚ {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘› (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| โ‰ฅ ๐œ€/2} โˆช {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘š (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| โ‰ฅ ๐œ€/2}, by the triangle inequality. If ๐‘“๐‘› is Cauchy in measure, there exists a sequence ๐‘›๐‘˜ such that |{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘›๐‘˜ (๐‘ฅ) โˆ’ ๐‘“๐‘›๐‘˜+1 (๐‘ฅ)| โ‰ฅ

1 1 }| < ๐‘˜ . ๐‘˜ 2 2

If we define ๐ด๐‘˜ =

โ‹ƒ

{๐‘ฅ โˆˆ โ„๐‘‘ โˆถ |๐‘“๐‘›๐‘— (๐‘ฅ) โˆ’ ๐‘“๐‘›๐‘—+1 (๐‘ฅ)| โ‰ฅ

๐‘—โ‰ฅ๐‘˜

then โˆž

1 1 = ๐‘˜โˆ’1 . ๐‘— 2 2 ๐‘—=๐‘˜

|๐ด๐‘˜ | โ‰ค โˆ‘

1 }, 2๐‘—

180

9. Hilbert transform

In particular, |๐ด๐‘˜ | โ†’ 0 and | โ‹‚๐‘˜ ๐ด๐‘˜ | = 0. If ๐‘ฅ โˆ‰ ๐ด๐‘˜ and ๐‘– โ‰ฅ ๐‘— โ‰ฅ ๐‘˜, ๐‘–

๐‘–

(9.13)

|๐‘“๐‘›๐‘— (๐‘ฅ) โˆ’ ๐‘“๐‘›๐‘– (๐‘ฅ)| โ‰ค โˆ‘ |๐‘“๐‘›๐‘™ (๐‘ฅ) โˆ’ ๐‘“๐‘›๐‘™+1 (๐‘ฅ)| < โˆ‘ ๐‘™=๐‘—

๐‘™=๐‘—

1 1 < ๐‘—โˆ’1 . ๐‘™ 2 2

Thus the sequence ๐‘“๐‘›๐‘— (๐‘ฅ) is Cauchy for each ๐‘ฅ โˆ‰ ๐ด๐‘˜ , and thus converges. We can then define ๐‘“(๐‘ฅ) = lim ๐‘“๐‘›๐‘— (๐‘ฅ) for each ๐‘ฅโˆˆ

(โ„๐‘‘ โงต ๐ด๐‘˜ ) = โ„๐‘‘ โงต ๐ด . โ‹ƒ โ‹‚ ๐‘˜ ๐‘˜

๐‘˜

Therefore the subsequence ๐‘“๐‘›๐‘˜ โ†’ ๐‘“ almost everywhere. By (9.13), |๐‘“๐‘›๐‘— (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)|
0, |{๐‘ฅ โˆˆ โ„ โˆถ |๐ป(๐‘“๐‘› โˆ’ ๐‘“๐‘š )(๐‘ฅ)| > ๐œ€}| โ‰ค

๐ด โ€–๐‘“ โˆ’ ๐‘“๐‘š โ€–๐ฟ1 โ†’ 0, ๐œ€ ๐‘›

so the sequence ๐ป๐‘“๐‘› is Cauchy in measure. Therefore, there exists a function, which we may call ๐ป๐‘“, such that ๐ป๐‘“๐‘› โ†’ ๐ป๐‘“ in measure. As we have seen before, ๐ป๐‘“ is well defined and there exists a subsequence ๐‘“๐‘›๐‘˜ such that ๐ป๐‘“๐‘›๐‘˜ โ†’ ๐ป๐‘“ almost everywhere (Exercise (8)). Theorem 9.12 implies that ๐ป๐‘“ thus defined is finite almost everywhere.

Exercises

181

Exercises (1) (a) The function ๐‘ž(๐‘ฅ, ๐‘ก) = ๐‘„๐‘ก (๐‘ฅ) =

1 ๐‘ฅ ๐œ‹ ๐‘ฅ2 + ๐‘ก 2

is harmonic in โ„2+ . (b) If ๐‘“ โˆˆ ๐ฟ1 (โ„), the function ๐‘ฃ(๐‘ฅ, ๐‘ก) = โˆซ ๐‘„๐‘ก (๐‘ฅ โˆ’ ๐‘ฆ)๐‘“(๐‘ฆ)๐‘‘๐‘ฆ โ„

is harmonic in โ„2+ . (c) If ๐‘“ โˆˆ ๐ฟ1 (โ„), the functions ๐‘ฃ(๐‘ฅ, ๐‘ก) and ๐‘ข(๐‘ฅ, ๐‘ก), the Poisson integral of ๐‘“, are conjugate harmonic. (2) Let ๐›พ be the lower semicircle of radius ๐‘ around the origin, and ๐œ‰ > 0. Then โˆซ ๐‘“(๐‘ง)๐‘‘๐‘ง = 2๐œ‹๐‘– Res๐‘ง=โˆ’๐‘–๐‘ก ๐‘“(๐‘ง) = โˆ’๐‘–๐‘’โˆ’2๐œ‹๐‘ก๐œ‰ , ๐›พ

where ๐‘“(๐‘ง) is the function defined in the proof of Lemma 9.4. (3) For ๐‘“ โˆˆ ๐ถ๐‘ (โ„), consider the Cauchy integral ๐น(๐‘ง) =

๐‘“(๐‘ก) 1 โˆซ ๐‘‘๐‘ก, ๐‘–๐œ‹ โ„ ๐‘ก โˆ’ ๐‘ง

for ๐‘ง = ๐‘ฅ + ๐‘–๐‘ฆ โˆˆ โ„2+ . (a) There exists a constant ๐ด > 0 such that |๐น(๐‘ง)| โ‰ค ๐ด/|๐‘ง|. (b) โˆซ ๐น(๐‘ฅ)2 ๐‘‘๐‘ฅ = 0. โ„

(c) Calculate โ„œ(๐น(๐‘ฅ)2 ) and conclude โ€–๐ป๐‘“โ€–๐ฟ2 = โ€–๐‘“โ€–๐ฟ2 . (4) Let ๐‘ฆ 1 โˆ’ |๐‘ฆ| โ‰ฅ 1 2+1 ๐‘ฆ ๐‘ฆ ฮฆ(๐‘ฆ) = { ๐‘ฆ |๐‘ฆ| < 1. ๐‘ฆ2 + 1 Then its collection {ฮฆ๐‘ก }๐‘ก>0 of dilations satisfies (a) โˆซ ฮฆ๐‘ก = 0 for all ๐‘ก > 0; โ„

(b) |ฮฆ๐‘ก (๐‘ฆ)| โ‰ค

1 for all ๐‘ก > 0 and ๐‘ฆ โˆˆ โ„; and ๐‘ก

182

9. Hilbert transform

๐‘ก for all ๐‘ก > 0 and ๐‘ฆ โˆˆ โ„, ๐‘ฆ โ‰  0. ๐‘ฆ2 (5) If ๐‘“ โˆˆ ๐ฟ1 (โ„) is differentiable at ๐‘ฅ โˆˆ โ„, then the limit (c) |ฮฆ๐‘ก (๐‘ฆ)| โ‰ค

lim โˆซ ๐‘กโ†’0

|๐‘ฆ|โ‰ฅ๐‘ก

๐‘“(๐‘ฅ โˆ’ ๐‘ฆ) ๐‘‘๐‘ฆ ๐‘ฆ

exists. (Hint: Use the identity, for any ๐›ฟ๐‘› > 0, โˆซ ๐‘กโ‰ค|๐‘ฆ| 0 such that โˆซ |๐‘‡๐‘“|๐‘ž โ‰ค ๐ด โˆซ |๐‘“|๐‘ž โ„

โ„

Notes

183 for some ๐‘ž > 1.

Notes The study of the conjugate function started in the work of several authors in the early 20th century, using complex variable methods (as in Exercise (3)). See the references in [Zyg02] for a list of such authors. The decomposition method in the proof of Theorem 9.12 is due to Alberto P. Calderรณn and Antoni Zygmund [CZ52], where they generalize to other kernels (see Exercises (9), (10) and (11)) and other dimensions. See [Ste70] for a further discussion of these methods, and a more extensive study of singular integrals.

Chapter 10

Mathematics of fractals

10.1. Hausdorff dimension The purpose of this chapter is to introduce the basic ideas in the study of fractals. We start with the Hausdorff dimension, which provides a means to quantify the complexity of a fractal set. For a set ๐ด โŠ‚ โ„๐‘‘ , we denote its diameter by diam ๐ด, diam ๐ด = sup{|๐‘ฅ โˆ’ ๐‘ฆ| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐ด}, the supremum over all distances between points in ๐ด. We observe, for example, that diam ๐ต๐‘Ÿ (๐‘ฅ) = 2๐‘Ÿ, and that, if ๐‘ฅ โˆˆ ๐ด and ๐‘Ÿ = diam ๐ด, ๐ด โŠ‚ ๐ต๐‘Ÿฬ„ (๐‘ฅ). For ๐›ฟ > 0, a ๐›ฟ-cover for ๐ด โŠ‚ โ„๐‘‘ is a collection ๐‘ˆ1 , ๐‘ˆ2 , . . . of subsets of โ„๐‘‘ such that diam ๐‘ˆ ๐‘— โ‰ค ๐›ฟ,

๐‘— = 1, 2, . . . ,

and

๐ดโŠ‚

โ‹ƒ

๐‘ˆ๐‘—.

๐‘—

The collection {๐‘ˆ ๐‘— } may be finite or infinite (as long as it is countable), and the sets ๐‘ˆ ๐‘— are arbitrary, as long as they cover ๐ด and have diameter not larger than ๐›ฟ. 185

186

10. Mathematics of fractals For ๐‘  โ‰ฅ 0, consider the number, for each ๐ด โŠ‚ โ„๐‘‘ ,

(10.1)

๐ป๐›ฟ๐‘  (๐ด) = inf { โˆ‘(diam ๐‘ˆ ๐‘— )๐‘  โˆถ {๐‘ˆ ๐‘— } is a ๐›ฟ-cover for ๐ด}. ๐‘—

In other words, we take all ๐›ฟ-covers {๐‘ˆ ๐‘— } for ๐ด, calculate for each one โˆ‘๐‘— (diam ๐‘ˆ ๐‘— )๐‘  , which is clearly nonnegative, and then take the infimum of all such sums. It is possible that all of them are divergent, and in that case we have ๐ป๐›ฟ๐‘  (๐ด) = โˆž. It is also possible that ๐ป๐›ฟ๐‘  (๐ด) = 0. For example, if ๐ด = โˆ…, or if ๐ด is finite. Indeed, if ๐ด = {๐‘ฅ1 , ๐‘ฅ2 , . . . , ๐‘ฅ๐‘˜ } โŠ‚ โ„๐‘‘ , consider the ๐›ฟ-cover for ๐ด given by the balls ๐ต๐œ€ (๐‘ฅ1 ), ๐ต๐œ€ (๐‘ฅ2 ), . . . , ๐ต๐œ€ (๐‘ฅ๐‘˜ ), with ๐œ€ > 0. Then, for ๐‘  > 0, ๐‘˜

โˆ‘ (diam ๐ต๐œ€ (๐‘ฅ๐‘— ))๐‘  = (2๐œ€)๐‘  ๐‘˜, ๐‘—=1

and thus ๐ป๐›ฟ๐‘  (๐ด) โ‰ค (2๐œ€)๐‘  ๐‘˜. As ๐œ€ > 0 is arbitrary, we have ๐ป๐›ฟ๐‘  (๐ด) = 0. It is not hard to see that, if ๐ด โŠ‚ ๐ต, then ๐ป๐›ฟ๐‘  (๐ด) โ‰ค ๐ป๐›ฟ๐‘  (๐ต), and, if ๐ด = โ‹ƒ๐‘— ๐ด๐‘— , then ๐ป๐›ฟ๐‘  (๐ด) โ‰ค โˆ‘ ๐ป๐›ฟ๐‘  (๐ด๐‘— ). ๐‘—

See Exercise (1). These are the same properties which are also satisfied by the outer measure | โ‹… |โˆ— . Now, observe that, if ๐›ฟ < ๐œ‚, then every ๐›ฟ-cover for a set ๐ด is also an ๐œ‚-cover for ๐ด. Hence, by the definition (10.1), we have ๐ป๐›ฟ๐‘  (๐ด) โ‰ฅ ๐ป๐œ‚๐‘  (๐ด). We see that ๐ป๐›ฟ๐‘  (๐ด) increases as ๐›ฟ decreases, so the limit exists as ๐›ฟ โ†’ 0 (it might be infinite). We thus define (10.2)

โ„‹ ๐‘  (๐ด) = lim ๐ป๐›ฟ๐‘  (๐ด) = sup{๐ป๐›ฟ๐‘  (๐ด) โˆถ ๐›ฟ > 0}. ๐›ฟโ†’0

โ„‹ (๐ด) is called the Hausdorff measure with exponent ๐‘  of ๐ด. As ๐ป๐›ฟ๐‘  , the Hausdorff measure also satisfies the properties of the outer measure | โ‹… |โˆ— (Exercise (2)). ๐‘ 

If {๐‘ˆ ๐‘— } is a ๐›ฟ-cover for ๐ด โŠ‚ โ„๐‘‘ , and ๐‘  > ๐‘ก โ‰ฅ 0, โˆ‘(diam ๐‘ˆ ๐‘— )๐‘  = โˆ‘(diam ๐‘ˆ ๐‘— )๐‘ โˆ’๐‘ก (diam ๐‘ˆ ๐‘— )๐‘ก โ‰ค ๐›ฟ๐‘ โˆ’๐‘ก โˆ‘(diam ๐‘ˆ ๐‘— )๐‘ก , ๐‘—

๐‘—

๐‘—

10.1. Hausdorff dimension

187

and, by the definition (10.1), (10.3)

๐ป๐›ฟ๐‘  (๐ด) โ‰ค ๐›ฟ๐‘ โˆ’๐‘ก ๐ป๐›ฟ๐‘ก (๐ด).

We thus have Theorem 10.4. Theorem 10.4. Let ๐ด โŠ‚ โ„๐‘‘ and ๐‘  > ๐‘ก โ‰ฅ 0. (1) If โ„‹ ๐‘ก (๐ด) < โˆž, then โ„‹ ๐‘  (๐ด) = 0. (2) If โ„‹ ๐‘  (๐ด) > 0, then โ„‹ ๐‘ก (๐ด) = โˆž. Proof. To prove (1), assume that โ„‹ ๐‘ก (๐ด) < โˆž. Thus ๐ป๐›ฟ๐‘ก (๐ด) is bounded in ๐›ฟ and, by (10.3), we obtain โ„‹ ๐‘  (๐ด) = lim ๐ป๐›ฟ๐‘  (๐ด) = 0. ๐›ฟโ†’0

(2) is the contrapositive of (1).

โ–ก

Theorem 10.4 allows us to conclude that there exists some number ๐ท such that โ„‹ ๐‘  (๐ด) = โˆž if ๐‘  < ๐ท, and โ„‹ ๐‘  (๐ด) = 0 if ๐‘  > ๐ท (see Figure H s (A) โˆž

D Figure 10.1. The number ๐ท, the Hausdorff dimension of ๐ด, satisfies that โ„‹ ๐‘  (๐ด) = โˆž if ๐‘  < ๐ท, and โ„‹ ๐‘  (๐ด) = 0 if ๐‘  > ๐ท.

10.1). Indeed, we see that ๐ท = inf{๐‘  โˆถ โ„‹ ๐‘  (๐ด) = 0}.

s

188

10. Mathematics of fractals

The value of โ„‹ ๐ท (๐ด) may be zero or infinity, or even a number 0 < โ„‹ ๐ท (๐ด) < โˆž. ๐ท is called the Hausdorff dimension of ๐ด, and is denoted by dim ๐ด. Example 10.5. For ๐‘  = 0, โ„‹ 0 (๐ด) = #๐ด, the number of elements of ๐ด, if ๐ด is finite, and thus dim ๐ด = 0. If ๐ด is infinite, โ„‹ 0 (๐ด) = โˆž. However, if ๐ด is countably infinite, its Hausdorff dimension is also zero (Exercise (3)). 10.6. We observe that, if ๐ด โŠ‚ โ„, โ„‹ 1 (๐ด) is the outer Lebesgue measure of ๐ด, |๐ด|โˆ— . Thus, for any interval [๐‘Ž, ๐‘], ๐‘ > ๐‘Ž, โ„‹ 1 ([๐‘Ž, ๐‘]) = ๐‘ โˆ’ ๐‘Ž, and thus intervals have Hausdorff dimension 1. 10.7. If ๐‘  > ๐‘‘ and ๐ด โŠ‚ โ„๐‘‘ , โ„‹ ๐‘  (๐ด) = 0. Indeed, consider first a unit cube ๐‘„ in โ„๐‘‘ and, given ๐›ฟ > 0, subdivide it in 2๐‘‘๐‘ subcubes ๐‘„๐‘— of sides 2โˆ’๐‘ < ๐›ฟ/๐‘‘. Each subcube has diameter diam(๐‘„๐‘— ) = 2โˆ’๐‘ โˆš๐‘‘ < ๐›ฟ, and hence {๐‘„๐‘— } is a ๐›ฟ-cover for ๐‘„. Thus โˆ‘(diam ๐‘„๐‘— )๐‘  = (2โˆ’๐‘ โˆš๐‘‘)๐‘  โ‹… 2๐‘‘๐‘ = ๐‘‘ ๐‘ /2 2(๐‘‘โˆ’๐‘ )๐‘ . ๐‘—

As ๐‘ is an arbitrary positive integer (as long as 2โˆ’๐‘ < ๐›ฟ/๐‘‘) and ๐‘  > ๐‘‘, then ๐ป๐›ฟ๐‘  (๐‘„) = 0. Hence โ„‹ ๐‘  (๐‘„) = 0, and therefore โ„‹ ๐‘  (๐ด) = 0 for any ๐ด โŠ‚ โ„๐‘‘ , because any ๐ด is contained in the union of countably many unit cubes. Therefore dim ๐ด โ‰ค ๐‘‘ for any ๐ด โŠ‚ โ„๐‘‘ . Example 10.8. Consider the Cantor ternary set, constructed by the removal of middle intervals starting from ๐ถ0 = [0, 1]. Thus ๐ถ1 = [0, 1/3] โˆช [2/3, 1], ๐ถ2 = [0, 1/9] โˆช [2/9, 1/3] โˆช [2/3, 7/9] โˆช [8/9, 1],

...

as seen in Figure 10.2. The Cantor set is

โˆž

๐ถ=

โ‹‚

๐ถ๐‘› .

๐‘›=0

The Cantor set is compact, perfect (does not have any isolated points), uncountable and of measure 0. Thus โ„‹ 1 (๐ถ) = 0. We show that its

10.1. Hausdorff dimension

189

Figure 10.2. The construction of the Cantor ternary set.

Hausdorff dimension is ๐ท=

log 2 โ‰ˆ 0.63. log 3

To see that dim ๐ถ โ‰ค ๐ท, we prove โ„‹ ๐ท (๐ถ) < โˆž. Given ๐›ฟ > 0, let ๐‘› so that 3โˆ’๐‘› < ๐›ฟ. Thus, ๐ถ๐‘› is the union of 2๐‘› intervals ๐ผ๐‘— of length 3โˆ’๐‘› < ๐›ฟ, so they form a ๐›ฟ-cover for ๐ถ. Hence 2๐‘›

๐ป๐›ฟ๐ท (๐ถ)

โ‰ค โˆ‘ (diam ๐ผ๐‘— )๐ท = (3โˆ’๐‘› )๐ท 2๐‘› = ( ๐‘—=1

2 ๐‘› ) = 1, 3๐ท

as ๐ท is the number such that 3๐ท = 2. Since ๐›ฟ > 0 is arbitrary, we have that โ„‹ ๐ท (๐ถ) โ‰ค 1, and therefore dim ๐ถ โ‰ค ๐ท. To show that dim ๐ถ โ‰ฅ ๐ท (and thus conclude they are equal), we verify that โ„‹ ๐ท (๐ถ) > 0. In fact, we prove that โ„‹ ๐ท (๐ถ) โ‰ฅ 1, and we will do it by contradiction. Assume โ„‹ ๐ท (๐ถ) < 1, so there exist intervals ๐ผ๐‘— so that ๐ถโŠ‚

โ‹ƒ

๐ผ๐‘—

and

โˆ‘(diam ๐ผ๐‘— )๐ท < 1. ๐‘—

๐‘—

By widening the intervals a little bit so that their sum is still smaller than 1, we can assume all ๐ผ๐‘— are open, and thus, by the compactness of ๐ถ, we can choose a finite number of them that still cover ๐ถ. Hence we have ๐‘

(10.9)

๐ถโŠ‚

โ‹ƒ ๐‘—=1

๐‘

๐ผ๐‘—

and

โˆ‘ (diam ๐ผ๐‘— )๐ท < 1. ๐‘—=1

Now, again by compactness of ๐ถ, we can find ๐‘€ large enough so that each interval of ๐ถ๐‘€ is completely contained in one of the ๐ผ๐‘— . We can now shorten each ๐ผ๐‘— to a closed interval in such a way that its extreme points coincide with extreme points of the intervals of ๐ถ๐‘€ , and they still cover

190

10. Mathematics of fractals

๐ถ. That is, if ๐ผ๐‘— contains the intervals ๐ท1 , ๐ท2 , . . . , ๐ท๐‘˜ in ๐ถ๐‘€ , we replace ๐ผ๐‘— with the smallest closed interval that contains ๐ท1 โˆช ๐ท2 โˆช . . . โˆช ๐ท๐‘˜ . Fix ๐ผ๐‘— . Let ๐ธ0 be the gap of largest length between the ๐ท๐‘– . By the construction of ๐ถ๐‘€ , such gap must exist. We write ๐ผ๐‘— โงต๐ธ0 as the union ๐ธ1 โˆช ๐ธ2 of closed intervals, as in Figure 10.3. We see that, by the construction, diam ๐ธ1 , diam ๐ธ2 โ‰ค diam ๐ธ0 .

Figure 10.3. The largest gap inside ๐ผ๐‘— , for the special case when ๐ผ๐‘— contains the intervals ๐ท1 , ๐ท2 , ๐ท3 . By the construction of ๐ถ ๐‘€ , such largest gap must exist and is unique.

Thus diam ๐ธ0 โ‰ฅ

diam ๐ธ1 + diam ๐ธ2 2

and diam ๐ผ๐‘— = diam ๐ธ0 + diam ๐ธ1 + diam ๐ธ2 3 3 1 1 โ‰ฅ diam ๐ธ1 + diam ๐ธ2 = 3( diam ๐ธ1 + diam ๐ธ2 ). 2 2 2 2 Now, since ๐ท < 1, the function ๐‘ฅ โ†ฆ ๐‘ฅ๐ท is concave (Exercise (6)), and thus ๐ท 1 1 (diam ๐ผ๐‘— )๐ท โ‰ฅ 3๐ท ( diam ๐ธ1 + diam ๐ธ2 ) 2 2 1 1 โ‰ฅ 3๐ท ( (diam ๐ธ1 )๐ท + (diam ๐ธ2 )๐ท ) 2 2 = (diam ๐ธ1 )๐ท + (diam ๐ธ2 )๐ท ,

where we have again used the fact that 3๐ท = 2. Therefore, we can replace ๐ผ๐‘— with the intervals ๐ธ1 and ๐ธ2 , and the new set of intervals still satisfies (10.9). We can continue replacing the intervals ๐ผ๐‘— until all gaps are removed, so we arrive to the cover formed precisely by the 2๐‘€ intervals of ๐ถ๐‘€ ,

10.2. Self-similar sets

191

which of course satisfies 2๐‘€

โˆ‘ (diam ๐ผ๐‘— )๐ท = 2๐‘€ (3โˆ’๐‘€ )๐ท = 1, ๐‘—=1

a contradiction with (10.9).

10.2. Self-similar sets Cantorโ€™s example shows us the difficulty of calculating the Hausdorff dimension of a given set ๐ด โŠ‚ โ„๐‘‘ . However, under certain conditions, it is possible to calculate the Hausdorff dimension a self-similar set. A function ๐น โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ is called a contraction if there exists a constant ๐›ผ < 1 such that |๐น(๐‘ฅ) โˆ’ ๐น(๐‘ฆ)| โ‰ค ๐›ผ|๐‘ฅ โˆ’ ๐‘ฆ| for all ๐‘ฅ, ๐‘ฆ โˆˆ โ„๐‘‘ . The number ๐›ผ is called the contraction constant of ๐น. A nonempty compact set ๐พ โŠ‚ โ„๐‘‘ is self-similar if there exist contractions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ such that (10.10)

๐พ = ๐‘“1 (๐พ) โˆช ๐‘“2 (๐พ) โˆช . . . โˆช ๐‘“๐‘ (๐พ).

One can show that ๐พ is determined by the contractions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ , that is, there exists only one nonempty compact set ๐พ that satisfies (10.10) [YHK97]. Let ๐›ผ1 , ๐›ผ2 , . . . , ๐›ผ๐‘ be the contraction constants of the functions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ in (10.10) and let ๐ท be the unique positive number such that (10.11)

๐ท ๐ท ๐›ผ๐ท 1 + ๐›ผ2 + . . . + ๐›ผ๐‘ = 1.

Such number exists because the function ๐‘ฅ โ†ฆ ๐›ผ๐‘ฅ1 + ๐›ผ๐‘ฅ2 + . . . + ๐›ผ๐‘ฅ๐‘ is strictly decreasing on [0, โˆž), equal to ๐‘ at ๐‘ฅ = 0 and goes to 0 as ๐‘ฅ โ†’ โˆž. We have the following result. Proposition 10.12. If ๐พ is the self-similar set determined by (10.10) and ๐ท satisfies (10.11), then dim ๐พ โ‰ค ๐ท. Proof. We prove that โ„‹ ๐ท (๐พ) < โˆž. For this, we will use the following notation. For a finite sequence ๐‘ค = ๐‘ค 1 ๐‘ค 2 โ‹ฏ ๐‘ค ๐‘š of length ๐‘š (we will call such sequences words), where each ๐‘ค๐‘— โˆˆ {1, 2, . . . , ๐‘}, let ๐‘“๐‘ค denote

192

10. Mathematics of fractals

the composition ๐‘“๐‘ค = ๐‘“๐‘ค1 โˆ˜ ๐‘“๐‘ค2 โˆ˜ โ‹ฏ โˆ˜ ๐‘“๐‘ค๐‘š , and ๐พ๐‘ค the set ๐‘“๐‘ค (๐พ). Each ๐พ๐‘ค is called a cell of level ๐‘š, and, by (10.10), ๐พ=

โ‹ƒ

๐พ๐‘ค ,

๐‘คโˆˆ๐‘Š๐‘š

where ๐‘Š๐‘š is the set of words of length ๐‘š. Also, since each ๐‘“๐‘— is a contraction with constant ๐›ผ๐‘— , we have diam ๐พ๐‘ค โ‰ค ๐›ผ๐‘ค1 ๐›ผ๐‘ค2 โ‹ฏ ๐›ผ๐‘ค๐‘š diam ๐พ, and thus diam ๐พ๐‘ค โ‰ค ๐ด๐‘š diam ๐พ, where ๐ด = max{๐›ผ1 , ๐›ผ2 , . . . , ๐›ผ๐‘ }. Since ๐ด < 1, given ๐›ฟ > 0 we can choose ๐‘š large enough so that the collection of cells of level ๐‘š, {๐พ๐‘ค }๐‘คโˆˆ๐‘Š๐‘š , is a ๐›ฟ-cover for ๐พ. Also, ๐ท ๐ท ๐ท โˆ‘ (diam๐พ๐‘ค )๐ท โ‰ค โˆ‘ ๐›ผ๐ท ๐‘ค1 ๐›ผ๐‘ค2 โ‹ฏ ๐›ผ๐‘ค๐‘š (diam ๐พ) ๐‘คโˆˆ๐‘Š๐‘š

๐‘คโˆˆ๐‘Š๐‘š ๐‘

๐‘

๐‘

๐ท ๐ท ๐ท = ( โˆ‘ ๐›ผ๐ท ๐‘ค1 )( โˆ‘ ๐›ผ๐‘ค2 ) โ‹ฏ ( โˆ‘ ๐›ผ๐‘ค๐‘š )(diam ๐พ) ๐‘ค1 =1

๐‘ค2 =1

๐‘ค๐‘š =1

๐ท

= (diam ๐พ) , by (10.11). Hence ๐ป๐›ฟ๐ท (๐พ) โ‰ค (diam ๐พ)๐ท and, as ๐›ฟ > 0 is arbitrary, we have that โ„‹ ๐ท (๐พ) โ‰ค (diam ๐พ)๐ท < โˆž. Therefore dim ๐พ โ‰ค ๐ท. โ–ก Consider the Cantor set ๐ถ discussed in Example 10.8. ๐ถ is selfsimilar with respect to the contractions ๐‘“1 , ๐‘“2 โˆถ โ„ โ†’ โ„ given by ๐‘“1 (๐‘ฅ) =

1 ๐‘ฅ 3

and

Thus, in this case, ๐›ผ1 = ๐›ผ2 =

๐‘“2 (๐‘ฅ) =

1 2 ๐‘ฅ+ . 3 3

1 and the number ๐ท that satisfies 3

1 ๐ท 1 ๐ท ( ) +( ) =1 3 3 is precisely ๐ท=

log 2 , log 3

which we proved to be the Hausdorff dimension of ๐ถ. However, (10.11) does not guarantee that ๐ท is the Hausdorff dimension of ๐พ for any self-similar set, as Example 10.13 shows.

10.2. Self-similar sets

193

Example 10.13. Consider the functions ๐‘”1 , ๐‘”2 โˆถ โ„ โ†’ โ„ given by 2 2 1 ๐‘ฅ and ๐‘”2 (๐‘ฅ) = ๐‘ฅ + . 3 3 3 ๐‘”1 and ๐‘”2 are contractions and, if ๐ผ = [0, 1], then ๐‘”1 (๐‘ฅ) =

๐ผ = ๐‘”1 (๐ผ) โˆช ๐‘”2 (๐ผ), so ๐ผ is self-similar with respect to ๐‘”1 and ๐‘”2 . Both functions have contraction constant 2/3, and the number ๐ท that satisfies 2 ๐ท 2 ๐ท ( ) +( ) =1 3 3 is

log 2 . log 3/2 Note that ๐ท > 1, so it is not equal to dim ๐ผ. ๐ท=

The problem with Example 10.13 is the overlap of the images ๐‘”1 (๐ผ) = [0, 2/3]

and

๐‘”2 (๐ผ) = [1/3, 1]

of ๐ผ under the contractions ๐‘”1 and ๐‘”2 , so we need conditions to avoid this problem. We say that a function ๐‘“ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ is a similitude if there exists a constant ๐›ผ such that |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฆ)| = ๐›ผ|๐‘ฅ โˆ’ ๐‘ฆ|. It can be verified that, if ๐‘“ is a similitude, then itโ€™s a composition of a dilation, a rotation and a translation (Exercise (7)). Thus, for any ball ๐ต๐‘Ÿ (๐‘ฅ) โŠ‚ โ„๐‘‘ , (10.14)

๐‘“(๐ต๐‘Ÿ (๐‘ฅ)) = ๐ต๐›ผ๐‘Ÿ (๐‘“(๐‘ฅ)).

Let ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ be contractive similitudes. They satisfy the open set condition if there exists a bounded open set ๐‘ˆ โŠ‚ โ„๐‘‘ such that (1) ๐‘“๐‘— (๐‘ˆ) โŠ‚ ๐‘ˆ for each ๐‘— = 1, 2, . . . , ๐‘; (2) ๐‘“๐‘– (๐‘ˆ) โˆฉ ๐‘“๐‘— (๐‘ˆ) = โˆ…, for ๐‘–, ๐‘— = 1, 2, . . . , ๐‘ such that ๐‘– โ‰  ๐‘—. One can verify that, since ๐‘“1 (๐‘ˆ) โˆช ๐‘“2 (๐‘ˆ) โˆช . . . โˆช ๐‘“๐‘ (๐‘ˆ) โŠ‚ ๐‘ˆ, then ๐พ โŠ‚ ๐‘ˆ,ฬ„ where ๐‘ˆฬ„ is the closure of ๐‘ˆ (Exercise (8)).

194

10. Mathematics of fractals

Condition (2) guarantees that there are no interior overlaps between the images ๐‘“๐‘— (๐พ) of the contractions. Theorem 10.15. Let ๐‘“๐‘— โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ , ๐‘— = 1, 2, . . . , ๐‘, be contractive similitudes with constants ๐›ผ๐‘— < 1, respectively, that satisfy the open set condition. If ๐พ is the self-similar set with respect to the ๐‘“๐‘— , then dim ๐พ = ๐ท, where ๐ท is the unique solution to (10.11). Proof. We already know that dim ๐พ โ‰ค ๐ท, by Proposition 10.12. To prove that dim ๐พ โ‰ฅ ๐ท, we verify that โ„‹ ๐ท (๐พ) > 0. Let ๐‘ˆ โŠ‚ โ„๐‘‘ be a bounded open set as in the open set condition, and let ๐›ผ, ๐›ฝ > 0 and ๐‘ฅ0 , ๐‘ฆ0 โˆˆ โ„๐‘‘ such that (10.16)

๐ต๐›ผฬ„ (๐‘ฅ0 ) โŠ‚ ๐‘ˆ โŠ‚ ๐ต๐›ฝ (๐‘ฆ0 ).

Since the ๐‘“๐‘— are similitudes, for each word ๐‘ค โˆˆ ๐‘Š๐‘š we have, by (10.14), (10.17)

๐ต๐›ผฬ„ ๐‘ค ๐›ผ (๐‘“๐‘ค (๐‘ฅ0 )) โŠ‚ ๐‘ˆ๐‘ค โŠ‚ ๐ต๐›ผ๐‘ค ๐›ฝ (๐‘“๐‘ค (๐‘ฆ0 )),

where ๐›ผ๐‘ค = ๐›ผ๐‘ค1 ๐›ผ๐‘ค2 โ‹ฏ ๐›ผ๐‘ค๐‘š and ๐‘ˆ๐‘ค = ๐‘“๐‘ค (๐‘ˆ). Let ๐›พ = min{๐›ผ1 , ๐›ผ2 , . . . , ๐›ผ๐‘ } and ๐œ=(

๐›ผ๐›พ ๐‘‘ ) . 2๐›ฝ + 1

We will prove that โ„‹ ๐ท (๐พ) โ‰ฅ ๐œ. In order to arrive to a contradiction, assume โ„‹ ๐ท (๐พ) < ๐œ. Let ๐›ฟ > 0 be small enough (๐›ฟ < ๐œ is sufficient) and {๐‘‰ ๐‘— } a ๐›ฟ-cover for ๐พ such that (10.18)

โˆ‘ diam(๐‘‰ ๐‘— )๐ท < ๐œ. ๐‘—

By widening the ๐‘‰ ๐‘— up we can assume they are open, and so we can also assume they are finite because ๐พ is compact. For each ๐‘—, consider the set of words ฮ›๐‘— = {๐‘ค โˆถ ๐›ผ๐‘ค โ‰ค diam ๐‘‰ ๐‘— < ๐›ผ๐‘คโ€ฒ } where, if ๐‘ค = ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š , then ๐‘คโ€ฒ = ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘šโˆ’1 . That is, ฮ›๐‘— is the set of words ๐‘ค of sequences ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š which make the product ๐›ผ๐‘ค1 ๐›ผ๐‘ค2 โ‹ฏ ๐›ผ๐‘ค๐‘š precisely smaller or equal to diam ๐‘‰ ๐‘— , and larger if we miss the last factor ๐›ผ๐‘ค๐‘š . The words in ฮ›๐‘— satisfy the following two properties.

10.2. Self-similar sets

195

โ€ข If ๐‘ข, ๐‘ค โˆˆ ฮ›๐‘— are different, then the letters of ๐‘ข cannot be the first letters of ๐‘ค, nor the other way around. This guarantees that, by the open set condition, ๐‘ˆแต† โˆฉ ๐‘ˆ๐‘ค = โˆ…

if ๐‘ข, ๐‘ค โˆˆ ฮ›๐‘— , ๐‘ข โ‰  ๐‘ค.

โ€ข If ๐‘ค โˆˆ ฮ›๐‘— , ๐›พ diam ๐‘‰ ๐‘— < ๐›ผ๐‘ค โ‰ค diam ๐‘‰ ๐‘— , so all the products ๐›ผ๐‘ค , for ๐‘ค โˆˆ ฮ›๐‘— , are essentially of the same size. Let ๐‘ฅ โˆˆ ๐‘‰ ๐‘— . Then ๐‘‰ ๐‘— โŠ‚ ๐ตdiam ๐‘‰๐‘— (๐‘ฅ). Using (10.17), we have ๐‘ˆ๐‘ค โŠ‚ ๐ต๐›ผ๐‘ค ๐›ฝ (๐‘“๐‘ค (๐‘ฆ0 )) โŠ‚ ๐ต2๐›ผ๐‘ค ๐›ฝ+diam ๐‘‰๐‘— (๐‘ฅ) ฬ„ โˆฉ ๐‘‰ ๐‘— โ‰  โˆ…. As ๐›ผ๐‘ค ๐›ฝ โ‰ค ๐›ฝ diam ๐‘‰ ๐‘— , we have for each ๐‘ค โˆˆ ฮ›๐‘— such that ๐‘ˆ๐‘ค (10.19)

๐‘ˆ๐‘ค โˆˆ ๐ต(2๐›ฝ+1) diam ๐‘‰๐‘— (๐‘ฅ).

Thus, the ball ๐ต(2๐›ฝ+1) diam ๐‘‰๐‘— (๐‘ฅ) contains all sets ๐‘ˆ๐‘ค such that ๐‘ค โˆˆ ฮ›๐‘— ฬ„ โˆฉ ๐‘‰ ๐‘— โ‰  โˆ…. Moreover, each set ๐‘ˆ๐‘ค contains the ball ๐ต๐›ผ ๐›ผ (๐‘“๐‘ค (๐‘ฅ0 )), and ๐‘ˆ๐‘ค ๐‘ค by (10.17), and all of them are disjoint, so the ball ๐ต(2๐›ฝ+1) diam ๐‘‰๐‘— (๐‘ฅ) contains, say, ๐‘ balls of radius ๐›ผ๐‘ค ๐›ผ = ๐›ผ๐‘คโ€ฒ ๐›ผ๐‘ค๐‘š ๐›ผ > ๐›พ๐›ผ diam ๐‘‰ ๐‘— , where ฬ„ โˆฉ ๐‘‰ ๐‘— โ‰  โˆ…}. ๐‘ = #{๐‘ค โˆˆ ฮ›๐‘— โˆถ ๐‘ˆ๐‘ค Since the measure |๐ต๐‘Ÿ (๐‘ฅ)| of a ball of radius ๐‘Ÿ is given by ๐œ”๐‘‘ ๐‘Ÿ๐‘‘ /๐‘‘, we have the inequality ๐‘‘

๐‘‘

((2๐›ฝ + 1) diam ๐‘‰ ๐‘— ) โ‰ฅ ๐‘(๐›พ๐›ผ diam ๐‘‰ ๐‘— ) , from the fact that we have a ball of radius (2๐›ฝ + 1) diam ๐‘‰ ๐‘— that contains ๐‘ disjoint balls of radius ๐›พ๐›ผ diam ๐‘‰ ๐‘— . Therefore ๐‘โ‰ค(

2๐›ฝ + 1 ๐‘‘ 1 ) = . ๐›พ๐›ผ ๐œ

Now, for each ๐‘š and ๐‘—, we consider the sum ๐ด๐‘š (๐‘—) =

โˆ‘ ๐‘ค=๐‘ค1 ๐‘ค2 . . .๐‘ค๐‘š ๐‘ˆฬ„ ๐‘ค โˆฉ๐‘‰๐‘— โ‰ 0

๐›ผ๐ท ๐‘ค.

196

10. Mathematics of fractals

๐ด๐‘š (๐‘—) is decreasing in ๐‘š because ๐‘ˆ๐‘ค๐‘– โŠ‚ ๐‘ˆ๐‘ค for each ๐‘– = 1, 2, . . . , ๐‘, and ๐‘ โˆ‘๐‘–=1 ๐›ผ๐ท ๐‘– = 1. Hence, if ๐ฟ = max{๐‘š โˆถ ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š โˆˆ ฮ›๐‘— }, we have, as each ๐›ผ๐‘– < 1, โˆ‘

๐ด๐ฟ (๐‘—) โ‰ค

๐›ผ๐ท ๐‘ค โ‰ค

๐‘คโˆˆฮ›๐‘— ๐‘ˆฬ„ ๐‘ค โˆฉ๐‘‰๐‘— โ‰ โˆ…

โˆ‘

(diam ๐‘‰ ๐‘— )๐ท = ๐‘(diam ๐‘‰ ๐‘— )๐ท

๐‘คโˆˆฮ›๐‘— ๐‘ˆฬ„ ๐‘ค โˆฉ๐‘‰๐‘— โ‰ โˆ…

1 (diam ๐‘‰ ๐‘— )๐ท . ๐œ However, as ๐พ โŠ‚ ๐‘ˆ,ฬ„ ๐พ โŠ‚ โ‹ƒ๐‘— ๐‘‰ ๐‘— , we conclude โ‰ค

โˆ‘

1=

๐‘ค=๐‘ค1 ๐‘ค2 . . .๐‘ค๐ฟ ๐‘ˆฬ„ ๐‘ค โˆฉ๐พโ‰ โˆ…


0. (a) If ๐ด โŠ‚ ๐ต, then ๐ป๐›ฟ๐‘  (๐ด) โ‰ค ๐ป๐›ฟ๐‘  (๐ต). (b) If ๐ด = โ‹ƒ๐‘— ๐ด๐‘— , then ๐ป๐›ฟ๐‘  (๐ด) โ‰ค โˆ‘ ๐ป๐›ฟ๐‘  (๐ด๐‘— ). ๐‘—

(2) Let ๐‘  โ‰ฅ 0. (a) If ๐ด โŠ‚ ๐ต, then โ„‹ ๐‘  (๐ด) โ‰ค โ„‹ ๐‘  (๐ต). (b) If ๐ด = โ‹ƒ๐‘— ๐ด๐‘— , then โ„‹ ๐‘  (๐ด) โ‰ค โˆ‘ โ„‹ ๐‘  (๐ด๐‘— ). ๐‘— ๐‘ 

(c) If dist(๐ด, ๐ต) > 0, then โ„‹ (๐ด โˆช ๐ต) = โ„‹ ๐‘  (๐ด) + โ„‹ ๐‘  (๐ต). (3) If ๐ด is countable, then dim ๐ด = 0. (4) There exist constants ๐‘, ๐ถ > 0 such that, for any measurable ๐ด โŠ‚ โ„๐‘‘ , ๐‘|๐ด| โ‰ค โ„‹ ๐‘‘ (๐ด) โ‰ค ๐ถ|๐ด|, and thus โ„‹ ๐‘‘ is comparable to Lebesgue measure on โ„๐‘‘ . (5) If ๐ด โŠ‚ โ„๐‘‘ is open, then dim ๐ด = ๐‘‘. (6) If 0 < ๐‘ < 1, the function ๐‘ฅ โ†ฆ ๐‘ฅ๐‘ is concave, that is, for ๐‘ฅ, ๐‘ฆ > 0 and 0 โ‰ค ๐‘ก โ‰ค 1, (๐‘ก๐‘ฅ + (1 โˆ’ ๐‘ก)๐‘ฆ)๐‘ โ‰ฅ ๐‘ก๐‘ฅ๐‘ + (1 โˆ’ ๐‘ก)๐‘ฆ๐‘ . (7) Let ๐‘“ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ be a similitude with coefficient ๐›ผ > 0: for every ๐‘ฅ, ๐‘ฆ โˆˆ โ„๐‘‘ , |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฆ)| = ๐›ผ|๐‘ฅ โˆ’ ๐‘ฆ| 1 Let ๐‘”(๐‘ฅ) = (๐‘“(๐‘ฅ) โˆ’ ๐‘“(0)). ๐›ผ (a) For all ๐‘ฅ, ๐‘ฆ โˆˆ โ„๐‘‘ , ๐‘”(๐‘ฅ) โ‹… ๐‘”(๐‘ฆ) = ๐‘ฅ โ‹… ๐‘ฆ. (b) If ๐‘’ 1 , ๐‘’ 2 , . . . , ๐‘’ ๐‘‘ is the standard basis for โ„๐‘‘ , then ๐‘”(๐‘’ 1 ), ๐‘”(๐‘’ 2 ), . . ., ๐‘”(๐‘’ ๐‘‘ ) is an orthonormal basis for โ„๐‘‘ . (c) For ๐‘ฅ, ๐‘ฆ โˆˆ โ„๐‘‘ and ๐‘— = 1, 2, . . . , ๐‘‘, ๐‘”(๐‘ฅ + ๐‘ฆ) โ‹… ๐‘”(๐‘’๐‘— ) = (๐‘”(๐‘ฅ) + ๐‘”(๐‘ฆ)) โ‹… ๐‘”(๐‘’๐‘— ). (d) For ๐‘ฅ โˆˆ โ„๐‘‘ , ๐œ† โˆˆ โ„ and ๐‘— = 1, 2, . . . , ๐‘‘, ๐‘”(๐œ†๐‘ฅ) โ‹… ๐‘”(๐‘’๐‘— ) = ๐œ†๐‘”(๐‘ฅ) โ‹… ๐‘”(๐‘’๐‘— ).

202

10. Mathematics of fractals (e) ๐‘” is an orthogonal linear transformation. (f) Conclude that every similitude in โ„๐‘‘ is of the form ๐‘“(๐‘ฅ) = ๐›ผ๐‘€๐‘ฅ + ๐‘ฅ0 , where ๐›ผ > 0, ๐‘€ โˆˆ ๐‘‚(๐‘‘) and ๐‘ฅ0 โˆˆ โ„๐‘‘ .

(8) Let ๐พ be the self-similar set with respect to the contractions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ โˆถ โ„๐‘‘ โ†’ โ„๐‘‘ , and suppose ๐ด โŠ‚ โ„๐‘‘ is nonempty and satisfies ๐‘“1 (๐ด) โˆช ๐‘“2 (๐ด) โˆช . . . โˆช ๐‘“๐‘ (๐ด) โŠ‚ ๐ด. Then ๐พ โŠ‚ ๐ด,ฬ„ where ๐ดฬ„ is the closure of ๐ด. (Hint: Prove that, if ๐ด๐œ€ = {๐‘ฅ โˆˆ โ„๐‘‘ โˆถ there exists ๐‘ฆ โˆˆ ๐ด such that |๐‘ฅ โˆ’ ๐‘ฆ| < ๐œ€}, then ๐พ โŠ‚ ๐ด๐œ€ for any ๐œ€ > 0.) (9) Describe explicitly the convex hull of the Hata tree set of Example 10.22, and prove that its interior is an open set under which the contractions that define the Hata set satisfy the open set conditions. (10) Describe explicitly an open set, under which the open set condition is satisfied, for the golden fractal of Example 10.23. (11) The polynomial ๐‘ฅ3 + 2๐‘ฅ โˆ’ 1 has only one real root, and it belongs to the interval (0, ๐›พ), where ๐›พ = 1/๐œ‘ is the reciprocal of the golden ratio.

Notes The Hausdorff dimension was introduced by Felix Hausdorff in [Hau18]. Theorem 10.15 was proven by John Hutchinson in [Hut81]. It is sometimes known as Moranโ€™s theorem due to Patrick Alfred Pierce Moranโ€™s related work in [Mor46]. The proof presented here is from [YHK97]. The Sierpiล„ski gasket was studied by Wacล‚aw Sierpiล„ski in [Sie15]. The snowflake set is presented in [Kig01]. The Hata set was introduced by Masayoshi Hata in [Hat85], and the golden fractal by Marc Frantz in [Fra09].

Chapter 11

The Laplacian on the Sierpiล„ski gasket

In the previous chapters, we studied harmonic funcions on regions given by open subsets of the Euclidean space โ„๐‘‘ . However, in the study of diffusion in disordered media, fractals sets, as the ones we studied in Chapter 10, are better suited as models to those systems. The purpose of this chapter is to construct a Laplacianโ€”and study the harmonic functions, on the Sierpiล„ski gasketโ€”the set ๐‘† described in Example 10.20. We will do this by working on discrete approximations to ๐‘†. To motivate our methods, we start by discussing a discrete approach to harmonic analysis on the interval [0, 1].

11.1. Discrete energy on the interval We start with the following model. Consider a spring placed between two nodes at points (0, ๐‘Ž) and (1, ๐‘) in the plane, as in Figure 11.1 (left). If we assume that it is in equilibrium when ๐‘Ž = ๐‘, then its energy is given by a constant multiple of (11.1)

โ„ฐ0 = (๐‘Ž โˆ’ ๐‘)2 .

Now, assume we put a node at the point (1/2, ๐‘ฅ), as in Figure 11.1 (right). We now have a pair of springs whose total energy is given by (11.2)

โ„ฐ1 = ๐›ผ((๐‘Ž โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘)2 ), 203

204

11. The Laplacian on the Sierpiล„ski gasket

Figure 11.1. The spring on the left has energy โ„ฐ0 = (๐‘Ž โˆ’ ๐‘)2 , while the pair of springs on the right have total energy โ„ฐ1 = ๐›ผ((๐‘Ž โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘)2 ).

for an appropriate constant ๐›ผ. It is not hard to see that the value of ๐‘ฅ that minimizes (11.2) is ๐‘Ž+๐‘ ๐‘ฅโˆ— = , 2 and at this value the energy โ„ฐ1 in (11.2) is equal to ๐›ผ โ„ฐ1 = . 2 Note that if the node is placed at (1/2, ๐‘ฅโˆ— ), then it is the midpoint of the line segment from (0, ๐‘Ž) to (1, ๐‘). Hence the total energy of the springs must be the same as โ„ฐ0 , and thus we must choose ๐›ผ = 2. If we continue subdividing dyadically the interval [0, 1], we obtain a partition 1 2 3 ๐‘‰๐‘š = {0, ๐‘š , ๐‘š , ๐‘š , . . . , 1} 2 2 2 for each ๐‘š โ‰ฅ 1. Given a real valued function ๐‘ข on ๐‘‰๐‘š , we define its energy โ„ฐ๐‘š (๐‘ข) as the quadratic form 2๐‘š

(11.3)

โ„ฐ๐‘š (๐‘ข) = 2

๐‘š

โˆ‘ (๐‘ข( ๐‘˜=1

๐‘˜โˆ’1 ๐‘˜ 2 โˆ’ ๐‘ข( ) )) . 2๐‘š 2๐‘š

If ๐‘ข(0) = ๐‘Ž, ๐‘ข(1) = ๐‘, then โ„ฐ0 (๐‘ข) = (๐‘Ž โˆ’ ๐‘)2 , and, as we observed above, min{โ„ฐ1 (๐‘ข) โˆถ ๐‘ข(0) = ๐‘Ž, ๐‘ข(1) = ๐‘} = โ„ฐ0 (๐‘ข) = (๐‘Ž โˆ’ ๐‘)2 , and is attained when ๐‘ข(1/2) = (๐‘Ž + ๐‘)/2. Inductively, we can verify that min{โ„ฐ๐‘š (๐‘ข) โˆถ ๐‘ข(0) = ๐‘Ž, ๐‘ข(1) = ๐‘} = โ„ฐ0 (๐‘ข) for each ๐‘š โ‰ฅ 1, and that is attained at the function ๐‘ข that satisfies 1 ๐‘˜โˆ’1 2๐‘˜ โˆ’ 1 ๐‘˜ (11.4) ๐‘ข( ๐‘š ) = (๐‘ข( ๐‘šโˆ’1 ) + ๐‘ข( ๐‘šโˆ’1 )) 2 2 2 2

11.1. Discrete energy on the interval

205

for each ๐‘š โ‰ฅ 1 and each 1 โ‰ค ๐‘˜ โ‰ค 2๐‘šโˆ’1 , and thus can be constructed inductively by the algorithm (11.4). In fact, (11.4) describes the dyadic points of the line segment from (0, ๐‘ข(0)) to (1, ๐‘ข(1)). By the mean value theorem, if ๐‘ข is a differentiable function in [0, 1], for each 1 โ‰ค ๐‘˜ โ‰ค 2๐‘š there is some ๐‘ก ๐‘˜ โˆˆ [(๐‘˜ โˆ’ 1)/2๐‘š , ๐‘˜/2๐‘š ] such that 1 ๐‘˜โˆ’1 ๐‘˜ ) โˆ’ ๐‘ข( ๐‘š ) = ๐‘ขโ€ฒ (๐‘ก ๐‘˜ ) โ‹… ๐‘š , 2๐‘š 2 2 and hence (11.3) can be written as ๐‘ข(

2๐‘š

2๐‘š

๐‘ขโ€ฒ (๐‘ก ) 2 1 โ„ฐ๐‘š (๐‘ข) = 2 โˆ‘ ( ๐‘š๐‘˜ ) = โˆ‘ ๐‘ขโ€ฒ (๐‘ก ๐‘˜ )2 โ‹… ๐‘š , 2 2 ๐‘˜=1 ๐‘˜=1 ๐‘š

which, if ๐‘ขโ€ฒ is Riemann-integrable on [0, 1], is a Riemann sum of the integral 1

โ„ฐ(๐‘ข) = โˆซ ๐‘ขโ€ฒ (๐‘ก)2 ๐‘‘๐‘ก,

(11.5)

0

the energy of the function ๐‘ข on [0, 1]. We note that the minimizers of this energy are the linear functions, which are the harmonic functions in [0, 1], and the continuous limit of algorithm (11.4). Moreover, we can observe that, if we polarize the quadratic form โ„ฐ๐‘š , we see that the bilinear form 2๐‘š

โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = 2

๐‘š

โˆ‘ (๐‘ข( ๐‘˜=1

๐‘˜โˆ’1 ๐‘˜ ๐‘˜โˆ’1 ๐‘˜ ) โˆ’ ๐‘ข( ๐‘š ))(๐‘ฃ( ๐‘š ) โˆ’ ๐‘ฃ( ๐‘š )) 2๐‘š 2 2 2

converges, for ๐‘ข, ๐‘ฃ differentiable in [0, 1] and ๐‘ขโ€ฒ , ๐‘ฃโ€ฒ Riemann-integrable, to 1

โ„ฐ(๐‘ข, ๐‘ฃ) = โˆซ ๐‘ขโ€ฒ (๐‘ก)๐‘ฃโ€ฒ (๐‘ก)๐‘‘๐‘ก, 0

the energy form on the interval studied in Section 1.4. If, say, ๐‘ข โˆˆ ๐ถ 2 ([0, 1]), and ๐‘ฃ is zero at the boundary points 0 and 1, then we can integrate by parts to obtain 1

โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ’ โˆซ ๐‘ขโ€ณ (๐‘ก)๐‘ฃ(๐‘ก)๐‘‘๐‘ก.

(11.6)

0 โ€ณ

Note that ๐‘ข is the Laplacian of ๐‘ข for a one-variable function, and that ๐‘ข is harmonic (๐‘ขโ€ณ (๐‘ก) = 0 for all ๐‘ก โˆˆ [0, 1]) if and only if it is a linear function.

206

11. The Laplacian on the Sierpiล„ski gasket We can also obtain ๐‘ขโ€ณ (๐‘ก) by a discrete limit. Note that we can rewrite 2๐‘š

โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ’ โˆ‘ ๐ป๐‘š ๐‘ข( ๐‘˜=0

๐‘˜ ๐‘˜ )๐‘ฃ( ๐‘š ), 2๐‘š 2

where ๐ป๐‘š ๐‘ข(

๐‘˜ 1 ๐‘˜โˆ’1 ๐‘˜+1 ๐‘˜ ) = โˆ’๐‘š (๐‘ข( ๐‘š ) + ๐‘ข( ๐‘š ) โˆ’ 2๐‘ข( ๐‘š )) ๐‘š 2 2 2 2 2

if 1 โ‰ค ๐‘˜ โ‰ค 2๐‘š โˆ’ 1, 1 1 (๐‘ข( ๐‘š ) โˆ’ ๐‘ข(0)), 2โˆ’๐‘š 2

๐ป๐‘š ๐‘ข(0) = and

2๐‘š โˆ’ 1 ) โˆ’ ๐‘ข(1)). 2๐‘š If, for 1 โ‰ค ๐‘˜ โ‰ค 2๐‘š โˆ’ 1, we define the piecewise linear function ๐‘ฃ by ๐ป๐‘š ๐‘ข(1) =

๐‘ฃ(

1

2โˆ’๐‘š

(๐‘ข(

1 if ๐‘— = ๐‘˜ ๐‘— )={ ๐‘š 2 0 otherwise,

then

๐‘˜ ). 2๐‘š In fact, one can verify explicitly (Exercise (1)) that โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ’๐ป๐‘š ๐‘ข( 1

โˆซ ๐‘ขโ€ณ (๐‘ก)๐‘ฃ(๐‘ก)๐‘‘๐‘ก = ๐ป๐‘š ๐‘ข( 0

๐‘˜ ). 2๐‘š

2

As ๐‘ข โˆˆ ๐ถ ([0, 1]), given ๐œ€ > 0 we can choose ๐‘š large enough so that |๐‘ขโ€ณ (๐‘ฅ) โˆ’ ๐‘ขโ€ณ (๐‘ฆ)| < ๐œ€ for |๐‘ฅ โˆ’ ๐‘ฆ| โ‰ค 2โˆ’๐‘š . Thus 1

1

1

| โˆซ ๐‘ขโ€ณ (๐‘ก)๐‘ฃ(๐‘ก)๐‘‘๐‘ก โˆ’ ๐‘ขโ€ณ ( ๐‘˜ ) โˆซ ๐‘ฃ(๐‘ก)๐‘‘๐‘ก| โ‰ค โˆซ |๐‘ขโ€ณ (๐‘ก) โˆ’ ๐‘ขโ€ณ ( ๐‘˜ )|๐‘ฃ(๐‘ก)๐‘‘๐‘ก | | | 2๐‘š 2๐‘š | 0

0

0

1

< ๐œ€ โˆซ ๐‘ฃ(๐‘ก)๐‘‘๐‘ก. 0

Since

1

โˆซ ๐‘ฃ(๐‘ก)๐‘‘๐‘ก = 0

1 , 2๐‘š

11.2. Harmonic structure on the Sierpiล„ski gasket

207

we have that, for any dyadic point ๐‘ฅ0 in [0, 1], 1 ๐ป๐‘š ๐‘ข(๐‘ฅ0 ) = ๐‘ขโ€ณ (๐‘ฅ0 ). lim ๐‘šโ†’โˆž 2โˆ’๐‘š The previous limit is also a known result from calculus. We call the ๐ป๐‘š the sequence of discrete Laplacians on the interval [0, 1].

11.2. Harmonic structure on the Sierpiล„ski gasket We now proceed to define a Laplacian on the Sierpiล„ski gasket ๐‘† introduced in Example 10.20. Recall that ๐‘† is the self-similar set in the plane that satisfies ๐‘† = ๐‘“1 (๐‘†) โˆช ๐‘“2 (๐‘†) โˆช ๐‘“3 (๐‘†), where the ๐‘“๐‘– โˆถ โ„2 โ†’ โ„2 are the contractions 1 ๐‘“๐‘– (๐‘ฅ) = (๐‘ฅ + ๐‘ ๐‘– ), 2 and the points ๐‘1 = (1/2, โˆš3/2), ๐‘2 = (0, 0) and ๐‘3 = (1, 0) are the vertices of an equilateral triangle (Figure 11.2). As in the case of the interval, we will do this by constructing a sequence of quadratic forms

Figure 11.2. The Sierpiล„ski gasket.

on approximating points to ๐‘†, whose limit will define a quadratic form on functions on ๐‘†. This quadratic form willl induce a Laplacian on the set ๐‘†, which will also be a limit of discrete difference operators. We start by considering three springs on the sides of an equilateral triangle, with nodes at each of its vertices, at heights ๐‘Ž, ๐‘ and ๐‘, as in

208

11. The Laplacian on the Sierpiล„ski gasket

Figure 11.3. Three springs with nodes at heights ๐‘Ž, ๐‘ and ๐‘.

Figure 11.3. The energy of this system of springs is now defined to be โ„ฐ0 = (๐‘Ž โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘Ž)2 . Suppose that we want to add nodes at their middle points, with heights ๐‘ฅ, ๐‘ฆ, ๐‘ง, as in Figure 11.4. The energy is now given by โ„ฐ1 = ๐›ผ((๐‘Ž โˆ’ ๐‘ง)2 + (๐‘ง โˆ’ ๐‘ฆ)2 + (๐‘ฆ โˆ’ ๐‘Ž)2 + (๐‘ง โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘ง)2 + (๐‘ฆ โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘ฆ)2 ), where the constant ๐›ผ is chosen so that, if ๐‘ฅ, ๐‘ฆ and ๐‘ง minimize โ„ฐ1 , then this minimum is equal to โ„ฐ0 . Thus, we can calculate (Exercise (3)) that the minimizing values are (11.7)

๐‘ฅโˆ— =

๐‘Ž + 2๐‘ + 2๐‘ , 5

๐‘ฆโˆ— =

2๐‘Ž + ๐‘ + 2๐‘ , 5

๐‘งโˆ— =

2๐‘Ž + 2๐‘ + ๐‘ , 5

and that

5 . 3 We observe that the values (11.7) are averages of the values at the vertices, with weights 2 to 1 depending on whether the corresponding vertex is adyacent or opposite from each middle point node. ๐›ผ=

Let ๐‘‰0 = {๐‘1 , ๐‘2 , ๐‘3 } and define, for each ๐‘š โ‰ฅ 1, the set ๐‘‰๐‘š = ๐‘“1 (๐‘‰๐‘šโˆ’1 ) โˆช ๐‘“2 (๐‘‰๐‘šโˆ’1 ) โˆช ๐‘“3 (๐‘‰๐‘šโˆ’1 ). Note that each ๐‘ฅ โˆˆ ๐‘‰๐‘š is of the form ๐‘“๐‘ค (๐‘ ๐‘– ), where, as in Chapter 10, ๐‘ค = ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š โˆˆ ๐‘Š๐‘š , where ๐‘Š๐‘š is the set of words of length ๐‘š (with

11.2. Harmonic structure on the Sierpiล„ski gasket

209

Figure 11.4. We add three more nodes, at the middle points of each previous spring, with heights ๐‘ฅ, ๐‘ฆ, ๐‘ง that minimize the energy โ„ฐ1 .

letters ๐‘ค๐‘— = 1, 2, 3), and ๐‘“๐‘ค = ๐‘“๐‘ค1 โˆ˜ ๐‘“๐‘ค2 โˆ˜ โ‹ฏ โˆ˜ ๐‘“๐‘ค๐‘š .

Figure 11.5. The sets ๐‘‰ 0 , ๐‘‰ 1 and ๐‘‰ 2 . The edges of the graphs join adjacent vertices at each level.

We thus have ๐‘‰๐‘š =

โ‹ƒ

๐‘“๐‘ค (๐‘‰0 ).

๐‘คโˆˆ๐‘Š๐‘š

We say that two vertices ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰๐‘š are adjacent, or neighbors, and write ๐‘ฅ โˆผ๐‘š ๐‘ฆ, if there exists a word ๐‘ค โˆˆ ๐‘Š๐‘š such that ๐‘ฅ = ๐‘“๐‘ค (๐‘ ๐‘– ) and ๐‘ฆ = ๐‘“๐‘ค (๐‘๐‘— ), for some ๐‘–, ๐‘— = 1, 2, 3. Figure 11.5 shows ๐‘‰0 , ๐‘‰1 and ๐‘‰2 , with edges joining neighboring vertices. For a function ๐‘ข defined on ๐‘‰๐‘š , define the quadratic form 2 5 ๐‘š (11.8) โ„ฐ๐‘š (๐‘ข) = ( ) โˆ‘ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)) . 3 ๐‘ฅโˆผ ๐‘ฆ ๐‘š

210

11. The Laplacian on the Sierpiล„ski gasket

The quadratic form โ„ฐ๐‘š defined in (11.8) is positive semidefinite, and from its definition on can verify that โ„ฐ๐‘š (๐‘ข) = 0 if and only if ๐‘ข is constant on ๐‘‰๐‘š . โ„ฐ๐‘š is called the discrete energy of level ๐‘š on the Sierpiล„ski gasket. If ๐‘ข is a function on ๐‘‰๐‘š , then, for each ๐‘— = 1, 2, 3, ๐‘ข โˆ˜ ๐‘“๐‘— is a function on ๐‘‰๐‘šโˆ’1 . From (11.8) we see that 3

(11.9)

โ„ฐ๐‘š (๐‘ข) =

5 โˆ‘โ„ฐ (๐‘ข โˆ˜ ๐‘“๐‘— ). 3 ๐‘—=1 ๐‘šโˆ’1

Thus, from the previous discussion, if ๐‘ฃ is a function on ๐‘‰๐‘šโˆ’1 , then (11.10) min{โ„ฐ๐‘š (๐‘ข) โˆถ functions ๐‘ข on ๐‘‰๐‘š such that ๐‘ข|๐‘‰๐‘šโˆ’1 = ๐‘ฃ} = โ„ฐ๐‘šโˆ’1 (๐‘ฃ), where the minimizing function ๐‘ข is calculated using (11.7). The sequence โ„ฐ๐‘š is called a harmonic structure on ๐‘†. Using (11.10), one can verify by induction that, given a function ๐œŒ on ๐‘‰0 , we have that min{โ„ฐ๐‘š (๐‘ข) โˆถ functions ๐‘ข on ๐‘‰๐‘š such that ๐‘ข|๐‘‰0 = ๐œŒ} = โ„ฐ0 (๐œŒ). Again, the minimizer function can be calculated recursively using the algorithm (11.7). The points in ๐‘‰0 are called the boundary of ๐‘†, and the minimizer function ๐‘ข is called harmonic with boundary values ๐œŒ. Figure 11.6 shows the harmonic function with boundary values given by ๐œŒ(๐‘1 ) = 1 and ๐œŒ(๐‘2 ) = ๐œŒ(๐‘3 ) = 0. If ๐‘ข is harmonic, we can show that there exists ๐ด > 0 such that, for ๐‘š โ‰ฅ 0, 3 ๐‘š (11.11) |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)| โ‰ค ๐ด( ) 5 whenever ๐‘ฅ โˆผ๐‘š ๐‘ฆ. Indeed, let ๐ด = max{|๐‘ข(๐‘1 ) โˆ’ ๐‘ข(๐‘2 )|, |๐‘ข(๐‘2 ) โˆ’ ๐‘ข(๐‘3 )|, |๐‘ข(๐‘3 ) โˆ’ ๐‘ข(๐‘1 )|}. Then clearly (11.11) is satisfied at ๐‘š = 0, and suppose it is true at ๐‘š โˆ’ 1, for some ๐‘š โ‰ฅ 1. If ๐‘ฅ โˆผ๐‘š ๐‘ฆ, then either ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 or, say, ๐‘ฅ โˆˆ ๐‘‰๐‘šโˆ’1 and ๐‘ฆ โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 . Let ๐‘ค โˆˆ ๐‘Š๐‘šโˆ’1 , and ๐‘ฅ๐‘– = ๐‘“๐‘ค (๐‘ ๐‘– ), for ๐‘– = 1, 2, 3. If ๐‘ฆ ๐‘– is the point in ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 in the cell bounded by the points ๐‘ฅ1 , ๐‘ฅ2 , ๐‘ฅ3 , opposite to ๐‘ฅ๐‘– , then, by (11.7), ๐‘ข(๐‘ฆ ๐‘– ) =

๐‘ข(๐‘ฅ๐‘– ) + 2๐‘ข(๐‘ฅ๐‘— ) + 2๐‘ข(๐‘ฅ๐‘˜ ) , 5

11.2. Harmonic structure on the Sierpiล„ski gasket 3

i 3

i

2

4

0 1

1 2

0 0 1 2

1

Figure 11.6. Harmonic function with boundary values ๐œŒ(๐‘ 1 ) = 1 and ๐œŒ(๐‘ 2 ) = ๐œŒ(๐‘ 3 ) = 0.

where ๐‘–, ๐‘—, ๐‘˜ are the three different numbers 1, 2, 3. Thus |๐‘ข(๐‘ฆ ๐‘– ) โˆ’ ๐‘ข(๐‘ฆ๐‘— )| ๐‘ข(๐‘ฅ๐‘– ) + 2๐‘ข(๐‘ฅ๐‘— ) + 2๐‘ข(๐‘ฅ๐‘˜ ) ๐‘ข(๐‘ฅ๐‘— ) + 2๐‘ข(๐‘ฅ๐‘– ) + 2๐‘ข(๐‘ฅ๐‘˜ ) | โˆ’ | 5 5 |๐‘ข(๐‘ฅ๐‘– ) โˆ’ ๐‘ข(๐‘ฅ๐‘— )| 1 3 ๐‘š 3 ๐‘šโˆ’1 = โ‰ค ๐ด( ) , โ‰ค โ‹… ๐ด( ) 5 5 5 5

= ||

by the induction hypothesis, and ๐‘ข(๐‘ฅ๐‘— ) + 2๐‘ข(๐‘ฅ๐‘– ) + 2๐‘ข(๐‘ฅ๐‘˜ ) | | 5 |๐‘ข(๐‘ฅ๐‘– ) โˆ’ ๐‘ข(๐‘ฅ๐‘— )| 2|๐‘ข(๐‘ฅ๐‘– ) โˆ’ ๐‘ข(๐‘ฅ๐‘˜ )| โ‰ค + 5 5 ๐‘šโˆ’1 1 3 2 3 ๐‘šโˆ’1 3 ๐‘š โ‰ค โ‹… ๐ด( ) + โ‹… ๐ด( ) = ๐ด( ) . 5 5 5 5 5

|๐‘ข(๐‘ฅ๐‘– ) โˆ’ ๐‘ข(๐‘ฆ๐‘— )| = ||๐‘ข(๐‘ฅ๐‘– ) โˆ’

Thus (11.11) holds for every ๐‘š โ‰ฅ 0. If ๐‘‰โˆ— =

โ‹ƒ

๐‘šโ‰ฅ0

๐‘‰๐‘š ,

211

212

11. The Laplacian on the Sierpiล„ski gasket

we can use estimate (11.11) to show that ๐‘ข is a uniformly continuous function on ๐‘‰โˆ— (Exercise (4)), which, since ๐‘‰โˆ— is dense in ๐‘† (Exercise (5)), can be extended to a continuous function on ๐‘† (Exercise (6)).

11.3. The Laplacian on the Sierpiล„ski gasket Let ๐‘ข โˆˆ ๐ถ(๐‘†), a continuous function on ๐‘†, and for each ๐‘š โ‰ฅ 0, let ๐‘ข๐‘š = ๐‘ข|๐‘‰๐‘š , its restriction to ๐‘‰๐‘š . By (11.10), the sequence โ„ฐ๐‘š (๐‘ข๐‘š ) is increasing. Let โ„ฑ = {๐‘ข โˆˆ ๐ถ(๐‘†) โˆถ โ„ฐ๐‘š (๐‘ข๐‘š ) is bounded}. Then, for each ๐‘ข โˆˆ โ„ฑ, the limit โ„ฐ(๐‘ข) = lim โ„ฐ๐‘š (๐‘ข๐‘š ) = sup โ„ฐ๐‘š (๐‘ข๐‘š ) ๐‘šโ†’โˆž

๐‘šโ‰ฅ0

exists. We call โ„ฐ(๐‘ข) the energy of ๐‘ข on ๐‘†. Example 11.12. If ๐‘ข is constant, then โ„ฐ๐‘š (๐‘ข๐‘š ) = 0 for all ๐‘š, and thus ๐‘ข โˆˆ โ„ฑ and โ„ฐ(๐‘ข) = 0. In fact, since โ„ฐ(๐‘ข) = sup โ„ฐ๐‘š (๐‘ข๐‘š ), ๐‘šโ‰ฅ0

then โ„ฐ(๐‘ข) = 0 only if ๐‘ข is a constant function. Example 11.13. If ๐‘ข is the continuous extension of a harmonic function, then โ„ฐ๐‘š (๐‘ข๐‘š ) = โ„ฐ0 (๐‘ข0 ) for all ๐‘š, and thus โ„ฐ(๐‘ข) = โ„ฐ0 (๐‘ข0 ). Moreover, since โ„ฐ(๐‘ข) = sup โ„ฐ๐‘š (๐‘ข๐‘š ), ๐‘šโ‰ฅ0

then ๐‘ข minimizes โ„ฐ among all functions in โ„ฑ with boundary values ๐‘ข0 . We thus also call ๐‘ข harmonic in ๐‘†. We say that ๐‘ข โˆˆ ๐ถ(๐‘†) is ๐‘š-harmonic if, for each ๐‘ค โˆˆ ๐‘Š๐‘š , the function ๐‘ข โˆ˜ ๐‘“๐‘ค is harmonic. Equivalently, ๐‘ข is the continuous extension of the function constructed using (11.7) starting from a given function on ๐‘‰๐‘š . Thus, for each ๐‘› โ‰ฅ ๐‘š, โ„ฐ๐‘› (๐‘ข๐‘› ) = โ„ฐ๐‘š (๐‘ข๐‘š ), and thus ๐‘ข โˆˆ โ„ฑ and โ„ฐ(๐‘ข) = โ„ฐ๐‘š (๐‘ข๐‘š ). Example 11.14. Let ๐‘š โ‰ฅ 0 and ๐‘ฅ โˆˆ ๐‘‰๐‘š . Consider the ๐‘š-harmonic function ๐œ“๐‘ฅ,๐‘š (if ๐‘š = 0 we just say harmonic) constructed from the function on ๐‘‰๐‘š given by 1 ๐‘ฆ=๐‘ฅ ๐œ’(๐‘ฆ) = { 0 ๐‘ฆ โ‰  ๐‘ฅ.

11.3. The Laplacian on the Sierpiล„ski gasket

213

๐œ“๐‘ฅ,๐‘š is called the ๐‘š-harmonic spline at ๐‘ฅ. Note that, for any ๐‘š-harmonic function ๐‘ข, we have ๐‘ข = โˆ‘ ๐‘ข(๐‘ฅ)๐œ“๐‘ฅ,๐‘š . ๐‘ฅโˆˆ๐‘‰๐‘š

In particular, if ๐‘š = 0, the splines of Example 11.14 correspond to the harmonic functions with boundary values 1, 0, 0, as the one in Figure 11.6. If we denote ๐œ“๐‘๐‘– ,0 simply by ๐œ“๐‘– , we see that ๐‘ข = ๐‘ข(๐‘1 )๐œ“1 + ๐‘ข(๐‘2 )๐œ“2 + ๐‘ข(๐‘3 )๐œ“3 , and thus ๐œ“1 , ๐œ“2 and ๐œ“3 form a basis for the (3-dimensional) space of harmonic functions on ๐‘†. 2๐œ‹ Note that, by the symmetry of ๐‘† and (11.7), each ๐œ“๐‘– is a rotation 3 of each other. Also, any other ๐œ“๐‘ฅ,๐‘š is either of the form ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1

or

โˆ’1 ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 + ๐œ’๐‘†๐‘คฬ„ โ‹… ๐œ“๐‘— โˆ˜ ๐‘“๐‘ค ฬ„ ,

where ๐‘ค, ๐‘คฬ„ โˆˆ ๐‘Š๐‘š , depending on whether ๐‘ฅ โˆˆ ๐‘‰0 or ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต๐‘‰0 (Exercise (8)). We want to obtain the analogous result of integration by parts in the interval, as in (11.6), for the Sierpiล„ski gasket. This will give us a way to define the Laplacian on ๐‘†. For this, we first need to define an integral of functions on ๐‘†. The usual Lebesgue integral that we have been using in the previous chapters is not useful for us here, since ๐‘† is a set of measure zero in the plane. So we have to start by defining a measure on subsets of ๐‘†. Recall that, for any word ๐‘ค โˆˆ ๐‘Š๐‘š , we define the cell ๐‘†๐‘ค = ๐‘“๐‘ค (๐‘†). For any given cell, we define 1 , 3๐‘š where ๐‘š is the length of the word ๐‘ค. We now define, for ๐ด โŠ‚ ๐‘†, (11.15)

(11.16)

๐œ‡(๐‘†๐‘ค ) =

๐œ‡(๐ด) = inf { โˆ‘ ๐œ‡(๐‘‡๐‘— ) โˆถ ๐‘‡๐‘— are cells and ๐ด โŠ‚ ๐‘—

โ‹ƒ

๐‘‡๐‘— }.

๐‘—

As in the case of the Hausdorff measure that we discussed in Chapter 10, ๐œ‡ satisfies the properties of the Lebesgue outer measure. Proposition 11.17. Let ๐œ‡ be defined on subsets of ๐‘† as in (11.16). (1) If ๐ด โŠ‚ ๐ต, then ๐œ‡(๐ด) โ‰ค ๐œ‡(๐ต).

214

11. The Laplacian on the Sierpiล„ski gasket (2) If ๐ด = โ‹ƒ๐‘— ๐ด๐‘— , then ๐œ‡(๐ด) โ‰ค โˆ‘๐‘— ๐œ‡(๐ด๐‘— ). (3) If dist(๐ด, ๐ต) > 0, ๐œ‡(๐ด โˆช ๐ต) = ๐œ‡(๐ด) + ๐œ‡(๐ต).

We leave the proof of Proposition 11.17 as an exercise (Exercise (9)). We can then define a set ๐ด โŠ‚ ๐‘† to be measurable if, for any ๐ต โŠ‚ ๐‘†, ๐œ‡(๐ต) = ๐œ‡(๐ต โˆฉ ๐ด) + ๐œ‡(๐ต โงต ๐ด). As in the case of the Lebesgue measure, part (3) of Proposition 11.17 implies that every open subset of ๐‘† is measurable (and hence every closed subset ). Also, every open ๐‘ˆ โŠ‚ ๐‘† can be written as an almost disjoint union โ‹ƒ๐‘— ๐‘‡๐‘— of cells, so ๐œ‡(๐‘ˆ) = โˆ‘ ๐œ‡(๐‘‡๐‘— ). ๐‘—

We, in fact, can prove more. Proposition 11.18. (1) If ๐ด1 , ๐ด2 , . . . are disjoint measurable subsets of ๐‘†, then ๐ด = โ‹ƒ๐‘— ๐ด๐‘— is measurable and ๐œ‡(๐ด) = โˆ‘ ๐œ‡(๐ด๐‘— ). ๐‘—

(2) If ๐ด โŠ‚ ๐‘† is measurable, its complement ๐‘† โงต ๐ด is also measurable and ๐œ‡(๐‘† โงต ๐ด) = 1 โˆ’ ๐œ‡(๐ด). The proofs of these statements are similar to their analogous results on Lebesgue measure and we again leave them as an exercise (Exercise (9)). We can now proceed to define the integral in an analogous way, first for characteristic functions of measurable sets, โˆซ ๐œ’๐ด ๐‘‘๐œ‡ = ๐œ‡(๐ด), ๐‘†

then for simple functions ๐œ™ = โˆ‘๐‘— ๐‘๐‘— ๐œ‡(๐ด๐‘— ), โˆซ ๐œ™๐‘‘๐œ‡ = โˆ‘ ๐‘๐‘— ๐œ‡(๐ด๐‘— ), ๐‘†

๐‘—

11.3. The Laplacian on the Sierpiล„ski gasket

215

and then by approximating any measurable ๐‘“ โ‰ฅ 0 with simple functions โˆซ ๐‘“๐‘‘๐œ‡ = sup { โˆซ ๐œ™๐‘‘๐œ‡ โˆถ ๐œ™ is simple and 0 โ‰ค ๐œ™ โ‰ค ๐‘“}. ๐‘†

๐‘†

However, while the Lebesgue measure is traslation invariant and dilations by a positive ๐›ฟ induce a factor of ๐›ฟ๐‘‘ , where ๐‘‘ is the dimension of the space, in this case we have the following properties. 11.19. If ๐œŽ is a symmetry of ๐‘†, say, a rotation or a reflection over an axis passing through one of its vertices ๐‘ ๐‘– , then โˆซ ๐‘“ โˆ˜ ๐œŽ๐‘‘๐œ‡ = โˆซ ๐‘“๐‘‘๐œ‡. ๐‘†

๐‘†

To prove this fact, observe that it is enough to prove it for characteristic functions ๐œ’๐ด , and in that case โˆซ ๐œ’๐ด โˆ˜ ๐œŽ๐‘‘๐œ‡ = โˆซ ๐œ’๐œโˆ’1 (๐ด) ๐‘‘๐œ‡ = ๐œ‡(๐œŽโˆ’1 (๐ด)), ๐‘†

๐‘†

so we just need to verify that ๐œ‡ is invariant under the symmetries of ๐‘†. This follows from the observation that, for any cell ๐‘†๐‘ค and any symmetry ๐œŽ of ๐‘†, then ๐œŽ(๐‘†๐‘ค ) is also a cell of level ๐‘š, and thus have the same measure. Indeed, if ๐œŽ is the reflection over the axis passing through ๐‘ ๐‘– , so it keeps ๐‘ ๐‘– fixed and switches ๐‘๐‘— and ๐‘ ๐‘˜ , then, for any ๐‘ค โˆˆ ๐‘Š๐‘š , ๐œŽ(๐‘†๐‘ค ) = ๐‘†๐‘คฬ„ , where ๐‘คฬ„ โˆˆ ๐‘Š๐‘š is the word obtained from ๐‘ค by switching the letters ๐‘— and ๐‘˜. If ๐œŽ is the rotation ๐‘1 โ†ฆ ๐‘2 , ๐‘2 โ†ฆ ๐‘3 and ๐‘3 โ†ฆ ๐‘1 , then ๐œŽ(๐‘†๐‘ค ) = ๐‘†๐‘คฬ„ where ๐‘คฬ„ is obtained by replacing any letter ๐‘– of ๐‘ค by the letter ๐‘– + 1 (mod 3). 11.20. For any ๐‘ค โˆˆ ๐‘Š๐‘š , โˆซ ๐‘“๐‘‘๐œ‡ = ๐‘†๐‘ค

1 โˆซ ๐‘“ โˆ˜ ๐‘“๐‘ค ๐‘‘๐œ‡. 3๐‘š ๐‘†

In order to verify this for characteristic functions, we have to verify that 1 ๐œ‡(๐‘“โˆ’1 (๐ด โˆฉ ๐‘†๐‘ค )), 3๐‘š ๐‘ค which follows from the fact that, if the ๐‘‡๐‘— are cells, ๐œ‡(๐ด โˆฉ ๐‘†๐‘ค ) =

๐‘“๐‘คโˆ’1 (๐ด โˆฉ ๐‘†๐‘ค ) โŠ‚

โ‹ƒ ๐‘—

๐‘‡๐‘—

if and only if

๐ด โˆฉ ๐‘†๐‘ค โŠ‚

โ‹ƒ ๐‘—

๐‘“๐‘ค (๐‘‡๐‘— ),

216

11. The Laplacian on the Sierpiล„ski gasket

and ๐œ‡(๐‘“๐‘ค (๐‘‡๐‘— )) =

1 ๐œ‡(๐‘‡๐‘— ) 3๐‘š

for each cell ๐‘‡๐‘— . Example 11.21. If ๐‘“ is a harmonic function on ๐‘†, then ๐‘“ = ๐‘“(๐‘1 )๐œ“1 + ๐‘“(๐‘2 )๐œ“2 + ๐‘“(๐‘3 )๐œ“3 , where each ๐œ“๐‘– is the harmonic function with boundary values ๐œ“๐‘– (๐‘ ๐‘– ) = 1 and ๐œ“๐‘– (๐‘๐‘— ) = 0, for ๐‘— โ‰  ๐‘–. As we have seen above, each ๐œ“๐‘– is a rotation of the others. Since ๐œ“1 + ๐œ“2 + ๐œ“3 = 1 and ๐œ‡(๐‘†) = 1, we have that โˆซ ๐œ“1 ๐‘‘๐œ‡ = โˆซ ๐œ“2 ๐‘‘๐œ‡ = โˆซ ๐œ“3 ๐‘‘๐œ‡ = ๐‘†

๐‘†

Therefore โˆซ ๐‘“๐‘‘๐œ‡ = ๐‘†

๐‘†

1 . 3

๐‘“(๐‘1 ) + ๐‘“(๐‘2 ) + ๐‘“(๐‘3 ) . 3

Example 11.22. Let ๐œ“๐‘ฅ,๐‘š be the ๐‘š-harmonic spline at ๐‘ฅ. As we have seen above, if ๐‘ฅ โˆˆ ๐‘‰0 , ๐œ“๐‘ฅ,๐‘š = ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 for some ๐‘– and ๐‘ค โˆˆ ๐‘Š๐‘š , and thus 1 1 โˆซ ๐œ“๐‘ฅ,๐‘š ๐‘‘๐œ‡ = โˆซ ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 ๐‘‘๐œ‡ = ๐‘š โˆซ ๐œ“๐‘– ๐‘‘๐œ‡ = ๐‘š+1 . 3 ๐‘† 3 ๐‘† ๐‘† ๐‘ค

โˆ’1 If ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 , then ๐œ“๐‘ฅ,๐‘š = ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 + ๐œ’๐‘†๐‘คฬ„ โ‹… ๐œ“๐‘— โˆ˜ ๐‘“๐‘ค ฬ„ for some ๐‘–, ๐‘— and ๐‘ค, ๐‘คฬ„ โˆˆ ๐‘Š๐‘š , and thus โˆ’1 โˆซ ๐œ“๐‘ฅ,๐‘š = โˆซ ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 ๐‘‘๐œ‡ + โˆซ ๐œ“๐‘— โˆ˜ ๐‘“๐‘ค ฬ„ ๐‘‘๐œ‡ ๐‘†

๐‘†๐‘ค

๐‘†๐‘คฬ„

1 1 = ๐‘š โˆซ ๐œ“๐‘– ๐‘‘๐œ‡ + ๐‘š โˆซ ๐œ“๐‘— ๐‘‘๐œ‡ 3 ๐‘† 3 ๐‘† 2 = ๐‘š+1 . 3 We are ready to define the Laplacian. First, consider the polarization โ„ฐ(๐‘ข, ๐‘ฃ) of the energy, defined for ๐‘ข, ๐‘ฃ โˆˆ โ„ฑ. We can see that โ„ฐ(๐‘ข, ๐‘ฃ) = lim โ„ฐ๐‘š (๐‘ข, ๐‘ฃ), where in the right-hand side we also denote by ๐‘ข and ๐‘ฃ their restrictions to ๐‘‰๐‘š , and, for functions ๐‘‰๐‘š , โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) is the polarization of the quadratic form โ„ฐ๐‘š , 5 ๐‘š โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = ( ) โˆ‘ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))(๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)). 3 ๐‘ฅโˆผ ๐‘ฆ ๐‘š

11.3. The Laplacian on the Sierpiล„ski gasket

217

In fact, one can show (Exercise (10)) that 5 ๐‘š (11.23) โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ’( ) โˆ‘ ฮ”๐‘š ๐‘ข(๐‘ฅ) ๐‘ฃ(๐‘ฅ), 3 ๐‘ฅโˆˆ๐‘‰ ๐‘š

where ฮ”๐‘š is the difference operator (11.24)

ฮ”๐‘š ๐‘ข(๐‘ฅ) = โˆ‘ (๐‘ข(๐‘ฆ) โˆ’ ๐‘ข(๐‘ฅ)). ๐‘ฆโˆผ๐‘š ๐‘ฅ

Note that this sum has two terms if ๐‘ฅ โˆˆ ๐‘‰0 , or four terms if ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 . ฮ”๐‘š is called the discrete Laplacian at level ๐‘š. We say that ๐‘ข โˆˆ dom ฮ” if ๐‘ข โˆˆ โ„ฑ and there exists ๐‘“ โˆˆ ๐ถ(๐‘†) such that, for all ๐‘ฃ โˆˆ โ„ฑ with ๐‘ฃ|๐‘‰0 = 0, (11.25)

โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ’ โˆซ ๐‘“๐‘ฃ๐‘‘๐œ‡. ๐‘†

We write ๐‘“ = ฮ”๐‘ข, and we call it the Laplacian of ๐‘ข. 11.26. If ๐‘ข is harmonic, for each ๐‘š โ‰ฅ 1 and all ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 , we have that ฮ”๐‘š ๐‘ข(๐‘ฅ) = 0 (Exercise (11)). Thus, if ๐‘ฃ โˆˆ โ„ฑ with ๐‘ฃ|๐‘‰0 = 0, โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = 0 for all ๐‘š, and thus โ„ฐ(๐‘ข, ๐‘ฃ) = 0. We thus conclude that, if ๐‘ข is harmonic, ๐‘ข โˆˆ dom ฮ” and ฮ”๐‘ข = 0. 11.27. If ๐‘ข is ๐‘š-harmonic, ฮ”๐‘› ๐‘ข(๐‘ฅ) = 0 for all ๐‘› > ๐‘š and ๐‘ฅ โˆˆ ๐‘‰๐‘› โงต ๐‘‰๐‘š . Thus, for any other ๐‘ฃ โˆˆ โ„ฑ, โ„ฐ(๐‘ข, ๐‘ฃ) = โ„ฐ๐‘š (๐‘ข, ๐‘ฃ). However, ๐‘ข โˆ‰ dom ฮ”, because ฮ”๐‘ข(๐‘ฅ) cannont be defined for ๐‘ฅ โˆˆ ๐‘‰๐‘š . Fix ๐‘ฅ0 โˆˆ ๐‘‰โˆ— โงต ๐‘‰0 , and let ๐‘š be so ๐‘ฅ0 โˆˆ ๐‘‰๐‘š . If ๐‘ฃ = ๐œ“๐‘ฅ0 ,๐‘š , then, as we have seen before, ๐‘ฃ โˆˆ โ„ฑ, and of course ๐‘ฃ|๐‘‰0 = 0, because ๐‘ฅ0 โˆ‰ ๐‘‰0 . Let ๐‘ข โˆˆ dom ฮ”. We first note that, since ๐‘ฃ is ๐‘š-harmonic, 5 ๐‘š โ„ฐ(๐‘ข, ๐‘ฃ) = โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ’( ) โˆ‘ ฮ”๐‘š ๐‘ข(๐‘ฅ) ๐‘ฃ(๐‘ฅ) 3 ๐‘ฅโˆˆ๐‘‰ ๐‘š

5 ๐‘š = โˆ’( ) ฮ”๐‘š ๐‘ข(๐‘ฅ0 ), 3 because ๐‘ฃ(๐‘ฅ0 ) = 1 and ๐‘ฃ(๐‘ฅ) = 0 for any other ๐‘ฅ โˆˆ ๐‘‰๐‘š . Also โˆซ ฮ”๐‘ข(๐‘ฅ)๐‘ฃ(๐‘ฅ)๐‘‘๐œ‡(๐‘ฅ) โˆ’ ฮ”๐‘ข(๐‘ฅ0 ) โˆซ ๐‘ฃ(๐‘ฅ)๐‘‘๐œ‡(๐‘ฅ) ๐‘†

๐‘†

= โˆซ (ฮ”๐‘ข(๐‘ฅ) โˆ’ ฮ”๐‘ข(๐‘ฅ0 ))๐‘ฃ(๐‘ฅ)๐‘‘๐œ‡(๐‘ฅ). ๐‘†

218

11. The Laplacian on the Sierpiล„ski gasket

Given ๐œ€ > 0, we can choose ๐‘š large enough so that |ฮ”๐‘ข(๐‘ฅ) โˆ’ ฮ”๐‘ข(๐‘ฅ0 )| < ๐œ€ for every ๐‘ฅ โˆˆ supp ๐‘ฃ, because ฮ”๐‘ข is continuous. Thus, for such ๐‘š, and using our result in Example 11.22, | โˆซ (ฮ”๐‘ข(๐‘ฅ) โˆ’ ฮ”๐‘ข(๐‘ฅ ))๐‘ฃ(๐‘ฅ)๐‘‘๐œ‡(๐‘ฅ)| < ๐œ€ โˆซ ๐‘ฃ๐‘‘๐œ‡ = 2 ๐œ€. 0 | | 3๐‘š+1 ๐‘† ๐‘† Putting the previous identities and inequalities together we obtain ๐‘š |( 5 ) ฮ” ๐‘ข(๐‘ฅ ) โˆ’ 2 ฮ”๐‘ข(๐‘ฅ )| < 2 ๐œ€, ๐‘š 0 0 | | 3 3๐‘š+1 3๐‘š+1

and thus

(11.28)

| 3 5๐‘š ฮ” ๐‘ข(๐‘ฅ ) โˆ’ ฮ”๐‘ข(๐‘ฅ )| < ๐œ€. ๐‘š 0 0 | |2

We have therefore proven Theorem 11.29, that states that the Laplacian is a limit of normalized difference operators. Theorem 11.29. If ๐‘ข โˆˆ dom ฮ” and ๐‘ฅ0 โˆˆ ๐‘‰โˆ— โงต ๐‘‰0 , then

ฮ”๐‘ข(๐‘ฅ0 ) = lim

๐‘šโ†’โˆž

3 ๐‘š 5 ฮ”๐‘š ๐‘ข(๐‘ฅ0 ). 2

We can also prove that the limit in Theorem 11.28 is uniform, in the sense that, given ๐œ€ > 0, we can choose ๐‘š such that (11.29) is true independently of the particular choice of ๐‘ฅ0 โˆˆ ๐‘‰๐‘š . Moreover, the existence of this uniform limit also implies that ๐‘ข โˆˆ dom ฮ”. We have left these facts as exercises (Exercises (12) and (13)).

Exercises

219

Exercises (1) Let ๐‘ฅ0 โˆˆ (0, 1) and โ„Ž > 0 so that (๐‘ฅ0 โˆ’ โ„Ž, ๐‘ฅ0 + โ„Ž) โŠ‚ (0, 1). Let ๐‘ฃ be the piecewise linear function given on [0, 1] by โŽง0 โŽช โŽช ๐‘ก โˆ’ ๐‘ฅ0 + โ„Ž ๐‘ฃ(๐‘ก) = โ„Ž โŽจ ๐‘ฅ0 + โ„Ž โˆ’ ๐‘ก โŽช โ„Ž โŽช โŽฉ0

๐‘ก < ๐‘ฅ0 โˆ’ โ„Ž ๐‘ฅ0 โˆ’ โ„Ž โ‰ค ๐‘ก < ๐‘ฅ 0 ๐‘ฅ0 โ‰ค ๐‘ก < ๐‘ฅ 0 + โ„Ž ๐‘ฅ0 + โ„Ž โ‰ค ๐‘ก.

Then, for ๐‘ข โˆˆ ๐ถ 2 ([0, 1]), 1

โˆซ ๐‘ขโ€ณ (๐‘ก)๐‘ฃ(๐‘ก)๐‘‘๐‘ก = 0

๐‘ข(๐‘ฅ0 โˆ’ โ„Ž) + ๐‘ข(๐‘ฅ0 + โ„Ž) โˆ’ 2๐‘ข(๐‘ฅ0 ) . โ„Ž

(2) For functions ๐‘ข and ๐‘ฃ as in the previous exercise, 1

1 โˆซ ๐‘ขโ€ณ (๐‘ก)๐‘ฃ(๐‘ก)๐‘‘๐‘ก = ๐‘ขโ€ณ (๐‘ฅ0 ). โ„Žโ†’0 โ„Ž 0 lim

(3) Let ๐‘“(๐‘ฅ, ๐‘ฆ, ๐‘ง) =(๐‘Ž โˆ’ ๐‘ง)2 + (๐‘ง โˆ’ ๐‘ฆ)2 + (๐‘ฆ โˆ’ ๐‘Ž)2 + (๐‘ง โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘ง)2 + (๐‘ฆ โˆ’ ๐‘ฅ)2 + (๐‘ฅ โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘ฆ)2 . Then ๐‘“ takes its minimum at ๐‘ฅโˆ— =

๐‘Ž + 2๐‘ + 2๐‘ , 5

๐‘ฆโˆ— =

2๐‘Ž + ๐‘ + 2๐‘ , 5

๐‘“(๐‘ฅโˆ— , ๐‘ฆโˆ— , ๐‘งโˆ— ) =

3 ((๐‘Ž โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘)2 + (๐‘ โˆ’ ๐‘Ž)2 ). 5

๐‘งโˆ— =

2๐‘Ž + 2๐‘ + ๐‘ , 5

and

(4) If ๐‘ข is a harmonic function, then it is uniformly continuous on ๐‘‰โˆ— , the set of all vertices in ๐‘†. (5) The union ๐‘‰โˆ— of all vertices in ๐‘† is dense in ๐‘†. (6) If ๐‘ข is a harmonic function, then it can be extended to a continuous function on ๐‘†. (Hint: Use Exercises (4) and (5).)

220

11. The Laplacian on the Sierpiล„ski gasket

(7) If ๐‘ข is the continuous extension of a harmonic function in ๐‘†, then ๐‘ข is a Hรถlder continuous function with exponent ๐›ผ=

log(5/3) . log 2

(8) Let ๐‘š โ‰ฅ 1 and ๐‘ฅ โˆˆ ๐‘‰๐‘š . (a) If ๐‘ฅ โˆˆ ๐‘‰0 , ๐œ“๐‘ฅ,๐‘š = ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 for some ๐‘– = 1, 2, 3, and ๐‘ค is the word ๐‘ค = ๐‘–๐‘– โ‹ฏ ๐‘– โˆˆ ๐‘Š๐‘š . (b) If ๐‘ฅ โˆ‰ ๐‘‰0 , there exist ๐‘–, ๐‘— = 1, 2, 3 and ๐‘ค, ๐‘คฬ„ โˆˆ ๐‘Š๐‘š such that โˆ’1 ๐œ“๐‘ฅ,๐‘š = ๐œ’๐‘†๐‘ค โ‹… ๐œ“๐‘– โˆ˜ ๐‘“๐‘คโˆ’1 + ๐œ’๐‘†๐‘คฬ„ โ‹… ๐œ“๐‘— โˆ˜ ๐‘“๐‘ค ฬ„ .

(9) (a) Prove Proposition 11.17. (b) Prove Propostion 11.18. (10) For functions ๐‘ข, ๐‘ฃ on ๐‘‰๐‘š , โˆ‘ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))(๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)) = โˆ’ โˆ‘ ฮ”๐‘š ๐‘ข(๐‘ฅ) ๐‘ฃ(๐‘ฅ), ๐‘ฅโˆผ๐‘š ๐‘ฆ

๐‘ฅโˆˆ๐‘‰๐‘š

where ฮ”๐‘š is the difference operator ฮ”๐‘š ๐‘ข(๐‘ฅ) = โˆ‘ (๐‘ข(๐‘ฆ) โˆ’ ๐‘ข(๐‘ฅ)). ๐‘ฆโˆผ๐‘š ๐‘ฅ

(11) If ๐‘ข is harmonic, then, for each ๐‘š โ‰ฅ 1 and ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 , ฮ”๐‘š ๐‘ข(๐‘ฅ) = 0. (12) Let ๐‘ข โˆˆ dom ฮ”. Then, for any ๐œ€ > 0, there exists ๐‘ such that, for any ๐‘š โ‰ฅ ๐‘, | 3 5๐‘š ฮ” ๐‘ข(๐‘ฅ) โˆ’ ฮ”๐‘ข(๐‘ฅ)| < ๐œ€ ๐‘š |2 | for any ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 . (13) Let ๐‘ข โˆˆ โ„ฑ and suppose there exists ๐‘“ โˆˆ ๐ถ(๐‘†) such that, for any ๐œ€ > 0, we can find ๐‘ such that, for any ๐‘š โ‰ฅ ๐‘, | 3 5๐‘š ฮ” ๐‘ข(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| < ๐œ€ ๐‘š | |2 for any ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 . Then ๐‘ข โˆˆ dom ฮ” and ๐‘“ = ฮ”๐‘ข.

Notes The calculus on the Sierpiล„ski gasket developed in this chapter was introduced by Jun Kigami in [Kig89], where the motivation from the discretization of the interval is also presented. The idea of using springs to define discrete energy was also used in [YHK97]. The discrete energy can also be obtained from the theory of electrical networks, as in

Notes

221

[Str06] or in [Kig01]. The Laplacian on the Sierpiล„ski gasket can also be introduced through Brownian motion, as in [Bar98].

Chapter 12

Eigenfunctions of the Laplacian

In Chapter 3, we studied the decomposition of functions in the interval in terms of trigonometric functions, as a means to solve the Dirichlet problem in the disk. The trigonometric functions sine and cosine satisfy the equations ๐‘‘2 sin ๐‘ฅ = โˆ’ sin ๐‘ฅ ๐‘‘๐‘ฅ2

and

๐‘‘2 cos ๐‘ฅ = โˆ’ cos ๐‘ฅ, ๐‘‘๐‘ฅ2

which imply that the trigonometric functions sin ๐‘˜๐‘ฅ and cos ๐‘˜๐‘ฅ are all eigenfunctions of the operator ๐‘‘ 2 /๐‘‘๐‘ฅ2 , as they satisfy ๐‘‘2 ๐œ™(๐‘ฅ) = โˆ’๐œ†๐œ™(๐‘ฅ) ๐‘‘๐‘ฅ2 for some nonnegative ๐œ† โˆˆ โ„. In this chapter we study the eigenfunctions and eigenvalues of the Laplacian on the Sierpiล„ski gasket, defined in Chapter 11. That is, we will study the solutions to the equation ฮ”๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘ข(๐‘ฅ), for some scalar ๐œ† and nonzero ๐‘ข โˆˆ dom ฮ”, and each ๐‘ฅ โˆˆ ๐‘† โงต ๐‘‰0 . The first natural question is whether such eigenfunctions can be constructed through an interpolation algorithm, as is the case of the harmonic functions, studied above. Though there is no a priori reason for the existence 223

224

12. Eigenfunctions of the Laplacian

of such algorithm, we first note that, in the case of the interval [0, 1], it indeed exists.

12.1. Discrete eigenfunctions on the interval We first observe that if ๐œ™ is an eigenfunction of the Laplacian ฮ” = ๐‘‘ 2 /๐‘‘๐‘ฅ2 on the interval, then, for a sufficiently large ๐‘š, ๐œ™|๐‘‰๐‘š is an eigenfunction of the difference operator 1 1 ) + ๐‘ข(๐‘ฅ + ๐‘š ) โˆ’ 2๐‘ข(๐‘ฅ), 2๐‘š 2 where ๐‘ข is a function on the dyadic partition ๐‘‰๐‘š of level ๐‘š, and ๐‘ฅ is of the form ๐‘˜/2๐‘š , 1 โ‰ค ๐‘˜ โ‰ค 2๐‘š โˆ’ 1. Indeed, the eigenfunctions of ฮ” are linear combinations of functions of the form ๐œ™(๐‘ฅ) = ๐‘’๐œ”๐‘ฅ , since ฮ”๐‘š ๐‘ข(๐‘ฅ) = ๐‘ข(๐‘ฅ โˆ’

๐‘‘ 2 ๐œ”๐‘ฅ ๐‘’ = ๐œ”2 ๐‘’๐œ”๐‘ฅ . ๐‘‘๐‘ฅ2 Now 1

1

ฮ”๐‘š ๐œ™(๐‘ฅ) = ๐‘’๐œ”(๐‘ฅโˆ’ 2๐‘š ) + ๐‘’๐œ”(๐‘ฅ+ 2๐‘š ) โˆ’ 2๐‘’๐œ”๐‘ฅ = โˆ’๐œ†๐‘š ๐‘’๐œ”๐‘ฅ , where ๐œ”

๐œ”

๐œ”

โˆ’๐œ†๐‘š = ๐‘’โˆ’ 2๐‘š + ๐‘’ 2๐‘š โˆ’ 2 = (๐‘’ 2๐‘š+1 โˆ’ ๐‘’

โˆ’

๐œ” 2๐‘š+1

)2 = 4 sinh

2

๐œ” 2๐‘š+1

.

Thus, the restriction of ๐œ™ to ๐‘‰๐‘š is an eigenfunction of the difference operator ฮ”๐‘š with respect to the eigenvalue โˆ’๐œ†๐‘š . Moreover, note that, as we have seen in Chapter 11, 4๐‘š ฮ”๐‘š ๐œ™(๐‘ฅ) โ†’ ๐œ™โ€ณ (๐‘ฅ) = ๐œ”2 ๐œ™(๐‘ฅ) as ๐‘š โ†’ โˆž, and ๐œ” โ†’ ๐œ”2 , 2๐‘š+1 so the discrete eigenvalues, when properly normalized, converge to the true eigenvalues of ฮ”. 2

โˆ’4๐‘š ๐œ†๐‘š = (2๐‘š+1 )2 sinh

Note that it may happen that ๐œ™|๐‘‰๐‘š โ‰ก 0 for some ๐‘š: for instance, if ๐œ™(๐‘ฅ) = sin(2๐‘› ๐œ‹๐‘ฅ), then ๐œ™|๐‘‰๐‘š โ‰ก 0 for every ๐‘š โ‰ค ๐‘›. Since an eigenfunction ๐œ™ is not identically zero, ๐œ™|๐‘‰๐‘š โ‰ข 0 for sufficiently large ๐‘š, so the argument above applies.

12.1. Discrete eigenfunctions on the interval

225

Conversely, we can construct the eigenfunctions of ฮ” by extending discrete eigenfunctions of the operators ฮ”๐‘š . Suppose we have a function ๐‘ข on ๐‘‰๐‘šโˆ’1 which is an eigenfunction of ฮ”๐‘šโˆ’1 with respect to the eigenvalue โˆ’๐œ†๐‘šโˆ’1 , that is ฮ”๐‘šโˆ’1 ๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘šโˆ’1 ๐‘ข(๐‘ฅ) for each ๐‘ฅ โˆˆ ๐‘‰๐‘šโˆ’1 โงต ๐‘‰0 . We want to extend ๐‘ข to a function on ๐‘‰๐‘š , that we also denote by ๐‘ข, that satisfies ฮ”๐‘š ๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘š ๐‘ข(๐‘ฅ) for all ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 , for some ๐œ†๐‘š . In particular, for ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 of the form (2๐‘˜ + 1)/2๐‘š , we want ฮ”๐‘š ๐‘ข(

2๐‘˜ + 1 2๐‘˜ + 1 ) = โˆ’๐œ†๐‘š ๐‘ข( ๐‘š ), 2๐‘š 2

and thus 2๐‘˜ 2๐‘˜ + 2 2๐‘˜ + 1 2๐‘˜ + 1 ) + ๐‘ข( ๐‘š ) โˆ’ 2๐‘ข( ๐‘š ) = โˆ’๐œ†๐‘š ๐‘ข( ๐‘š ), 2๐‘š 2 2 2 so ๐‘ข satisfies, for each ๐‘˜, 2๐‘˜ + 1 ๐‘˜ ๐‘˜+1 (12.1) (2 โˆ’ ๐œ†๐‘š )๐‘ข( ๐‘š ) = ๐‘ข( ๐‘šโˆ’1 ) + ๐‘ข( ๐‘šโˆ’1 ). 2 2 2 Note that ๐‘˜/2๐‘šโˆ’1 , (๐‘˜ + 1)/2๐‘šโˆ’1 โˆˆ ๐‘‰๐‘šโˆ’1 so, provided ๐œ†๐‘š โ‰  2, we have the extension algorithm ๐‘ข(

(12.2)

๐‘˜ ๐‘˜+1 ๐‘ข( ๐‘šโˆ’1 ) + ๐‘ข( ๐‘šโˆ’1 ) 2๐‘˜ + 1 2 2 . ๐‘ข( ๐‘š ) = 2 2 โˆ’ ๐œ†๐‘š

The equation (12.2) tells us how to extend ๐‘ข from ๐‘‰๐‘šโˆ’1 to ๐‘‰๐‘š , if we want ๐‘ข to be an eigenfunction of ฮ”๐‘š with respect to โˆ’๐œ†๐‘š . Now, the eigenfunction equation must also be satisfied at points in ๐‘‰๐‘šโˆ’1 โงต ๐‘‰0 , so we also want ๐‘˜ ๐‘˜ ฮ”๐‘š ๐‘ข( ๐‘šโˆ’1 ) = โˆ’๐œ†๐‘š ๐‘ข( ๐‘šโˆ’1 ) 2 2 for 1 โ‰ค ๐‘˜ โ‰ค 2๐‘šโˆ’1 โˆ’ 1, which gives the equation (12.3)

(2 โˆ’ ๐œ†๐‘š )๐‘ข(

๐‘˜ 2๐‘˜ โˆ’ 1 2๐‘˜ + 1 ) = ๐‘ข( ๐‘š ) + ๐‘ข( ๐‘š ). 2 2 2๐‘šโˆ’1

By (12.2), ๐‘˜โˆ’1 ๐‘˜ ๐‘ข( ๐‘šโˆ’1 ) + ๐‘ข( ๐‘šโˆ’1 ) 2๐‘˜ โˆ’ 1 2 2 ๐‘ข( ๐‘š ) = , 2 2 โˆ’ ๐œ†๐‘š

226

12. Eigenfunctions of the Laplacian

and substituting in (12.3) we obtain ๐‘˜โˆ’1 ๐‘˜+1 ๐‘˜ ) + ๐‘ข( ๐‘šโˆ’1 ) = (2 โˆ’ 4๐œ†๐‘š + ๐œ†2๐‘š )๐‘ข( ๐‘šโˆ’1 ). 2๐‘šโˆ’1 2 2 Since ๐‘ข is assumed to be an eigenfunction of ฮ”๐‘šโˆ’1 with respect to the eigenvalue โˆ’๐œ†๐‘šโˆ’1 , we also have ๐‘ข(

๐‘˜โˆ’1 ๐‘˜+1 ๐‘˜ ) + ๐‘ข( ๐‘šโˆ’1 ) = (2 โˆ’ ๐œ†๐‘šโˆ’1 )๐‘ข( ๐‘šโˆ’1 ), 2๐‘šโˆ’1 2 2 so we obtain the equation ๐‘ข(

2 โˆ’ ๐œ†๐‘šโˆ’1 = 2 โˆ’ 4๐œ†๐‘š + ๐œ†2๐‘š , which can be written as (12.4)

๐œ†๐‘šโˆ’1 = ๐œ†๐‘š (4 โˆ’ ๐œ†๐‘š ).

We have proven that we can always extend an eigenfunction of ฮ”๐‘šโˆ’1 with respect to the eigenvalue โˆ’๐œ†๐‘šโˆ’1 to an eigenfunction of ฮ”๐‘š with respect to โˆ’๐œ†๐‘š , provided ๐œ†๐‘š โ‰  2 and the condition (12.4) holds. Together with (12.2), this provides an algorithm to construct all eigenfunctions of ฮ”. Indeed, since the roots of (12.4) are ๐œ†๐‘š = 2 ยฑ โˆš4 โˆ’ ๐œ†๐‘šโˆ’1 , we can start from an eigenfunction of ฮ”๐‘š0 with respect to ๐œ†๐‘š0 , for some ๐‘š0 , and extend to each ๐‘š > ๐‘š0 choosing ๐œ†๐‘š as any of these roots. Note that, if ๐‘ค = 2โˆ’โˆš4 โˆ’ ๐‘ง, using the principal branch of โˆš4 โˆ’ ๐‘ง on โ„‚โงต[4, โˆž), then ๐‘ง ๐‘ค = + ๐‘‚(|๐‘ง|2 ), 4 and we can conclude that 4๐‘š ๐œ†๐‘š converges if we choose the minus sign for all but finitely many ๐‘š (Exercise (1)). Note that, if ๐œ†๐‘š0 โ‰  2, the starting eigenfunction on ๐‘‰๐‘š0 will be an extension of its restriction to ๐‘‰๐‘š0 โˆ’1 , with respect to the eigenvalue โˆ’๐œ†๐‘š0 โˆ’1 given by (12.4). If ๐œ†๐‘š0 = 2, then, by (12.1) and (12.3), ๐‘ข( and

๐‘˜ ๐‘˜+1 ) + ๐‘ข( ๐‘š โˆ’1 ) = 0, 2๐‘š0 โˆ’1 2 0

0 โ‰ค ๐‘˜ โ‰ค 2๐‘š0 โˆ’1 โˆ’ 1,

2๐‘˜ โˆ’ 1 2๐‘˜ + 1 1 โ‰ค ๐‘˜ โ‰ค 2๐‘š0 โˆ’1 โˆ’ 1, ) + ๐‘ข( ๐‘š0 ) = 0, 2 ๐‘š0 2 so we either have ๐‘ข|๐‘‰๐‘š โˆ’1 โ‰ก 0 and ๐‘ข|๐‘‰๐‘š โงต๐‘‰๐‘š โˆ’1 equal to an alternating 0 0 0 sequence of ยฑ1, or the other way around, as is shown in Figure 12.1. ๐‘ข(

12.2. Discrete eigenfunctions on the Sierpiล„ski gasket

227

Their limit corresponds to the eigenfunctions of the form sin(๐‘›๐œ‹๐‘ฅ) and

Figure 12.1. Eigenfunctions of ฮ”3 with respect to the eigenvalue ๐œ†3 = 2. Note that they correspond to the restrictions to ๐‘‰ 3 of the functions sin(4๐œ‹๐‘ฅ) (top) and cos(4๐œ‹๐‘ฅ) (bottom).

cos(๐‘›๐œ‹๐‘ฅ), respectively.

12.2. Discrete eigenfunctions on the Sierpiล„ski gasket We now study the eigenfunctions and eigenvalues of the Laplacian on the Sierpiล„ski gasket ๐‘† and, in particular, the possiblity of constructing them through an interpolation process as in the case of the interval. This time we donโ€™t have any explicit formulae nor identities for the eigenfunctions, so we cannot prove directly that their restrictions to each ๐‘‰๐‘š are discrete eigenfunctions of the difference operators ฮ”๐‘š given by (11.24). However, in this section we show how to extend an eigenfunction of ฮ”๐‘šโˆ’1 on ๐‘‰๐‘šโˆ’1 to the next level, provided certain conditions are satisfied, and we discuss whether this process generates all eigenfunctions of the Laplacian ฮ” on ๐‘†. Assume ๐‘ข is an eigenfunction of ฮ”๐‘šโˆ’1 on ๐‘‰๐‘šโˆ’1 with respect to the eigenvalue โˆ’๐œ†๐‘šโˆ’1 , so it satisfies the equation (12.5)

ฮ”๐‘šโˆ’1 ๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘šโˆ’1 ๐‘ข(๐‘ฅ)

for every ๐‘ฅ โˆˆ ๐‘‰๐‘šโˆ’1 โงต๐‘‰0 . We want to extend to a function on ๐‘‰๐‘š , which we also denote by ๐‘ข, which will be an eigenfunction of ฮ”๐‘š with eigenvalue โˆ’๐œ†๐‘š . Let ๐‘ฅ1 , ๐‘ฅ2 , ๐‘ฅ3 โˆˆ ๐‘‰๐‘šโˆ’1 be the boundary points of a cell ๐‘†๐‘ค , for some

228

12. Eigenfunctions of the Laplacian

๐‘ค โˆˆ ๐‘Š๐‘šโˆ’1 . Let ๐‘ฆ1 , ๐‘ฆ2 , ๐‘ฆ3 โˆˆ ๐‘‰๐‘š โˆฉ ๐‘†๐‘ค be the points in ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 inside the cell, where each ๐‘ฆ ๐‘– is opposite to each ๐‘ฅ๐‘– , as in Figure 12.2.

Figure 12.2. The vertices in the cell ๐‘† ๐‘ค , for ๐‘ค โˆˆ ๐‘Š๐‘š . ๐‘ฅ1 , ๐‘ฅ2 , ๐‘ฅ3 โˆˆ ๐‘‰๐‘šโˆ’1 , while ๐‘ฆ 1 , ๐‘ฆ 2 , ๐‘ฆ 3 โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 .

We want to construct ๐‘ข so that, for each ๐‘ฆ ๐‘– , ฮ”๐‘š ๐‘ข(๐‘ฆ ๐‘– ) = โˆ’๐œ†๐‘š ๐‘ข(๐‘ฆ ๐‘– ). By (11.24), this gives us the three equations (12.6a)

๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฅ3 ) + ๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 ) = (4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฆ1 ),

(12.6b)

๐‘ข(๐‘ฅ1 ) + ๐‘ข(๐‘ฅ3 ) + ๐‘ข(๐‘ฆ1 ) + ๐‘ข(๐‘ฆ3 ) = (4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฆ2 ),

(12.6c)

๐‘ข(๐‘ฅ1 ) + ๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฆ1 ) + ๐‘ข(๐‘ฆ2 ) = (4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฆ3 ),

which can be added to obtain (12.7)

(2 โˆ’ ๐œ†๐‘š )(๐‘ข(๐‘ฆ1 ) + ๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 )) = 2(๐‘ข(๐‘ฅ1 ) + ๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฅ3 )).

If ๐œ†๐‘š โ‰  2, we obtain the relation ๐‘ข(๐‘ฆ1 ) + ๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 ) =

2(๐‘ข(๐‘ฅ1 ) + ๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฅ3 )) , 2 โˆ’ ๐œ†๐‘š

which can be used, by adding ๐‘ข(๐‘ฆ ๐‘– ) to its corresponding equation in (12.6a)โ€“(12.6c), to obtain (12.8)

2๐‘ข(๐‘ฅ๐‘– ) + (4 โˆ’ ๐œ†๐‘š )(๐‘ข(๐‘ฅ๐‘— ) + ๐‘ข(๐‘ฅ๐‘˜ )) = (5 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฆ ๐‘– ), 2 โˆ’ ๐œ†๐‘š

12.2. Discrete eigenfunctions on the Sierpiล„ski gasket

229

where ๐‘–, ๐‘—, ๐‘˜ are the three distinct numbers 1, 2, 3. Thus, if ๐œ†๐‘š โ‰  5, we have the extension algorithm (12.9)

๐‘ข(๐‘ฆ ๐‘– ) =

2๐‘ข(๐‘ฅ๐‘– ) + (4 โˆ’ ๐œ†๐‘š )(๐‘ข(๐‘ฅ๐‘— ) + ๐‘ข(๐‘ฅ๐‘˜ )) . (2 โˆ’ ๐œ†๐‘š )(5 โˆ’ ๐œ†๐‘š )

Note that (12.9) reduces to the extension algorithm (11.7) for harmonic functions if ๐œ†๐‘š = 0. The resulting function ๐‘ข on ๐‘‰๐‘š will be an eigenfunction of ฮ”๐‘š if it also satifies ฮ”๐‘š ๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘š ๐‘ข(๐‘ฅ) for each ๐‘‰๐‘šโˆ’1 โงต ๐‘‰0 . Suppose that ๐‘ฅ1 above is not in ๐‘‰0 . Then ๐‘ฅ1 belongs to two cells in level ๐‘š โˆ’ 1, as in

Figure 12.3. If ๐‘ฅ1 โˆˆ ๐‘‰๐‘šโˆ’1 โงต ๐‘‰ 0 , then it belongs to two cells of level ๐‘š โˆ’ 1, with vertices ๐‘ฅ2 , ๐‘ฅ3 , ๐‘ฅ4 , ๐‘ฅ5 โˆˆ ๐‘‰๐‘šโˆ’1 and ๐‘ฆ 1 , ๐‘ฆ 2 , ๐‘ฆ 3 , ๐‘ฆ 4 , ๐‘ฆ5 , ๐‘ฆ 6 โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 .

Figure 12.3. Thus, ฮ”๐‘š ๐‘ข(๐‘ฅ1 ) = โˆ’๐œ†๐‘š ๐‘ข(๐‘ฅ1 ) implies the equation ๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 ) + ๐‘ข(๐‘ฆ4 ) + ๐‘ข(๐‘ฆ5 ) = (4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฅ1 ).

230

12. Eigenfunctions of the Laplacian

As each ๐‘ข(๐‘ฆ ๐‘– ), ๐‘– = 2, 3, 4, 5, satisfies (12.9), we see that (12.10)

๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 ) + ๐‘ข(๐‘ฆ4 ) + ๐‘ข(๐‘ฆ5 ) =

4(4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฅ1 ) + (6 โˆ’ ๐œ†๐‘š )(๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฅ3 ) + ๐‘ข(๐‘ฅ4 ) + ๐‘ข(๐‘ฅ5 )) . (2 โˆ’ ๐œ†๐‘š )(5 โˆ’ ๐œ†๐‘š )

Now, ๐‘ข(๐‘ฅ1 ) also satisfies ฮ”๐‘šโˆ’1 ๐‘ข(๐‘ฅ1 ) = โˆ’๐œ†๐‘šโˆ’1 ๐‘ข(๐‘ฅ1 ), so we have ๐‘ข(๐‘ฅ2 ) + ๐‘ข(๐‘ฅ3 ) + ๐‘ข(๐‘ฅ4 ) + ๐‘ข(๐‘ฅ5 ) = (4 โˆ’ ๐œ†๐‘šโˆ’1 )๐‘ข(๐‘ฅ1 ), which substituting in (12.10) gives us ๐‘ข(๐‘ฆ2 ) + ๐‘ข(๐‘ฆ3 ) + ๐‘ข(๐‘ฆ4 ) + ๐‘ข(๐‘ฆ5 ) =

4(4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฅ1 ) + (6 โˆ’ ๐œ†๐‘š )(4 โˆ’ ๐œ†๐‘šโˆ’1 )๐‘ข(๐‘ฅ1 ) , (2 โˆ’ ๐œ†๐‘š )(5 โˆ’ ๐œ†๐‘š )

so 4(4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฅ1 ) + (6 โˆ’ ๐œ†๐‘š )(4 โˆ’ ๐œ†๐‘šโˆ’1 )๐‘ข(๐‘ฅ1 ) = (4 โˆ’ ๐œ†๐‘š )๐‘ข(๐‘ฅ1 ). (2 โˆ’ ๐œ†๐‘š )(5 โˆ’ ๐œ†๐‘š ) It is not hard to see that this must hold for every ๐‘ฅ โˆˆ ๐‘‰๐‘šโˆ’1 โงต ๐‘‰0 , and, since ๐‘ข is an eigenfunction of ฮ”๐‘šโˆ’1 , it cannot be zero at all points, we obtain the relation 4(4 โˆ’ ๐œ†๐‘š ) + (6 โˆ’ ๐œ†๐‘š )(4 โˆ’ ๐œ†๐‘šโˆ’1 ) = (4 โˆ’ ๐œ†๐‘š )(2 โˆ’ ๐œ†๐‘š )(5 โˆ’ ๐œ†๐‘š ), which we can rewrite as (6 โˆ’ ๐œ†๐‘š )๐œ†๐‘šโˆ’1 = (6 โˆ’ ๐œ†๐‘š )๐œ†๐‘š (5 โˆ’ ๐œ†๐‘š ). If ๐œ†๐‘š โ‰  6, we obtain the condition (12.11)

๐œ†๐‘šโˆ’1 = ๐œ†๐‘š (5 โˆ’ ๐œ†๐‘š ).

We summarize the previous analysis in Proposition 12.12. Proposition 12.12. Suppose that ๐œ†๐‘š โ‰  2, 5, 6, and that we have the condition (12.11). (1) If ๐‘ข is an eigenfunction of ฮ”๐‘šโˆ’1 on ๐‘‰๐‘šโˆ’1 with respect to the eigenvalue โˆ’๐œ†๐‘šโˆ’1 and is extended to ๐‘‰๐‘š by (12.9), then we obtain an eigenfunction of ฮ”๐‘š with respect to the eigenvalue โˆ’๐œ†๐‘š . (2) If ๐‘ข is an eigenfunction of ฮ”๐‘š on ๐‘‰๐‘š with respect to the eigenvalue โˆ’๐œ†๐‘š , then ๐‘ข|๐‘‰๐‘šโˆ’1 is an eigenfunction of ฮ”๐‘šโˆ’1 with respect to โˆ’๐œ†๐‘šโˆ’1 .

12.2. Discrete eigenfunctions on the Sierpiล„ski gasket

231

The discrete process described in Proposition 12.12 can be used to construct eigenfunctions of the Laplacian ฮ” on ๐‘†. Indeed, note that the roots of (12.11) are given by ๐œ†๐‘š =

5 ยฑ โˆš25 โˆ’ 4๐œ†๐‘šโˆ’1 . 2

Using the principal branch of โˆš5 โˆ’ ๐‘ง on โ„‚ โงต [5, โˆž), we observe that, if ๐‘ค=

5 โˆ’ โˆš25 โˆ’ 4๐‘ง , 2

then ๐‘ค=

๐‘ง + ๐‘‚(|๐‘ง|2 ) 5

as ๐‘ง โ†’ 0, and thus 3 ๐‘š 5 ๐œ†๐‘š 2 converges if we start from some |๐œ†๐‘š0 | < 5 and choose the minus sign for all but infinitely many ๐‘š (Exercise (3)). Moreover, the functions ๐‘ข defined by Proposition 12.12 converge to a uniformly continuous function on ๐‘‰โˆ— (Exercise (4)), which satisfies (12.13)

3 ๐‘š 3 5 ฮ”๐‘š ๐‘ข(๐‘ฅ) = โˆ’ 5๐‘š ๐œ†๐‘š ๐‘ข(๐‘ฅ) 2 2 3 for all ๐‘ฅ โˆˆ ๐‘‰โˆ— โงต ๐‘‰0 . Thus, if 5๐‘š ๐œ†๐‘š โ†’ ๐œ†, we can continuously extend ๐‘ข 2 to a function on ๐‘† in dom ฮ” that satisfies ฮ”๐‘ข = โˆ’๐œ†๐‘ข, and thus is an eigenfunction with respect to the eigenvalue โˆ’๐œ†. We described how to construct an eigenfunction of ฮ” through this interpolation process, so we now dicuss whether we can obtain all eigenfunctions of ฮ” in this way. First, we deal with the question of whether we can obtain any eigenvalue โˆ’๐œ† as a limit (12.13) starting from a proper ๐œ† ๐‘š0 . An eigenfunction ๐‘ข that satisfies ๐‘ข|๐‘‰0 = 0 is called a Dirichlet eigenfunction, and we say that an eigenvalue is a Dirichlet eigenvalue if it has a corresponding Dirichlet eigenfunction. Note that, if โˆ’๐œ† is an eigenvalue that is not a Dirichlet eigenvalue, its corresponding eigenspace is at most 3-dimensional, because #๐‘‰0 = 3

232

12. Eigenfunctions of the Laplacian

and otherwise we could obtain a linear combination of 4 or more of its eigenfunctions to obtain 0 on ๐‘‰0 . Moreover, for any ๐œ†1 โ‰  2, (12.9) defines an eigenfunction of ฮ”1 for any choice of ๐‘ข(๐‘1 ), ๐‘ข(๐‘2 ) and ๐‘ข(๐‘3 ), so it defines a 3-dimensional space of eigenfunctions and thus, by taking the minus sign for every ๐‘š in the roots of (12.11), defines a 3-dimensional space of eigenfunctions with respect to each eigenvalue โˆ’๐œ†, if ๐œ† is a limit (12.13). To see which points in โ„‚ are such limits, define 5 โˆ’ โˆš25 โˆ’ 4๐‘ง 2 and, for |๐‘ง| < 2 (so to avoid ๐‘ง = 2), ๐œ“(๐‘ง) =

ฮจ(๐‘ง) = lim 5๐‘š ๐œ“๐‘š (๐‘ง), ๐‘šโ†’โˆž

๐‘š

where ๐œ“ is the composition of ๐œ“ with itself ๐‘š times. ฮจ is well defined in this way (Exercise (3)), and by the above discussion, for any |๐œ†1 | < 2, if 3 ๐œ† = 5ฮจ(๐œ†1 ), 2 then โˆ’๐œ† is an eigenvalue of ฮ”. The convergence of the limit that defines ฮจ is uniform in compact subsets of ๐ต2 (0), so ฮจ is holomorphic in ๐ต2 (0). Moreover, since ๐œ“โ€ฒ (0) = 1/5, we see that each 5๐‘š ๐œ“๐‘š (๐‘ง) has derivative 1 at 0, so ฮจโ€ฒ (0) = 1. By the open mapping theorem of complex analysis (see, for instance, [Ull08, Theorem 5.7]), we see that ฮจ(๐ต2 (0)) contains a ball ๐ต๐‘Ÿ (0) for some ๐‘Ÿ > 0. Therefore, given any non-Dirichlet eigenvalue in the disc of radius 15๐‘Ÿ/2, this eigenvalue and its corresponding eigenfunctions are the result of the discrete algorithm starting from some |๐œ†1 | < 2. Now, for a larger non-Dirichlet eigenvalue โˆ’๐œ†, we choose ๐‘š0 large enough so that |๐œ†| < 3 โ‹… 5๐‘š0 ๐‘Ÿ/2. Thus 3 ๐œ† = 5๐‘š0 ฮจ(๐œ†๐‘š0 ) 2 for some ๐œ†๐‘š0 โˆˆ ๐ต2 (0), so we need to make sure that we can construct a 3dimensional eigenspace of ฮ”๐‘š0 with respect to โˆ’๐œ†๐‘š0 . However, observe that the system of #(๐‘‰๐‘š0 โงต ๐‘‰0 ) linear equations ฮ”๐‘š0 ๐‘ข(๐‘ฅ) = โˆ’๐œ†๐‘š0 ๐‘ข(๐‘ฅ), can be written as ๐’Ÿ๐‘ข(๐‘ฅ) = ๐‘๐‘ฅ ,

12.3. Dirichlet eigenfunctions

233

where ๐’Ÿ is the linear operator given, for each ๐‘ฅ โˆˆ ๐‘‰๐‘š0 โงต ๐‘‰0 , by ๐’Ÿ๐‘ข(๐‘ฅ) = โˆ‘ ๐‘ข(๐‘ฆ) โˆ’ (4 โˆ’ ๐œ†๐‘š0 )๐‘ข(๐‘ฅ), ๐‘ฆโˆผ๐‘š ๐‘ฅ ๐‘ฆโˆ‰๐‘‰0

and โˆ’๐‘ข(๐‘ ๐‘– ) ๐‘๐‘ฅ = { 0

๐‘ฅ โˆผ๐‘š ๐‘ ๐‘– ๐‘ฅ is not a neighbor of a point ๐‘ ๐‘– โˆˆ ๐‘‰0 .

Thus, either we have a unique solution for each choice of ๐‘ข(๐‘1 ), ๐‘ข(๐‘2 ), ๐‘ข(๐‘3 ), so we have a 3-dimensional space of eigenfunctions with respect to โˆ’๐œ†๐‘š0 , or โˆ’๐œ†๐‘š0 is a Dirichlet eigenvalue, because in that case we would have solutions when all ๐‘๐‘ฅ = 0, and thus โˆ’๐œ† would be a Dirichlet eigenvalue. We have proven that any eigenfunction with respect to a non-Dirichlet eigenvalue can be constructed through the discrete process. The Dirichlet eigenfunctions will be discussed in Section 12.3.

12.3. Dirichlet eigenfunctions In this section we discuss the construction of the Dirichlet eigenfunctions of ฮ” on the Sierpiล„ski gasket ๐‘†. For this, we will analyze more carefully the discrete process, keeping track on the number of linearly independent eigenfunctions constructed, in order to conclude that we have constructed all of them. We first construct all discrete Dirichlet eigenfunctions of the operators ฮ”๐‘š , ๐‘š โ‰ฅ 1. We start with the Dirichlet eigenfunctions of ฮ”1 on ๐‘‰1 , so we need to solve the three equations ฮ”1 ๐‘ข(๐‘ฅ) = โˆ’๐œ†1 ๐‘ข(๐‘ฅ), one for each of the three points ๐‘ฅ โˆˆ ๐‘‰1 โงต ๐‘‰0 , with the Dirichlet condition ๐‘ข(๐‘1 ) = ๐‘ข(๐‘2 ) = ๐‘ข(๐‘3 ) = 0. By Exercise (5), we have three linearly independent solutions shown in Figure 12.4, with corresponding eigenvalues โˆ’2 and โˆ’5, the latter with multiplicity 2. Note that these are two of the forbidden eigenvalues of Proposition 12.12, which makes sense as their restrictions to ๐‘‰0 are not eigenfunctions of ฮ”0 . Also, observe that one of the Dirichlet eigenfunctions with ๐œ†1 = 5 is the rotation by 2๐œ‹/3 of the other, and that we have

234

12. Eigenfunctions of the Laplacian

Figure 12.4. The Dirichlet eigenfunctions of ฮ”1 on ๐‘‰ 1 , corresponding to ๐œ†1 = 2, 5 and 5, respectively. For simplicity, we only show the values of the eigenfunction at the vertices, and we donโ€™t show the ones that have value zero.

a third eigenfunction corresponding to the next rotation. However, the latter is not linearly independent of the other two. If we remove the Dirichlet condition, we have two more linearly independent solutions for ๐œ†1 = 2 and one more for ๐œ†1 = 5, shown in Figure 12.5. Note that, for the case ๐œ†1 = 2,

Figure 12.5. Eigenfunctions of ฮ”1 on ๐‘‰ 1 corresponding to ๐œ†1 = 2, 2 and 5, respectively. The third rotation of the first two of them is also an eigenfunction corresponding to ๐œ†1 = 2, but is not linearly independent.

๐‘ข(๐‘1 ) + ๐‘ข(๐‘2 ) + ๐‘ข(๐‘3 ) = 0, which is a necessary condition given by (12.7). For ๐œ†1 = 5, we have ๐‘ข(๐‘1 ) = ๐‘ข(๐‘2 ) = ๐‘ข(๐‘3 ), which follows from (12.8) (Exercise (6)). We now move on to the eigenvalues of ฮ”2 on ๐‘‰2 . As #(๐‘‰2 โงต ๐‘‰0 ) = 12, we need to construct twelve linearly independent Dirichlet eigenfunctions. Although we could just solve explicitly the system of equations (12.14)

ฮ”2 ๐‘ข(๐‘ฅ) = โˆ’๐œ†2 ๐‘ข(๐‘ฅ)

12.3. Dirichlet eigenfunctions

235

for ๐‘ฅ โˆˆ ๐‘‰2 โงต ๐‘‰0 , with ๐‘ข(๐‘ ๐‘– ) = 0, we will construct them in such a way that we can generalize for any ๐‘š โ‰ฅ 2. First, six of the Dirichlet eigenfunctions correspond to the extensions of the eigenfunctions of ฮ”1 shown in Figure 12.4, with eigenvalues โˆ’๐œ†2 where 5 ยฑ โˆš25 โˆ’ 4๐œ†1 ๐œ†2 = . 2 Thus, we have six eigenfunctions with ๐œ†2 given by 5 + โˆš17 5 โˆ’ โˆš17 5 + โˆš5 5 โˆ’ โˆš5 , , , , 2 2 2 2 the last two with multiplicity 2. The other six Dirichlet eigenfunctions are not extensions from any eigenfunction on ๐‘‰1 , so they must correspond to any of the forbidden values ๐œ†2 = 2, 5 or 6; otherwise, their restrictions would be Dirichlet eigenfunctions on ๐‘‰1 , by Proposition 12.12. For ๐œ†2 = 2, note that the restriction of such eigenfunction to each cell ๐‘† ๐‘– โˆฉ ๐‘‰2 must correspond to one of eigenfunctions with ๐œ†1 = 2 shown in Figures 12.4 or 12.5, since the same difference equation must be satisfied. However, since (12.14) must also be satisfied in the points ๐‘‰1 โงต ๐‘‰0 , this impose an extra restriction of how the ๐‘‰1 -eigenfunctions are pasted into ๐‘‰2 . In fact, one sees that it is impossible to do this because, when one starts at one of the corners, the next one is either determined by this condition if one chooses to start with the eigenfunction of Figure 12.4, and then it is impossible to past the third one, or it is already impossible to paste the second one if we choose one of those shown in Figure 12.5. See Figure 12.6. Therefore, there are no Dirichlet eigenfunctions of ฮ”2 with respect to the eigenvalue โˆ’2. For ๐œ†2 = 5, we now see that it is possible to paste two consecutive eigenfunctions shown in Figure 12.4, and their rotations, so we obtain the three linearly independent Dirichlet eigenfunctions shown in Figure 12.7. For ๐œ†2 = 6, we first consider the eigenfunctions of ฮ”1 with ๐œ†1 = 6. These are shown in Figure 12.8. These are not Dirichlet eigenfunctions, of course, but they can be pasted consecutively (as in the case of ๐œ†2 = 5) to form Dirichlet eigenvalues of ฮ”2 , as is shown in Figure 12.9. Note that not only each is a rotation of the other, but each one is โ€œcenteredโ€ at a point in ๐‘‰1 , the point where two of the eigenfunctions from Figure 12.8

236

12. Eigenfunctions of the Laplacian

Figure 12.6. Failed attempts to construct a Dirichlet eigenfunction on ๐‘‰ 2 with ๐œ†2 = 2.

Figure 12.7. Three linearly independent Dirichlet eigenfunctions with ๐œ†2 = 5, constructed by pasting two consecutive eigenfunctions on ๐‘‰ 1 .

Figure 12.8. The three linearly independent eigenfunctions ฮ”1 with ๐œ†1 = 6. Note that each is a rotation of the other.

where pasted. These form the last three linearly independent Dirichlet eigenfunctions of ฮ”2 on ๐‘‰2 . We can now generalize these constructions for all ๐‘š โ‰ฅ 3. Since #(๐‘‰๐‘š โงต ๐‘‰0 ) =

3๐‘š+1 โˆ’ 3 2

12.3. Dirichlet eigenfunctions

237

Figure 12.9. Three linearly independent Dirichlet eigenfunctions of ฮ”2 on ๐‘‰ 2 with ๐œ†2 = 6. Note that each is centered at a point in ๐‘‰ 1 .

(Exercise (8)), there must be (3๐‘š+1 โˆ’ 3)/2 linearly independent Dirichlet eigenfunctions of ฮ”๐‘š . First, note that, as in the case for ๐‘š = 2, there are no Dirichlet eigenfunctions with ๐œ†๐‘š = 2, for ๐‘š โ‰ฅ 3. The argument is the same as above, and we leave it as an exercise (Exercise (7)). Also, note that if ๐œ†๐‘šโˆ’1 = 6, then 5 ยฑ โˆš25 โˆ’ 4๐œ†๐‘šโˆ’1 5ยฑ1 = = 2, 3, 2 2 so we cannot extend such Dirichlet eigenfunctions of ฮ”๐‘šโˆ’1 using the negative sign above. Thus, in order to count the eigenfunctions of ฮ”๐‘š that are extensions from ๐‘‰๐‘šโˆ’1 , we first need to count the discrete Dirichlet eigenfunctions with respect to the eigenvalue โˆ’6. These are constructed as in the case ๐‘š = 2 above. Indeed, for each ๐‘‰๐‘š , we can paste two of the eigenfunctions in Figure 12.8 centered at each point ๐‘ฅ โˆˆ ๐‘‰๐‘šโˆ’1 , as in Figure 12.9. Hence, we have (3๐‘š โˆ’ 3)/2 such eigenfunctions. On ๐‘‰๐‘šโˆ’1 , we have then (3๐‘šโˆ’1 โˆ’ 3)/2 Dirichlet eigenfunctions with ๐œ†๐‘šโˆ’1 = 6, so they extend to Dirichlet eigenfunctions of ฮ”๐‘š with ๐œ†๐‘š = 3. The number of Dirichlet eigenfunctions with ๐œ†๐‘šโˆ’1 โ‰  6 is then 3๐‘š โˆ’ 3 3๐‘šโˆ’1 โˆ’ 3 โˆ’ = 3๐‘šโˆ’1 , 2 2 and they extend to 2 โ‹… 3๐‘šโˆ’1 eigenfunctions of ฮ”๐‘š with ๐œ†๐‘š =

5 ยฑ โˆš25 โˆ’ 4๐œ†๐‘šโˆ’1 . 2

Thus, the number of Dirichlet eigenfunctions of ฮ”๐‘š extended from ๐‘‰๐‘šโˆ’1 is (3๐‘šโˆ’1 โˆ’ 3)/2 + 2 โ‹… 3๐‘šโˆ’1 = (5 โ‹… 3๐‘šโˆ’1 โˆ’ 3)/2 which, together with the

238

12. Eigenfunctions of the Laplacian

ones with ๐œ†๐‘š = 6 that we already constructed, make a total of 8 โ‹… 3๐‘šโˆ’1 โˆ’ 6 5 โ‹… 3๐‘šโˆ’1 โˆ’ 3 3๐‘š โˆ’ 3 + = . = 4 โ‹… 3๐‘šโˆ’1 โˆ’ 3. 2 2 2 Thus, we need to construct 3๐‘š+1 โˆ’ 3 3๐‘šโˆ’1 + 3 โˆ’ (4 โ‹… 3๐‘šโˆ’1 โˆ’ 3) = 2 2 more eigenfunctions. They will correspond to ๐œ†๐‘š = 5. To construct these, first note that we can paste together a chain of eigenfunctions with ๐œ†1 = 5 (as the last two in Figure 12.4 and their rotations) in each of the cells ๐‘†๐‘ค , for ๐‘ค โˆˆ ๐‘Š๐‘šโˆ’1 , surrounding each downward triangular cycle formed with points in ๐‘‰ ๐‘˜ โงต ๐‘‰ ๐‘˜โˆ’1 , for ๐‘˜ = 1, 2, . . . , ๐‘š โˆ’ 1. We show this for the largest and one of the next largest downward triangular cycles for ๐‘‰3 in Figure 12.10. As #๐‘‰ ๐‘˜ โงต ๐‘‰ ๐‘˜โˆ’1 = 3๐‘˜ ,

Figure 12.10. Two of the Dirichlet eigenfunctions of ฮ”3 with ๐œ†3 = 5, chained around triangles with vertices in ๐‘‰ 1 โงต ๐‘‰ 0 and ๐‘‰ 2 โงต ๐‘‰ 1 , respectively.

we have that the number of downward triangular cycles with vertices in ๐‘‰ ๐‘˜ โงต ๐‘‰ ๐‘˜โˆ’1 is 3๐‘˜โˆ’1 , and thus the total number of such cycles is 3๐‘šโˆ’1 โˆ’ 1 . 2 We obtain two more linearly independent eigenfunctions by pasting a chain if eigenfunctions along the edge from vertex ๐‘1 to ๐‘2 , and another one from ๐‘2 to ๐‘3 . We thus have a total of (3๐‘šโˆ’1 + 3)/2 eigenfunctions with ๐œ†๐‘š = 5, as required. 1 + 3 + 32 + 3๐‘šโˆ’2 =

We have thus constructed all Dirichlet eigenfunctions of the discrete Laplacian ฮ”๐‘š on ๐‘‰๐‘š . As in Section 12.2, starting from each eigenfunction on ๐‘‰๐‘š0 with ๐œ†๐‘š0 , the sequence obtained from discrete converges to

12.3. Dirichlet eigenfunctions

239

a uniformly continuous function on ๐‘‰โˆ— , which extends continuously to a Dirichlet eigenfunction ๐‘ข of ฮ” on ๐‘† with respect to the eigenvalue โˆ’๐œ†, where 3 ๐œ† = lim 5๐‘š ๐œ†๐‘š . 2 ๐‘šโ†’โˆž However, recall that, for each ๐‘š > ๐‘š0 , 5 + ๐œ–๐‘š โˆš25 โˆ’ 4๐œ†๐‘šโˆ’1 , 2 where all but infinitely many ๐œ–๐‘š = โˆ’1. This means that, for each finite sequence ๐œ–๐‘š0 +1 , ๐œ–๐‘š0 +2 , . . . , ๐œ–๐‘€ of ยฑ1, we obtain a distinct eigenvalue โˆ’๐œ† with a distinct eigenfunction ๐‘ข when choosing ๐œ–๐‘š = โˆ’1 for all ๐‘š > ๐‘€. Therefore, we obtain a sequence of Dirichlet eigenvalues and eigenfunctions from each starting ๐œ†๐‘š0 . ๐œ†๐‘š =

In Figure 12.11 we show the Dirichlet eigenfunctions that we ob-

Figure 12.11. Dirichlet eigenfunctions of ฮ” on ๐‘† with respect to the eigenvalues โˆ’๐œ†, with ๐œ† โ‰ˆ 16.816, ๐œ† โ‰ˆ 240.1686 and ๐œ† โ‰ˆ 920.6197, respectively, obtained from ๐œ†1 = 2 when choosing ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 2 in the first case, ๐œ–2 = 1 and ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 3 in the second, and ๐œ–2 = ๐œ–3 = 1 and ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 4 in the third case.

tain when we start from the Dirichlet eigenfunction with respect to the eigenvalue โˆ’2 shown in Figure 12.4, where ๐‘ข(๐‘ฅ) = 1 for ๐‘ฅ โˆˆ ๐‘‰1 โงต ๐‘‰0 . We obtain the first eigenfunction shown when we choose ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 2. The first terms of the sequence ๐œ†๐‘š in this case are 5 โˆ’ โˆš17 5 โˆ’ โˆš15 + 2โˆš17 5 โˆ’ โˆš15 + 2โˆš15 + 2โˆš17 , , , ... , 2 2 2 and one can approximate its normalized limit by 2,

3 lim 5๐‘š ๐œ†๐‘š โ‰ˆ 16.816. 2 ๐‘šโ†’โˆž

240

12. Eigenfunctions of the Laplacian

We obtain the second eigenfunction when choosing ๐œ–2 = 1 and ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 3, so the first terms of the sequence ๐œ†๐‘š are 5 + โˆš17 5 โˆ’ โˆš15 โˆ’ 2โˆš17 5 โˆ’ โˆš15 + 2โˆš15 โˆ’ 2โˆš17 , , , ... , 2 2 2 and its limit 3 lim 5๐‘š ๐œ†๐‘š โ‰ˆ 240.1686. 2 ๐‘šโ†’โˆž We obtain the third eigenfunction in Figure 12.11 when choosing ๐œ–2 = ๐œ–3 = 1 and ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 4, and in this case 3 lim 5๐‘š ๐œ†๐‘š โ‰ˆ 920.6197. 2 ๐‘šโ†’โˆž In Figure 12.12 we show the eigenfunctions obtained starting with ๐œ†1 = 2,

Figure 12.12. Dirichlet eigenfunctions of ฮ” on ๐‘† with respect to the eigenvalues โˆ’๐œ†, with ๐œ† โ‰ˆ 55.8858, ๐œ† โ‰ˆ 172.3645 and ๐œ† โ‰ˆ 1032.0357.

5, corresponding to the second eigenfunction in Figure 12.4, and the same choice of starting sequences ๐œ–2 , ๐œ–3 as above. In this case we obtain the limits ๐œ† โ‰ˆ 55.8858, ๐œ† โ‰ˆ 172.3645, ๐œ† โ‰ˆ 1032.0357, respectively. 12.15. The argument in Section 12.2 can also be applied here to conclude that, if โˆ’๐œ† is a Dirichlet eigenvalue of ฮ” on ๐‘†, then ๐œ† is the limit of (3/2)5๐‘š ๐œ†๐‘š , starting from ๐œ†1 = 2, ๐œ†๐‘š0 = 5 or ๐œ†๐‘š0 = 6, from an appropiately chosen ๐‘š0 and a particular choice of ๐œ–๐‘š , ๐‘š > ๐‘š0 (Exercise (5)). Our construction above produces all Dirichlet eigenfunctions with respect to a Dirichlet eigenvalue โˆ’๐œ†. 12.16. If โˆ’๐œ† is a Dirichlet eigenvalue, so is โˆ’5๐‘š ๐œ†, for each ๐‘š โ‰ฅ 1. Indeed, if ๐œ† is the discrete limit starting from ๐œ†๐‘š0 , then ๐œ† ฬ„ = 5๐‘š ๐œ† is the limit ฬ„ +๐‘š = ๐œ†๐‘š . starting from ๐œ†๐‘š 0 0

Exercises

241

It is also possible to prove that we can construct all eigenfunctions ๐‘ข with respect to a Dirichlet eigenvalue through this discrete algorithm, even those where ๐‘ข|๐‘‰0 โ‰  0. The argument involves Neumann derivatives at the vertices, which we wonโ€™t discuss here. The reader can find the argument in [Str06].

Exercises (1) Let ๐œ™(๐‘ง) = 2 โˆ’ โˆš4 โˆ’ ๐‘ง, for |๐‘ง| < 4, using the principal branch of โˆš4 โˆ’ ๐‘ง on โ„‚ โงต [4, โˆž). 1 (a) ๐œ™(๐‘ง) = ๐‘ง + ๐‘‚(|๐‘ง|2 ) as ๐‘ง โ†’ 0. 4 (b) Given |๐œ†1 | < 4, the sequence defined by ๐œ†๐‘š = ๐œ™(๐œ†๐‘šโˆ’1 ) and ๐‘ง๐‘š = 4๐‘š ๐œ†๐‘š for ๐‘š โ‰ฅ 1 satisfies ๐‘ง๐‘š โˆ’ ๐‘ง๐‘šโˆ’1 = ๐‘‚(2โˆ’๐‘š ). (c) ๐‘ง๐‘š is Cauchy and hence converges. (2) Let ๐œ†๐‘š be the sequence defined in Exercise (1) with ๐œ†1 = 2. Then 4๐‘š ๐œ†๐‘š โ†’ ๐œ‹2 . 5 โˆ’ โˆš25 โˆ’ 4๐‘ง 25 , for |๐‘ง| < , with the principal branch of 2 4 โˆš5 โˆ’ ๐‘ง on โ„‚ โงต [5, โˆž). 1 (a) ๐œ“(๐‘ง) = ๐‘ง + ๐‘‚(|๐‘ง|2 ) as ๐‘ง โ†’ 0. 5 25 (b) Given |๐œ†1 | < , the sequence defined by ๐œ†๐‘š = ๐œ“(๐œ†๐‘šโˆ’1 ) and 4 ๐‘ง๐‘š = 5๐‘š ๐œ†๐‘š for ๐‘š โ‰ฅ 1 satisfies

(3) Let ๐œ“(๐‘ง) =

๐‘ง๐‘š โˆ’ ๐‘ง๐‘šโˆ’1 = ๐‘‚(5โˆ’๐‘š ). (c) ๐‘ง๐‘š is Cauchy and hence converges. (4) The sequence functions resulting from the discrete process on the Sierpiล„ski gasket constructs a uniformly continuous function on ๐‘‰โˆ— . (5) Calculate the eigenvalues and eigenvectors of the matrix 4 โˆ’1 โˆ’1 (โˆ’1 4 โˆ’1) . โˆ’1 โˆ’1 4

242

12. Eigenfunctions of the Laplacian

(6) If ๐œ†๐‘š = 5, the equation (12.8) implies that ๐‘ข(๐‘ฅ1 ) = ๐‘ข(๐‘ฅ2 ) = ๐‘ข(๐‘ฅ3 ). (7) For ๐‘š โ‰ฅ 2, ฮ”๐‘š does not have any Dirichlet eigenfunctions on ๐‘‰2 with respect to the eigenvalue โˆ’2. (8) (a) For any ๐‘š โ‰ฅ 0, #๐‘‰๐‘š =

3๐‘š+1 + 3 . 2

(b) For any ๐‘š โ‰ฅ 1, 3๐‘š+1 โˆ’ 3 . 2 (9) If โˆ’๐œ† is a Dirichlet eigenvalue of ฮ” on ๐‘†, then there exists ๐‘š0 and a sequence ๐œ–๐‘š such that 3 ๐œ† = lim 5๐‘š ๐œ†๐‘š , 2 ๐‘šโ†’โˆž with ๐œ†๐‘š0 = 2, 5 or 6. #(๐‘‰๐‘š โงต ๐‘‰0 ) =

(10) Let ๐‘ข be the Dirichlet eigenfunction obtained by discrete with ๐œ†1 = 2, and choosing ๐œ–๐‘š = โˆ’1 for all ๐‘š โ‰ฅ 2. This is the first eigenfunction shown in Figure 12.11. Then ๐‘ข(๐‘ฅ) > 0 for all ๐‘ฅ โˆˆ ๐‘† โงต ๐‘‰0 .

Notes The algorithm described in this chapter for the construction of the eigenfunctions, as well as the relation for the discrete eigenvalues between levels, is known as the spectral decimation method and is due to Tadashi Shima [Shi91], who based his work from the results in [Ram84]. Our discussion follows the analysis presented in [Str06].

Chapter 13

Harmonic functions on post-critically finite sets

In this chapter we discuss how to construct the Laplacian on other fractal sets, analogous to the construction of the Laplacian on the Sierpiล„ski gasket.

13.1. Post-critically finite sets Let ๐พ โŠ‚ โ„๐‘‘ be a self-similar set with respect to the contractions ๐‘“1 , ๐‘“2 , . . . , ๐‘“๐‘ . Recall that ๐พ is nonempty, compact, and satisfies ๐พ = ๐‘“1 (๐พ) โˆช ๐‘“2 (๐พ) โˆช . . . โˆช ๐‘“๐‘ (๐พ). We assume that the contractions ๐‘“๐‘— are restricted to ๐พ, so we see them as functions ๐‘“๐‘— โˆถ ๐พ โ†’ ๐พ. We use the notation of Chapter 10, so ๐‘Š๐‘š , for ๐‘š โ‰ฅ 1, is the set of words ๐‘ค = ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š of length ๐‘š with each ๐‘ค ๐‘– โˆˆ {1, 2, . . . , ๐‘}. We define ๐‘“๐‘ค = ๐‘“๐‘ค1 โˆ˜ ๐‘“๐‘ค2 โˆ˜ โ‹ฏ โˆ˜ ๐‘“๐‘ค๐‘š and ๐พ๐‘ค = ๐‘“๐‘ค (๐พ). For ๐‘š = 0 we define ๐‘Š0 = {โˆ…}, where โˆ… is the empty word, ๐‘“โˆ… is the identity function, and ๐พโˆ… = ๐พ. Each ๐พ๐‘ค , with ๐‘ค โˆˆ ๐‘Š๐‘š , is called a cell of level ๐‘š. We have that ๐พ = โ‹ƒ๐‘คโˆˆ๐‘Š ๐‘“๐‘ค (๐พ) where, if ๐›ผ๐‘— ๐‘š is the contraction constant of ๐‘“๐‘— , diam(๐พ๐‘ค ) โ‰ค ๐›ผ๐‘ค diam(๐พ), 243

244

13. Harmonic functions on post-critically finite sets

with ๐›ผ๐‘ค = ๐›ผ๐‘ค1 ๐›ผ๐‘ค2 โ‹ฏ ๐›ผ๐‘ค๐‘š . Since each ๐›ผ๐‘— < 1, then diam(๐พ๐‘ค ) โ†’ 0 if ๐‘š โ†’ โˆž. Thus, for each sequence ๐œ” = ๐‘ค 1 , ๐‘ค 2 , . . ., if we denote ๐œ”๐‘š = ๐‘ค 1 ๐‘ค 2 . . . ๐‘ค ๐‘š , then โ‹‚๐‘šโ‰ฅ1 ๐พ๐œ”๐‘š contains exactly one point of ๐พ. Conversely, if ๐‘ฅ โˆˆ ๐พ, there exists a sequence ๐œ” such that (13.1)

โ‹‚

๐พ๐œ”๐‘š = {๐‘ฅ}.

๐‘šโ‰ฅ1

For a self-similar set ๐พ, define the set (13.2)

๐’ž=

โ‹ƒ

๐พ๐‘– โˆฉ ๐พ๐‘— .

๐‘–โ‰ ๐‘—

Then ๐’ž is the set of all overlaps between the images ๐พ๐‘— of ๐พ under the contractions. ๐’ž is called the critical set. We assume ๐’ž โ‰  โˆ…, and define the post-critical set of ๐พ by (13.3)

๐‘‰0 =

โ‹ƒ โ‹ƒ

๐‘“๐‘คโˆ’1 (๐’ž),

๐‘šโ‰ฅ1 ๐‘คโˆˆ๐‘Š๐‘š

Observe that ๐‘‰0 is the union of all possible pre-images, under any number of iterations of the ๐‘“๐‘— , of the critical set. We say that ๐พ is a post-critically finite set, denoted as PCF set, if ๐‘‰0 is finite. Example 13.4. Consider the Sierpiล„ski gasket ๐‘† discussed in the previous chapters. Its critical set ๐’ž consists of the three middle points of its sides, which clearly are the images of the triangle vertices ๐‘1 , ๐‘2 , ๐‘3 (see Figure 11.2) under the contractions ๐‘“1 , ๐‘“2 , ๐‘“3 . Thus ๐‘1 , ๐‘2 , ๐‘3 โˆˆ ๐‘‰0 . Now, since ๐‘1 โˆ‰ ๐‘† 2 nor ๐‘1 โˆ‰ ๐‘† 3 , and ๐‘1 is the fixed point of ๐‘“1 , the further preimages of ๐‘1 contain only the point ๐‘1 . Similarly for ๐‘2 and ๐‘3 , which are the fixed points of ๐‘“2 and ๐‘“3 , respectively, we have that ๐‘‰0 = {๐‘1 , ๐‘2 , ๐‘3 }. Therefore ๐‘† is a PCF set. Example 13.5. The interval ๐ผ = [0, 1], seen as a self-similar set with contractions ๐‘“1 (๐‘ฅ) = ๐‘ฅ/2 and ๐‘“2 (๐‘ฅ) = ๐‘ฅ/2+1/2, is also a PCF set. Indeed, ๐’ž = {1/2} and ๐‘‰0 = {0, 1}. Note that, in both of the previous examples, ๐‘‰0 is known as the boundary of ๐‘† and ๐ผ, respectively. In general, ๐‘‰0 is called the boundary of the PCF set ๐พ. Example 13.6. The Hata tree set ๐พ of Example 10.22, with contractions 2 1 ๐‘“1 (๐‘ง) = ๐‘๐‘ง,ฬ„ ๐‘“2 (๐‘ง) = ๐‘ง ฬ„ + , 3 3

13.2. Harmonic structures and discrete energy

245

โˆš3 1 where ๐‘ = + ๐‘–. Its critical set consists only of the point 1/3 (see 2 6 Figure 10.6), which is given by 1 = ๐‘“1 (๐‘) = ๐‘“2 (0). 3 Thus ๐‘, 0 โˆˆ ๐‘‰0 . Now ๐‘ = ๐‘“1 (1), so 1 โˆˆ ๐‘‰0 . Since 0, ๐‘ โˆˆ ๐พ1 โงต ๐พ2 and 1 โˆˆ ๐พ2 โงต ๐พ1 , and 0 and 1 are the fixed points of ๐‘“1 and ๐‘“2 , respectively, we conclude that ๐‘‰0 = {๐‘, 0, 1}. For each ๐‘š โ‰ฅ 1, we define ๐‘‰๐‘š = ๐‘“1 (๐‘‰๐‘šโˆ’1 ) โˆช ๐‘“2 (๐‘‰๐‘šโˆ’1 ) โˆช . . . โˆช ๐‘“๐‘ (๐‘‰๐‘šโˆ’1 ). ๐‘‰๐‘š is the set of vertices of level ๐‘š. We see that ๐‘‰๐‘š =

โ‹ƒ

๐‘“๐‘ค (๐‘‰0 )

๐‘คโˆˆ๐‘Š๐‘š

and, by (13.3), each ๐‘‰๐‘šโˆ’1 โŠ‚ ๐‘‰๐‘š . The set of all vertices is denoted by ๐‘‰โˆ— , so ๐‘‰โˆ— =

โ‹ƒ

๐‘‰๐‘š .

๐‘šโ‰ฅ0

By (13.1), ๐‘‰โˆ— is dense in ๐พ (Exercise (1)). If we denote ๐‘‰0 = {๐‘1 , ๐‘2 , . . . , ๐‘๐‘€ }, where ๐‘€ = #๐‘‰0 , then each ๐‘ฅ โˆˆ ๐‘‰๐‘š is of the form ๐‘ฅ = ๐‘“๐‘ค (๐‘ ๐‘– ) for some ๐‘ค โˆˆ ๐‘Š๐‘š and ๐‘— = 1, 2, . . . , ๐‘. We say that two vertices ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰๐‘š are adjacent, or neighbors, and write ๐‘ฅ โˆผ๐‘š ๐‘ฆ, if there exists ๐‘ค โˆˆ ๐‘Š๐‘š such that ๐‘ฅ = ๐‘“๐‘ค (๐‘ ๐‘– ) and ๐‘ฆ = ๐‘“๐‘ค (๐‘๐‘— ), for some ๐‘–, ๐‘—. That is, ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘“๐‘ค (๐‘‰0 ), so they belong to the same cell ๐พ๐‘ค .

13.2. Harmonic structures and discrete energy For a nonempty finite set ๐‘‰, let ๐‘™(๐‘‰) be the vector space of real valued functions on ๐‘‰. Thus, if #๐‘‰ = ๐‘€, then ๐‘™(๐‘‰) โ‰… โ„๐‘€ . ๐‘™(๐‘‰) has the inner product โŸจ๐‘ข, ๐‘ฃโŸฉ = โˆ‘ ๐‘ข(๐‘ฅ)๐‘ฃ(๐‘ฅ), ๐‘ฅโˆˆ๐‘‰

so โŸจโ‹…, โ‹…โŸฉ is essentially the dot product in โ„๐‘€ .

246

13. Harmonic functions on post-critically finite sets

The standard basis of ๐‘™(๐‘‰) is given by the functions ๐œ’{๐‘ฅ} , ๐‘ฅ โˆˆ ๐‘‰. A linear operator ๐‘‡ on ๐‘™(๐‘‰) can be identified with the ๐‘€ ร—๐‘€ matrix which, with respect to the standard basis, has entries ๐‘‡๐‘ฅ๐‘ฆ = โŸจ๐‘‡๐œ’{๐‘ฅ} , ๐œ’{๐‘ฆ} โŸฉ. Any bilinear form ๐ต on ๐‘™(๐‘‰) is induced by an operator ๐‘‡ via ๐ต(๐‘ข, ๐‘ฃ) = โˆ’โŸจ๐‘‡๐‘ข, ๐‘ฃโŸฉ. We use the negative sign for technical reasons that will be clear below. If ๐ต is symmetric, that is ๐ต(๐‘ข, ๐‘ฃ) = ๐ต(๐‘ฃ, ๐‘ข) for any ๐‘ข, ๐‘ฃ โˆˆ ๐‘™(๐‘‰), then ๐‘‡ is symmetric. If ๐ต is symmetric, we denote ๐ต(๐‘ข, ๐‘ข) simply by ๐ต(๐‘ข), so ๐‘ข โ†ฆ ๐ต(๐‘ข) is a quadratic form on ๐‘™(๐‘‰). The symmetric bilinear form โ„ฐ is a Dirichet form if it satisfies (1) โ„ฐ(๐‘ข) โ‰ฅ 0 for all ๐‘ข โˆˆ ๐‘™(๐‘‰); (2) โ„ฐ(๐‘ข) = 0 if and only if ๐‘ข is constant; and (3) โ„ฐ(๐‘ข)ฬ„ โ‰ค โ„ฐ(๐‘ข) for every ๐‘ข โˆˆ ๐‘™(๐‘‰), where ๐‘ข(๐‘ฅ) < 0 โŽง0 ๐‘ข(๐‘ฅ) ฬ„ = ๐‘ข(๐‘ฅ) 0 โ‰ค ๐‘ข(๐‘ฅ) โ‰ค 1 โŽจ ๐‘ข(๐‘ฅ) > 1 โŽฉ1 is the cut of ๐‘ข by [0, 1]. 13.7. Any bilinear form โ„ฐ given by โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ‘ ๐‘๐‘ฅ๐‘ฆ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))(๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)), ๐‘ฅ,๐‘ฆโˆˆ๐‘‰

with ๐‘๐‘ฅ๐‘ฆ โ‰ฅ 0, is a Dirichlet form if and only if, for each ๐‘ฅ0 , ๐‘ฆ0 โˆˆ ๐‘‰, we can find ๐‘ฅ1 , ๐‘ฅ2 , . . . , ๐‘ฅ๐‘˜ = ๐‘ฆ0 such that ๐‘๐‘ฅ๐‘—โˆ’1 ๐‘ฅ๐‘— > 0. We just need to observe that โ„ฐ(๐‘ข) = โˆ‘ ๐‘๐‘ฅ๐‘ฆ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))2 ๐‘ฅ,๐‘ฆโˆˆ๐‘‰

is a nonnegative linear combination of the squares (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))2 , so (1) and (3) are clearly satisfied, and that โ„ฐ(๐‘ข) = 0 when, for each pair ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰, ๐‘๐‘ฅ๐‘ฆ = 0 or ๐‘ข(๐‘ฅ) = ๐‘ข(๐‘ฆ), so the condition implies that this can only happen if ๐‘ข is constant. Note that โ„ฐ(๐‘ข, ๐‘ฃ) is the polarization of โ„ฐ(๐‘ข).

13.2. Harmonic structures and discrete energy

247

Let ๐พ be a PCF set and consider the sequence ๐‘‰๐‘š of vertices of level ๐‘š, defined for ๐‘š โ‰ฅ 0 as above. For each ๐‘š โ‰ฅ 0, let โ„ฐ๐‘š be a Dirichlet form on ๐‘™(๐‘‰๐‘š ). We say that the sequence โ„ฐ๐‘š is a harmonic structure on ๐พ if is satisfies: โ€ข (self-similarity) there exist numbers 0 < ๐‘Ÿ1 , ๐‘Ÿ2 , . . . , ๐‘Ÿ๐‘ < 1 such that ๐‘

(13.8)

1 โ„ฐ๐‘šโˆ’1 (๐‘ข โˆ˜ ๐‘“๐‘— , ๐‘ฃ โˆ˜ ๐‘“๐‘— ) ๐‘Ÿ ๐‘—=1 ๐‘—

โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ‘

for all ๐‘ข, ๐‘ฃ โˆˆ (๐‘‰๐‘š ) and ๐‘š โ‰ฅ 1; and โ€ข (compatibility) for each ๐‘š โ‰ฅ 1 and ๐‘ฃ โˆˆ ๐‘™(๐‘‰๐‘šโˆ’1 ), (13.9)

โ„ฐ๐‘šโˆ’1 (๐‘ฃ) = min{โ„ฐ๐‘š (๐‘ข) โˆถ ๐‘ข โˆˆ ๐‘™(๐‘‰๐‘š ), ๐‘ข|๐‘‰๐‘šโˆ’1 = ๐‘ฃ}.

We call each Dirichlet form โ„ฐ๐‘š on ๐‘™(๐‘‰๐‘š ) the energy on ๐‘‰๐‘š . For each ๐‘ข โˆˆ ๐‘™(๐‘‰๐‘š ) and ๐‘— = 1, 2, . . . , ๐‘, ๐‘ข โˆ˜ ๐‘“๐‘— defines a function on ๐‘‰๐‘šโˆ’1 , so selfsimilarity of the harmonic structure โ„ฐ๐‘š means that the energy on each level can be distributed over each cell ๐พ๐‘— , weighted by the 1/๐‘Ÿ๐‘— (although they are not really weights). We canโ€”inductivelyโ€”verify that the selfsimilarity property (13.8) implies that (13.10)

โ„ฐ๐‘š (๐‘ข, ๐‘ฃ) = โˆ‘ ๐‘คโˆˆ๐‘Š๐‘š

1 โ„ฐ (๐‘ข โˆ˜ ๐‘“๐‘ค , ๐‘ฃ โˆ˜ ๐‘“๐‘ค ), ๐‘Ÿ๐‘ค 0

for any ๐‘ข, ๐‘ฃ โˆˆ ๐‘™(๐‘‰๐‘š ), where ๐‘Ÿ๐‘ค = ๐‘Ÿ๐‘ค1 ๐‘Ÿ๐‘ค2 โ‹ฏ ๐‘Ÿ๐‘ค๐‘š . Hence, the harmonic structure is determined by the initial energy โ„ฐ0 on ๐‘‰0 and the numbers ๐‘Ÿ1 , ๐‘Ÿ2 , . . . , ๐‘Ÿ๐‘ . The compatibility property (13.9) means that the energy is preserved between levels. Inductively, using and (13.10) we can see that, given ๐‘ฃ โˆˆ ๐‘™(๐‘‰0 ), (13.11)

โ„ฐ0 (๐‘ฃ) = min{โ„ฐ๐‘š (๐‘ข) โˆถ ๐‘ข โˆˆ ๐‘™(๐‘‰๐‘š ), ๐‘ข|๐‘‰0 = ๐‘ฃ}

for every ๐‘š โ‰ฅ 1. We can then prove by induction that, given โ„ฐ0 and ๐‘Ÿ1 , ๐‘Ÿ2 , . . . , ๐‘Ÿ๐‘ , if โ„ฐ๐‘š is given by (13.10) and (13.11) is satisfied with ๐‘š = 1, then (13.11) is satisfied for every ๐‘š โ‰ฅ 1 (Exercise (4)). If ๐‘ข โˆˆ ๐‘™(๐‘‰๐‘š ) satifies (13.11), we say that ๐‘ข is harmonic on ๐‘‰๐‘š . A harmonic function is a function ๐‘ข on ๐‘‰โˆ— such that each ๐‘ข|๐‘‰๐‘š is harmonic on ๐‘‰๐‘š , for every ๐‘š.

248

13. Harmonic functions on post-critically finite sets

Example 13.12. Consider the interval [0, 1] with contractions as above. The vertices ๐‘‰๐‘š of level ๐‘š correspond to the dyadic partition 1 2 2๐‘š โˆ’ 1 , , . . . , , 1}. 2๐‘š 2๐‘š 2๐‘š The sequence of energies defined in (11.3), ๐‘‰๐‘š = {0,

2๐‘š

โ„ฐ๐‘š (๐‘ข) = 2

๐‘š

โˆ‘ (๐‘ข( ๐‘˜=1

๐‘˜โˆ’1 ๐‘˜ 2 ) โˆ’ ๐‘ข( ๐‘š )) , 2๐‘š 2

is a harmonic structure on ๐ผ, as discussed in Section 11.1. Harmonic functions are restrictions to ๐‘‰โˆ— of linear functions. Example 13.13. Similarly, the sequence of energies on the vertices of the Sierpiล„ski gasket ๐‘† discussed in Section 11.2, 2 5 ๐‘š โˆ‘ (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)) , โ„ฐ๐‘š (๐‘ข) = ( ) 3 ๐‘ฅ,๐‘ฆโˆˆ๐‘‰ ๐‘ฅโˆผ๐‘ฆ

๐‘š

is a harmonic structure on ๐‘†. Harmonic functions can be constructed by interpolation, as discussed in that section. Example 13.14. Consider now the Hata tree set ๐พ described in Example 13.6. We have that ๐‘‰0 = {๐‘, 0, 1}. For โ„Ž > 0, we consider the Dirichlet form on ๐‘™(๐‘‰0 ) given by (13.15)

โ„ฐ0 (๐‘ข) = (๐‘ข(0) โˆ’ ๐‘ข(1))2 + โ„Ž(๐‘ข(0) โˆ’ ๐‘ข(๐‘))2 .

โ„ฐ0 is a Dirichlet form, as discussed in 13.7. For ๐‘Ÿ1 , ๐‘Ÿ2 > 0, we have, for ๐‘ข โˆˆ ๐‘™(๐‘‰1 ), โ„ฐ1 (๐‘ข) =

1 ((๐‘ข(0) โˆ’ ๐‘ข(๐‘))2 + โ„Ž(๐‘ข(0) โˆ’ ๐‘ข(1/3))2 ) ๐‘Ÿ1 1 + ((๐‘ข(1/3) โˆ’ ๐‘ข(1))2 + โ„Ž(๐‘ข(1/3) โˆ’ ๐‘ข(๐‘‘))2 ), ๐‘Ÿ2

where ๐‘‘ = ๐‘“2 (๐‘) (see Figure 13.1). Given ๐‘ข(๐‘) = ๐›ผ, ๐‘ข(0) = ๐›ฝ and ๐‘ข(1) = ๐›พ, we want to find ๐‘ข(1/3) = ๐‘ฅฬ„ and ๐‘ข(๐‘‘) = ๐‘ฆ ฬ„ such that the quadratic function 1 1 ๐‘“(๐‘ฅ, ๐‘ฆ) = ((๐›ผ โˆ’ ๐›ฝ)2 + โ„Ž(๐›ฝ โˆ’ ๐‘ฅ)2 ) + ((๐‘ฅ โˆ’ ๐›พ)2 + โ„Ž(๐‘ฅ โˆ’ ๐‘ฆ)2 ) ๐‘Ÿ1 ๐‘Ÿ2 ฬ„ and ๐‘Ÿ1 , ๐‘Ÿ2 such that 0 < ๐‘Ÿ1 , ๐‘Ÿ2 < 1 and takes its minimum at (๐‘ฅ,ฬ„ ๐‘ฆ), ๐‘“(๐‘ฅ,ฬ„ ๐‘ฆ)ฬ„ = โ„Ž(๐›ผ โˆ’ ๐›ฝ)2 + (๐›ฝ โˆ’ ๐›พ)2 .

13.2. Harmonic structures and discrete energy

249

c

1/ 3

0

1 d

Figure 13.1. The set ๐‘‰ 1 of vertices of level 1 of the Hata tree set.

By elementary calculus ๐‘“ takes its minimum at ๐‘ฅฬ„ = ๐‘ฆ ฬ„ =

(13.16)

โ„Ž๐‘Ÿ2 ๐›ฝ + ๐‘Ÿ1 ๐›พ โ„Ž๐‘Ÿ2 + ๐‘Ÿ1

and ๐‘“(๐‘ฅ,ฬ„ ๐‘ฆ)ฬ„ =

1 โ„Ž (๐›ผ โˆ’ ๐›ฝ)2 + (๐›ฝ โˆ’ ๐›พ)2 . ๐‘Ÿ1 โ„Ž๐‘Ÿ2 + ๐‘Ÿ1

Thus 1 =โ„Ž ๐‘Ÿ1

and

โ„Ž = 1, โ„Ž๐‘Ÿ2 + ๐‘Ÿ1

so 1 1 and ๐‘Ÿ2 = 1 โˆ’ 2 = 1 โˆ’ ๐‘Ÿ12 . โ„Ž โ„Ž Note that 0 < ๐‘Ÿ1 , ๐‘Ÿ2 < 1 if and only if โ„Ž > 1. Thus, we have a family of harmonic structures on the Hata set, one for each number โ„Ž > 1. (13.17)

๐‘Ÿ1 =

By (13.16), a harmonic function ๐‘ข on ๐‘‰1 satisfies 1 1 )๐‘ข(0) + 2 ๐‘ข(1). 2 โ„Ž โ„Ž Note that (13.18) is a convex combination of ๐‘ข(0) and ๐‘ข(1), and that it does not depend on the value of ๐‘ข(๐‘). By the observations above, (13.18) provides an algorithm to construct a harmonic function on ๐‘‰โˆ— . Figure 13.2 shows harmonic functions with โ„Ž = 2 and boundary values ๐‘ข = ๐œ’{1} and ๐‘ข = ๐œ’{๐‘} . (13.18)

๐‘ข(1/3) = ๐‘ข(๐‘‘) = (1 โˆ’

250

13. Harmonic functions on post-critically finite sets

Figure 13.2. Harmonic functions on the Hata tree set with โ„Ž = 2 and boundary values ๐‘ข(๐‘) = ๐‘ข(0) = 0 and ๐‘ข(1) = 1 on the left, and ๐‘ข(๐‘) = 1 and ๐‘ข(0) = ๐‘ข(1) = 0 on the right.

13.3. Discrete Laplacians Let ๐‘‰ be a finte set and ๐ป a symmetric operator on ๐‘™(๐‘‰). ๐ป is a Laplacian on ๐‘‰ if (a) ๐ป is nonpositive definite; (b) ๐ป๐‘ข = 0 if and only if ๐‘ข is a constant; and (c) ๐ป๐‘ฅ๐‘ฆ โ‰ฅ 0 for all ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰, ๐‘ฅ โ‰  ๐‘ฆ. Let โ„ฐ be the bilinear form induced by ๐ป, so โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ’โŸจ๐ป๐‘ข, ๐‘ฃโŸฉ. It is clear that โ„ฐ(๐‘ข) โ‰ฅ 0 for all ๐‘ข if and only if ๐ป is nonpositive definite, by definition. By the spectral theorem, โ„ฐ(๐‘ข) = 0 if and only if ๐‘ข is a constant is equivalent to saying that ๐ป๐‘ข = 0 if and only if ๐‘ข is a constant (Exercise (6)). We have Proposition 13.19. Proposition 13.19. The bilinear form โ„ฐ induced by ๐ป is a Dirichlet form if and only if ๐ป is a Laplacian on ๐‘‰.

13.3. Discrete Laplacians

251

Proof. We have seen that (1) and (2) of the definition of a Dirichlet form are equivalent to (a) and (b) of the definition of a Laplacian. Now, โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ’โŸจ๐ป๐‘ข, ๐‘ฃโŸฉ = โˆ’ โˆ‘ ๐ป๐‘ข(๐‘ฅ)๐‘ฃ(๐‘ฅ) ๐‘ฅโˆˆ๐‘‰

= โˆ’ โˆ‘ ๐ป๐‘ฅ๐‘ฆ ๐‘ข(๐‘ฆ)๐‘ฃ(๐‘ฅ) ๐‘ฅ,๐‘ฆโˆˆ๐‘‰

=

1 โˆ‘ ๐ป (๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ))(๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)), 2 ๐‘ฅ,๐‘ฆโˆˆ๐‘‰ ๐‘ฅ๐‘ฆ

using the fact that โˆ‘ ๐ป๐‘ฅ๐‘ฆ = 0 ๐‘ฆโˆˆ๐‘‰

for each ๐‘ฅ, because ๐ป is zero on a constant. Hence, if ๐ป๐‘ฅ๐‘ฆ โ‰ฅ 0 for all pairs ๐‘ฅ โ‰  ๐‘ฆ, โ„ฐ is a Dirichlet form by 13.7. Suppose some ๐ป๐‘ฅ๐‘ฆ < 0. By multiplying by a positive constant, we can assume ๐ป๐‘ฅ๐‘ฆ = โˆ’1. Let ๐‘ก < 0 and define ๐‘ข on ๐‘‰ by setting ๐‘ข(๐‘ฅ) = 1, ๐‘ข(๐‘ฆ) = ๐‘ก and ๐‘ข(๐‘ง) = 0 for all ๐‘ง โˆˆ ๐‘‰, ๐‘ง โ‰  ๐‘ฅ, ๐‘ฆ. Then โ„ฐ(๐‘ข) =

1 โˆ‘ ๐ป (๐‘ข(๐‘ง) โˆ’ ๐‘ข(๐‘ค))2 = โˆ’(1 โˆ’ ๐‘ก)2 + ๐ด๐‘ก2 + ๐ต, 2 ๐‘ง,๐‘คโˆˆ๐‘‰ ๐‘ง๐‘ค

for some ๐ด, ๐ต โ‰ฅ 0. Since ๐‘ก < 0, the cut ๐‘ขฬ„ of ๐‘ข by [0, 1] is ๐‘ข(๐‘ฅ) = 1 and ๐‘ข(๐‘ง) = 0 for all ๐‘ง โ‰  ๐‘ฅ, so โ„ฐ(๐‘ข)ฬ„ = โˆ’1 + ๐ต. Thus โ„ฐ(๐‘ข) โˆ’ โ„ฐ(๐‘ข)ฬ„ = 2๐‘ก + (๐ด โˆ’ 1)๐‘ก2 < 0 if ๐ด โ‰ค 1 or 0>๐‘ก>โˆ’

2 , ๐ดโˆ’1

so โ„ฐ doesnโ€™t satisfy (3) of the definition of Dirichlet form.

โ–ก

Hence, if โ„ฐ๐‘š is a harmonic structure on the PCF set ๐พ, then we have a sequence of Laplacians ๐ป๐‘š on ๐‘‰๐‘š , where each ๐ป๐‘š induces โ„ฐ๐‘š . The self-similarity condition (13.8) gives a relation between each ๐ป๐‘šโˆ’1 and ๐ป๐‘š , which then defines an extension of ๐ป๐‘šโˆ’1 on ๐‘‰๐‘šโˆ’1 to ๐‘‰๐‘š . We leave this as an exercise (Exercise (9)). To write the compatiblity condition (13.9) in terms of Laplacians, we make the following observations. Let ๐‘‰ a finite set and ๐‘ˆ โŠ‚ ๐‘‰ a proper

252

13. Harmonic functions on post-critically finite sets

subset. If ๐ป is a Laplacian on ๐‘‰, we write the matrix of ๐ป, which we also denote by ๐ป, in blocks ๐‘‡ ๐ป=( ๐ฝ

(13.20)

๐ฝ๐‘ก ), ๐‘‹

where ๐‘‡ โˆถ ๐‘™(๐‘ˆ) โ†’ ๐‘™(๐‘ˆ), ๐ฝ โˆถ ๐‘™(๐‘ˆ) โ†’ ๐‘™(๐‘‰ โงต๐‘ˆ) and ๐‘‹ โˆถ ๐‘™(๐‘‰ โงต๐‘ˆ) โ†’ ๐‘™(๐‘‰ โงต๐‘ˆ). Hence, if we write ๐‘ข โˆˆ ๐‘™(๐‘‰) as ๐‘ข ๐‘ข = ( 0) , ๐‘ข1 where ๐‘ข0 = ๐‘ข|๐‘ˆ and ๐‘ข1 = ๐‘ข|๐‘‰ โงต๐‘ˆ , then (13.21)

๐ป๐‘ข = (

๐‘‡๐‘ข0 + ๐ฝ ๐‘ก ๐‘ข1 ). ๐ฝ๐‘ข0 + ๐‘‹๐‘ข1

๐‘‹ is invertible. Indeed, for ๐‘ฃ โˆˆ ๐‘™(๐‘‰ โงต ๐‘ˆ), consider its extension ๐‘ข โˆˆ ๐‘™(๐‘‰) with ๐‘ข|๐‘ˆ = 0 and ๐‘ข|๐‘‰ โงต๐‘ˆ = ๐‘ฃ. Then, if โ„ฐ is the Dirichlet form induced by ๐ป, by (13.21) we have โ„ฐ(๐‘ข) = โˆ’โŸจ๐ป๐‘ข, ๐‘ขโŸฉ = โˆ’โŸจ (

๐ฝ๐‘ก๐‘ฃ 0 ) , ( ) โŸฉ = โˆ’โŸจ๐‘‹๐‘ฃ, ๐‘ฃโŸฉ. ๐‘‹๐‘ฃ ๐‘ฃ

Thus ๐‘‹๐‘ฃ = 0 implies โ„ฐ(๐‘ข) = 0, so ๐‘ข must be constant and therefore ๐‘ฃ = 0. Note that this also implies that ๐‘‹ is negative definite. We can thus write โ„ฐ(๐‘ข) = โˆ’โŸจ๐‘‡๐‘ข0 + ๐ฝ ๐‘ก ๐‘ข1 , ๐‘ข0 โŸฉ โˆ’ โŸจ๐ฝ๐‘ข0 + ๐‘‹๐‘ข1 , ๐‘ข1 โŸฉ = โŸจ(๐‘‡ โˆ’ ๐ฝ ๐‘ก ๐‘‹ โˆ’1 ๐ฝ)๐‘ข0 , ๐‘ข0 โŸฉ โˆ’ โŸจ๐‘‹(๐‘ข1 + ๐‘‹ โˆ’1 ๐ฝ๐‘ข0 ), ๐‘ข1 + ๐‘‹ โˆ’1 ๐ฝ๐‘ข0 โŸฉ Given ๐‘ข0 on ๐‘ˆ, we see that โ„ฐ(๐‘ข) is minimal when โŸจ๐‘‹(๐‘ข1 + ๐‘‹ โˆ’1 ๐ฝ๐‘ข0 ), ๐‘ข1 + ๐‘‹ โˆ’1 ๐ฝ๐‘ข0 โŸฉ = 0, because ๐‘‹ is negative definite. Thus we require (13.22)

๐‘ข1 + ๐‘‹ โˆ’1 ๐ฝ๐‘ข0 = 0

and โ„ฐ(๐‘ข) = โ„ฐโ€ฒ (๐‘ข0 ), where โ„ฐโ€ฒ is the quadratic form induced by ๐‘‡ โˆ’ ๐ฝ ๐‘ก ๐‘‹ โˆ’1 ๐ฝ. Equations (13.21) and (13.22) imply that a minimizer function ๐‘ข satifies (13.23)

๐ป๐‘ข|๐‘‰ โงต๐‘ˆ = 0.

13.3. Discrete Laplacians

253

13.24. If ๐‘ข โˆˆ ๐‘™(๐‘‰) satisfies (13.23), then, for all ๐‘ฅ โˆˆ ๐‘‰, min ๐‘ข(๐‘ฆ) โ‰ค ๐‘ข(๐‘ฅ) โ‰ค max ๐‘ข(๐‘ฆ). ๐‘ฆโˆˆ๐‘ˆ

๐‘ฆโˆˆ๐‘ˆ

If we have any of the above equalities for some ๐‘ฅ โˆˆ ๐‘‰ โงต ๐‘ˆ, then ๐‘ข is constant. Thus, a minimizer function satisfies the maximum principle. Indeed, for any ๐‘ฅ โˆˆ ๐‘‰, ๐ป๐‘ข(๐‘ฅ) = โˆ‘ ๐ป๐‘ฅ๐‘ฆ ๐‘ข(๐‘ฆ) = โˆ‘ ๐ป๐‘ฅ๐‘ฆ (๐‘ข(๐‘ฆ) โˆ’ ๐‘ข(๐‘ฅ)), ๐‘ฆโˆˆ๐‘‰

๐‘ฆโˆˆ๐‘‰

because โˆ‘๐‘ฆโˆˆ๐‘‰ ๐ป๐‘ฅ๐‘ฆ = 0 for all ๐‘ฅ. Thus, if ๐‘ฅ โˆˆ ๐‘‰ โงต ๐‘ˆ, ๐ป๐‘ข(๐‘ฅ) = 0 and thus (13.25)

โˆ‘ ๐ป๐‘ฅ๐‘ฆ (๐‘ข(๐‘ฆ) โˆ’ ๐‘ข(๐‘ฅ)) = 0. ๐‘ฆโˆˆ๐‘‰

If ๐‘ข takes its maximum at ๐‘ฅ โˆˆ ๐‘‰ โงต ๐‘ˆ, (13.25) implies that ๐‘ข(๐‘ฆ) = ๐‘ข(๐‘ฅ) whenever ๐ป๐‘ฅ๐‘ฆ > 0, because each ๐ป๐‘ฅ๐‘ฆ โ‰ฅ 0 and ๐‘ข(๐‘ฆ) โˆ’ ๐‘ข(๐‘ฅ) โ‰ค 0, so ๐‘ข = ๐‘ข(๐‘ฅ) on the set of points ๐‘ฆ such that ๐ป๐‘ฅ๐‘ฆ > 0. By 13.7, for any ๐‘ฆ โˆˆ ๐‘‰ there exists a sequence ๐‘ฆ0 = ๐‘ฅ, ๐‘ฆ1 , ๐‘ฆ2 , . . ., ๐‘ฆ๐‘› = ๐‘ฆ such that ๐ป๐‘ฆ๐‘–โˆ’1 ๐‘ฆ๐‘– > 0, so recursively the above argument shows that ๐‘ข(๐‘ฆ) = ๐‘ข(๐‘ฅ), and ๐‘ข is a constant. If we apply the previous analysis with ๐‘‰ = ๐‘‰๐‘š and ๐‘ˆ = ๐‘‰๐‘šโˆ’1 , we see that the compatibility condition (13.9) is equivalent to (13.26)

๐ป๐‘šโˆ’1 = ๐‘‡ โˆ’ ๐ฝ ๐‘ก ๐‘‹ โˆ’1 ๐ฝ,

where ๐‘‡, ๐ฝ and ๐‘‹ are the blocks of ๐ป๐‘š as in (13.20). Note that the minimizer function ๐‘ข โˆˆ ๐‘™(๐‘‰๐‘š ), for a given ๐‘ฃ โˆˆ ๐‘™(๐‘‰๐‘šโˆ’1 ), is given by extending (13.27)

๐‘ข|๐‘‰๐‘š โงต๐‘‰๐‘šโˆ’1 = โˆ’๐‘‹ โˆ’1 ๐ฝ๐‘ฃ.

By (13.21), ๐‘ข is a minimizer for a given ๐‘ฃ โˆˆ ๐‘™(๐‘‰๐‘šโˆ’1 ) if (13.28)

๐ป๐‘š ๐‘ข|๐‘‰๐‘š โงต๐‘‰๐‘šโˆ’1 = 0.

By (13.11) and (13.28), ๐‘ข is harmonic if (13.29)

๐ป๐‘š (๐‘ข|๐‘‰๐‘š )(๐‘ฅ) = 0

for all ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 . Since (13.29) is a linear equation, the set of harmonic functions is a vector space. If #๐‘‰0 = ๐‘€, this vector space is ๐‘€-dimensional. Moreover, (13.27) provides an algorithm to construct a harmonic function by interpolation to each level ๐‘š from level ๐‘š โˆ’ 1.

254

13. Harmonic functions on post-critically finite sets

13.30. As a consequence of (13.28) and 13.24, we have that harmonic functions satisfy the maximum principle: if ๐‘ข is harmonic, then min ๐‘ข โ‰ค ๐‘ข(๐‘ฅ) โ‰ค max ๐‘ข ๐‘‰0

๐‘‰0

for all ๐‘ฅ โˆˆ ๐‘‰โˆ— , and we have any of the equalities for some ๐‘ฅ โˆˆ ๐‘‰โˆ— โงต ๐‘‰0 only if ๐‘ข is a constant. 13.31. If ๐‘ข is harmonic, then it is uniformly continuous in ๐‘‰โˆ— , so it has a unique continuous extension to ๐พ. To see this, choose first ๐‘š large enough so that, if ๐‘ฅ โˆผ๐‘š ๐‘ฆ, at most one of them belong to ๐‘‰0 . If ๐‘ข is harmonic and not a constant, then (13.32) max{|๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)| โˆถ ๐‘ฅ โˆผ๐‘š ๐‘ฆ} < max{|๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 }, where ๐‘ฃ = ๐‘ข|๐‘‰0 is the boundary value of ๐‘ข. By iterating (13.27), for each ๐‘ค โˆˆ ๐‘Š๐‘š there exists an operator ๐‘ƒ๐‘ค โˆถ ๐‘™(๐‘‰0 ) โ†’ ๐‘™(๐‘“๐‘ค (๐‘‰0 )) such that ๐‘ข|๐‘“๐‘ค (๐‘‰0 ) = ๐‘ƒ๐‘ค ๐‘ฃ. Note that, if ๐‘ฃ(๐‘ฅ) ฬƒ = ๐‘ฃ(๐‘ฅ) โˆ’

1 โˆ‘ ๐‘ฃ(๐‘ฆ), ๐‘€ ๐‘ฆโˆˆ๐‘‰ 0

then max{|๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 } = max{|๐‘ฃ(๐‘ฅ) ฬƒ โˆ’ ๐‘ฃ(๐‘ฆ)| ฬƒ โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 }, and โˆ‘๐‘ฅโˆˆ๐‘‰ ๐‘ฃ(๐‘ฅ) ฬƒ = 0. By Exercise (12) and (13.32), there exists ๐›พ๐‘ค < 1 0 such that max{|๐‘ƒ๐‘ค ๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ƒ๐‘ค ๐‘ฃ(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 } โ‰ค ๐›พ๐‘ค max{|๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 }, and thus max{|๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)| โˆถ ๐‘ฅ โˆผ๐‘š ๐‘ฆ} โ‰ค ๐›พ max{|๐‘ฃ(๐‘ฅ) โˆ’ ๐‘ฃ(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰0 }, where ๐›พ = max{๐›พ๐‘ค โˆถ ๐‘ค โˆˆ ๐‘Š๐‘š } < 1. Thus, for any ๐‘˜ โ‰ฅ 1, if ๐‘ฅ โˆผ๐‘˜๐‘š ๐‘ฆ then |๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)| โ‰ค ๐ด๐›พ๐‘˜ , which implies that ๐‘ข is uniformly continuous because ๐ด๐›พ๐‘˜ โ†’ 0. We say that ๐‘ข โˆˆ ๐ถ(๐พ) is harmonic if ๐‘ข|๐‘‰โˆ— is a harmonic function.

13.4. The Laplacian on a PCF set

255

13.4. The Laplacian on a PCF set A Laplacian on a PCF set can be constructed in the same way that we constructed a Laplacian on the Sierpiล„ski gasket in Section 11.3. For a given ๐‘ข โˆˆ ๐ถ(๐พ), the sequence โ„ฐ๐‘š (๐‘ข), where we are denoting the restriction of ๐‘ข to ๐‘‰๐‘š simply by ๐‘ข, is increasing, so we define โ„ฑ = {๐‘ข โˆˆ ๐ถ(๐พ) โˆถ โ„ฐ๐‘š (๐‘ข) is bounded} and define the bilinear form โ„ฐ on โ„ฑ by โ„ฐ(๐‘ข, ๐‘ฃ) = lim โ„ฐ๐‘š (๐‘ข, ๐‘ฃ). We also call โ„ฐ(๐‘ข) the energy of ๐‘ข on ๐พ. As in the case of the Sierpiล„ski gasket, โ„ฑ contains constant and harmonic functions, as well as ๐‘š-harmonic functions, where a function ๐‘ข โˆˆ ๐ถ(๐พ) is called ๐‘š-harmonic if ๐‘ข โˆ˜ ๐‘“๐‘ค is harmonic for each ๐‘ค โˆˆ ๐‘Š๐‘š . Let ๐‘‘ > 0 be the unique number such that ๐‘‘ ๐‘Ÿ1๐‘‘ + ๐‘Ÿ2๐‘‘ + . . . + ๐‘Ÿ๐‘ = 1.

(13.33)

For each cell ๐พ๐‘ค , we define ๐œ‡(๐พ๐‘ค ) = ๐‘Ÿ๐‘ค๐‘‘ . By (13.33) we have that โˆ‘ ๐œ‡(๐พ๐‘ค ) = 1 ๐‘คโˆˆ๐‘Š๐‘š

and, as in the case of the Sierpiล„ski gasket, if we define for each ๐ด โŠ‚ ๐พ ๐œ‡(๐ด) = inf { โˆ‘ ๐œ‡(๐‘‡๐‘— ) โˆถ ๐‘‡๐‘— are cells and ๐ด โŠ‚ ๐‘—

โ‹ƒ

๐‘‡๐‘— },

๐‘—

the ๐œ‡ satisfies the analogous to Propositions 11.17 and 11.18 in Chapter 11. We thus obtain a self-similar measure on ๐พ, and hence an integral on ๐พ with respect to the measure ๐œ‡. As above, we say that ๐‘ข โˆˆ dom ฮ” if ๐‘ข โˆˆ โ„ฑ and there exists ๐‘“ โˆˆ ๐ถ(๐พ) such that, for all ๐‘ฃ โˆˆ โ„ฑ with ๐‘ฃ|๐‘‰0 = 0, (13.34)

โ„ฐ(๐‘ข, ๐‘ฃ) = โˆ’ โˆซ ๐‘“๐‘ฃ๐‘‘๐œ‡. ๐พ

We write ๐‘“ = ฮ”๐‘ข, and we call it the Laplacian of ๐‘ข. If ๐‘ข is harmonic, for each ๐‘š โ‰ฅ 1 we have ๐ป๐‘š ๐‘ข(๐‘ฅ) = 0 for all ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 , and thus โ„ฐ(๐‘ข, ๐‘ฃ) = โ„ฐ0 (๐‘ข, ๐‘ฃ)

256

13. Harmonic functions on post-critically finite sets

for any ๐‘ฃ โˆˆ โ„ฑ. Hence, if further ๐‘ฃ|๐‘‰0 = 0, we have โ„ฐ(๐‘ข, ๐‘ฃ) = 0 and therefore ฮ”๐‘ข(๐‘ฅ) = 0 for all ๐‘ฅ โˆˆ ๐พ โงต ๐‘‰0 . For ๐‘ฅ0 โˆˆ ๐‘‰โˆ— โงต ๐‘‰0 , let ๐‘š large enough so that ๐‘ฅ0 โˆˆ ๐‘‰๐‘š and let ๐œ“๐‘ฅ0 ,๐‘š be the ๐‘š-harmonic function such that ๐œ“๐‘ฅ0 ,๐‘š |๐‘‰๐‘š = ๐œ’{๐‘ฅ0 } . As in the case of the Sierpiล„ski gasket, we call ๐œ“๐‘ฅ0 ,๐‘š the ๐‘š-harmonic spline on ๐‘ฅ0 . Write ๐œ‡๐‘ฅ0 ,๐‘š = โˆซ ๐œ“๐‘ฅ0 ,๐‘š ๐‘‘๐œ‡. ๐พ

Thus, if ๐‘ข โˆˆ dom ฮ”, for large ๐‘š we have โˆซ ฮ”๐‘ข ๐œ“๐‘ฅ0 ,๐‘š ๐‘‘๐œ‡ โ‰ˆ ฮ”๐‘ข(๐‘ฅ0 )๐œ‡๐‘ฅ0 ,๐‘š , ๐พ

because ฮ”๐‘ข is continuous. Also โ„ฐ(๐‘ข, ๐œ“๐‘ฅ0 ,๐‘š ) = โ„ฐ๐‘š (๐‘ข, ๐œ“๐‘ฅ0 ,๐‘š ) = โˆ’ โˆ‘ ๐ป๐‘š ๐‘ข(๐‘ฅ) ๐œ“๐‘ฅ0 ,๐‘š (๐‘ฅ) = โˆ’๐ป๐‘š ๐‘ข(๐‘ฅ0 ), ๐‘ฅโˆˆ๐‘‰๐‘š

so we have that ฮ”๐‘ข(๐‘ฅ0 ) โ‰ˆ

1 ๐ป (๐‘ฅ ). ๐œ‡๐‘ฅ0 ,๐‘š ๐‘š 0

We can make this argument precise to prove Theorem 13.35, the generalization of Theorem 11.29 to PCF sets. Theorem 13.35. Let ๐‘ข โˆˆ โ„ฑ. ๐‘ข โˆˆ dom ฮ” if and only if there exists ๐‘“ โˆˆ ๐ถ(๐พ) such that 1 lim max {|| ๐ป๐‘š (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)|| โˆถ ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰0 } = 0. ๐œ‡

๐‘šโ†’โˆž

๐‘ฅ,๐‘š

In such case, ๐‘“ = ฮ”๐‘ข. We leave the details as an exercise (Exercise (13)).

Exercises

257

Exercises (1) ๐‘‰โˆ— is dense in ๐พ. (2) Let ๐‘† be the Sierpiล„ski gasket, ๐‘š โ‰ฅ 1, and ๐‘ฅ โˆˆ ๐‘‰๐‘š โงต ๐‘‰๐‘šโˆ’1 . Then there exists a unique ๐‘ค โˆˆ ๐‘Š๐‘šโˆ’1 and ๐‘–, ๐‘— = 1, 2, 3 such that ๐‘ฅ = ๐‘“๐‘ค๐‘– (๐‘๐‘— ) = ๐‘“๐‘ค๐‘— (๐‘ ๐‘– ). (3) Let ๐‘† be the Sierpiล„ski gasket, ๐‘š โ‰ฅ 1, 0 โ‰ค ๐‘› < ๐‘š, and ๐‘ฅ โˆˆ ๐‘‰๐‘š such that ๐‘ฅ โˆˆ ๐‘‰๐‘›+1 โงต ๐‘‰๐‘› . Then there exists a unique ๐‘ค โˆˆ ๐‘Š๐‘› and ๐‘–, ๐‘— = 1, 2, 3 such that ๐‘ฅ = ๐‘“๐‘ค๐‘–๐‘—. . .๐‘— (๐‘๐‘— ) = ๐‘“๐‘ค๐‘—๐‘–. . .๐‘– (๐‘ ๐‘– ), where the words ๐‘ค๐‘–๐‘— . . . ๐‘— and ๐‘ค๐‘—๐‘– . . . ๐‘– of length ๐‘š have ๐‘š โˆ’ 1 โˆ’ ๐‘› repeated ๐‘—โ€™s and ๐‘–โ€™s, respectively. (4) Let โ„ฐ0 be a Dirichlet form on ๐‘™(๐‘‰0 ) and 0 < ๐‘Ÿ1 , ๐‘Ÿ2 , . . . , ๐‘Ÿ๐‘ < 1. If โ„ฐ๐‘š is given by (13.10) and (13.11) is satisfied with ๐‘š = 1, then (13.11) is satisfied for every ๐‘š โ‰ฅ 1. (5) Calculate all the harmonic structures of the interval, self-similar with the contractions ๐‘“1 (๐‘ฅ) = ๐‘ฅ/2 and ๐‘“2 (๐‘ฅ) = ๐‘ฅ/2 + 1/2 on the real line. (6) Let ๐‘‰ be a finite set, ๐ป a symmetric nonpositive definite operator on ๐‘™(๐‘‰), and โ„ฐ(๐‘ข, ๐‘ฃ) โˆ’ โŸจ๐ป๐‘ข, ๐‘ฃโŸฉ the bilinear form induced by ๐ป. The following are equivalent. (a) โ„ฐ(๐‘ข) = 0 if and only if ๐‘ข is a constant. (b) ๐ป๐‘ข = 0 if and only if ๐‘ข is a constant. (Hint: Use the spectral theorem.) (7) Calculate the initial Laplacian ๐ป0 of the harmonic structure of the Sierpiล„ski gasket. (8) Calculate the Laplacian ๐ป0 of the harmonic structure of the Hata tree set of Example 13.14. (9) For each ๐‘— = 1, 2, . . . , ๐‘, let ๐‘…๐‘— โˆถ ๐‘™(๐‘‰๐‘š ) โ†’ ๐‘™(๐‘‰๐‘šโˆ’1 ) be given by ๐‘…๐‘— ๐‘ข = ๐‘ข โˆ˜ ๐‘“๐‘— . Then the self-similarity condition (13.8) is equivalent to ๐‘

1 ๐‘ก ๐‘…๐‘— ๐ป๐‘šโˆ’1 ๐‘…๐‘— . ๐‘Ÿ ๐‘—=1 ๐‘—

๐ป๐‘š = โˆ‘

258

13. Harmonic functions on post-critically finite sets

(10) For each ๐‘ค โˆˆ ๐‘Š๐‘š , let ๐‘…๐‘ค โˆถ ๐‘™(๐‘‰๐‘š ) โ†’ ๐‘™(๐‘‰0 ) be given by ๐‘…๐‘ค ๐‘ข = ๐‘ข โˆ˜ ๐‘“๐‘ค . Then (13.10) is equivalent to 1 ๐‘ก ๐ป๐‘š = โˆ‘ ๐‘… ๐‘ค ๐ป0 ๐‘… ๐‘ค . ๐‘Ÿ ๐‘ค ๐‘คโˆˆ๐‘Š ๐‘š

(11) Calculate the Laplacian ๐ป1 of the harmonic structure of the Hata tree set and verify (13.26) explicitly. (12) (a) Let ๐‘‰ a finite set and โ„’ be the set of equivalence classes of ๐‘™(๐‘‰) modulo constants, that is, under the equivalence relation ๐‘ข โˆผ ๐‘ฃ if and only if ๐‘ข โˆ’ ๐‘ฃ is a constant. Then โ„’ is finite dimensional, โ€–๐‘ขโ€– = max{|๐‘ข(๐‘ฅ) โˆ’ ๐‘ข(๐‘ฆ)| โˆถ ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‰} and is a norm on โ„’. (b) The set {๐‘ข โˆˆ ๐‘™(๐‘‰) โˆถ โˆ‘๐‘ฅโˆˆ๐‘‰ ๐‘ข(๐‘ฅ) = 0, โ€–๐‘ขโ€– โ‰ค 1} is compact. (13) Prove Theorem 13.35.

Notes The development of the Laplacian on post-critically finite sets was done by Kigami in [Kig93], where he proves that harmonic functions satisfy the linear equations involving discrete Laplacians, and hence form a finite dimensional space. The number ๐‘‘ defined by equation (13.33) is equal to the Hausdorff dimension of ๐พ with respect to the effective resistence metric, which was developed in [Kig94]. If ๐ต๐œ€ (๐‘ฅ) is a ball in ๐พ with respect to this metric, then we can prove that ๐œ‡(๐ต๐œ€ (๐‘ฅ)) โˆผ ๐œ€๐‘‘ for small ๐œ€, where ๐œ‡ is the measure defined in this chapter, and thus ๐œ‡ satisfies the analogous result to the Hardyโ€“Littlewood Theorem 7.4 on a PCF set [Sรกe12]. Further results on the analysis on PCF sets can be found in the text [Kig01].

Appendix A

Some results from real analysis

In this appendix we review some basic results from real analysis. With a few exceptions that may be useful for the discussion in the text, we wonโ€™t include the proofs of the results listed here. All can be found in standard introductory texts as [Gau09] or [BS92] for the case of the real line, [Spi65] or [Fle77] for Euclidean space, and [Fol99] or [MH93] for metric and Banach spaces.

A.1. The real line The set of real numbers โ„ is a complete ordered field. It is unique up to isomorphism. With complete we mean that it satisfies the least upper bound axiom: Axiom (Completeness). If ๐ด โŠ‚ โ„ is nonempty and bounded, then it has a least upper bound. A sequence ๐‘ฅ๐‘› in โ„ converges to ๐‘ฅ, and we write ๐‘ฅ๐‘› โ†’ ๐‘ฅ, if, given any ๐œ€ > 0, there exists ๐‘ such that ๐‘› โ‰ฅ ๐‘ implies |๐‘ฅ๐‘› โˆ’ ๐‘ฅ| < ๐œ€. Any convergent sequence is a Cauchy sequence: given ๐œ€ > 0, there exists ๐‘ such that ๐‘›, ๐‘š โ‰ฅ ๐‘ implies |๐‘ฅ๐‘› โˆ’ ๐‘ฅ๐‘š | < ๐œ€. The converse is also true. Theorem A.1. A sequence in โ„ converges if and only if it is a Cauchy sequence. 259

260

A. Some results from real analysis

In fact, Theorem A.1 is equivalent to the completeness axiom. It is clear that convergent sequences are bounded. The converse is not true, but we have the following fact. Theorem A.2 (Bolzanoโ€“Weierstrass). If ๐‘ฅ๐‘› is bounded, then it has a convergent subsequence. It turns out that the Bolzanoโ€“Weierstrass theorem is also equivalent to the completeness axiom. A function ๐‘“ โˆถ ๐ด โ†’ โ„, where ๐ด โŠ‚ โ„, is continuous at ๐‘ฅ0 โˆˆ ๐ด if given ๐œ€ > 0, there exists ๐›ฟ > 0 such that |๐‘ฅ โˆ’ ๐‘ฅ0 | < ๐›ฟ implies |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ0 )| < ๐œ€, whenever ๐‘ฅ โˆˆ ๐ด. ๐‘“ is continuous at ๐‘ฅ0 if, and only if, for every sequence ๐‘ฅ๐‘› in ๐ด that converges to ๐‘ฅ0 , then we have ๐‘“(๐‘ฅ๐‘› ) โ†’ ๐‘“(๐‘ฅ0 ). Sums and products of continuous functions are continuous. Quotients are also continuous, provided they are well defined. ๐‘“ is continuous on ๐ด if it is continuous at each point of ๐ด. ๐‘“ is uniformly continuous on ๐ด if, for each ๐œ€ > 0, there exist ๐›ฟ > 0 such that |๐‘ฅ โˆ’ ๐‘ฆ| < ๐›ฟ, for any ๐‘ฅ, ๐‘ฆ โˆˆ ๐ด, implies |๐‘“(๐‘ฅ) โˆ’ ๐‘“(๐‘ฆ)| < ๐œ€. If ๐‘“ is uniformly continuous and ๐‘ฅ๐‘› is a Cauchy sequence in ๐ด, then ๐‘“(๐‘ฅ๐‘› ) is also a Cauchy sequence. This has the following consequence. A limit point of ๐ด is a point ๐‘ฅ โˆˆ โ„ such that there exists a sequence ๐‘ฅ๐‘› in ๐ด such that ๐‘ฅ๐‘› โ†’ ๐‘ฅ. ๐‘ฅ may or may not be in ๐ด. Theorem A.3. If ๐‘“ is uniformly continuous on ๐ด and ๐‘ฅ is a limit point of ๐ด, then ๐‘“ can extended to a continuous function on ๐ด โˆช {๐‘ฅ}. Given a sequence ๐‘ฅ๐‘› in ๐ด that converges to ๐‘ฅ, the extension of ๐‘“ to ๐‘ฅ is well-defined by the limit of ๐‘“(๐‘ฅ๐‘› ), which converges because it is a Cauchy sequence. The union ๐ดฬ„ of ๐ด and its limit points is called the closure of ๐ด. Thus, by Theorem A.3, if ๐‘“ is uniformly continuous on ๐ด, then it can be extended continuously to its closure ๐ด.ฬ„ Let ๐‘“๐‘› โˆถ ๐ด โ†’ โ„ be a sequence of functions on ๐ด. We say that ๐‘“๐‘› converges pointwise to ๐‘“ โˆถ ๐ด โ†’ โ„ if, for each ๐‘ฅ โˆˆ ๐ด, ๐‘“๐‘› (๐‘ฅ) โ†’ ๐‘“(๐‘ฅ). We usually just write to ๐‘“๐‘› โ†’ ๐‘“ to denote that ๐‘“๐‘› converges pointwise to ๐‘“. We say that ๐‘“๐‘› converges uniformly to ๐‘“ if, given ๐œ€ > 0, there exists ๐‘ such that ๐‘› โ‰ฅ ๐‘ implies |๐‘“๐‘› (๐‘ฅ) โˆ’ ๐‘“(๐‘ฅ)| < ๐œ€ for all ๐‘ฅ โˆˆ ๐ด. In this case we write ๐‘“๐‘› โ‡‰ ๐‘“. Theorem A.4. If each ๐‘“๐‘› is continuous and ๐‘“๐‘› โ‡‰ ๐‘“, then ๐‘“ is continuous.

A.2. Topology

261

A series โˆ‘ ๐‘Ž๐‘› converges to ๐‘  if the sequence ๐‘ ๐‘› of partial sums, ๐‘ ๐‘› = ๐‘Ž1 + ๐‘Ž2 + . . . + ๐‘Ž๐‘› , converges to ๐‘ . By Theorem A.1, if โˆ‘ |๐‘Ž๐‘› | converges, then โˆ‘ ๐‘Ž๐‘› converges. If โˆ‘ |๐‘Ž๐‘› | converges, we write โˆ‘ |๐‘Ž๐‘› | < โˆž and say that โˆ‘ ๐‘Ž๐‘› converges absolutely. Thus, any absolutely convergent series converges. Not every convergent series is absolutely convergent; we say that such a series converges conditionally. A test for conditional convergence is the following: Theorem A.5 (Dirichletโ€™s test). Let ๐‘Ž๐‘› , ๐‘๐‘› be sequences in โ„ satisfying the following. (1) Each ๐‘Ž๐‘› > 0, ๐‘Ž๐‘›+1 โ‰ค ๐‘Ž๐‘› and ๐‘Ž๐‘› โ†’ 0 (we write ๐‘Ž๐‘› โ†˜ 0). (2) The sequence ๐‘ ๐‘› = ๐‘1 + ๐‘2 + . . . ๐‘๐‘› is bounded. Then the series โˆ‘ ๐‘Ž๐‘› ๐‘๐‘› converges. A series โˆ‘ ๐‘“๐‘› of functions converges pointwise or uniformly if the corresponding sequence of partial sums ๐‘ ๐‘› (๐‘ฅ) = ๐‘“1 (๐‘ฅ) + . . . + ๐‘“๐‘› (๐‘ฅ) converges pointwise or uniformly, respectively. Thus, if the series โˆ‘ ๐‘“๐‘› converges uniformly to ๐‘“ and each ๐‘“๐‘› is continuous, then ๐‘“ is also continuous. We have the following test for uniform convergence of a series. Theorem A.6 (Weierstrass ๐‘€-test). Let ๐‘“๐‘› be a sequence of functions on ๐ด satisfying the following: (1) Each ๐‘“๐‘› is bounded, with |๐‘“๐‘› (๐‘ฅ)| โ‰ค ๐‘€๐‘› for all ๐‘ฅ โˆˆ ๐ด. (2) โˆ‘ ๐‘€๐‘› < โˆž. Then the series โˆ‘ ๐‘“๐‘› converges uniformly.

A.2. Topology A set ๐ด โŠ‚ โ„ is closed if it contains its limit points. Equivalently, ๐ด is closed if ๐ดฬ„ = ๐ด. ๐ด is open if, for each ๐‘ฅ โˆˆ ๐ด, there exists ๐œ€ > 0 such that (๐‘ฅ โˆ’ ๐œ€, ๐‘ฅ + ๐œ€) โŠ‚ ๐ด. A set is open if, and only if, itโ€™s complement is closed. Any union of open sets in โ„ is open, and any intersection of closed sets is closed. Finite intersections of open sets are open, while finite unions of closed sets are closed. A countable intersection of open sets is called a ๐บ ๐›ฟ set, and a countable union of closed set is called an ๐น๐œ set.

262

A. Some results from real analysis

A set ๐ต โŠ‚ ๐ด is said to be open in ๐ด is there exists an open set ๐‘ˆ in โ„ such that ๐ต = ๐ด โˆฉ ๐‘ˆ. Similarly, ๐ต is closed in ๐ด if there exists a closed set ๐ธ in โ„ such that ๐ต = ๐ด โˆฉ ๐ธ. ๐ต is open in ๐ด if, and only if, ๐ด โงต ๐ต is closed in ๐ด. If ๐‘“ โˆถ ๐ด โ†’ โ„ is a function, then ๐‘“ is continuous on ๐ด if, and only if, for any open set ๐‘ˆ in โ„ its preimage ๐‘“โˆ’1 (๐‘ˆ) is open in ๐ด, and equivalently, if for any closed set ๐ธ in โ„ its preimage ๐‘“โˆ’1 (๐ธ) is closed in ๐ด. By the Bolzanoโ€“Weierstrass theorem, if ๐ด is closed and bounded, then any sequence in ๐ด has a convergent subsequence, and its limit is also in ๐ด (we say it converges in ๐ด). In fact, if any sequence in ๐ด has a subsequence that converges in ๐ด, then ๐ด must be closed and bounded. As a consequence, if ๐‘“ is continuous on ๐ด and ๐ด is closed and bounded, then its image ๐ด is also closed and bounded. An open cover for a set ๐ด โŠ‚ โ„ is a collection {๐‘ˆ๐›ผ } of open sets such that ๐ด โŠ‚ โ‹ƒ๐›ผ ๐‘ˆ๐›ผ . We say that the open cover {๐‘ˆ๐›ผ } for ๐ด has a finite subcover if we can choose a finite subcollection ๐‘ˆ๐›ผ1 , ๐‘ˆ๐›ผ2 , . . . , ๐‘ˆ๐›ผ๐‘˜ that also covers ๐ด. We have the following result. Theorem A.7 (Heineโ€“Borel). Let ๐ด โŠ‚ โ„. The following are equivalent. (1) ๐ด is closed and bounded. (2) Every open cover for ๐ด has a finite subcover. A set that satisfies the statements of the Heineโ€“Borel theorem is called compact. We also have the following results for continuous functions on compact sets. Theorem A.8. Let ๐ด โŠ‚ โ„ be compact and ๐‘“ โˆถ ๐ด โ†’ โ„ continuous on ๐ด. Then: (1) ๐‘“ is uniformly continuous. (2) ๐‘“(๐ด) is compact. In particular, if ๐ด is compact and ๐‘“ is continuous on ๐ด, then ๐‘“ takes its maximum and its minimum value in ๐ด.

A.3. Riemann integration Let ๐‘“ โˆถ [๐‘Ž, ๐‘] โ†’ โ„ be bounded and ๐’ซ = {๐‘ฅ0 = ๐‘Ž < ๐‘ฅ1 < . . . < ๐‘ฅ๐‘› = ๐‘} a partition of [๐‘Ž, ๐‘]. The lower and upper sums of ๐‘“ with respect to ๐’ซ are

A.3. Riemann integration

263

given by ๐‘›

๐‘›

๐ฟ(๐‘“, ๐’ซ) = โˆ‘ ๐‘š๐‘– (๐‘ฅ๐‘– โˆ’ ๐‘ฅ๐‘–โˆ’1 ) and

๐‘ˆ(๐‘“, ๐’ซ) = โˆ‘ ๐‘€๐‘– (๐‘ฅ๐‘– โˆ’ ๐‘ฅ๐‘–โˆ’1 ),

๐‘–=1

๐‘–=1

respectively, where ๐‘š๐‘– = inf{๐‘“(๐‘ฅ) โˆถ ๐‘ฅ โˆˆ [๐‘ฅ๐‘–โˆ’1 , ๐‘ฅ๐‘– ]} and ๐‘€๐‘– = sup{๐‘“(๐‘ฅ) โˆถ ๐‘ฅ โˆˆ [๐‘ฅ๐‘–โˆ’1 , ๐‘ฅ๐‘– ]}. We say that ๐‘“ is Riemann-integrable on [๐‘Ž, ๐‘] if, for any ๐œ€ > 0, there exists a partition ๐’ซ such that ๐‘ˆ(๐‘“, ๐’ซ) โˆ’ ๐ฟ(๐‘“, ๐’ซ) < ๐œ€. In such case, the unique number ๐ผ that satisfies ๐ฟ(๐‘“, ๐’ซ) < ๐ผ < ๐‘ˆ(๐‘“๐’ซ) for all partitions is called the integral of ๐‘“, and is denoted by ๐‘

๐‘

โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ, ๐‘Ž

โˆซ ๐‘“, or simply โˆซ ๐‘“. ๐‘Ž

Linear combinations and products of Riemann-integrable functions are Riemann-integrable, and we have ๐‘›

๐‘›

โˆซ โˆ‘ ๐‘“๐‘˜ = โˆ‘ โˆซ ๐‘“๐‘˜ . ๐‘˜=1

๐‘˜=1

Also, if ๐‘“ โ‰ค ๐‘” on [๐‘Ž, ๐‘], then โˆซ ๐‘“ โ‰ค โˆซ ๐‘”. If ๐‘“ is Riemann-integrable, then so is |๐‘“| and | โˆซ ๐‘“| โ‰ค โˆซ |๐‘“|. | | We also have that, for any ๐‘ โˆˆ (๐‘Ž, ๐‘), ๐‘

๐‘

๐‘

โˆซ ๐‘“ = โˆซ ๐‘“ + โˆซ ๐‘“. ๐‘Ž

๐‘Ž

๐‘

Continuous and monotone functions are Riemann-integrable, as well as piecewise continuous or piecewise monotone functions. A set ๐ด โŠ‚ โ„ is of measure zero if for each ๐œ€ > 0 there exists intervals {๐ผ๐‘› } such that ๐ด โŠ‚ โ‹ƒ๐‘› ๐ผ๐‘› and โˆ‘ |๐ผ๐‘› | < ๐œ€, where |๐ผ๐‘› | denotes the length ot ๐ผ๐‘› . Finite and countable infinite sets are of measure zero. There exist also uncountable sets of measure zero, as the Cantor set (see Chapter 5). We have the following criterion for Riemann integrability. Theorem A.9. A bounded function ๐‘“ โˆถ [๐‘Ž, ๐‘] โ†’ โ„ is Riemann-integrable if, and only if, the set where ๐‘“ is not continuous is of measure zero.

264

A. Some results from real analysis

A Riemann-integrable function can be approximated by continuous functions, in the following sense. Theorem A.10. Let ๐‘“ โˆถ [๐‘Ž, ๐‘] โ†’ โ„ be Riemann-integrable, and let ๐‘€ such that |๐‘“(๐‘ฅ)| โ‰ค ๐‘€ for all ๐‘ฅ โˆˆ [๐‘Ž, ๐‘]. For any ๐œ€ > 0, there exists a continuous function ๐‘” on [๐‘Ž, ๐‘] such that |๐‘”(๐‘ฅ)| โ‰ค ๐‘€

and

โˆซ |๐‘“ โˆ’ ๐‘”| < ๐œ€.

We include the proof of this theorem here, as its ideas are useful for the discussion in the text (see Chapter 3). Proof. Given ๐œ€ > 0, let ๐’ซ = {๐‘ฅ0 = ๐‘Ž < ๐‘ฅ1 < . . . < ๐‘ฅ๐‘› = ๐‘} be a partition so that ๐‘ˆ(๐‘“, ๐’ซ) โˆ’ ๐ฟ(๐‘“, ๐’ซ) < ๐œ€/2. In particular, |๐‘ˆ(๐‘“, ๐’ซ) โˆ’ โˆซ ๐‘“| < ๐œ€ . | | 2 Note that ๐‘ˆ(๐‘“, ๐’ซ) is the integral of the function โ„Ž on [๐‘Ž, ๐‘] given by โ„Ž(๐‘ฅ) = ๐‘€๐‘– , if ๐‘ฅ โˆˆ [๐‘ฅ๐‘–โˆ’1 , ๐‘ฅ๐‘– ), which satisfies โ„Ž โ‰ฅ ๐‘“. Hence |๐‘ˆ(๐‘“, ๐’ซ) โˆ’ โˆซ ๐‘“| = โˆซ |โ„Ž โˆ’ ๐‘“|. | | Let ๐›ฟ > 0 so that

๐œ€ . 4๐‘€๐‘› Let ๐‘” be the continuous function on [๐‘Ž, ๐‘] such that it is equal to ๐‘€๐‘– on each interval [๐‘ฅ๐‘–โˆ’1 + ๐›ฟ, ๐‘ฅ๐‘– โˆ’ ๐›ฟ], linear from ๐‘€๐‘–โˆ’1 to ๐‘€๐‘– on each [๐‘ฅ๐‘–โˆ’1 โˆ’ ๐›ฟ, ๐‘ฅ๐‘–โˆ’1 +๐›ฟ], ๐‘– = 2, . . . , ๐‘›, linear from ๐‘“(๐‘Ž) to ๐‘€1 on [๐‘ฅ0 , ๐‘ฅ0 +๐›ฟ], and linear from ๐‘€๐‘› to ๐‘“(๐‘) on [๐‘ฅ๐‘› โˆ’๐›ฟ, ๐‘ฅ๐‘› ]. As each |๐‘€๐‘– | โ‰ค ๐‘€, we have that |๐‘”| โ‰ค ๐‘€. Also ๐›ฟ
0, there exists a partition ๐’ซ such that ๐‘ˆ(๐‘“, ๐’ซ) โˆ’ ๐ฟ(๐‘“, ๐’ซ) < ๐œ€. If ๐‘“ is Riemann integrable, its integral is denoted by โˆซ ๐‘“(๐‘ฅ)๐‘‘๐‘ฅ, ๐‘…

โˆซ ๐‘“, or simply โˆซ ๐‘“. ๐‘…

As in the case of the Riemann integral of single variable functions on an interval, linear combinations and products of Riemann-integrable func๐‘› ๐‘› tions are Riemann-integrable, we have โˆซ โˆ‘๐‘˜=1 ๐‘“๐‘˜ = โˆ‘๐‘˜=1 โˆซ ๐‘“๐‘˜ , and, if ๐‘“ โ‰ค ๐‘” on ๐‘…, then โˆซ ๐‘“ โ‰ค โˆซ ๐‘”. If ๐‘“ is Riemann-integrable on ๐‘…, then so is |๐‘“| and | โˆซ ๐‘“| โ‰ค โˆซ |๐‘“|. We also have that, for any partition ๐’ซ of ๐‘… with subrectangles, each ๐‘“|๐‘† is Riemann-integrable for each ๐‘† โˆˆ ๐’ซ and โˆซ ๐‘“ = โˆ‘ โˆซ ๐‘“. ๐‘…

๐‘†โˆˆ๐’ซ ๐‘†

As above, a set ๐ด โŠ‚ โ„๐‘‘ is of measure zero if for each ๐œ€ > 0 there exist rectangles {๐‘…๐‘› } such that ๐ด โŠ‚ โ‹ƒ๐‘› ๐‘…๐‘› and โˆ‘ vol(๐‘…๐‘› ) < ๐œ€. The analog

A.5. Complete metric spaces

267

of Theorem A.9 is true for multivariable functions: a bounded function ๐‘“ โˆถ ๐‘… โ†’ โ„ is Riemann-integrable if and only if the set where ๐‘“ is not continuous is of measure zero. If we write ๐‘… = ๐‘† ร— ๐‘‡, where ๐‘† and ๐‘‡ are rectangles in โ„๐‘™ and โ„๐‘˜ , respectively, with ๐‘‘ = ๐‘™ + ๐‘˜, we have the identity, for a continuous ๐‘“ โˆถ ๐‘… โ†’ โ„, (A.12)

โˆซ ๐‘“ = โˆซ ( โˆซ ๐‘“๐‘ฅ (๐‘ฆ)๐‘‘๐‘ฆ)๐‘‘๐‘ฅ, ๐‘…

๐‘†

๐‘‡

where, for each ๐‘ฅ โˆˆ ๐‘† and ๐‘ฆ โˆˆ ๐‘‡, ๐‘“๐‘ฅ (๐‘ฆ) = ๐‘“(๐‘ฅ, ๐‘ฆ). The assumption that ๐‘“ is continuous is needed because ๐‘“๐‘ฅ might not be a Riemann-integrable function on ๐‘‡ for all ๐‘ฅ โˆˆ ๐‘†, given a Riemann-integrable function ๐‘“ on ๐‘…. See [Spi65, Theorem 3.10] for a precise version of (A.12) for general Riemann-integrable functions. If ๐‘… โŠ‚ โ„๐‘‘ is a closed rectangle, ๐ด โŠ‚ ๐‘… and ๐‘“ โˆถ ๐‘… โ†’ โ„ is Riemannintegrable, we define โˆซ ๐‘“ = โˆซ ๐‘“ โ‹… ๐œ’๐ด , ๐ด

๐‘…

where ๐œ’๐ด is the characteristic function of the set ๐ด, provided ๐œ’๐ด is Riemann integrable, which occurs when the set of boundary points1 of ๐ด is of measure zero.

A.5. Complete metric spaces The Euclidean space โ„๐‘‘ with the Euclidean distance induced by the norm ๐‘ฅ โ†ฆ |๐‘ฅ| is an example of a metric space. A metric space is a set provided with a metric ๐‘‘ โˆถ ๐‘‹ ร— ๐‘‹ โ†’ [0, โˆž) that satisfies, for all ๐‘ฅ, ๐‘ฆ, ๐‘ง โˆˆ ๐‘‹, (1) ๐‘‘(๐‘ฅ, ๐‘ฆ) = 0 if and only if ๐‘ฅ = ๐‘ฆ, (2) ๐‘‘(๐‘ฅ, ๐‘ฆ) = ๐‘‘(๐‘ฆ, ๐‘ฅ), and (3) ๐‘‘(๐‘ฅ, ๐‘ฆ) โ‰ค ๐‘‘(๐‘ฅ, ๐‘ง) + ๐‘‘(๐‘ง, ๐‘ฆ). A sequence ๐‘ฅ๐‘› in a metric space ๐‘‹ converges to ๐‘ฅ if, for any ๐œ€ > 0, there exists ๐‘ such that ๐‘‘(๐‘ฅ๐‘› , ๐‘ฅ) < ๐œ€. A convergent sequence ๐‘ฅ๐‘› is clearly a Cauchy sequence: given ๐œ€ > 0, there exists ๐‘ such that ๐‘‘(๐‘ฅ๐‘› , ๐‘ฅ๐‘š ) < ๐œ€ for any ๐‘š, ๐‘› โ‰ฅ ๐‘. 1

A point ๐‘ฅ is in the boundary of ๐ด if, for all ๐œ€ > 0, ๐ด โˆฉ ๐ต๐œ€ (๐‘ฅ) โ‰  โˆ… and ๐ต๐œ€ (๐‘ฅ) โงต ๐ด โ‰  โˆ….

268

A. Some results from real analysis

In general, though, not every Cauchy sequence converges in a metric space. A complete metric space is a metric space in which every Cauchy sequence converges. By Theorem A.1, the Euclidean space is complete. If ๐‘‹ is a real or complex vector space, โ€– โ‹… โ€– โˆถ ๐‘‹ โ†’ [0, โˆž) is a norm if it satisfies, for any vectors ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‹, (1) โ€–๐‘ฅโ€– = 0 if and only if ๐‘ฅ = 0, (2) โ€–๐œ†๐‘ฅโ€– = |๐œ†|โ€–๐‘ฅโ€– for any scalar ๐œ†, and (3) โ€–๐‘ฅ + ๐‘ฆโ€– โ‰ค โ€–๐‘ฅโ€– + โ€–๐‘ฆโ€–. The properties above imply that, if โ€– โ‹… โ€– is a norm on ๐‘‹, then ๐‘‘(๐‘ฅ, ๐‘ฆ) = โ€–๐‘ฅ โˆ’ ๐‘ฆโ€– is a metric, and thus ๐‘‹ is metric space. If ๐‘‹ is a complete normed vector space, we call it a Banach space. A series โˆ‘ ๐‘ฅ๐‘› in the normed space ๐‘‹ converges to ๐‘ฅ if the sequence ๐‘› of partial sums ๐‘ ๐‘› = โˆ‘๐‘˜=1 ๐‘ฅ๐‘˜ converges to ๐‘ฅ. Equivalently, ๐‘›

|| โˆ‘ ๐‘ฅ โˆ’ ๐‘ฅ|| โ†’ 0. ๐‘˜ || || ๐‘˜=1

A series converges absolutely if the series โˆ‘ โ€–๐‘ฅ๐‘› โ€– converges. If ๐‘‹ is a Banach space and the series โˆ‘ ๐‘ฅ๐‘› converges absolutely, then it converges. It turns out that this fact is equivalent to the completeness of ๐‘‹, as stated in Theorem A.13 due to Banach. Theorem A.13. The normed vector space ๐‘‹ is a Banach space if and only if every absolutely convergent series converges in ๐‘‹. An inner product space in the vector space ๐‘‹ is a scalar form โŸจโ‹…, โ‹…โŸฉ on ๐‘‹ ร— ๐‘‹ that satisfies (1) โŸจ๐‘ฅ, ๐‘ฅโŸฉ โ‰ฅ 0, and โŸจ๐‘ฅ, ๐‘ฅโŸฉ = 0 if and only if ๐‘ฅ = 0, (2) โŸจ๐‘ฅ, ๐‘ฆโŸฉ = โŸจ๐‘ฆ, ๐‘ฅโŸฉ for any ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‹, and (3) โŸจ๐œ†๐‘ฅ + ๐œ‡๐‘ฆ, ๐‘งโŸฉ = ๐œ†โŸจ๐‘ฅ, ๐‘งโŸฉ + ๐œ‡โŸจ๐‘ฆ, ๐‘งโŸฉ for any scalars ๐œ†, ๐œ‡ and ๐‘ฅ, ๐‘ฆ, ๐‘ง โˆˆ ๐‘‹. Two vectors ๐‘ฅ, ๐‘ฆ โˆˆ ๐‘‹ are called orthogonal, and we write ๐‘ฅ โŸ‚ ๐‘ฆ, if โŸจ๐‘ฅ, ๐‘ฆโŸฉ = 0. If ๐‘† โŠ‚ ๐‘‹, ๐‘ฅ is orthogonal to ๐‘†, and we write ๐‘ฅ โŸ‚ ๐‘†, if

A.5. Complete metric spaces

269

๐‘ฅ is orthogonal to every vector in ๐‘†. The orthogonal complement of ๐‘†, denoted as ๐‘†โŸ‚ , is given by all the orthogonal vectors to ๐‘†. ๐‘† โŸ‚ is always a vector subspace of ๐‘‹. If ๐‘‹ is a real vector space, then an inner product is a positive symmetric bilinear form. An inner product induces the norm โ€–๐‘ฅโ€– = โˆšโŸจ๐‘ฅ, ๐‘ฅโŸฉ. The norm satisfies the parallelogram identity โ€–๐‘ฅ + ๐‘ฆโ€–2 + โ€–๐‘ฅ โˆ’ ๐‘ฆโ€–2 = 2โ€–๐‘ฅโ€–2 + 2โ€–๐‘ฆโ€–2 , and the polarization identity, given by 1 โŸจ๐‘ฅ, ๐‘ฆโŸฉ = (โ€–๐‘ฅ + ๐‘ฆโ€–2 โˆ’ โ€–๐‘ฅ โˆ’ ๐‘ฆโ€–2 ) 4 in the real case, and by 1 โŸจ๐‘ฅ, ๐‘ฆโŸฉ = (โ€–๐‘ฅ + ๐‘ฆโ€–2 โˆ’ โ€–๐‘ฅ โˆ’ ๐‘ฆโ€–2 + ๐‘–โ€–๐‘ฅ + ๐‘–๐‘ฆโ€–2 โˆ’ ๐‘–โ€–๐‘ฅ โˆ’ ๐‘–๐‘ฆโ€–2 ) 4 in the complex case. If ๐‘‹ is a complete inner product space, then it is called a Hilbert space. Note that a Hilbert space is a Banach space whose norm is induce by an inner product. If ๐‘‹ is a Hilbert space and ๐‘Œ is a closed vector subspace, then, for each ๐‘ฅ โˆˆ ๐‘‹, there is ๐‘ฆ0 โˆˆ ๐‘Œ closest to ๐‘ฅ, that is, โ€–๐‘ฆ0 โˆ’ ๐‘ฅโ€– โ‰ค โ€–๐‘ฆ โˆ’ ๐‘ฅโ€– for all ๐‘ฆ โˆˆ ๐‘Œ . ๐‘ฆ0 is called the orthogonal projection of ๐‘ฅ in ๐‘Œ , and satisfies ๐‘ฅ โˆ’๐‘ฆ0 โŸ‚ ๐‘Œ . As a consequence, if ๐‘Œ is a proper closed subspace of ๐‘‹, then ๐‘Œ โŸ‚ is nontrivial.

Acknowledgments

I would like to thank all the students who have taken this course, both in IAS/PCMI and Colima, who provided the motivation to work on this text. Their questions and commentary to the manuscript were very valuable during its writing. I would also like to thank the reviewers of the first version of the manuscript, as their observations lead to several improvements to this book. I finally want to thank Eriko Hironaka, Ina Mette, Marcia C. Almeida, John F. Brady Jr., and Abigail Lawson at the AMS, who took care of all the logistics in the publication of this book.

271

Bibliography

[ABR01] S. Axler, P. Bourdon, and W. Ramey, Harmonic function theory, 2nd ed., Graduate Texts in Mathematics, vol. 137, Springer-Verlag, New York, 2001, DOI 10.1007/978-1-4757-8137-3. MR1805196 [Axl15]

S. Axler, Linear algebra done right, 3rd ed., Undergraduate Texts in Mathematics, Springer, Cham, 2015, DOI 10.1007/978-3-319-11080-6. MR3308468

[Ban24] St. Banach, Sur le thรฉorรจme de M. Vitali., Fund. Math. 5 (1924), no. 1924, 130โ€“136. [Bar98]

M. T. Barlow, Diffusions on fractals, Lectures on probability theory and statistics (SaintFlour, 1995), Lecture Notes in Math., vol. 1690, Springer, Berlin, 1998, pp. 1โ€“121, DOI 10.1007/BFb0092537. MR1668115

[Ber14]

S. Bernstein, Sur la convergence absolue des sรฉries trigonomรฉtriques, C. R. Acad. Sci., Paris 158 (1914), 1661โ€“1663 (French).

[Bre07]

D. M. Bressoud, A radical approach to real analysis, 2nd ed., Classroom Resource Materials Series, Mathematical Association of America, Washington, DC, 2007. MR2284828

[Bre08]

D. M. Bressoud, A radical approach to Lebesgueโ€™s theory of integration, MAA Textbooks, Cambridge University Press, Cambridge, 2008. MR2380238

[BS92]

Robert G. Bartle and Donald R. Sherbert, Introduction to real analysis, New York, NY: Wiley, 1992 (English).

[CZ52]

A. P. Calderon and A. Zygmund, On the existence of certain singular integrals, Acta Math. 88 (1952), 85โ€“139, DOI 10.1007/BF02392130. MR52553

[du 76]

P. du Bois-Reymond, Zusรคtze zur vorstehenden Abhandlung, Math. Ann. 10 (1876), 431โ€“445 (German).

[Eva10]

L. C. Evans, Partial differential equations, 2nd ed., Graduate Studies in Mathematics, vol. 19, American Mathematical Society, Providence, RI, 2010, DOI 10.1090/gsm/019. MR2597943

[Fat06]

P. Fatou, Sรฉries trigonomรฉtriques et sรฉries de Taylor (French), Acta Math. 30 (1906), no. 1, 335โ€“ 400, DOI 10.1007/BF02418579. MR1555035

[Fej00]

L. Fejรฉr, Sur les fonctions bornรฉes et intรฉgrables, Comptes Rendus Hebdomadaries, Seances de lโ€™Academie de Sciences, Paris 131 (1900), 984โ€“987.

[Fle77]

W. Fleming, Functions of several variables, 2nd ed., Undergraduate Texts in Mathematics, Springer-Verlag, New York-Heidelberg, 1977. MR0422527

[Fol95]

G. B. Folland, Introduction to partial differential equations, 2nd ed., Princeton University Press, Princeton, NJ, 1995. MR1357411

273

274 [Fol99]

Bibliography G. B. Folland, Real analysis: Modern techniques and their applications, 2nd ed., Pure and Applied Mathematics (New York), John Wiley & Sons, Inc., New York, 1999. A WileyInterscience Publication. MR1681462

[Fou55] J. Fourier, Analytical theory of heat, Dover Publications, Inc., New York, 1955. Translated, with notes, by Alexander Freeman. MR0075136 [Fra09]

M. Frantz, A fractal made of golden sets, Math. Mag. 82 (2009), no. 4, 243โ€“254, DOI 10.4169/193009809X468670. MR2571788

[Gam01] T. W. Gamelin, Complex analysis, Undergraduate Texts in Mathematics, Springer-Verlag, New York, 2001, DOI 10.1007/978-0-387-21607-2. MR1830078 [Gau40] C. F. Gauss, Allgemeine lehrsรคtze in beziehung auf die verkehrten verhรคltnisse des quadrats der entfernung wirkenden anziehungs- und abstossungs- krรคfte, Weidmann, 1840. [Gau09] E. D. Gaughan, Introduction to analysis, 3rd ed., Brooks/Cole Publishing Co., Pacific Grove, CA, 1987. MR1010646 [Hat85]

M. Hata, On the structure of self-similar sets, Japan J. Appl. Math. 2 (1985), no. 2, 381โ€“414, DOI 10.1007/BF03167083. MR839336

[Hau18] F. Hausdorff, Dimension und รคuรŸeres MaรŸ (German), Math. Ann. 79 (1918), no. 1-2, 157โ€“179, DOI 10.1007/BF01457179. MR1511917 [HL30]

G. H. Hardy and J. E. Littlewood, A maximal theorem with function-theoretic applications, Acta Math. 54 (1930), no. 1, 81โ€“116, DOI 10.1007/BF02547518. MR1555303

[Hut81] J. E. Hutchinson, Fractals and self-similarity, Indiana Univ. Math. J. 30 (1981), no. 5, 713โ€“747, DOI 10.1512/iumj.1981.30.30055. MR625600 [K8ฬˆ8]

T. W. Kรถrner, Fourier analysis, Cambridge University Press, Cambridge, 1988, DOI 10.1017/CBO9781107049949. MR924154

[Kel67]

O. D. Kellogg, Foundations of potential theory, Die Grundlehren der mathematischen Wissenschaften, Band 31, Springer-Verlag, Berlin-New York, 1967. Reprint from the first edition of 1929. MR0222317

[Kig89]

J. Kigami, A harmonic calculus on the Sierpiล„ski spaces, Japan J. Appl. Math. 6 (1989), no. 2, 259โ€“290, DOI 10.1007/BF03167882. MR1001286

[Kig93]

J. Kigami, Harmonic calculus on p.c.f. self-similar sets, Trans. Amer. Math. Soc. 335 (1993), no. 2, 721โ€“755, DOI 10.2307/2154402. MR1076617

[Kig94]

J. Kigami, Effective resistances for harmonic structures on p.c.f. self-similar sets, Math. Proc. Cambridge Philos. Soc. 115 (1994), no. 2, 291โ€“303, DOI 10.1017/S0305004100072091. MR1277061

[Kig01]

J. Kigami, Analysis on fractals, Cambridge Tracts in Mathematics, vol. 143, Cambridge University Press, Cambridge, 2001, DOI 10.1017/CBO9780511470943. MR1840042

[Leb98]

H. Lebesgue, Sur lโ€™approximation des fonctions, Bull. Sci. Math., II. Sรฉr. 22 (1898), 278โ€“287 (French).

[Leb99a] H. Lebesgue, Sur la dรฉfinition de lโ€™aire dโ€™une surface, C. R. Acad. Sci., Paris 129 (1899), 870โ€“873 (French). [Leb99b] H. Lebesgue, Sur quelques surfaces non rรฉglรฉes applicables sur le plan, C. R. Acad. Sci., Paris 128 (1899), 1502โ€“1505 (French). [Leb01a] H. Lebesgue, Sur la dรฉfinition de certaines intรฉgrales de surface, C. R. Acad. Sci., Paris 131 (1901), 867โ€“870 (French). [Leb01b] H. Lebesgue, Sur le minimum de certaines intรฉgrales, C. R. Acad. Sci., Paris 131 (1901), 935โ€“ 937 (French). [Leb01c] H. Lebesgue, Sur une gรฉnรฉralisation de lโ€™intรฉgrale dรฉfinie, C. R. Acad. Sci., Paris 132 (1901), 1025โ€“1028 (French). [Leb02]

H. Lebesgue, Intรฉgrale, longueur, aire, Annali di Mat. (3) 7 (1902), 231โ€“359 (French).

[Leb03]

H. Lebesgue, Sur les sรฉries trigonomรฉtriques (French), Ann. Sci. ร‰cole Norm. Sup. (3) 20 (1903), 453โ€“485. MR1509032

Bibliography

275

[Leb04]

H. Lebesgue, Leรงons sur lโ€™intรฉgration et la recherche de fonctions primitives, Paris: GauthierVillars. VII u. 136 S. 8โˆ˜ (1904), 1904.

[Leb10]

H. Lebesgue, Sur lโ€™intรฉgration des fonctions discontinues (French), Ann. Sci. ร‰cole Norm. Sup. (3) 27 (1910), 361โ€“450. MR1509126

[Lev06]

B. Levi, Sopra lโ€™integrazione delle serie, Ist. Lombardo, Rend., II. Ser. 39 (1906), 775โ€“780 (Italian).

[MH93] J. E. Marsden and M. J. Hoffman, Elementary classical analysis, 2 ed., New York, W. H. Freeman and Company, 1993 (English). [Min96] H. Minkowski, Geometrie der Zahlen. In 2 Lieferungen. Lfg. 1, Leipzig: B. G. Teubner. 240 S. 8โˆ˜ . (1896)., 1896. [Mor46] P. A. P. Moran, Additive functions of intervals and Hausdorff measure, Proc. Cambridge Philos. Soc. 42 (1946), 15โ€“23, DOI 10.1017/s0305004100022684. MR14397 [MS13]

C. Muscalu and W. Schlag, Classical and multilinear harmonic analysis. Vol. I, Cambridge Studies in Advanced Mathematics, vol. 137, Cambridge University Press, Cambridge, 2013. MR3052498

[Nel61]

E. Nelson, A proof of Liouvilleโ€™s theorem, Proc. Amer. Math. Soc. 12 (1961), 995, DOI 10.2307/2034412. MR259149

[PdC06] M.-A. Parseval des Chรชnes, Mรฉmoire sur les sรฉries et sur lโ€™intรฉgration complรจte dโ€™une รฉquation aux differences partielle linรฉaires du second ordre, ร  coefficiens constans, Mรฉmoires prรฉsentรฉs ร  lโ€™Institut des Sciences, Lettres et Arts, par divers savants, et lus dans ses assemblรฉes. Sciences mathรฉmatiques et physiques. (Savants รฉtrangers.), 1806. [Pla10]

M. Plancherel, Contribution ร  lโ€™รฉtude de la reprรฉsentation dโ€™une fonction arbitraire par des intรฉgrales dรฉfinies, Rend. Circ. Mat. Palermo 30 (1910), 289โ€“335.

[Poi20]

S. D. Poisson, Mรฉmoire sur lโ€™intรฉgration de quelques รฉquations linรฉaires aux diffรฉrences partielles, et particuliรจrement de lโ€™รฉquation gรฉnรฉrale du mouvement des fluides รฉlastiques, Mรฉmoires Acad. Sci. Inst. France 3 (1820), 121โ€“176.

[Ram84] R. Rammal, Spectrum of harmonic excitations on fractals, J. Physique 45 (1984), no. 2, 191โ€“ 206, DOI 10.1051/jphys:01984004502019100. MR737523 [Rie51]

B. Riemann, Grundlagen fรผr eine allgemeine theorie der functionen einer verรคnderlichen complexen grรถsse, Ph.D. thesis, Gรถttingen, 1851.

[Rie10]

F. Riesz, Untersuchungen รผber Systeme integrierbarer Funktionen (German), Math. Ann. 69 (1910), no. 4, 449โ€“497, DOI 10.1007/BF01457637. MR1511596

[Rie13]

F. Riesz, Les systรจmes dโ€™รฉquations linรฉaires ร  une infinite dโ€™inconnues, Paris: Gauthier-Villars, VI + 182 S. 8โˆ˜ . (Collection Borel.) (1913)., 1913.

[Rie32]

F. Riesz, Sur lโ€™existence de la dรฉrivรฉe des fonctions monotones et sur quelques problรจmes qui sโ€™y rattachent, Acta Litt. Sci. Szeged 5 (1932), 208โ€“221 (French).

[Sรกe12]

R. A. Sรกenz, Nontangential limits and Fatou-type theorems on post-critically finite self-similar sets, J. Fourier Anal. Appl. 18 (2012), no. 2, 240โ€“265, DOI 10.1007/s00041-011-9194-1. MR2898728

[Shi91]

T. Shima, On eigenvalue problems for the random walks on the Sierpiล„ski pre-gaskets, Japan J. Indust. Appl. Math. 8 (1991), no. 1, 127โ€“141, DOI 10.1007/BF03167188. MR1093832

[Sie15]

W. Sierpiล„ski, Sur une courbe dont tout point est un point de ramification, C. R. Acad. Sci., Paris 160 (1915), 302โ€“305 (French).

[Spi65]

M. Spivak, Calculus on manifolds. A modern approach to classical theorems of advanced calculus, W. A. Benjamin, Inc., New York-Amsterdam, 1965. MR0209411

[SS03]

E. M. Stein and R. Shakarchi, Fourier analysis: An introduction, Princeton Lectures in Analysis, vol. 1, Princeton University Press, Princeton, NJ, 2003. MR1970295

[SS05]

E. M. Stein and R. Shakarchi, Real analysis: Measure theory, integration, and Hilbert spaces, Princeton Lectures in Analysis, vol. 3, Princeton University Press, Princeton, NJ, 2005. MR2129625

276

Bibliography

[Ste70]

E. M. Stein, Singular integrals and differentiability properties of functions, Princeton Mathematical Series, No. 30, Princeton University Press, Princeton, N.J., 1970. MR0290095

[Str06]

R. S. Strichartz, Differential equations on fractals: A tutorial, Princeton University Press, Princeton, NJ, 2006. MR2246975

[SW71]

E. M. Stein and G. Weiss, Introduction to Fourier analysis on Euclidean spaces, Princeton Mathematical Series, No. 32, Princeton University Press, Princeton, N.J., 1971. MR0304972

[Ull08]

D. C. Ullrich, Complex made simple, Graduate Studies in Mathematics, vol. 97, American Mathematical Society, Providence, RI, 2008, DOI 10.1090/gsm/097. MR2450873

[Vit08]

G. Vitali, Sui gruppi di punti e sulle funzioni di variabili reali, Torino Atti 43 (1908), 229โ€“246 (Italian).

[YHK97] M. Yamaguti, M. Hata, and J. Kigami, Mathematics of fractals, Translations of Mathematical Monographs, vol. 167, American Mathematical Society, Providence, RI, 1997. Translated from the 1993 Japanese original by Kiki Hudson, DOI 10.1090/mmono/167. MR1471705 [Zyg02]

A. Zygmund, Trigonometric series. Vol. I, II, 3rd ed., Cambridge Mathematical Library, Cambridge University Press, Cambridge, 2002. With a foreword by Robert A. Fefferman. MR1963498

Index

๐น๐œ set, 261 ๐บ ๐›ฟ set, 261 Abel means, 43 summable, 43, 55 theorem, 43 adyacent vertex, 209, 245 almost everywhere, 98 average, 54, 137, 143 Banach space, 126, 268 theorem, 268 Bernsteinโ€™s theorem, 69 Besselโ€™s inequality, 51 better kernels, 148, 158, 165, 173 Bolzanoโ€“Weierstrass theorem, 260 boundary, 210, 244 Cantor set, 188 Cauchy sequence, 259, 267 cell, 192, 243 Cesร ro summable, 55 summation, 54 sums, 54 Chebyshevโ€™s inequality, 140 closed set, 261 closure, 260 compact, 262

complete, 259, 268 connected set, 266 continuous, 260 uniformly, 260 contraction, 191 contraction constant, 191 convergence absolute, 46, 48, 261 conditional, 261 in measure, 179 mean-square, 59, 164, 168 nontangential, 146 pointwise, 260 radial, 29 uniform, 260 convolution, 128 cover ๐›ฟ-cover, 185 critical set, 244 cube, 85 dyadic, 85 strictly thinner, 88 diameter, 185 Dirichlet eigenfunction, 231, 233 eigenvalue, 231 form, 246 principle, 11 problem, 10, 27

277

278 test, 261 discrete energy, 204, 210 discrete Laplacian, 207, 217 domain, 6 dominated convergence theorem, 118 dyadic decomposition, 51 eigenfunction, 223 eigenvalue, 223 energy form, 9, 212, 216, 247, 255 minimizers, 9 extended real line, 96 Fatouโ€™s lemma, 111 Fejรฉr kernel, 58 theorem, 55 flounces function, 42 Fourier coefficients, 39 inversion formula, 157 series, 40 transform, 151 Fubiniโ€™s theorem, 123

Index Hilbert space, 269 transform, 171, 173 Hopf lemma, 33 integrable function, 113, 115 locally integrable, 138 Riemann-integrable, 263, 266 square integrable, 130 integral complex valued function, 115 extended real valued function, 113 improper, 74 nonnegative function, 108 nonnegative simple function, 105 on the Sierpiล„ski gasket, 214 Riemann, 263, 266 self-similarity, 215 symmetry, 215 isometry, 160, 172

Gaussian kernel, 154, 158 golden fractal, 199 good kernels, 27, 58, 79, 81

Laplace equation, 3 Laplacian, 3, 217, 250, 255 polar coordinates, 11, 35 Lebesgue differentiation theorem, 143 measure, 94 space, 124, 130 Lipschitz function, 53

Hรถlder continuous, 53, 69 Hardy-Littlewood maximal function, 138 theorem, 139 harmonic ๐‘š-harmonic function, 212, 217 conjugate, 4, 167 function, 3, 13, 210, 212, 217, 247, 254 harmonic structure, 210, 247 compatibility, 247 self-similarity, 247 Harnack inequality, 33 Hata tree set, 197, 244, 248 Hausdorff dimension, 185, 188 measure, 186 heat kernel, 81, 154 Heineโ€“Borel theorem, 262

maximal function, 138 uncentered, 142 maximum princple, 20, 253, 254 measurable ๐น๐œ set, 94 ๐บ ๐›ฟ set, 94 closed cube, 92 closed set, 93, 214 complement, 89, 214 countable intersection, 92 countable union, 90, 214 function, 96 open set, 93, 214 pointwise limit, 97 set, 89, 214 union, 90 measure countable additivity, 94, 214 countable subadditivity, 87

Index exterior, 86 invariance, 96 Lebesgue, 94 monotone continuity, 94 monotonicity, 87 on the Sierpiล„ski gasket, 213 outer, 85, 186 regularity, 95 zero, 86, 263, 266 metric space, 267 Minkowskiโ€™s inequality, 124, 132 monotone convergence theorem, 109 negative part, 102 neighbor, 209, 245 nontangential limit, 146 norm, 268 open cover, 262 open set, 261 open set condition, 193 orthogonal transformation, 31 orthogonality, 49, 59, 268 Parsevalโ€™s identity, 60 PCF set, 244 Plancherelโ€™s theorem, 162 Poisson integral, 27, 75, 164, 167 kernel, 24, 71, 154, 158, 167 polarization identity, 160 positive part, 102 post-critical set, 244 post-critically finite set, 244 punctured ball, 30 Riemannโ€“Lebesgue lemma, 53, 152 sawtooth function, 40, 46 section of a function, 123 of a set, 120 self-similar, 191, 243 sharkteeth function, 41, 46 Sierpiล„ski gasket, 196, 207, 244 similitude, 193 simple function, 99 approximation, 100 reduced form, 100

279 singularity isolated, 29 removable, 30 snowflake set, 197 spherical coordinates, 7 spline, 213, 256 integral, 216 step function, 100 subordination, 81 symmetry lemma, 25, 32 Tonelliโ€™s theorem, 124 translation, 31 trigonometric polynomial, 37, 58 unitary operator, 162 upper half-space, 5, 71 Vitalliโ€™s covering lemma, 140 volume, 85 weak type, 139, 174, 182 Weierstrass ๐‘€-test, 261 word, 191, 208, 243

Published Titles in This Subseries 105 Ricardo A. Sยด aenz, Introduction to Harmonic Analysis, 2023 93 Iva Stavrov, Curvature of Space and Time, with an Introduction to Geometric Analysis, 2020 66 Thomas Garrity, Richard Belsho๏ฌ€, Lynette Boos, Ryan Brown, Carl Lienert, David Murphy, Junalyn Navarra-Madsen, Pedro Poitevin, Shawn Robinson, Brian Snyder, and Caryn Werner, Algebraic Geometry, 2013 63 Marยดฤฑa Cristina Pereyra and Lesley A. Ward, Harmonic Analysis, 2012 ยด 58 Alvaro Lozano-Robledo, Elliptic Curves, Modular Forms, and Their L-functions, 2011 51 Richard S. Palais and Robert A. Palais, Di๏ฌ€erential Equations, Mechanics, and Computation, 2009 49 33 32 7

Francis Bonahon, Low-Dimensional Geometry, 2009 Rekha R. Thomas, Lectures in Geometric Combinatorics, 2006 Sheldon Katz, Enumerative Geometry and String Theory, 2006 Judy L. Walker, Codes and Curves, 2000

3 Roger Knobel, An Introduction to the Mathematical Theory of Waves, 2000 2 Gregory F. Lawler and Lester N. Coyle, Lectures on Contemporary Probability, 1999