Author: Wong M.W.  

Tags: mathematics   physics  

ISBN: 978-9814583084

Year: 2014

Text
                    Series on Analysis. Application* and Computation - Vol.6
ISAAC
* M W Wong
An Introduction to
Pseudo-Differential
Operators
3rd Edition


Series on Analysis, Applications and Computation - Vol. 6 An Introduction to Pseudo-Differential Operators 3rd Edition
Series on Analysis, Applications and Computation Series Editors: Heinrich G W Begehr (Freie Univ. Berlin, Germanyj Robert Pertsch Gilbert (Univ Delaware, USA) M. W. Wong (York Univ., Canada) Advisory Board Members: Mikhail S Agranovich (Moscow Inst, of Elec. & Math., Russia), Ryuichi Ashino (Osaka Kyoiku Univ, Japan), Alain Bourgeat (Univ. de Lyon, France), Victor Burenkov (Cardiff Univ, UK), Jinyuan Du (Wuhan Univ., China), Antonio Fasano (Univ di Firenez, Italy), Massimo Lanza de Cristoforis (Univ. di Padova, Italy), Bert-Wolfgang Schulze (Univ. Potsdam, Germany), Masahiro Yamamoto (Univ of Tokyo, Japan) & Armand Wirgin (CNRS-Marseille, France) Published Vol. 1: Boundary Values and Convolution in Ultradistribution Spaces by R D Carmichael, A Kaminski & S Pilipovic Vol. 2: Complex Analysis by MW Wong Vol. 3: Topics in Mathematical Analysis edited by P Ciatti, E Gonzalez, M L de Cristoforis & G P Leonardi Vol. 4: Nonlinear Waves: An Introduction by P Popivanov & A Slavova Vol. 5: Asymptotic Behavior of Generalized Functions by S Pilipovic, B Stankovic & J Vindas Vol. 6: An Introduction to Pseudo-Differential Operators, 3rd Edition by MW Wong
Series on Analysis, Applications and Computation - Vol. 6 ISAAC! An Introduction to Pseudo'Differential Operators 3rd Edition o M W Wong York University, Canada ^ World Scientific NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE Library of Congress Cataloging-in-Publication Data Wong, M. W. (Man Wah), 1951- An introduction to pseudo-differential operators / by M.W. Wong (York University, Canada). ~ 3rd edition. pages cm. -- (Series on analysis, applications and computation ; vol. 6) Includes bibliographical references and index. ISBN 978-9814583084 (hardcover: alk. paper) 1. Pseudodifferential operators. I. Title. QA329.7.W658 2014 515'.7242-dc23 2014003783 British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Copyright © 2014 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the publisher. For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher. Printed in Singapore
Preface There have been a lot of developments in pseudo-differential operators since the first edition was published in 1991. The second edition, published in 1999, has served well as an introduction to pseudo-differential operators in the capacity of a textbook. The prerequisites for the first two editions are minimal and can be seen from the prefaces for the first and second editions, which follow this one. The third edition is intended to contain not only improvements of some of the contents and additional exercises to some of the existing chapters in the second edition, but also new chapters to make the book more useful without losing the original intent of keeping the book as elementary as possible. The new chapters notwithstanding, the whole book remains to be a textbook primarily for beginning graduate students in mathematics. It is also useful to mathematicians aspiring to do research in pseudo-differential operators and related topics. The new chapters are the last seven chapters, Chapters 17-23, of the book. The focus of Chapters 17-21 is on the class of pseudo-differential operators studied in the first two editions. The theme underlying Chap¬ ters 17-19 is Garding’s inequality, which is used to prove the existence and uniqueness of solutions of pseudo-differential equations. In particular, the Hille-Yosida-Phillips theorem on one-parameter semigroups is used to prove the existence and uniqueness of solutions of initial value problems for heat equations governed by pseudo-differential operators. After a chap¬ ter, Chapter 20, on the general theory of Fredholm operators that we need for this book, the ellipticity and Fredholmness of pseudo-differential oper¬ ators are developed in Chapter 21. The ellipticity, Fredholmness, an index formula and the spectral invariance for another class of pseudo-differential operators, dubbed symmetrically global pseudo-differential operators in this
vi An Introduction to Pseudo-Differential Operators, 3rd Edition book, are studied in Chapters 22 and 23. The emphasis of the book, as in the first two editions, is on the global theory of elliptic pseudo-differential operators on Lp(Rn), 1 < p < oo. As we are now well into the new millennium and moving forward with increasing acceleration, many advanced topics in any area of science and engineering in 1991 are now being taught in basic courses to students. The prerequisites for a complete understanding of the book can be succinctly described as a first course in functional analysis including the Riesz theory of compact operators. The book contains ample material to be studied leisurely and carefully for a two-semester course. One-semester courses can be designed by omitting certain topics in order to fulfil the needs of the students and the duration of the semester. Preface to the Second Edition The first edition of the book has been used as the textbook for the standard graduate course in partial differential equations at York University since its publication in 1991. The motivation for writing the second edition stems from the desire to remove several deficiencies and obscurities, and to incor¬ porate the improvements that I can see through many years of teaching the subject to graduate students and discussions of the subject with colleagues. Notwithstanding the many changes I have in mind, I am convinced that the elementary character of the book has served and will serve well as an ideal introduction to the study of pseudo-differential operators. Thus, the basic tenet of the second edition is to retain the style and the scope of the first edition. Notable in the second edition is the addition of two chapters to the book. Experience in teaching pseudo-differential operators reveals the fact that many graduate students are still not comfortable with the interchange of order of integration and differentiation. The new chapter added to the beginning of the book is to prove a theorem to this effect which can cope with every interchange of order of integration and differentiation encoun¬ tered in the book. Another new chapter, added as the final chapter in the second edition, is to prove a theorem on the existence of weak solutions of pseudo-differential equations. The inclusion of this chapter, in my opinion, enhances the value of the book as a book on partial differential equations. Furthermore, it provides a valuable connection with the chapter on minimal
Preface vii and maximal operators and the chapter on global regularity. Other new features in the second edition include a deeper study of elliptic operators and parametrices, more details on the proof of the Lp- boundedness of pseudo-differential operators, additional exercises in several chapters of the book, a slightly expanded bibliography and an index. Preface to the First Edition The aim of the book is to give a straightforward account of a class of pseudo-differential operators. The prerequisite for understanding the book is a course in real variables. It is hoped that the book can be used in courses in functional analysis, Fourier analysis and partial differential equations. The first eight chapters of the book contain the basic formal calculus of pseudo-differential operators. The remaining five chapters are devoted to some topics of a more functional analytic character. It is clear to the expert that the book takes up a single theme in a wide subject and many important topics are omitted. It is my belief that this approach is in fact a more effective introduction of pseudo-differential operators to mathematicians and graduate students beginning to learn the subject. Exercises are included in the text. They are useful to anyone who wants to understand and appreciate the book better. The actual writing of the book was essentially carried out and completed at the University of California at Irvine while I was on sabbatical leave from York University in the academic year in 1987-88. The preliminary drafts of the book have been used in seminars and graduate courses at the University of California at Irvine and York University. Many colleagues and students have helped me improve the contents and organization of the book. In particular, I wish to thank Professor William Margulies at the California State University at Long Beach, Professor Mar¬ tin Schechter at the University of California at Irvine, Professor Tuan Vu and Mr. Zhengbin Wang at York University for their stimulating conver¬ sations and critical comments about my book. I also wish to thank Mr. Lian Pi, my Ph.D. research student at York University, who has worked out every exercise in the book.
This page intentionally left blank
Contents Preface v 1. Introduction, Notation and Preliminaries 1 2. Differentiation of Integrals Depending on Parameters 5 3. The Convolution 9 4. The Fourier Transform 17 5. Tempered Distributions 27 6. Symbols, Pseudo-Differential Operators and Asymptotic Expansions 31 7. A Partition of Unity and Taylor’s Formula 43 8. The Product of Two Pseudo-Differential Operators 55 9. The Formal Adjoint of a Pseudo-Differential Operator 61 10. The Parametrix of an Elliptic Pseudo-Differential Operator 69 11. Lp-Boundedness of Pseudo-Differential Operators 75 12. The Sobolev Spaces Hs'p, —oo <s<oo, l<p<oo 87 ix
x An Introduction to Pseudo-Differential Operators, 3rd Edition 13. Closed Linear Operators 95 14. Minimal and Maximal Pseudo-Differential Operators 99 15. Global Regularity of Elliptic Partial Differential Equations 105 16. Weak Solutions of Pseudo-Differential Equations 109 17. Garding’s Inequality 113 18. Strong Solutions of Pseudo-Differential Equations 121 19. One-Parameter Semigroups Generated by Pseudo-Differential Operators 129 20. Fredholm Operators 143 21. Fredholm Pseudo-Differential Operators 155 22. Symmetrically Global Pseudo-Differential Operators 167 23. Spectral Invariance of Symmetrically Global Pseudo-Differential Operators 177 Bibliography 181 Index 183
Chapter 1 Introduction, Notation and Preliminaries Let W1 be the usual Euclidean space given by En = {(#i, £2,..., xn) : Xj’s are real numbers}. We denote points in En by x,y,£,rj etc. Let x = (#1,^2,... ,xn) and y = (2/172/2, - - -,2fn) be any two points in En. The inner product x • y of x and y is defined by n * • y = E 3=1 and the norm \x\ of x is defined by On En, the simplest differential operators are , j = 1,2,..., n. We some¬ times denote by dj. For reasons we shall see later in the book, we usu¬ ally find the operator Dj given by Dj = —idj, i2 = — 1, better in expressing certain formulas. The most general linear partial differential operator of order m on En treated in this book is of the form E 1D? ■Dn”, (i-i) <*i+0!2H \-an<m where c*i, «2,...,an are nonnegative integers and aalya2,...,an (x) is an in_ finitely differentiable complex-valued function on En. To simplify the ex¬ pression (1.1), we let Oi — (c^i, OL2, •••, Oifi)j 1
2 An Introduction to Pseudo-Differential Operators, 3rd Edition M = ai 3=1 and The a, given by an n-tuple of nonnegative integers, is called a multi-index. We call \x\ the length of the multi-index a. With the help of multi-indices, we can rewrite our differential operator (1.1) in the better form aa(x)Da. (1.2) |a|<m For each fixed x in Mn, the operator (1.2) is a polynomial in D\, D%,..., Dn. Therefore it is natural to denote the operator (1.2) by P(x,D). If we re¬ place D in (1.2) by a point £ = (£i,£2, • • • ,£n) in Kn, then we obtain a polynomial E|c*|<min where £a = £“'£22 Naturally, this polynomial is denoted by P(#,£). We call P(x,£) the symbol of the operator P(x,D). In this book, we shall study the partial differential operators (1.2) and their generalizations called pseudo-differential operators. To do this, we find it convenient to introduce in Chapters 2-5 certain aspects of analysis pertinent to our need. The following list of remarks, notation and formulas will be useful to us. (i) We denote the set of all real numbers by M and the set of all complex numbers by C. (ii) All vector spaces are assumed to be over the field of complex numbers. All functions are assumed to be complex-valued unless otherwise specified. (iii) We do not bother to distinguish a function / from its value f(x) at x. In other words, we shall occasionally use the symbol f(x) to denote the function / without any warning. (iv) Although the differential operator Da = D^D^2 • • • D%n is more useful to us, we still use the differential operator da = d“1 d%2 • • • d%n very often in the book. In case we want to emphasize the variable x (or £) with respect to which we differentiate, we write d% (or d£) for da and D% (or D^) for Da. (v) We denote the set of all infinitely differentiable functions on W1 by C°°(Rn).
Introduction, Notation and Preliminaries 3 (vi) The Lp norm of a function / in Lp(Rn), 1 < p < oo, is denoted by II/IIp- Let a = (ai,a2,...,an) and P = (A,ft,...,ft) be any two multi¬ indices. (vii) p < a means that ft < aj for j = 1,2,..., n. (viii) a - P is the multi-index (<*i - ft,c*2 - ft,... ,an - ft) whenever P < a. (ix) a! = aila^! * * * oin\. W (p) = (£) (¾) • • • (TJ whenever (xi) The formula £“(/$)=£ (a(1-3) /3<a W is known as Leibniz’s formula. It is a special case of the following more general Leibniz’s formula. (xii) Let P{D) = £w<m aaDa be a linear partial differential operator with constant coefficients, and P(Q its symbol. Then P(D)(fg) = £ 1(^(0)/)(0^), \ti\<m where PM (D) is the linear partial differential operator with symbol P^ (¢) given by ^(0 = (^)(0. (etf*. (xiii) Let /eC“(r). Then n*/T|_Vr (^/)(^/).-(^/) where Ca<i)>a(2),...>a(fc)’s are constants and the sum is taken over all possible multi-indices ..., a^k\ which form a partition of a. The formula (1.4) is valid at all points x in Rn for which f(x) ^ 0. (xiv) Let / be a measurable function on Rn x Rn. Then for 1 < p < oo, {[ \[ f(x>y)dy\ < [ {[ \f(xiy)\pdx\ dy. (i.s) URn I JRn | ) J R» t JR" J This inequality is the well-known Minkowski's inequality in integral form.
4 An Introduction to Pseudo-Differential Operators, 3rd Edition (xv) The inequality |xa| < |*|W (1.6) for all x £ Rn and multi-indices a, which will be used quite often in this book, is an inequality in terms of the absolute value of a real number, the norm of a point in Rn and the length of a multi-index. Its proof is left as an exercise. Exercises 1.1. Find the symbol of each of the following partial differential operators onR2. (i) a2 , a2 dx\ ' dx\ (») 6-6 M ^ + 6 (iv)a§7 + *6 (v) dxi 557+^ d dX2 1.2. For each of the partial differential operators in Exercise 1.1, find the zero set of the symbol, i.e., the set of zeros of the symbol. 1.3. Find the symbol of the partial differential operator P{X'D) = ^+X ldx\ on R2. For each fixed x € R2, find the zero set {£ £ R2 : P(x, £) = 0}. 1.4. What is the analog of Minkowski’s inequality in (1.5) when p = oo? 1.5. Prove inequality (1.6).
Chapter 2 Differentiation of Integrals Depending on Parameters The aim of this chapter is to prove a theorem on how to differentiate an integral depending on parameters in order to justify every interchange of integration and differentiation throughout the book. I hope analysts can find the theorem, or some variant of it, useful. Other criteria can be found on, e.g., p.288 in [Friedman (1971)], pp.94-95 in [Royden (1988)], and p.85 in [Wheeden and Zygmund (1977)]. Theorem 2.1. Let (F,/x) be a measure space and f : Rn xY -* C be a measurable function such that (i) /(x, •) € LX(Y) for all x in Rn, (ii) /(*,?/) E C°°(Rn) for almost all y in Y, (iii) sup^Kn fY \(9%f)(x,y)\dii < oo for all multi-indices a. Then the integral JY f(x,y)dn, as a function of x, is in C°°(Rn) and for all multi-indices f). We begin with a lemma. Lemma 2.2. Let f : W1 x Y -+ C be such that the hypotheses of Theo¬ rem 2.1 are satisfied. Then for any multi-index a and j = 1,2,..., n, the integrals and as functions of Xj, are continuous on R. 5
6 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Using the mean value theorem and hypothesis (iii), we obtain J -..,Xj + h,...,xn,y)\ ~ for all x in Rn, where Ma,j= sup [ \(d%dXjf){x, y) \ dp, xeRnJY and hence the lemma. □ Proof of Theorem 2.1 In view of of the proof of Lemma 2.2, the theorem is valid for the zero multi-index. Suppose that the theorem is valid for any multi-index with length l and let 7 be a multi-index with length i + 1. If we write 7 as /3 4- e, where /? is a multi-index with length l, and € is a multi-index with length one and the only nonzero entry in the jth. position, then, by the fundamental theorem of calculus, Lemma 2.2 and the Fubini theorem, d7 f L f(x,y)dy = 9, J^(d%f)(x,y)dy = £“ \JY{(^f)(xi,---,Xj xn,y) - 1, • • • • • • ,a;„,y)} dy = h™ol,JY {/ = lim ^ / ... ,s,... ,xn,y) dy^ ds = f (%lf)(x,v)dy, a; € M.". Thus, by induction, the proof is complete. From the proof of Theorem 2.1, we obtain the following result. Corollary 2.3. The conclusions of Theorem 2.1 remain valid if hypothesis (iii) is replaced by the hypothesis that, for every multi-index a, the integral /y ($?/)(#> 2/) dfi, as a function of x, is continuous on Rn.
Differentiation of Integrals Depending on Parameters 7 Exercises 2.1. Let / be a bounded function defined on the strip Q = {(xi,x2) : x\ € M, 0 < x2 < 1} in M2 such that for each fixed X\ in M, the function f(xi, •) is measurable on [0,1]. Furthermore, suppose that (9f f)(x\, x2) exists for all nonnegative integers k and all (xi,x2) 6 Q, and for each nonnegative integer k, there exists a positive constant Ck such that 1(^1/)(^1^2)1 < C*, (xi,x2)eQ. Prove that (^1) L ^Xl,X2')dX2 = J (^1/)(^1^2)^2 for all xi € R 2.2. Let / € L1 (M.n) and let g € C°°(lRn) be such that dag 6 L°°(ln) for all multi-indices a. Prove that the function h on K" defined by Hx) = / f(y)g(x - y) dy, x e R\ J R" is in C°°(Rn) and (dah)(x) = [ f(y)(dag)(x — y) dy, x&W1. J Rn 2.3. Let / be the function on R defined by oo /(*) = 53 cne*nx, X € R, n=—oo where ..., c_2, c_i, Co, ci, C2,... are constants such that oo $3 lc"l < °°- n=—oo Prove that, if oo 53 I0"!71* < 00 n=—oo for all nonnegative integers k, then / € ¢70°^) and for k = 0,1,2,..., we have oo /(fc)(*)= 53 Cn«*eina, areR
This page intentionally left blank
Chapter 3 The Convolution In this chapter we introduce two important subsets of C°°{Rn), usually denoted by Cq°(W1) and S. The aim of this chapter is to prove that they are dense in Lp(Rn), 1 < p < oo. To this end, we need the notion of convolution. Theorem 3.1. (Young’s Inequality) Let f £ L1(Wl) and g € Lp(Rn), 1 < p < oo. Then the integral exists for almost every i G In. // the value of the integral is denoted by (/ * 9)(x)i then f *g e Lp(W1) and II/*»IIp<II/IIiIWIp- Remark 3.2. We usually call f * g the convolution of / and g. Proof of Theorem 3.1 For p = 1, let h(x)= [ \f{x-y)\\g(y)\dy. J Rn Then, by Fubini’s theorem, = IMIill/lli- 9
10 An Introduction to Pseudo-Differential Operators, 3rd Edition Hence h(x) < oo for almost every x e Rn. Using Fubini’s theorem again, we have II/* plli =/ / f(x-y)a{y)dy dx JRn \JRn <[ (/ l/(^-y)l|p(p)l%)rfa; JRn \JRn / = (/ 19(y)\dy) ( [ \f(x — y)\ dx\ \JRn / \JRn J = IIpIIiII/IIi- This proves the theorem for p = 1. For 1 < p < oo, let M*) = / 1/(1 - v) I Ip(p)Ip%- «/Rn Then, as in the previous case, hp(x) < oo for almost every x G Rn. Let p' be the conjugate index of p. Then, by Holder’s inequality, f \f(x-y)\\g(y)\dy JRn = / 1/(31 — y)|1/,p" 1/(^ — 2/)|1/p|p(2/)| dy JRn < !/(* - y)l dpj 1/(^ - p)| |p(j/)lp<fyj = ll/lli/p'{M*)}1/p- (3.1) Hence fRn f(x - y)g(y)dy exists for almost every x € Rn. Moreover, by (3.1), we have 11/ * Pllp = JRnf(x~y)9(y)dy| <2xj -(/ (/ l/(x_y)llp(y)lrfp) dx\ Ultn \JRn J J < ll/lli/p <ll/lli/p'{ii/llillG||i}1/p, where G(x) = |ff(x)|,> for all i £ t". Hence ll/*p||P<ll/llilMIP.
The Convolution 11 To prove the theorem for p = oo, note that [ \f(x-y)\\g(y)\dy <\\g\\oo f |/(x-y)|dy = IMIooll/lli. (3.2) JRn JRn Hence the integral fRn f(x — y) g(y) dy exists for every x G Rn. Moreover, by (3.2), we have 11/*0 Hoc < Mlooll/lll- □ Proposition 3.3. (Lp-Continuity of Translations) Let f G Lp(Rn), 1 < p < oo. Then lim ll/x-/llp = °, |aj|—>0 where fx is the function defined by fx{y) = f(x + y), yet". Before proving Proposition 3.3, let us define Co(Rn) to be the set of all continuous functions on Rn with compact support. The support of a continuous function h is defined to be the closure in Rn of the set {x G Rn : h(x) £ 0} and is denoted by supp(h). We give one property of the set Co(Rn) in the following proposition. Proposition 3.4. Co(Rn) is dense in Lp(Rn) for 1 < p < oo. Proposition 3.4 is a measure-theoretic result which I ask you to believe. Otherwise, see p.69 in [Rudin (1987)] for a proof. Proof of Proposition 3.3 Let S > 0 and / € Lp(Rn). Then, by Proposi¬ tion 3.4, there is a function g in Co(Rn) such that 11/-»llp <§• (3.3) Now, using the triangle inequality and (3.3), 6 6 S Wfx — /lip ^ II fx — 9x\\p + \\9x ~~ 9\\p + II# “ /Up <"3"^3"^3=<^ if \x\ is small enough. This completes the proof. Theorem 3.5. Let cp G Ll(Rn) be such that □
12 An Introduction to Pseudo-Differential Operatorst 3rd Edition l/P For e > 0, define the function <p£ by (pe{x) — £~n(f , X G Then for any function f G Lp(Wl)11 < p < oo, we have f *(pe-+ af in Lp{En) as e -* 0. Proof Since I <pe (x) dx — a for all e > 0, it follows from Minkowski’s inequality in integral form given by (1.5) that we have 11/ *ipe- af\\p = (jf ^ 1(/ * <Pe)(x) - af(x)\pdxj = J^n{f(x-y)-f(x)}<Pe{y)dy^ dxj = (j^ |j^{/(* - ey) - dxj </ {/ \f{x - ey) - f(x)\p\<p{y)\pdx\ dy = / \<P(y)\{ [ |/(x - ey) - f(x)\pdx\ dy J Rn UR" J = / kp(»)l 11/-^-/My. (3.4) J R» By Proposition 3.3, ||/_e2, — /||P -»■ 0 as e -¥ 0. Also, by the triangle inequality, \\f-ey — f\\p < 2||/||p. Hence an application of the Lebsegue dominated convergence theorem to the last integral in (3.4) implies that ||/ *<pt- af ||p -)• 0 as £ -¥ 0. □ i/p i i/p We introduce two important function spaces. It is customary to denote by C^E71) the set of all infinitely differentiable functions on Rn with com¬ pact support and by S the set of all infinitely differentiable functions ip on Rn such that for all multi-indices a and /9, sup \xa{D^p)(x)\ < 00. x€Rn
The Convolution 13 The space S is usually called the Schwartz space in deference to Laurent Schwartz. Obviously, C§° (W1) is included in S. That the inclusion is proper can be seen easily by noting that the function e~W is in S but not in Co°(Mn). We want to prove that Co°(En) is dense in Lp(Wl) for 1 < p < oo. To this end, we need two preliminary results. Proposition 3.6. Let ip € S and f € Lp(En), 1 < p < oo. Then En) and d«(f*v) = f*(d«v) for every multi-index a. Proof Let p G S. Then for every multi-index a, dacp e Lp (En), where p' is the conjugate index of p. Hence, by Holder’s inequality, (/ * (dap))(x) exists for every x G En. Therefore differentiation and integration can be interchanged. □ Proposition 3.7. Let f and g be continuous functions on En with compact support. Then the convolution f *g also has compact support. In fact, supp(/ *g)c supp(/) + supp(g). Remark 3.8. Let A and B be subsets of En. Then the vector sum A 4- B is defined by A + B = {x + y:x€A and y € B}. In fact, the vector sum can be defined for any two subsets of a vector space. Proof of Proposition 3.7 Since (/ * 9)(x) =/ f(x - y)g(y) dy, Jnn it follows that if (/ * g){x) ± 0, then there exists a ye supp(^) such that x — y G supp(/). Hence, by Remark 3.8, x e supp(/) + supp(^). This completes the proof. Theorem 3.9. Co°(En) dense in Lp(En) for 1 < p < oo. □
14 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Let (p £ Co°(Rn) be such that I (p(x)dx = 1. JRn Such a function exists by Exercise 3.2. For e > 0, define <p£ by <p£(x) = £~nip , X £ En. Then for all functions g £ Co (Rn), we have, by Propositions 3.6 and 3.7, g*(p£ £ Co°(Mn). Also, by Theorem 3.5, 9*<Pe->9 (3.5) in Lp(Rn) as e -¥ 0. Let S > 0 and / £ Lp(Mn). Then, by Proposition 3.4, there is a function g £ Co(Mn) such that 11/-0llp <§• (3-6) By (3.5), we can find a function ip £ Co°(Rn) such that llfl - V’llp < (3.7) Hence, by the triangle inequality, (3.6) and (3.7), we have 11/ - i>\\p < 11/ - slip + lb - V’llp <\ + \=5- This proves that C7q°(IRT1') is dense in Lp(Rn) for 1 < p < oo. □ Remark 3.10. An immediate consequence of Exercise 3.3 and Theorem 3.9 is that the Schwartz space S is also dense in Lp(Rn) for 1 < p < oo. Exercises 3.1. Let (p and (pe be the functions given in the hypotheses of Theorem 3.5. Let / be a bounded function on Rn which is continuous on an open subset V of Rn. Prove that f *<pe a f uniformly on every compact subset of V as e ->• 0. 3.2. Let ip be the function on Rn defined by Prove that tp £ Co°(Rn).
The Convolution 15 3.3. Prove that every function in S is in Lp(Rn), 1 < p < oo. 3.4. Is S dense in L°°(En)? Explain your answer. 3.5. Determine whether or not the Lp-continuity of translations is true for p = oo. 3.6. Use Minkowski’s inequality in integral form in (1.5) to prove Young’s inequality for 1 < p < oo.
This page intentionally left blank
Chapter 4 The Fourier Transform The Fourier transform will be used in Chapter 6 to define pseudo-differential operators. Two important results in the theory of Fourier transforms are the Fourier inversion formula for Schwartz functions in S and the Plancherel theorem for functions in L2(Rn). They are very useful for the study of pseudo-differential operators. Proof By Theorem 3.1, f *g G L1(Mn). Then, using the definition of the Fourier transform, we have Let / G Lx{IT). We define f by The function / is called the Fourier transform of / and is sometimes denoted by Tf. Proposition 4.1. Let f and g be in L1(Mn). Then f*g = (2n)n'2fg. = (27r)-"/ f f{x-y)g(y)dy]dx. JRn \JRn / 17
18 An Introduction to Pseudo-Differential Operators, 3rd Edition So, by Fubini’s theorem, we get (27:)-^/¾) = (2ir)~n f f e~i<'x~y^'if(x-y)e~iyig(y)dydx Jun JRn = (27r)_n f e~ty ig(y) ( f e~t(x~yHf(x - y) dx\ dy jRn \JRn / = (2tt)-"/2 ([Rne-ivMy)dy) m=mm- □ Proposition 4.2. Let Then (i) (Da(p)A(£) = €a<p(Q for every multi-index a, (ii) (D0<p)(£) = {{—x)^(p)A(^) for every multi-index /3, (iii) (f> € S. Proof Integrating by parts, we get (D»A(0 = (27t)-"/2 [ e-ix<(Day){x)dx J Rn = (2tt)-"/2 f Ce~ix^(x) dx = em- JRn This proves part (i). For part (ii), we have (D*<p) (0 = (27r)~n/2D0 (jf e~ix ^(x)dxj = (27r)-"/2 f (-x)t}e-ix *<p(x)dx= ((-x)0v)A (0- JRn The interchange of the order of differentiation and integration is valid be¬ cause (-x)^(f E 5. To prove part (iii), let a and /? be any two multi-indices. Then, by parts (i) and (ii), m^xoi = r((-*yV)A(o i = k^((-^)V)}a(oi- Since Dot((-x)l3(p) is in 5, hence in L1(Wl), it follows that sup 1^(^0)(01 = sup |{J?°((-*)M}A(0l £6Rn £€R" < (277)-^11^((-0^)111 < 00- □ That the Fourier transform turns functions in Lx(IRn) into continuous functions vanishing at infinity is the content of the following proposition.
The Fourier Transform 19 Proposition 4.3. (The Riemann-Lebesgue Lemma) Let f e L^K71). Then (i) f is continuous on Rn, (ii) limkKoo /(0 = 0, (iii) fj -+ f in L1(Rn) =½ fj f uniformly on Rn. Proof Let fj -+ f in L1(Rn). Then 1^(0 - /COI < - /|U. Hence fj —> f uniformly on Mn. This proves part (iii). To prove parts (i) and (ii), let p € <S. Then, by part (iii) of Proposition 4.2, p £ S. Hence parts (i) and (ii) are satisfied for functions in S. Let / G L1(Rn). Since S is dense in Ll(Rn), it follows that there is a sequence {pj} of functions in S such that Pj -+ f in L1(Rn). By part (iii) which we have proved, p] -+ f uniformly on Rn. This proves parts (i) and (ii). □ Let / be a measurable function defined on Rn. For any fixed y e Mn, we define functions Tyf and Myf by (Tvf)(x) = f(x + 2/), x € Rn, (4.1) and (Myf)(x)=eixvf(x), ier. (4.2) Let a be a nonzero real number. Then we define the function Daf by (Daf)(x) = f(ax), x € Rn. (4.3) Proposition 4.4. Let f e L^E"). Then the functions Tyf,Myf and Daf defined by (4.1), (4.2) and (4.3) respectively are in Ll(Mn). Moreover, (i) (Tvfno = (Mjm ir, (ii) (Myfno = (71,/)(0, e g Rn, (iii) (Daf)A(0 = \a\ n(D1/o/)(^), £ e Kn. Proof Obviously, Tyf,Myf and Daf axe in L1 (Rn). By a simple change
20 An Introduction to Pseudo-Differential Operators, 3rd Edition of variable, we have (Tyf)A(0 = (27T)-"/2 f e~ix't(Tyf)(x)dx JRn = (2tr)-"/2 / e~ix<f{x + y)dx jRn = (2tt)-”/2 / e~^-^f(x)dx = eiy i(^)~n/2 f e~ix<f(x)dx J Rn = (^y/)(0- Also, (Af,/)A(0 = (2tt)-”/2 f e-ix<{Myf)(x)dx JRn = (2tt)"”/2 / e~ix ieiv xf{x)dx JRn = fit - y) = (r_,/)«). Finally, by another change of variable, we have (A./m = (2tt)-"/2 / e-ix<(Daf)(x)dx J Rn = (2tt)~n/2 f e-ix*f(ax)dx J Rn = (2x)“n/2 / e-i(5)'€/(*)|orn<fa = |o|-"(i?1/a/)(0. Proposition 4.5. Le£ </>(#) = e l®l2/2. T/aen <£(£) = e l£l2/2. Proof We first compute (2?r)"n/2 f e~ix't-\Xfdx. J Rn □
The Fourier Transform 21 Note that r i n r°° (2?r)-n/2 / e-ix<-^dx = T\(2n)-^2 e~ix^-xidxj. (4.4) JRn jJl J — OO Hence it is sufficient to compute /OO e~lt<>~t2dkt, (, e (—00,00). -00 But /00 pOO e-it(-t2 (ft = I e-V +^dt -00 J—OO /00 e-(t2+it(-(,c2/i))dt -00 = e"^4 /°° e-(‘+i«/2))2dt J —OO = e~(*2^ f e~z*dz, (4.5) where L is the contour Imz = | in the complex z-plane. Using Cauchy’s integral theorem and the fact that the integrand goes to zero very fast as \z\ -+ 00, we have f e~z2dz= j e~t2 dt = \pK. (4.6) J L J—oo Hence, by (4.5) and (4.6), /oo e-^dt = 2-1^^4. (4.7) -OO By (4.4) and (4.7), we get (27r)~n/2 [ e~ix't-W2dx = 2-n/2e_|(Jl2/4. (4.8) J Rn Now, note that (27r)-"/2 e-to-€-d*l2/2>dx = (D1/y^p)A (¢), where ¢(x) = e"^2. Therefore, by (4.8) and part (iii) of Proposition 4.4, we get ¢(0 = (27r)-n/2 f e-ix-t-W2Mdx = e”^2/2, JRn as asserted □ Proposition 4.6. (The Adjoint Formula) Let f and g be functions in L\Rn). Then / f(x)9(x) dx = / f(x)g(x) dx. JRn JRn (4.9)
22 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof By Proposition 4.3, the Fourier transform of a function in L1(En) is bounded on Rn. Hence the integrals in (4.9) exist. Moreover, f f(x)g{x)dx = (27r)_n/2 [ (f e~lx'vf(y)dy\ g(x)dx JRn JRn \JRn J = (2tt)-"/2 f f(y) ( f e-ix vg(x)dx) dy JRn \J Rn J = f f(y)g(y) dy. JRn The interchange of the order of integration can obviously be justified by Fubini’s theorem. □ We are now prepared to prove the first important result in the theory of the Fourier transform. Theorem 4.7. (The Fourier Inversion Formula) (/)v = f for all func¬ tions f € <S. Here, the operation V is defined by g(x) = (2tt)-”/2 f eix tg(Odt, g&S. J Rn Remark 4.8. The function g is usually called the inverse Fourier trans¬ form of g. Proof of Theorem 4.7 We have (/)v(x) = (2x)-"/2 f eix<mdt. JRn Let e > 0. Define Ie(x) = (2ir)-n/2 f ^-(^1=72)/(^. J R" Let g(0 = e^€-(-2i€l2/2) = (MxDeym, where Then, by Propositions 4.4 and 4.5, g(V) = (T-x£~nDi/e<p)(r)) = £-ne-^2^ (4.10) (4.11) (4.12) (4.13)
The Fourier Transform 23 Hence, by (4.10), (4.11), (4.13) and Proposition 4.6, Is(x) = (2tt)-"/2 f g(0f(0 (% J Rn = (271-)-^/ g(T])f(ri)dri JRn = 6-^(277)-^2 / e-l”-*l2/(2£2>/(77)d77 JRn = (27r)-"/2(/*¥>,)(*), (4.14) where ip£(x) = £_rV ( f) • Since / G 5, it follows that / is in Lp(Rn), 1 < p < oo. Therefore, by (4.12), (4.14) and Theorem 3.5, Ie -► (27r)-n/2 (^ e-l*'2/2^) / = / in Lp(Mn) as e -» 0. Hence there exists a sequence {en} of positive real numbers such that I£n{x) f(x) for almost every x G Rn as en 0. By (4.10) and Lebesgue’s dominated convergence theorem, I'(x)->(27T)-n/2 f e*<f(0dt J Rn for every x G Mn as e -» 0. Hence (2^)-"/2/ eto«/(0 # = /(*) «/Rw for every x € Rn, and the proof is complete. □ Remark 4.9. An immediate consequence of the Fourier inversion formula is that the Fourier transformation / —> f is a one to one mapping of S onto <S. If we define / by /(*) = /(-*)> then the Fourier inversion formula is equivalent to the formula /=/, /€5. The next important result is the Plancherel theorem. Theorem 4.10. (The Plancherel Theorem) The mapping f / de¬ fined on S can he extended uniquely to a unitary operator on L2(Rn).
24 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Using the fact that S is dense in L2(Rn) and the Fourier inversion formula, it is sufficient to prove that ll^lb = IMh, e S. Let *ip be the function defined by ^p(x) = </?(-#), x£Rn. (4.15) Then rp eS and = (2tt)-"/2 / e^^xjdx J Rn = (2tt)“n/2 / eix S^x)dx J Rn = m- (4.16) Thus, by (4.15), (4.16), the adjoint formula and the Fourier inversion for¬ mula, \mi= [ JRn = / J Rn JRn = / v(OV’d) <*£ = / ¥>(-£M-£) J Rn J Rn = / <P(OW)d(, = Ml JRn and this completes the proof. □ Remark 4.11. The Plancherel theorem states that the Fourier transform of a function in L2(Rn) can be defined. If / G L2(Rn), then we shall denote its Fourier transform by / or Tf. The inverse of T : L2(Rn) -¥ L2(Rn) is of course denoted by : L2(Rn) -> L2(Rn). Exercises 4.1. Here is another elegant proof of Proposition 4.5. (i) Let ip be the function defined on R by <p{x) = e-*2/2 , xeK.
The Fourier Transform 25 Let y = <p. Prove that v'(0 + ftK0 = o, sen (ii) Use the result in part (i) to prove that £(S) = e-«2/2, Sen 4.2. Let {ipn} be the sequence of functions defined on R by for all x G R and n = 0,1,2, We call ipn the Hermite function of order n. (i) Prove that for all x e R and n = 0,1,2, — (ii) Prove that = i~n<pn for n = 0,1,2, — 4.3. Prove that if /6 L1(Mn) and / G Ll(Rn), then (/)v = / a.e. 4.4. Find a function / in Ll{R*) such that / is not in L1(En). 4.5. Prove that for all functions / in Lx(Mn), tp0(x) = e *2/2 and <Pn+1(*) = *y>n(*) - ¥>»(*) /(0 = /(-0, 4.6. Prove that for all functions / in L1(Mn), / = /• 4.7. Prove that T : 5 ->> 5 is bijective.
This page intentionally left blank
Chapter 5 Tempered Distributions We only give the rudiments of the theory of tempered distributions in this chapter. More details on this subject will be introduced in later chapters as the need arises. Definition 5.1. A sequence {cpj} of functions in the Schwartz space S is said to converge to zero in S (denoted by (pj —> 0 in S) if for all multi-indices a and /3, we have sup \xa(D0<fij)(x)\ -»• 0 x€Rn as j -¥ oo. Definition 5.2. A linear functional T on S is called a tempered distribution if for any sequence {ipj} of functions in S converging to zero in S, we have T(<Pj) 0 as j —» oo. Definition 5.3. Let / be a measurable function defined on Rn such that l/(*)l / Jr* rdx < OO /r- (1 + M)" for some positive integer N. Then we call / a tempered function. Proposition 5.4. Let f be a tempered function defined on W1. Then the linear functional Tf on S defined by Tf(<p)= [ f(x)ip(x)dx, <p€S, J Rn is a tempered distribution. 27
28 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Let N be a positive integer such that !/(*)! (i + |z|)" dx < oo. Then for all functions ip e <S, the integral fRn f(x)ip(x) dx exists. Indeed, we have f \f(x)\\<p(x)\dx J Rn = X-(iTRF(1 + w)''l,’w|,ix s (/,. (TTSF*) ,s6Tf(1 + W’''w,)l) < “• Let {ifj} be a sequence of functions in S converging to zero in S as j -+ oo. Then obviously, Since sup {(i + IsD^W*)!} ->o. \Tf(<fij)\< [ \f(x)\\<pj(x)\dx J Rn (5.1) (5.2) for all j, it follows from (5.1) and (5.2) that Tf(ipj) -+ 0 as j -+ oo. □ Proposition 5.5. Let f G Lp(En), 1 < p < oo. Then the linear functional Tf on S defined by Tf(>p) = f f(x)<p(x)dx, tpeS, JRn is a tempered distribution. Proof Note that / is a tempered function. See Exercise 5.1. □ Remark 5.6. It is customary to identify the tempered distribution Tf with the function / and to say that such tempered distributions are functions. Definition 5.7. Let T be a tempered distribution. Then the Fourier trans¬ form of T is defined to be the linear functional T on S given by TV) = T(£), tpeS.
Tempered Distributions 29 Proposition 5.8. T is also a tempered distribution. Proof Let {ipj} be a sequence of functions in S converging to zero in S. We need only prove that the sequence (¾} also converges to zero in S. To this end, let a and /3 be any two multi-indices. Then, by Proposition 4.2, sup \C{D09j)(O\ €eRn = sup i£“((-zyVj)A(oi = Sup nx>“((-*yVj)}A(oi < (27r)-"/2||D“((-x)Vi)lli- (5.3) Since (fj —> 0 in S as j oo, it follows that for any positive integer N, we have sup {(1 + 1^)^1(^((-^))(3)1} -*• 0 (5.4) xeRn as j —► oo. Since for any positive integer N greater than n, we have 11(^((-^)11, < sup {(1 + |x|)w|(D“((-x)V))(3)l} /(1 + (5.5) xeRn JRn for all j. Hence, by (4.3), (4.4) and (4.5), we conclude that (pj -> 0 in S as j -¥ oo. □ Theorem 5.9. (The Fourier Inversion Formula) Let T be a tempered distribution. Then T = T, where T is defined by T(<p) = T{<p), <p€S. Proof Let <p € S. Then, by Definition 5.7 and the Fourier inversion formula for S, we have t{<p) = Ttf) = T(<p) = T(0) = T{fp). □
30 An Introduction to Pseudo-Differential Operators, Zrd Edition Exercises 5.1. Prove that any function in Lp(Mn), 1 < p < oo, is a tempered function. 5.2. Let S : S -> C be the mapping defined by S(ip) = <p(0), (p£S. (i) Prove that 5 is a tempered distribution. (ii) Prove that S is not a tempered function. (See Remark 5.6.) 5.3. Let f £ Ll{Wl) and T be the tempered distribution which is equal to /. Prove that T is equal to /. 5.4. Do Exercise 5.3 again for f G L2(Mn). 5.5. Find the Fourier transform of the tempered distribution S defined in Exercise 5.2. 5.6. Prove that there exist positive constants C\ and (¾ such that ¢71(1 + If!) < (1 + |d2)1/2 < C7a(l + |d), £ € K*.
Chapter 6 Symbols, Pseudo-Differential Operators and Asymptotic Expansions In this chapter we give the definition and the most elementary properties of a pseudo-differential operator and its symbol. We begin by recalling that a linear partial differential operator P(x, D) on Rn is given by P(x,D)= £ aa(x)Da, (6.1) | a | <ra where the coefficients aa (#) are functions defined on Rn. If we replace the Da in (6.1) by the monomial in Rn, then we obtain the so-called symbol p(*,o = £ M*)r (6.2) I O'| <771 of the operator (6.1). In order to get another representation of the operator P(x,D), let us take any function ip in S. Then, by (6.1), (6.2), Proposition 4.2 and the Fourier inversion formula for Schwartz functions, we have (P{x,D)tp)(x) = £ aa(x)(Datp)(x) | O'| <777 = aa(x)(D°v)v(x) \a\<m = £ oa(*)(r£)v(s) I a I <777 = £ oa(a:)(27r)-n/2 |a:|<m [ e***Fmdt JRn = (2tx)~n'2 f eix*P{x,t)mdt. JRn (6.3) 31
32 An Introduction to Pseudo-Differential Operators, 3rd Edition So, we have represented the partial differential operator P(x, D) in terms of its symbol by means of the Fourier transform. This represen¬ tation immediately suggests that we can get operators more general than partial differential operators if we replace the symbol P(x, £) by more gen¬ eral symbols <r(x,£), which are no longer polynomials in £. The operators so obtained are called pseudo-differential operators. We shall do this in due course. Meanwhile, we should point out that in order to get a useful and tractable class of operators, it is necessary to impose certain conditions on the functions cr(x,£). Many different sets of conditions have been pro¬ posed, resulting in many different classes of pseudo-differential operators. Our discussion in this book is first restricted to the following class. Definition 6.1. Let m G (—00,00). Then we define Sm to be the set of all functions <t(x,£) in C°° (Rn x Rn) such that for any two multi-indices a and /?, there is a positive constant (?<*,/?, depending on a and P only, for which \(D%Dl*)(x,01 < ¢7^(1 + |f|)TO-M € Kn. We call any function a in UmGRSm a symbol Definition 6.2. Let <7 be a symbol. Then the pseudo-differential operator Ta associated to a is defined by (!»(*) = (27T)-n'2 [ eix<o(x,0<p(0dt, <peS. (6.4) JRn We give some examples. Example 6.3. Let P(x, D) = £|o|<m aa (x)Da be a linear partial differen¬ tial operator on W1. If all the coefficients aa(x) are C°° and have bounded derivatives of all orders, then the polynomial p(x,o= £ |a|<m is in Sm and hence P(x, D) is a pseudo-differential operator. (See formulas (6.3) and (6.4).) Proof Let 7 and S be multi-indices. Then \(D2D*P)(x,0\ < £ <?a,7|3fn |a|<m for all 1, { € ln, where Ca,7 = sup |(L>7aa)(a:)|. *€Rn (6.5)
Symbols, Pseudo-Differential Operators and Asymptotic Expansions 33 (6.6) It can be shown easily that <*<<*, otherwise, for all £ G Mn. The proof of (6.6) is left as an exercise. See Exercise 6.5. Hence, by (1.6), (6.5) and (6.6), \(D2D*P)(x,Q\< £ Ca,75!(“WH«l < <?;,*(!+lar-141 for all x, £ G Mn, where c;,»= E |a|<m V 7 Example 6.4. Let <r(£) = (1 +|£|2)m/2, — oo <m< oo. Then cr G Sm and hence Ta is a pseudo-differential operator. It is sometimes advantageous to denote Ta by (/ - A)m/2, where I is the identity operator and A the Laplacian, i.e., A A 92 A j=i Proof We need only prove that for any m G (—oo, oo) and multi-index /?, there exists a positive constant Cm,p such that K^)(oi <cm^(i+Kir-i« (6.7) for all £ G Rn. (6.7) is obviously true for the zero multi-index. Suppose that (6.7) is true for all m G (—oo, oo) and multi-indices /3 of length at most equal to l. Let 7 be a multi-index of length l +1. Then DT = D^Dj for some j = 1,2, ..., n and some multi-index /3 of length l. Hence |(I>7ff)(OI = \(d%a)(0\ = 1(0^)(01 (6.8) for all £Gln, where r(0=n^(l + Kla)(m/2)_1 for all £ e Mn. Therefore, by Leibniz’s formula, (d^m = m £ (f) (d%)^~s{(l + Ifl2)*"*/2)-1} (6.9) d<0 ' 7
34 An Introduction to Pseudo-Differential Operators, 3rd Edition for all £ G Rn. Hence, by (6.9) and the induction hypothesis, there exists a positive constant Cm,p such that i(0V)(s)i < cm3 £ (f) (i++Kir-2-w+i*i = cJ}(i+Kir-171 for all £ € I71, where cm,0 = cm,0Yl 6</3 Thus, by (6.8) and the principle of mathematical induction, the proof is complete. □ We give two very simple properties of pseudo-differential operators. Proposition 6.5. Let a and r be two symbols such that Ta — Tt. Then a = r. We first prove a lemma. Lemma 6.6. Let f be a continuous tempered function such that Tf(<p) = 0, (6.10) Then f is identically zero on Rn. Proof Without loss of generality, we assume that / is real-valued. Suppose f(xo) 0 for some Xo € Rn. Then there is an open ball B(xo,r) with center xo and radius r on which / is strictly positive (or negative). Choose a nonzero function <po € Co°(Rn) such that <po(aO > 0 for all x 6 Rn and supp(<£o) C B(xo, r). Such a function exists by Exercise 3.2. It is clear that ipo € S and Tf(<Po)= / f{x)cpo(x)dx «/Rn is strictly positive (or negative). This contradicts (6.10). □ Proof of Proposition 6.5 By hypothesis and Definition 6.2, we have [ eix<{<x(x,(,)-t(x, €)}$(£) d£ = 0, <p 6 S. JRn Since T : S —» S is one to one and onto by the Fourier inversion formula, it follows that f eix<{a(x, 0 - t(x, = 0, <p&S. JR”
Symbols, Pseudo-Differential Operators and Asymptotic Expansions 35 Now, for any fixed x € W1, etx ^{a(x, £) - r(x, £)} is a continuous tempered function in the variable £. This follows easily from the definition of a symbol in Definition 6.1. Hence, by Lemma 6.6, e**^{<r(x,4) — r(x,£)} = 0 for all £ £ Rn. Since x is fixed but arbitrary, it follows that <r(a;,£) -t(x,€) =0 for all x, £ £ W1. This proves that a = r. □ Proposition 6.7. Let a be a symbol. Then Ta maps the Schwartz space S into itself. Proof Let <p € S. Then for any two multi-indices a and /?, we need only prove that sup \xa(D^(Ta(p))(x)\ < oo. xeRn But, using integration by parts and Leibniz’s formula, we have x^D^T^ix) = x“(27r)-"/2 f D0{eix<a(x,O}mdt JRn = xa(2n)'~7 = (27T)-n/2 = (2tt)-"/2(-i)|“| f y, = (27T)-n/2(-l)W f £ (f\eixHD^-SDt^)(x,0 Dl(t?m)dt. (6.11) Using (6.11) and the fact that a is a symbol, say a £ Sm, we can find positive constants depending on a, /?,7 and <5 only, such that sup \xa{D0{Taip)){x)\ x€Rn < E E (!) (/ <'+Kir-'*1'1 \DKem)\dt. E r)ceixHDt^)(x, mtw 7</3 ''' /„ E (J)rpfefa-€)(^-v)(*,f)0(Ode (6.12)
36 An Introduction to Pseudo-Differential Operators, 3rd Edition Since ip € <S, it follows from (6.12) that sup \xa(D0(Taip))(x)\ < 00, and hence the proof is complete if we can justify the interchange of the order of differentiation and integration in (6.11). But, using the same argument as in the derivation of (6.12), we see that the last integral in (6.11) is absolutely convergent. This completes the proof. □ Remark 6,8. In general, a pseudo-differential operator does not map Co°(Rn) into Co°(Rn). For a way of showing this, see Exercise 6.4. An important notion in the theory of pseudo-differential operators is the asymptotic expansion of a symbol. Definition 6.9. Let a e Sm. Suppose we can find gj € Sm’, where m = mo > mi > m2 > • • • > rrij -> -oo, j oo, such that N-1 a - £ Oj € Sm" (6.13) j=0 for every positive integer N. Then we call YlJLo °j an asymptotic expansion of a and we write oo a ~ ^crj. j=o An important result in this connection is the following theorem. Theorem 6.10. Let mo > vn\ > m2 > * * • > nij -+ -00 as j 00. Suppose oj e Smj. Then there exists a symbol a G 5m° such that 00 3=0 Moreover, if r is another symbol with the same asymptotic expansion, then 0 — t £ Proof Let if € C°°(En) be such that 0<V>(£)<1> £eEn, ^(0 = o.
Symbols, Pseudo-Differential Operators and Asymptotic Expansions 37 and ¢(0 = 1, 10 >2. That such a function exists will be proved at the end of this chapter. Let (¾} be a sequence of positive numbers such that l > £q > £\ > £2 > * * * > Sj y 0 as j ->• 00. Define the function 0 on Rn x Mn by 00 <r(,x, 0 = ¢(^0^-(*, 0, *, I e Mn. (6.14) 1=0 Note that for each (x0,£o) E ln x f, there exists a neighborhood U of (#o j £0) and a positive integer N such that (*,0 = 0 for all (x,£) G U and j > N. Hence a G C00^71 x Rn). Furthermore, for any e G (0,1] and nonzero multi-index a, ¢(¢0=0, 10 <\, ¢(¢0 = 1, 10 >§, ^{¢(¢0) = e|a,W)(e0 = 0, |0 < -e or |0 > and |afWeO}l<^W, £6®", where Ca = SUp | (^¢)(01- «eRn If i < |£| < |, then e < -||| < Hence, for any nonzero multi-index a, we have i^rwe0}i<^iai(i+ior|o|=ci(i+ior|01, £€»% (6.15) where C'a = Ca4laL It is obvious that (6.15) is also true for the zero multi¬ index a. Now, using (6.15), Leibniz’s formula and the fact that crj G Sm>,
38 An Introduction to Pseudo-Differential Operators, 3rd Edition we can find positive constants Ca 7 and C^7 such that = |0?MeO(i?jto)(*,O}| < 7<a V /7 E + l£l)-,a|+,7|Cj,0,7(1 + KIH-W y<a V E +ior-|a| = CW(1 + I0)_1(1 + I0)mi+1-|a| (6.16) for all x,£ 6 E", where £,>,/? = £7<Q (“JCa^C^,,. Now, we choose et such that Cj,c,,0£j < 2~j (6.17) for all multi-indices a and /3 such that |a +/?| < j. By the definition of ip, we have </>(e,0 = 0 (6.18) whenever 1 + |£| < ej1. Hence, by (6.16), (6.17) and (6.18), 1^-^(^0^,0)1 < 2_i(l + |^|)mi+l-|a| (6.i9) whenever i,(ef and \a + fi\ <j. Now, for any two multi-indices a0 and A), we take j0 so large that jo > |a0 + Po\ and mjo + 1 < mo- Write JO — 1 oo *(*>0 = E ^(ejOo'iCaf.O + E ^0^,0 J=° J=J0 = /(®,0 + J(*,0. (6.20) Since 7(x,0 is a finite sum, it follows that / e Sm». By (6.19), |(£>^°£>f0J)(x,01 < E 1^^(6,0^,0)1 3=30 oo < 52 2_j(! + k|)m»'+1-l“ol 3=30 oo < 52 2_i(i+io)m<>-i“oi 3=30 = 2_-7o+1(1 + |0)«o-l*ol. (6.21)
Symbols, Pseudo-Differential Operators and Asymptotic Expansions 39 So, J is also in Sm°. Hence, by (6.20), <j G iSm°. We need to verify condition (6.13). To do this, we write TV-1 - Y <7j(x,0 3=0 oo TV— 1 = - y j=o j=0 N-1 oo = Y + Y i=o i=TV As before, we can show that oo Y € Smiv. j=N Since - 1 = 0, j < AT - 1, if l£l > , it follows from Exercise 6.3 that TV—1 Y - ftofoO € nm€R5m i=o and consequently TV-1 v-Y°i€SmN- j=o Finally, if r is another symbol such that r ~ then £ gmN for every positive integer N. Since tun —> — oo as iV oo, it follows that ¢7 — r G nmeR5m. This completes the proof of the theorem. □ We have used the following result in the proof of Theorem 6.10. Proposition 6.11. There exists a function ^ G £7°°(En) such that 0 < V>(0 < 1, ^) = 0, iei < l. and m = i, ki>2. TV-1 TV-1 a — r = «■ - S <*i - T~Yai © II © II
40 An Introduction to Pseudo-Differential Operators t 3rd Edition Proof We need only construct a function <£o £ Co°(Mn) suc^ that o<MO<h ^eRn, M0 = h Kl<i, and M0 = o, Kl>2. For then the function ?/> = 1—<po will satisfy all the conditions of Proposition 6.11. To construct <po, let / be any continuous function on Rn such that 0 </(*)<!, t e Rn, and /(*) = !. 1*1 < 2» /(*)= 0. 1*1 >4- Let <p € Co°( Rn) be a real-valued and nonnegative function such that Ifi(s) = 0, \s\ > -. and / J l«l<i <p(s) ds = 1. (6.22) That such a function ip exists is an immediate consequence of Exercise 3.2. Let (p0 = f * (p. Then, by Propositions 3.6 and 3.7, € Co°(Rn). Furthermore, using the location of the supports of / and <p, we see that Mt) = 0, \t\ > 2. Finally, for \t\ < 1, (po(t) = / f{t- s)ip(s)ds = / f(t- s)ip(s)ds. JRn J\«|<4 (6.23) JRn J\s\<\ Since, for \t\ < 1 and \s\ < we have \s - t\ < f and hence f(t — s) = 1. Therefore, by (6.22) and (6.23), Mt) = 1, 1*1 < 1- Again, by (6.22) and (6.23), we can prove easily that 0<<po(*)<l, *eRn. This completes the proof of the proposition. □ Remark 6.12. For another proof of Proposition 6.11, see Exercise 6.6.
Symbols, Pseudo-Differential Operators and Asymptotic Expansions 41 Exercises 6.1. Prove that if a E Smi and r E 5m2, then err E Sm 1+m2. 6.2. Prove that if cr E Sm, then D^D^a E for all multi-indices a and /?. 6.3. Let a be any symbol and p any function in S. Prove that the function r defined by r(x,0 = <r(x,Ov>(0» x’t e E”> is a symbol in nm€R5m. 6.4. Let a be the symbol defined by <r(0 = e-'«l2/2 for all £ € ]Rn. Show that the pseudo-differential operator Ta does not map Cg°(«”) into CjFiW1). 6.5. Let a and 5 be any two multi-indices. Prove that for all £ G R", S < a, otherwise. 6.6. (i) Let p E Co°(E) be such that p(x) > 0 for x E M, p(x) = 0 for x £ [-2, -1], and /oo p(x) dx — 1. -oo Define a function \ on (—oo,0] by X(0=f <p(x)dx, £ € (—oo,0]. Prove that x is infinitely differentiable on (—oo,0), O<X(0<1, £ € (—oo,0], X(0=O, I <-2, and x(0 = i, Ce[-i,o]. (ii) Let p E Cq°(E) be such that p(x) > 0 for x E R, p(x) = 0 for x ¢ [1,2] and
42 An Introduction to Pseudo-Differential Operators, 3rd Edition Define a function x on [0? °°) by x(0 = j <p(x) dx, £ € [0, oo). Prove that \ is infinitely differentiable on (0, oo), O<X(0<1, ¢€[0,00), x(0 = 0, £>2, and x(0 = i, ¢€[0,1]. (iii) Construct a function ip € C°°(Kn) such that 0 < ^(¢) < 1, ¢6^, ¢(0 = 0, 10 <1, and ¢(¢) = 1, 10 >2. 6.7. Let a £ C°°(Rn x W1) and let m £ (—00,00). Prove that a £ Sm if and only if for all multi-indices a and /?, there exist positive constants Ca,p and Ra,p such that \(D%Dfa)(x,Q\ < 0,3(1 + Kir"1*1, 10 > Ra,0-
Chapter 7 A Partition of Unity and Taylor’s Formula It is convenient to devote a chapter to several technical results which will be of particular importance for us in the next two chapters. In Theorem 7.1 we construct a partition of unity. Then we use this partition of unity to decompose a symbol a{x,£) into a family {0* (#,£)} of symbols with compact support in the £ variable. We are able to obtain good estimates on the partial Fourier transforms (with respect to the £ variable) of all the symbols (Jk{x, £). The precise estimates are given in Theorem 7.2. In Theorem 7.3 we prove a multi-dimensional version of Taylor’s formula with integral remainder. We begin with the construction of a partition of unity. Theorem 7.1. There is a sequence of functions in Co°(Mn) such that (i) 0<Wk(£)<l, £ E Rn, fc = 0,1,2,..., (ii) E2Low(f) = i, 4er\ (iii) at each £ E Mn, at least one and at most three of the tp'ks are nonzero, (iv) supp(^o) C {£ € R" : |0 < 2}, (v) supp(^) C {£ € R” : 2k~2 < |0 < 2fc+x}, k = 1,2,..., (vi) for each multi-index a, there is a constant Aa > 0 such that sup KdVjfeHOI <AQ2~kW, k = 0,1,2,.... $€Rn Proof We pick i/io to be any function in Co°(Rn) such that O<*>(0<1, £ G *>(0 = 1, I0<1, and *>(0=o. 10 >2. 43
44 An Introduction to Pseudo-Differential Operators, 3rd Edition For the existence of such a function, see Proposition 6.11 and Remark 6.12. Let ip be any other function in C§°(Mn) such that m = i < i€i < 2, and ¢(0 = 0, |0 < ^ or \0 > 4. The proof that such a function ip exists is left as Exercise 7.1. Now, for k = 1,2,..., define ipk by and then define by oo ¢(0 = £^*(6, ^€Rn- k=o Obviously, we have supp(V>o) c {£ € E” : |£| < 2} and supp(V»fc) c {£ € E" : 2*-2 < |£| < 2*+1} for k = 1,2, For each £ € K", the series defining ¢(0 contains at most three nonzero consecutive terms. This implies that for each 0 £ Rn, there exists a neighborhood U of 0 and a positive integer N such that ipk(0 = 0 for all £ € U and k > N. Hence V £ C°°(Rn). It is easy to see that for each £ £ 1", the series defining ¢(0 contains at least one nonzero term. Hence ¢(0 # 0 for all £ 6 IT. Now, for fc = 0,1,2,..., we define ipk by ¥>*( 0 = ipkiO ¢(0 ’ f € Rn. It is easy to see that <pk £ C{j°(Rn) for k = 0,1,2,... and the sequence {VfcJtLo satisfies the first five conditions. Hence it remains to prove that the last condition is true for <pk, k = 1,2, — To do this, we note that for each multi-index a, we have, by Leibniz’s formula, (ffv*)(0 = E (¾ {«* (¢) ®} (2^) <”>
A Partition of Unity and Taylor's Formula 45 Now, for each multi-index /?, by formula (1.4), d*3 (i)- £ 73^),...,0(1)' ¢/+1 (7.2) 00)+-+/3(0=/3 where ¢73(1) ^(0 is a constant and the sum is taken over all possible multi¬ indices /3(1),..., /?(*) which partition /3. Next, for every multi-index 7, there exists a constant C1 > 0 such that l(97*)(£)l < C72“*W (7.3) for all £ £ supp(<£fe) and k = 0,1,2, Let us assume (7.3) for a moment. Then, by (7.2) and (7.3), there exist positive constants Cp(i),..., Cpw such that (i) tt) - 1^0(0,...,0(01 ^(1)+...+^0)=^ 1 vs/l (7.4) < C£2"*l/J| for all £ G supp(^ft), where = 1^0(0 ,...,0(01 |C^d) ---^0(1)1- /3(1)+...+0(0=0 Hence, by (7.1) and (7.4), there exist positive constants Ca,p such that |(9<V*) (¢)1 < 5] (2)^2-^1(7^/,2-^-1)^ =i4a2-fcl“l 0<a V*/ for all fc = 0,1,2,... and all £ € E", where 0<O! It remains to prove (7.3). To do this, we consider three cases. Case 1: Suppose k = 0. Then for all £ € supp(<po), £ is in supp(^) for some j = 0,1,2 and hence 2 ¢(£) = ^o(0 + 53 ^f(0- 3=1 Therefore (9^)(0 = (9^o)(£) + 53^) (2P1) 2 j=i v y -(i-i)lil
46 An Introduction to Pseudo-Differential Operators, 3rd Edition for all £ € supp(<p0)- Thus, there is a constant C7 > 0 such that |(d7$)(OI < Cy, £ € supp(v>o)- (7.5) Case 2: Suppose k = 1. Then for all £ G supp(v?i), £ is in supp(^) for some j = 0,1,2,3 and hence 3 *(© = *>(© + !>*(©• J=1 Therefore (^¢)(0 = (d^0)(£) + E(»V) (^pr) 2-0-1)^ j=1 V 7 for all £ G supp((pi). Thus, there is a constant C" > 0 such that |(d7*)(£)l < C;[2|T| + 1 + 2-1^1 + 2-2!7'] = c;[22'71 + 2^ + 1 + 2_|7|]2“|71 (7.6) for all £ € supp(v?i). Case 3: Suppose k > 2. Then for all £ € supp(^fe), £ is in supp(V’j) for some j = k - 2, k - 1, k, k + 1, k + 2 and hence ¢(0 = £ *,(©. j=k-2 Therefore A+2 / £ \ (^¢)(0 = £ (0V) (2pr) 2-(j_1)l71 j=fc-2 ' 7 for all f € supp(^fc). Thus, there is a constant C" > 0 such that < C"[2~(k~3^ + 2-^-2^7^ + 2“^-1^7' + 2~k^ + 2”(*+1^7!] = (7"[23I7I + 22'71 + 2^ + 1 + 2"|7|]2"*|7I (7.7) for all £ G supp(<pife). Hence, by (7.5), (7.6) and (7.7), (6.3) follows and the proof of Theorem 7.1 is complete. □ Let a G Sm. For k = 0,1,2,..., we write <r*(s,0 =0^,0^(0 (7.8)
A Partition of Unity and Taylor’s Formula 47 for all x, £ E Rn and Kk(x,z) = (2n)~nV [ e^Or.fldf (7.9) for all i,2GEn, where {(£&} is the partition of unity constructed in Theo¬ rem 7.1. Theorem 7.2. For a// nonnegative integers N, and multi-indices a and /3, there exists a constant A, depending on m, n, N, a and /3 only, such that f \z\N\{d%d?Kk)(x,z)\dz < A2^m+ jRn for all k = 0,1,2, In the proof of Theorem 7.2, we make use of the following inequality given by \z\2N <nN |*>f, z€ir. M=JV (7.10) The proof of (7.10) is left as an exercise. See Exercise 7.2. Proof of Theorem 7.2 Let 7 be any multi-index. Then, by (7.8), (7.9), Plancherel’s theorem, Proposition 4.2, Leibniz’s formula, parts (iv) and (v) of Theorem 7.1, /' JR* zJ(d^d^Kk)(x,z)\2dz =/ Jwk 7'<7 V ' ' 2 di, (7.11) where W0 = {f 6 R" : |f | < 2} (7.12) and Wk = {f € Rn : 2k~2 < |f| < 2*+1} (7.13) for k - 1,2,— Hence, by (7.11), part (vi) of Theorem 7.1 and the fact that €a(d§<r) is a symbol in Sm+M, we get positive constants Ca,/3,y and
48 An Introduction to Pseudo-Differential Operators, 3rd Edition C7>7/ such that f \z1{d^Kk){x,z)\2dz jRn < j (y)cW(l + I^l)m+lal-l7,l^'7i7'2~fcl7-7,l | de- (7.14) Hence, by (7.12), (7.13) and (7.14), f \z'r(dgd?Kk)(ic,z)\2dz J Rn < J (fycajt,Y2(fe+2)(m+l“l-IVI)(77i7(2-fc|7-Vl| dj ^7#<7 _ 2*(2m+2|a|-2|7l) :(i8>y<77)y22(TO+l“HVI) ( (¾ 2fe(n+2m+2| a | - 2171) ^yC77,y22(m+l“l-lV ’}• (7.15) where C„ is the constant with the property that the volume of the ball in Rn with radius r is equal to Cnrn. Let Aa.dn.m.n be defined by 2 n2” jE (y)^./3.7'^,7'22(m+H-|VI)| • Then (7.15) becomes f |Z7(^0“^fc)(a:,2)|2d2<^>%m)n2*(n+2m+2|a|-2|7l). (7-16) J Rn Let N be any nonnegative integer. Then, by (7.10) and (7.16), f \z\2N\(d^Kk)(x,z)\2dz JR" <nN / \^(^Kk)(x,z)\2dz < ^412^n+2m+2lal“2iV)
A Partition of Unity and Taylor’s Formula 49 for all k = 0,1,2,..., where M=N I X/2 -N)k By taking square roots, we get {/ \z\2N\{d^d^Kk){x,z)\2dzy < ,422(("/2)+m+|<* for k = 0,1,2,..., where A2 = A}/2. Now write / kn(^a?^)(*,z)|&= / + / JR" V|z|<2-fc «/|z|>2-fe (7.17) (7.18) '|z|<2-fc «/|z|>2- By (7.17) and the Cauchy-Schwarz inequality, there is a constant A3 > 0, depending on m, n, AT, a and /? only, such that 1/2 dz! >l<2"fc < A32((n/2)+m+lal~N)k2~nk/2 = A32(m+lal~N)k (7.19) f <{[ \zn(d0d?Kk)(x,z)\>dz\1/2 ( / J\z\<2~k U Rn J [j\z\ for k = 0,1,2, Next, by (7.17) and the Cauchy-Schwarz inequality again, there is another constant A4 > 0, depending on m, n, N, a and jS only, such that L |z|>2- < y^_k\z\-2ndz^ < AA2«n/V+m+M-N-n)k |y £ r-2nrn_1drdcr) 7 , where da is the surface measure on the unit sphere 5n_1. Hence 1/2 < A42^n/2)+m+'a*-'/v-n)fe|Sn-1|1/2rc'"1/22nfe/2 = ^2(^+1«!-^)* (7.20) for k = 0,1,2,..., where A5 = ^4|Sn_1|1/2n_1/2 and |5n_1| is the surface area of 5n_1. Hence, by (7.18), (7.19) and (7.20), we complete the proof of Theorem 7.2. □ The following version of Taylor’s formula with integral remainder plays an important role in the study of pseudo-differential operators.
50 An Introduction to Pseudo-Differential Operators, 3rd Edition Theorem 7.3. Let f £ ¢7^(Rn). Then for all positive integers N, /(£ + v) - £ + £ 21 f\i-0)"-Hay)K + »v)<ie |a|<iV |l|=JV 7‘ J° (7.21) for all £, 77 £ Rn. Proof The proof is by induction on N. For N = 1, we need to prove that m+v) = m + E S / V/)«+h m (7.22) 171=17- Jo for all £, 77 £ Rn. To do this, we define a function (p : R C by V?(x) = /½ + xrj), x £ R. (7.23) Then, by the fundamental theorem of calculus, V(l) =^(0)+ [\f(t)dt. (7.24) ./o Hence, by (7.23), (7.24) and the chain rule, we get the formula m+v) = ho+E fvsftit+*•»)»»*. e E"> 7 = 1 70 which is exactly the same as (7.22). We now assume that (7.21) is true for the positive integer N. Then, by the induction hypothesis, we have /<£ + *)- £ a\ = /(1 + 9)- E a|</V+l (dam)„ v (dafm„ a! v ^ ot\ V a\<N |a|=N = N E Hf t (1 - 9)^(^m + 0v)de- E |7|=iV |a|=AT for all f, 7] £ Rn. Now, using the fact that iV [ (1-0)^40 = 1 Jo (7.25)
A Partition of Unity and Taylor’s Formula 51 and the formula (7.21) for N = 1, the formula (7.25) becomes /«+•>>- £ |o|<N+l al = E 0)"_1{(d7/)(£ + Ov) - (dy)(t)}de ItHv7' 70 = £ E {f(^+,/)« + ^) M=JV 7' |«|=l - J |7|=JV Let p = </90 in (7.26). Then /«+■>)- £ 2¾¾° (7.26) M<jv+i a: |7|=JV 1(51=1 r’ * l 0 J (7.27) By interchanging the order of integration in (7.27), we have /({+D- £ |a|<AT+l = £ £^w/V+V)«+OT){/,(i-0)"_,<»}<if> |-y|=iV |tf|=l 7’ ‘ ^ ^ 7+<5 /»l = E E ^r/0 (^/)½+w)(i-p)Ar4p. (7.28) |7|=JV|«S|=1 For all multi-indices 7 and <5 with |7| = N and |5| = 1, we have (7 + &)'■ 7W! = 7-5 + 1, (7.29) where 7 • S is the inner product of 7 and 5. Hence, by (7.28) and (7.29), (dafm. fit+v)~ El |a|<JV+l al -V 7,7+^ r1 = E E (7^(7 • <5+1) jf (9^/)½+^)(1-^. |7|=JV|i|=l (7.30)
52 An Introduction to Pseudo-Differential Operators, 3rd Edition For each multi-index a of length iV + 1, we can write a in the form 7 + 5, where |7| = N and |5| = 1 by picking 7= (<*1,£*2,- 1,...,On) (7.31) and 5 = (0,0,...,1,...,0) (7.32) in which S has 1 in the jth position whenever aj > 1. Then, by (7.31) and (7.32), we have 7 • S = aj — 1 (7.33) whenever ctj > 1. Hence, by (7.30) and (7.33), /(4 + 1)- E |a|<JV+l = E E / (9“/)(£+wX1 |a|=iV+lQj>l a* = (^ + 1) E ^j\dam + (ro)(l-p)Ndp. (7.34) |o|=N+l Therefore (7.21) is also true for the integer N + 1. Hence, by the principle of mathematical induction, the proof of Theorem 7.3 is complete. □ Exercises 7.1. Construct a function ?/> in Co°(l.n) such that 0<lKO<l, £eRn, m = h i<ki<2, and ¢(0=0, |{| < i or |f | > 4. 7.2. For any nonnegative integer AT, prove that Iz\2N<nN £ l*T hl=jv for all z G Mn.
A Partition of Unity and Taylor’s Formula 53 7.3. Let / G C00^71) be such that sup \(daf){x)\ < oo xeRn for all multi-indices a. Prove that for every positive integer AT, there exists a positive constant Cn such that /(*)- £ H<jv (dvm, a! * < x €
This page intentionally left blank
Chapter 8 The Product of Two Pseudo-Differential Operators In this chapter we prove that the product (or composition) of two pseudo¬ differential operators is again a pseudo-differential operator. We also give an asymptotic expansion for the symbol of the product. The main result is the following theorem. Theorem 8.1. Let a G Smi and r G S™2. Then the product TaTr of the pseudo-differential operators Ta and Tr is again a pseudo-differential operator T\, where A is in Sm 1+m2 and has the asymptotic expansion A~£^^(^)(w (8-1) Here (8.1) means that A- £ ^r-(%°)(dST) is a symbol in 5mi+m2-JV for every positive integer N. To motivate the proof, let us begin with an argument to find out how we should proceed. For any function tp in <S, we have (T*k<p)(x) = (2tt)-”/2 f e***akMmdt Jrn for all x G Kn, where and {pk} is the partition of unity constructed in Theorem 7.1. Hence f^T^Xx) = (2Tr)""/2 f eix< (£ ak{x,¢) 1 <p($) ^ *=o ■'R" lfe=o ) = (27r)-"/2/ eix*cr{x,Ov(0<% Jr* 55
56 An Introduction to Pseudo-Differential Operators, 3rd Edition for all x G tn. The interchange of YlkLo and JK« can of course be justified by Fubini’s theorem. Hence oo k=0 and the convergence of the series can be shown to be absolute and uniform for all x e W1. Our goal is to compute the symbol of TaTr. But let us first compute the symbol of T«kTr- You will see why in a minute. Let (p € S. Then, by the definition of a pseudo-differential operator, the Fourier transform, Fubini’s theorem and (7.9), we get {Ta„Tr<p)(x) = (2<Tn/2 [ eix^k(a:,0(TTv)*(Qdt J Rn = (2*)~n [ eix<ak{x,£)\ f e~*y (Tr<p)(y)dy\d£ JRn lJRn J = (27T)-n f (f ei(x-^<ak(x,Odt\ (Trv)(y)dy JRn UR» J = (27r)-"/2 f Kk(x, X - y)(Tr<p)(y) dy (8.2) J Rn for all x e W1. Hence, by the definition of a pseudo-differential operator and Fubini’s theorem again, the formula (8.2) becomes 0^(Tk^rP) (%) = (27r)“n f Kk(x,x-y) ( f eiy^T{y,r})p(ri)dq\dy JRn UR» J = (2tt)~n [ etxr> ( f e~t{x~y'>T>Kk(x,x-y)T{y,T})dy\<p(ri)dr) JRn KJRn J = (277)-^/ eixr)\k{x,Tj)if>{ri)dr] JRn for all x £ Mn, where A*(a?,77) - (2ir)~n/2 f e~t<'x~v'> ”Kk(x,x - y)r(y,i])dy. JRn By a simple change of variable, •^*(x, »7) — (27r)-n/2 f e~tz ,,Kk(x,z)T(x - z,rj)dz (8.3) J R» for all 2, 77 6 Wl. This suggests that (IVT^Xx) = (2t7)-"/2 f eix n(x, Tj)ip{rj) dn
The Product of Two Pseudo-Differential Operators 57 for all iGln, where oo A(®, »?) = £ A* (a:, rj) (8.4) k=0 for all x,rj £ Mn. Hence all we need to do is to show that A(x,rj), defined by (8.3) and (8.4), is a symbol in Smi+m2 an(j satisfies (8.1). Remark 8.2. Had we begun with (TaTTip)(x) instead of (TakTT(p)(x), we would have the divergent integral fRn e^x~y^ t<j(x,£)d£ instead of fRnez(x~y)'t<Jk(x,Qd£ in (8.2). This is why we cut off the symbol cr(x,£) in the £ variable by the partition of unity {<£>&}. Proof of Theorem 8.1 For k = 0,1,2,..., we define A* by Afe(x,0 = (27r)“n/2 f e~tz<Kk{x,z)r(x - z,£)dz (8.5) JRn for all x, £ £ En. Now, by the Taylor’s formula with integral remainder given in Theorem 7.3, we get t(x-z,0= £ (8.6) ImKWi where Rn1(x,z,0 = N1 £ ^r~ A1 -0)Ni~H^t)(x -6z,0<W (8.7) N=/Vi Jo for all x, z,£ £ En. Replacing r(x - z,£) in (8.5) by the right hand side of (8.6), and using (7.8), (7.9), Proposition 4.2 and the Fourier inversion formula, we get A*(*,0 = £ (d^k)(x,0(d^T)(x,0+ (*,0, (8.8) ImK^i where T$(x,£) = (2n)~n/2 [ e iz<Kk(x,z)RNl{x,z,£)dz (8.9) J Rn for all ar, £ £ W1. For any positive integer N, the function A given by (8.4) satisfies A- £ = A_ £ (^W(^)(^r)+ £ (8.10)
58 An Introduction to Pseudo-Differential Operators, Zrd Edition where Ni is any integer larger than N. Obviously, £ ^7-(d^a)(d^T) € Smi+ma-Jv. N<\ti\<Ni Hence, if we can prove that for all multi-indices a and /?, there exists a constant Ca^ > 0 such that \ DaxDl 1 (*,0 l \p\<Ni ^ J < Ca,p(l + |fl)-i+-»-*-l0l (8.11) for all a;,£ G Rn, then we can conclude that A is in Smi+m2 and has an asymptotic expansion given by (8.1). To this end, we first note that, by (7.8), (8.4) and (8.8), A- £ tD^(^amr) = f)T«. (8.12) \p\<Ni k=0 Then for any two multi-indices a and /3, we need to estimate D%D^T$ for all k = 0,1,2, We have the following estimate. Lemma 8.3. For all nonnegative integers M, there exists a positive con- stant such that \(D%DfT$)(x ,01 < Cajt.tiM1 + \Z\)m*-2M2(m'+2M-Nl)k (8.13) for all x, £ G Rn and k = 0,1,2, — Let us assume Lemma 8.3 for a moment. Then for all positive integers iV, and multi-indices a and /3, we can choose a positive integer M such that (1 + |£|)m2-2M < (i + |£|)mi+m2-AM/?l (8.14) for all £ G Rn. With this M fixed, we can choose another positive integer N\ so large that mi+2M-N1< 0. (8.15) By (8.12)-(8.15), D^DflX- ^-^(^)(^))1(^0 W<Ni < Ca,$M,Nl{l + |^|)mi+ra2-Af-|0l2(rai+2W-^)fc k=0 = CaS{ 1 + (¢1)-1+-^-101
The Product of Two Pseudo-Differential Operators 59 for all x,£ G Mn, where oo Cajt = Ca,0,M,NlYi 2(m'+2M-N'^k. □ It remains to prove Lemma 8.3. To this end, we need another lemma. Lemma 8.4. Let iJjVi(x,2,£) be the function given by (8.7). Then for all multi-indices a, /3 and 7, there exists a constant ¢^,/3,7 > 0 such that for all x,Zj£ G Rn. The proof that Lemma 8.4 implies Lemma 8.3 is by Leibniz’s formula, integration by parts, Exercise 4.5 and Theorem 7.2. We leave it as an exercise. See Exercise 8.1. Proof of Lemma 8.4 By (8.7), we have {d%d%RNl)(x,z,£) = ¾ V f\i -O)N'-l(dZ+»d0iT)(X -6z,0d6 (8.16) N=w. Jo for all x,z,£ G Rn. Hence, by (8.16) and Leibniz’s formula, for all G En. So, by Exercise 4.5, (8.17) and the fact that r G Sm2, there exist positive constants C7> and such that 7'<7 (d2c%dfRNl)(x,z,o (arr-y+a+M^T)(a, _ 0Z, (8.17) |(«afi?7Vl)(x,z,OI M=Wi 7' <7 7;<7 7712 —|/31 (1 + KI):
60 An Introduction to Pseudo-Differential Operators, 3rd Edition for all where Ss{G)cvc“Ay'} M=Ni and this completes the proof of Lemma 8.4. □ Exercises 8.1. Prove that Lemma 8.4 implies Lemma 8.3. 8.2. Let P(x,D)= Y1 a«(x)Da \a\<mi and Q(x, D) = Y, ba{x)Da, \ot\<m2 where the aa's and ba’s are in C°° (W1) and all their partial deriva¬ tives are bounded functions on Rn. Compute the symbol of the product P(x,D)Q{x,D) directly. Compare the answer with the symbol obtained by Theorem 8.1. 8.3. Let q € C°°(Rn) be such that sup \(Daq)(x)\ < oo for all multi-indices a. Let a be the symbol defined by o(x, £) = q(x) for all x,f € K". Let r be any other symbol. Use Theorem 8.1 to compute the symbols of the operators TaTr and TTTa. 8.4. Let cr 6 Smi and r € Sm2. Prove that the symbol of the pseudo¬ differential operator TaTr — TTTa is in Smi+m2-1.
Chapter 9 The Formal Adjoint of a Pseudo-Differential Operator We begin with a notation. For any pair of functions ip and ^ in S, we define by Let a be a symbol in Sm and Ta its associated pseudo-differential op¬ erator. Suppose there exists a linear operator T* : S -¥ S such that Then we call T* a formal adjoint of the operator Ta. It is very easy to see that a pseudo-differential operator has at most one formal adjoint. Three problems arise. (1) Does a formal adjoint exist? (2) If it exists, is it a pseudo-differential operator? (3) If it is a pseudo-differential operator, can we find an asymptotic expan¬ sion for its symbol? The aim of this chapter is to prove that the formal adjoint of a pseudo¬ differential operator exists and is a pseudo-differential operator. Moreover, we can obtain a useful asymptotic expansion for the symbol of the formal adjoint. To be more precise, let us prove the following theorem. Theorem 9.1. Let a be a symbol in Sm. Then the formal adjoint of the pseudo-differential operator Ta is again a pseudo-differential operator Tt, where r is a symbol in Sm and has the asymptotic expansion (9.1) (Taip,¢) = (<p,T*il>), ip,i>eS. (9.2) t(*,£) ~ ^ ,?i (^^)(^0- U/, II. ' (9.3) 61
62 An Introduction to Pseudo-Differential Operators, 3r<* Edition Here (9.3) means that m<n v- is a symbol in Sm~N for every positive integer N. Before the proof, let us show how the symbol r can be constructed. For & = 0,1,2,..., we define ak and Kk by (7.8) and (7.9) respectively. Then, by the definition of a formal adjoint, = (<P,T;^) (9.4) for all ip and ^ in S. By (9.1), Proposition 4.4 (ii), Proposition 4.6 and the definition of a pseudo-differential operator, where = f {Takip)(x)il)(x) dx J R» = (2ir)~n/2 f {[ eix nak(x,Ti)<p(Tj) drA ip(x) dx = (27r)_n/2 f if ai{x,y- x)<p(y) dj/| rp(x) dx, JRn \JRn > (9.5) &k(x,y) = (2ir)-n/2 f e~iy'v<7k{x,f?)dt) (9.6) J R« for all x,y € En. Therefore, by (7.9), (9.5), (9.6) and Fubini’s theorem, (^V* VO = (2n)~n/2f if Ok(x,y-x)i>(x)dx\ip(y)dy JR» l JR" J = (2ir)~n/2 f if Kk(x, x - y)ip(x) da:} <p(y) dy (9.7) Jr* ur" ) for all tpiip e S. Therefore, by (9.4) and (9.7), CC*V0(z) = (2ir)_n/2 f ~K^(y,y - x)i>{y) dy (9.8) JR* for all x € En. Hence, applying the Fourier inversion formula to the function tf) on the right hand side of (9.8), Fubini’s theorem and a change of variables,
The Formal Adjoint of a Pseudo-Differential Operator 63 we have (T;k4>)(x) = (2*)-» f Kk(y,y-x)\ I ety'nip(T]) dr] 1 dy JRn KJRn J = (27T)-" [If eiy vK^(y,y - x) dyl ¢(7)) d7) JRn URn ) = (2n)~n f eix'n | f eiv (-y~x)~Kk(y,y — x) dyl ¢(7)) dr) JRn URn J = (2*)"» f etX Tt{[ etTI zKk(x + z,z) dz\^(rj) drj jRn lJRn ) for all x E En. It is clear from (9.9) that T;k=TTk, (9.10) where rk(x,7)) = (2n)~n/2 [ etv'zKk(x + z,z)dz. (9.11) J R» Since oo (2^,^) = ^(2^,^) k=0 for all (p, *0 £ £, it is clear from (9.4) and (9.10) that a good candidate for r is given by oo r(x, v) = n (x, V) (9-12) k=0 for all x, f) e Mn. Hence it remains to prove that r is a symbol of order m with an asymptotic expansion given by (9.3) and = ivM) (9.13) for all </?, ^ £ S. Proof of Theorem 9.1 For fc = 0,1,2,..., define T& by (9.11). Let N\ be any positive integer. Then, by the Taylor’s formula with integral remainder given in Theorem 7.3, Kk(x + z,z) = ^ —x{d^Kk)(x,z) + R^l{x,z), ImKJVi ^ (9.14)
64 An Introduction to Pseudo-Differential Operators, 3rd Edition where, by (7.9), r(n! (z. z) N1 E 7T t(l-O)N'-\d>!Kk)(x + 9z,z)d0 ImNJVi -7° iVi E / (1 — 6)Nl_1 (2,K)~n/2 f eiz Hd2ak)(x + dz,0dZM. y° •/Rn (9.15) Hence, by (7.9), (9.11), (9.14) and Proposition 4.2, n<*,,>= ^ (-ow"~— - ■ l/*l<JVi ^ ;re T$(x, r,) = (2^)-n/2 f e^R^(x, z) dz J Rn all x,r] e En. For any positive integer AT, r_ v ,,, (9.16) (9.17) . £ n\ x V I »\<N ^ = r- E ^rW + E <»•"» iVj is any integer larger than JV. Obviously, E we can prove that for all multi-indices a and /?, there exis Cat& > 0 such that i exists a E {-ddd^A\(x,n) { M<* ^ JJ . aC\ -I- Inh^-^-l^l < ca,0( i + ii?ir-w- Rn, then we can conclude that r € Sm and has an asymptotic iven by (9.3). To this end, we first note that, by (7.8), (9.12) *•- E t^w=E«. I^KJVi k=o (9.19) (9.20)
The Formal Adjoint of a Pseudo-Differential Operator 65 Let a and () be any two multi-indices. Then, by (9.17) and an integration by parts, (DZD0T$)(z,t,) = (27T)-"/2J eir,'zZP (x, z) dz = (1 + M2r*(27r)-"/2 ^ e^(l - A,)* [z* {x,z)}dz, (9.21) where K is any positive integer. Let P(D) = (1 — A)K. Then, by (9.15), Leibniz’s formulas and an integration by parts, (1-A2)*{z^[d{^) (x,z)} r. ___ (3 /*1 = (1 - &z)KNx V —_ / (1-0)^-1(270-^2 M=*. Jo f (_oMe--«(d"+"<Tfc)(z + 0z,() dttdB JRn = JVi s |mI=JVi |«|<2K ' Jo f (—)d£dO, J R“ (9.22) where (.•.) = (-01°1 £ (fV* p](Dze iz i)(Di~pd2+,lak)(x + 9z,0- (9.23) p<6 Let 7 be any multi-index. Then, by (9.22), (9.23), an integration by parts and Leibniz’s formula, (l-A,)*{*/» (DZR\#)(*,*)} fll'-'lpw)-"/2 (***)<», (924) where z7(* * *) is equal to £.-V0M,_ i,w£(2) {DJ (£pDsx~Pd*+,1a)}(x 4- 9z^)(Dy~'r2k)(0 d£. (9.25)
66 An Introduction to Pseudo-Differential Operators, 3rd Edition Using the fact that a is a symbol in Sm, (9.25) and Theorem 7.1 (vi), we can find a constant C > 0, depending on 7',p,<S,a,p but not on k, such that |z7(***)l< / V ¢(1 HH£|)ro+'pHV'2-fch'-vl<i£) (9.26) Jwk ^ where Wo and Wjfe, k = 1,2,..., are given by (7.12) and (7.13) respectively. Let M be any positive integer. Then, by (9.26) and Exercise 7.2, there is another constant C > 0, depending on M, p, 8, a, fjt but not on fc, such that |(* * *)| < C\z\~2A*2fe(m+W-2M+n). (9.27) Hence, by (9.24), (9.27) and Exercise 9.5, there is a constant C > 0, de¬ pending on a, /?, K, M, iVi but not on k, such that \a-Az)K{ze (d:<)(x,Z)}| < i E \z\m+Nr-\s\[2Hm+2K-2M+n)^ (9.28) l|i|<2AT J We choose K so large that (1 + \n\)-2K < (1 + (9.29) for all rj £ Rn. Then we choose M so large, say, equal to M\ that m + 2K - 2M' + n < 0. (9.30) Then we choose Ni so large that / \z\-2M'< E 1 \6\<2K ) dz < 00. (9.31) Hence, by (9.28) and (9.31), there exists a constant C\ > 0, depending on a, /3, K, M', N\ but not on k, such that J |(1 - A2)* {z0 (a;,z)}| dz < <712k(m+Mf-2M'+n>. 'M<i Now, we choose M so large, say, equal to M", that m + 2K- 2M" + n < 0 and (9.32) (9.33) / |z\-2M" | E M|/J|+JVl_|{| 1 dz < oo. |j£|<2JC J (9.34)
The Formal Adjoint of a Pseudo-Differential Operator 67 Hence, by (9.28) and (9.34), there exists a constant C2 > 0, depending on a, /3, K, M", Ni but not on k, such that J |tt - &z)K {z0 (¾¾ (*,2!)} | dz < C2 2k(<m+2K-2M"+n). (9.35) Hence, by (9.21), (9.32) and (9.35), we get another constant C > 0, de¬ pending on a, /3, K, M', M", Ni but not on k, such that < _|_ |7/|2)“K{2*(m+2^-2M,+n) (9.36) Hence, for any two multi-indices a and /3, we have, by (9.20), (9.29), (9.30), (9.33) and (9.36), a constant > 0 for which (9.19) is valid for all x,rj £ Rn. Therefore the function r defined by (9.12) is a symbol in Sm and has an asymptotic expansion given by (9.3). That (9.13) is true should be by now obvious. At any rate, it is a simple consequence of Theorem 7.1 (ii), (7.8), (9.4), (9.10) and (9.12). □ Exercises 9.1. Prove that a pseudo-differential operator has a unique formal adjoint. 9.2. Let a and r be any two symbols. Prove that (t;y = ra and (TaTry = t;t;. 9.3. Let P(x,D) = J]|Q|<maa(x)HQ, where the aa’s are in 0°°^) and all their partial derivatives are bounded functions on Mn. Compute the symbol of the formal adjoint of P(x, D) directly. Compare the answer with the symbol obtained by Theorem 9.1. 9.4. Let a and r be as in Exercise 8.3. Use Theorem 9.1 to compute the symbols of the formal adjoints of Ta + Tr, TaTr and TrTa. 9.5. In deriving (9.28) from (9.24) and (9.27), we use the fact that there exists a positive constant (7, depending on /x, /3 and K only, such that \5\<2K \S\<2 K for all z £ W1. Prove the fact.
This page intentionally left blank
Chapter 10 The Parametrix of an Elliptic Pseudo-Differential Operator Among all pseudo-differential operators there exists a class of operators which come up frequently in applications and are particularly easy to work with. They are called elliptic operators. They are nice because they have approximate inverses (or parametrices) which are also pseudo-differential operators. Our first task is to make these concepts precise. A symbol a in Sm is said to be elliptic if there exist positive constants C and R such that k(*,oi>c(i + i€ir, m>R. Of course, a pseudo-differential operator Ta is said to be elliptic if its symbol is elliptic. Theorem 10.1. Let a be an elliptic symbol in Sm. Then there exists a symbol r in S~m such that TtT„ — I + r (10.1) T„Tt = I+ s, (10.2) where R and S are pseudo-differential operators with symbols in C\keR.Sk, and I is the identity operator. Remark 10.2. In other words, Theorem 10.1 says that if Ta is an elliptic pseudo-differential operator, then it can be inverted modulo some error terms R and S with symbols in n*€R5fe. In the theory of regularity of solutions of partial differential equations, operators with symbols in are called infinitely smoothing and can be neglected. (See Chapter 15 for a discussion of infinitely smoothing operators and regularity theory.) For these reasons, we call Tr an approximate inverse, or more often in the literature, a parametrix of Ta. 69
70 An Introduction to Pseudo-Differential Operators, 3rd Edition We first give a proof of (10.1). The idea is to find a sequence of symbols Tj G S~m~i, j = 0,1,2, Let us assume that this has been done. Then, by Theorem 6.10, there exists a symbol r G S~m such that r ~ Tv Then, by the product formula given in Theorem 8.1, the symbol A of the product TrTa is in 5° and such that A- £ S'" (10-3) |7| <N 7* for every positive integer N. But r ~ Yl^Lo tj imP^es that N-1 T-^Tje s~m~N (10.4) j=0 for every positive integer N. Hence, by (10.3) and (10.4), A" E EW>)e S~N (10-5) ft\<N j=0 for every positive integer N. But we can write £ h\<N 1 j=0 N-l = w + E «=i ncr + + £ l7|+i>AT h\<N, j<N |7| +3=1 3<l {-Z^(dJrj)(d2a). (10.6) To simplify (10.6), we choose rh j = 0,1,2,..., in the following way. Define T0 by To(x,0 ( *te) J o(x£) ’ \o, If I > If I < Rl (10.7) where ip is any function in ¢7^(171) such that ip( f) = 1 for If I - 2R and V’(f) = 0 f°r Ifl < R, and we define rj for / > 1 inductively by |7| +i=l r* n~-{ >70. (10.8)
The Parametrix of an Elliptic Pseudo-Differential Operator 71 Then it can be checked easily that Tj G j = 0,1,2, — (See Exercise 10.7.) Now, by (10.7), r0<7 = 1 for |£| > 2R. The second term on the right hand side of (10.6) vanishes for |£| > 2R by (10.7) and (10.8). As for the third term there, we see easily that € S~N whenever |7| + j > N. Hence, by (10.6), £ 44 -1 € S~N (10.9) \y\<N ' 3=0 for every positive integer N. Thus, by (10.5) and (10.9), A - 1 € S~N for every positive integer N. Hence, if we pick R to be the pseudo¬ differential operator with symbol A — 1, then the proof of (10.1) is complete. By a similar argument, we can find another symbol k in S~m such that TaTK = I + &, (10.10) where R1 is a pseudo-differential operator with symbol in C\keuSk. (See Exercise 10.8.) By (10.1) and (10.10), Tk + RTk = Tt + TtR!. Since RTk and TtR' are pseudo-differential operators with symbols in HjkeRSk, it follows that Tk = Tr + R", (10.11) where R" = TrR! - RTk is another pseudo-differential operator with symbol in C\keuSk. Hence, by (10.10) and (10.11), T'Tt^I + S, where S = R' - TaR”. Since S is a pseudo-differential operator with symbol in flkeuSk, it follows that (10.2) is proved. The following theorem tells us that only elliptic pseudo¬ differential operators have parametrices. Theorem 10.3. Let a G Sm be such that there exists are S~m for which (10.1) or (10.2) is true, where R and S are infinitely smoothing pseudo- differential operators, and I is the identity operator. Then a is elliptic.
72 An Introduction to Pseudo-Differential Operators, 3rd Edition Remark 10.4. If (10.1) (or (10.2)) is true, then we call Tr a left (or right) parametrix of 7^. Thus, a consequence of Theorems 10.1 and 10.3 is that if a pseudo-differential operator Ta has a left (or right) parametrix Tr, then Tt is also a right (or left) parametrix of Ta. Proof of Theorem 10.3 Let us first assume that (10.1) is valid and let r be the symbol of R. Then, by Theorem 8.1, 1 + r - or = <5, (10.12) where S is some symbol in S-1. Since r G S-m, we can find a positive constant C\ such that |r(*,fl| <Ci(l + |£|)-m, x,^P. (10.13) Thus, by (10.12) and (10.13), |1 + r(*,0 - *(*,0| < Ci\o(x,01(1 + IO)-m, *,$€»*, and hence k(*.OI>T(i + |or(i-|*(*,oi-|r(*,oi), (10.14) Since * € S-1, it follows that there is a positive constant Ci such that W*,0l<3»(l + I0)_1, 2,(6 1”. (10.15) Since r € C\k^ftSk, we can find a positive constant C% such that |r(*,OI < ¢3(1 + Kir1. € R”. (10.16) So, by (10.14)-(10.16), k(*.01 > T(1 + K|)ro{l - (Ca + C3)(l +Ifir1}, x,£€Rn. (10.17) Now, let R be any positive number such that 1 - (C-2 + c3)(l + Kl)"1 > 10 > R. (10.18) Then, by (10.17) and (10.18), we get k(x,OI> 2^(1+ I0)ro, I0>-R, and this completes the proof under the assumption that (10.1) is valid. The proof for the case when (10.2) is valid is similar and hence left as an exercise. See Exercise 10.9. □
The Parametrix of an Elliptic Pseudo-Differential Operator 73 Exercises 10.1. (i) Let P(x,D) = J2^<maa(x)Da, where the aa’s are in C00^71) and all their partial derivatives are bounded functions on Rn. We call S|a|=m aot(x)€a principal symbol of P(x, D) and denote it by Pm(xi 0* Prove that P(x, D) is elliptic if and only if there exist positive constants C and R such that \PmM\>C(l + \^\)m, 14 >R. (ii) LetP(D)=E|a|<ma aDa be a linear partial differential operator with constant coefficients. Let Pm(0 be the principal symbol of P{D). Prove that P(D) is elliptic if and only if Pm(0 = o, £ € Rn =» t = 0. 10.2. Let Ta and Tr be elliptic pseudo-differential operators. Prove that the product TaTr is also elliptic. 10.3. Let Ta be an elliptic pseudo-differential operator. Prove that the formal adjoint T* of Ta is also elliptic. 10.4. Prove that any two parametrices of an elliptic pseudo-differential operator differ by an infinitely smoothing operator. 10.5. Is a parametrix of an elliptic pseudo-differential operator elliptic? Explain your answer. 10.6. Prove that, in the proof of Theorem 10.1, (10.3) and (10.4) imply (10.5). 10.7. Let {rj}^ o be the sequence of functions defined by (10.7) and (10.8). Prove that Tj is a symbol in S~m~j for j = 0,1,2, — 10.8. Prove that there exists a symbol k in S~m such that (10.10) is valid. 10.9. Let a G Sm be such that there exists a r G S~m for which (10.2) is true, where S is an infinitely smoothing operator and I is the identity operator. Prove that a is elliptic. 10.10. Find a parametrix of an elliptic linear partial differential operator with constant coefficients on Rn.
This page intentionally left blank
Chapter 11 Zp-Boundedness of Pseudo-Differential Operators Let a be a symbol. Then, by Proposition 6.7, the pseudo-differential opera¬ tor Ta maps the Schwartz space S into S. In fact, the following proposition is true. Proposition 11.1. Ta maps S continuously into S. More precisely, if <Pk 0 in S, then ->• 0 in S as k ^ oo. To prove Proposition 11.1, we need some preliminary results. Lemma 11.2. If ifk -> 0 in S as k -» oo, then ipu -» 0 in Lp(Rn) as k —y oo for 1 < p < oo. Proof If (fk -* 0 in S as k -► oo, then <pk 0 uniformly on En as k -> oo. This proves the lemma for p = oo. So, consider 1 < p < oo. Then for any positive integer N with N > , we have Since (1 + |a?|) Np 6 L1(Rn), it follows from Lebesgue’s dominated conver¬ gence theorem that Lemma 11.3. The Fourier transformation T maps S continuously into S. More precisely, if tpk -> 0 in S as k ->• oo, then cpi —> 0 in S as k -¥ oo. (1+ 1^1^(^)1-^0 uniformly on Rn as k oo. Hence for k large enough, hk(*)l<(i + M)"N, as k -> oo. This completes the proof. □ 75
76 An Introduction to Pseudo-Differential Operators, Zrd Edition Proof Let a and /3 be any two multi-indices. Then, by Proposition 4.2, IH^WkKOI = |{r>“((-*yV)}A(£)l < (27r)-n/2||2?“((-*yv*)||i, 4eE”. (11.1) Since ipk —*■ 0 in S, it follows that Da((—x)^tpk) -4 0 in 5 as k -¥ oo. By Lemma 11.2, ||2)"((—x)&tpk)\\i -*• 0 as k -* oo. By (11.1), sup 1^(2)^)(01 -t 0 £eRn as A; -¥ oo. This proves that ^->Oin<Sasfc^oo. □ Proof of Proposition 11.1 Suppose a G Sm. Then for any two multi¬ indices a and /3, we have, by (6.12), positive constants depending on a, /3,7 and <5 only, such that sup 1^(2^(1^))(^)1 xeRn < |2)f(ov*(0)|<*e ■M+M (11.2) Since -* 0 in <S as k —> oo, it follows easily from Lemma 11.3 that (i+io2)(mHa|+|{|)/20|(rv*(O) -»■ o in S as k oo. Hence, by Lemma 11.2, the integral on the right hand side of (11.2) goes to zero as k -» oo. This proves that Taipk 0 in S as k -* oo. □ The pseudo-differential operator TV, initially defined on the Schwartz space <S, can be extended to a linear mapping defined on the space <S' of tempered distributions. To wit, take a distribution u in S' and define Tau by {Tau){v)=u{T&), <peS, (11.3) where T* is the formal adjoint of Ta introduced in Chapter 9. Proposition 11.4. T'<? is a linear mapping from S' into S'. Proof Let u€ S'. Then for any sequence {<£*} of functions in S converging to zero in 5, we have, by (11.3), (Tau)((pk) = w(T*^), k = 1,2,.... (11.4)
Lp-Boundedness of Pseudo-Differential Operators 77 By Proposition 11.1, T*Jpk 0 in S as k -» oo. Hence, using (11.4) and the fact that u is a tempered distribution, we conclude that (Tffu){ipk) 0 as k -> oo. Hence Tau e S'. □ To enquire whether Ta maps S’ continuously into Sf or not, we need a notion of convergence in S'. Definition 11.5. A sequence of distributions {uk} in S' is said to converge to zero in S' (denoted by Uk -> 0 in S') if Uk{<p) -» 0 as k —> oo for all <peS. Proposition 11.6. Ta maps S' continuously into S'. More precisely, if Uk 0 in S1 as k -» oo, then TaUk 0 in S' as k oo. Proof We need only check Definition 11.5 for the sequence {TaUk}- To do this, let ip e S. Then, using (11.3) and the fact that Uk -+ 0 in 5' as k -¥ oo, (TaUk)(ip) = Uk(T*Jp) -* 0 as k oo. Hence TaUk 0 in 5' as k oo, and the proof is complete.D Let us recall that, by Proposition 5.5 and Remark 5.6, every function / in Lp(Rn) is a tempered distribution. Hence, by Proposition 11.4, Taf is also a tempered distribution. What sort of a tempered distribution is it? We answer this question in the following theorem and Theorem 12.9 in Chapter 12. Theorem 11.7. Let a be a symbol in S°. Then Ta : Lp(Rn) -> Lp(Rn) is a bounded linear operator for 1 < p < oo. The following result plays an important role in our proof of Theorem 11.7. It is a special case of Theorem 2.5 in [Hormander (I960)]. Its proof is outside the scope of this book and hence omitted. Theorem 11.8. Let m € Ck(Rn — {0}), k > be such that there is a positive constant B for which |(0°m)(0| < ¢#0, for all multi-indices a with \a\ < k. Then for 1 < p < oo, there is a positive constant C, depending on p and n only, such that II^VIIp < CB\\tp\\p, veS, where (T<p)(x) = (27r)-"/2 / efa-«m(fl0(0d£, x€l".
78 An Introduction to Pseudo-Differential Operatorsf 3rd Edition Now, we can give a proof of Theorem 11.7. Proof of Theorem 11.7 Let Zn be the set of all n-tuples in Rn with integer coordinates. We write Rn as a union of cubes with disjoint interiors, i.e., Rn = UmeznQm^ where Qm is the cube with center at m, edges of length one and parallel to the coordinate axes. Let Qo be the cube with center at the origin. Let rj be any function in Co°(Rn) such that rj(x) = 1 for all x e Qo- For m € Zn, define am by 0Vn(z,£) = r]{x-m)cr(xy£), x,£eRn. Obviously, T*m — rj(x — m)Ta, and f \(T.<p)(x)\*dx< [ |(Ta„tp)(x)\*dx, <peS. (11.5) JQm JR" Since crm(x,f) has compact support in x, it follows from Theorem 4.7 and Fubini’s theorem that CTrm¥>)(*) = (2ir)~n/2 f eix<om{x, £)<?>(£) d£ J Rn = {2*)~n ( eix<\[ eixXa^(X,0dx\<p(0^ JRn UR" ) = (2ir)~n f eixX | f eix ^(X,Omd^}d\, (11.6) JRn URn ) where <^(A,£) = (2tt)-"/2 f e-iX x<rm(x,Z) dx, A, £ € Mn. JRn Lemma 11.9. For all multi-indices a and positive integers N, there is a positive constant Ca,s, depending on a and N only, such that |(£>^)(A,£)| < CaMl + KI)-|o|(l + |A|)-W, A,£ € 1". The proof of Lemma 11.9, though easy, will be given later. This lemma and Theorem 11.8 imply that the operator p i-» T\ip, defined on S by (Tx<p)(x) = (27r)-"/2 f (A, £)*>(£) d£, (11.7) JRn can be extended to a bounded linear operator on Lp(Rn). Moreover, for any positive integer N, there is a positive constant Cm such that \\txAp<cn(i + 1^)-^11^, cpes. (11.8)
Lp-Boundedness of Pseudo-Differential Operators 79 Using (11.6)-(11.8) and Minkowski’s inequality in integral form, \\T*m>p\\P = (2tr)-"/2 {j(_ dAp*} ^ < (2tt)-"/2 /(/ 1(^)(3:)1^) JR™ J = (27r)-"/2/ IMIpdA JRn < ^(277)-^2 (jT(l + |A|)-^a} |M|p, ^e5. By choosing N sufficiently large, we can get another positive constant Cn such that l|IVmv>||P < CnIMIp, ip € 5. (11.9) Hence, by (11.5) and (11.9), / \(Tatp)(x)\pdx < C'^llyllp, <p€S. (11.10) J Qm Now, we represent Ta as a singular integral operator. Precisely, we have Lemma 11.10. Let K(x,z) = (27r)”n/2 f elz’^a{x^)d£t J Rn in the distribution sense. Then (i) for each fixed x G Rn, K(x, •) is a function defined on W1 — {0}, (ii) for each sufficiently large positive integer N, there is a positive constant Cn such that \K(x,z)\<Cn\z\-n, z? 0, (iii) for each fixed x G Mn and ip G S vanishing on a neighborhood of x, (Taip)(x) = (27r)_n/2 f K(x,x- z)ip(z) dz. J Rn Let us assume Lemma 11.10 for a moment. Let Qbe the double of Qm, i.e., Qm the same center as Qm and edges parallel to the coordinate axes and twice the edge length of Qm. Let Q*m be another cube concentric with Qm and Q™ such that Qm C Q*m C Qm- Furthermore, we assume that there is a positive number S such that \x — z\ > 6 for all x G Qm
80 An Introduction to Pseudo-Differential Operators, 3rd Edition and z € Rn — Q*m. The geometry is illustrated by the following figure. Let ^ € Cq°(Rn) be such that 0 < ip(x) <1, xEf, supp(^) c Q” and ip(x) — l on a neighborhood of Write p = pi + P2, where ipi = 'tpp, and P2 = (1 - Then Tap = Taipi +Ta(p2- Write | (Ta<p)(x)\rdx \(Tap2)(x)\Pdx. and
Lp-Boundedness of Pseudo-Differential Operators 81 Then for any sufficiently large positive integer N, inequality (11.10) implies that there is a positive constant Cn such that Im= f \(Ta<pi)(X) + (Ta<p2){x)\*dx JQm <2p [ \(Ta(fi)(x)\pdx + 2PJm JQm <2pC&|Mp + 2pJm. (11.11) By Lemma 11.10, there is a positive constant C2n such that for all x € Qm, I K(x,x — z)ip2(z)dz J]R" I K(x,x — z)ip2(z)dz jRn-Q^ < C2N f \x- z\~2N\ip2(z)\dz. JRn-Q* |(7>2)(s)| = (27T)-”/2 = (2tt)-"/2 (11.12) Let A > y/n + 1. Then there exists a positive constant C\yN, depending on A and N only, such that \x-z\~™ _ (A + |s - z\)2N ^ „ (A + |x - z\)~2N \x - z\™ ~ K ,N (11.13) for all x 6 Qm and z6l“- QHence, by (11.12) and (11.13), m<p2)(x)\ < C2nC\,n f (A + |ar - z\)~2N\<p2{z)\dz, x G Qm. (11.14) Next, we note that for all x £ Qm and z £ Mn — Q*m, \+\x-z\ = \ + \x-m + m-z\ > A + |m — z\ — |a: — m\ “ (A ~ + ~ z\ >H+\m-z\, (11.15) where ^ = ^ + 1. By (11.14) and (11.15), m<P2)(x)\ <c,NcK„f jR*-Q-m (» + \m-z\)N
82 An Introduction to Pseudo-Differential Operators, 3rd Edition By Minkowski’s inequality in integral form and Holder’s inequality, (jf |(r^2)(*)|Pd*y/P (// + \x-z\)~N\<p2(z)\ 'P ' /P <C2NCxN\f / [jQm |./R”- < C2NCx,N f (/ JUL'-O' Uq -dz p \ 1 dx > = C2nC\n X»-Qi Q,m (ji + \m- z\r (fi+\x-z\)-Np\Mz)\Pjy/P Q-m UQm ip + \m- z\)NP |ya(-g)l Q.m (ii + \m-z\)N \JQ v dx > dz {y* (/x+|x-2;|) dz < C2nC\,n If (fi + \m- z\) [JRn-Q'm J l<P2(z)|P 1/P r dz \JwL"-Q'm + z\)np/2 j Hence for any sufficiently large positive integer iV, there is a positive con¬ stant C\,N,p, depending on A, N and p only, such that \<P2(Z)\P Jm~Cx’N’PL-Ql,^+\m-z\)^ dz. (11.16) By (11.11) and (11.16), Summing over all m in Zn, we get a positive constant <7, depending only on n,p, N and A, such that ) dx f l(7»(*)|P J Rn < 2pC% £ f \>p(x)\pdx meZn ^ + 2PC\, ,jv.p 5Z / jr»-c l¥’2(z)|P m6z„ (// + |m - zD^p/2 < c/r_ w.)p*+zi_Q_ =CLlvl:,)rd*+2'CA,iv-’)"p,td‘' dz (11.17)
Lp -Boundedness of Pseudo-Differential Operators 83 But using the same argument as in the derivation of (11.15), we get // + \m - z\ > 1+ \m — l\ (11.18) for all z € Qi and l ± m. By (11.18), Y Y f dz Jk-l&h (/t + |m_zl)Np/2 =,£lteW^.5.a^ = .5 (11.19) Hence, by (11.17) and (11.19), L i»4 s {c+ro»-|.[i+M^} L i^r^- Since S is dense in Lp(Rn) by Remark 3.10, it follows that Ta can be extended to a bounded linear operator on Lp(Rn). □ Remark 11.11. We leave it as an exercise to prove that the bounded extension coincides with Ta : S' S' restricted to the space Lp(Rn). See Exercise 11.2. We now come to the proofs of Lemmas 11.9 and 11.10.
84 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof of Lemma 11.9 Let /? be an arbitrary multi-index. Then, by integration by parts and Leibniz’s formula, = 2n)~n'2 f e-ix X*m(x,Odx J Rn = (—iA)^D?(27r)_n/2 f e~tx Xri(x — m)(r(x1 £) dx J R* = D%(2ir)~n/2 f - m)a(x,£) dx Jun = (2tt)-"/2 f {d^e~ixX}V(x - m)(D%a)(x,0 dx JRn = (—1)1^1 (2it)~n/2 f e-ix Xd%{n(x -m)(D%a)(x,€)}dx JRn = (-1)1^(2^)-^2 £ (13) f e~ix X(d2n)(x - m)(dt'lD%<j)(x, £) dx. ££ W Jk* Using the properties of 7} and the fact that a E S°, we can find a positive constant C^,/?, depending on a and /3 only, such that |(-iA)^(^)(A,OI < CaA 1 + |f|)-l“l, A,* e IT. The lemma follows easily from this estimate. □ Proof of Lemma 11.10 Let a be an arbitrary multi-index with length greater than n. Then (~iz)aK(x,z) = (27r)-"/2 f e*x(8?a){x,$dt (11.20) J Rn in the distribution sense. Since a E 5°, it follows from (11.20) and Propo¬ sition 4.3 that (iz)aK(x, z) is a continuous function on W1 and there is a positive constant Ca such that \z*\\K{z,z)\<Ca for all xy z E Rn. Hence part (i) follows immediately and part (ii) follows if we use the inequality in Exercise 7.2. To prove part (iii), we define the tempered distribution Lx by Lz(il>)= [ <r(s.0^(£)d£> 1>eS. J R"
Lp-Boundedness of Pseudo-Differential Operators 85 Then, by the definition of a pseudo-differential operator, Proposition 4.4 and the definition of the Fourier transform of a tempered distribution, M(x) = (27r)-"/2 [ J Rn = (2n)-n'2Lx(Mx<p) = (2tt)-”/2L*((T^)a) = (2w)-n'2Lx(Txlp). (11.21) By part (i), Lx(ip) = f K(x,-z)i>(z)dz (11.22) J R" for all ip e S vanishing on a neighborhood of the origin. Hence, by (11.21) and (11.22), (Taip)(x) = (27r) n/2 f K(x,-z)(Tx(p){z)dz JRn = (27r)"n/2 f K{pc, -z)(p(x + z) dz J Rn = (27r)”n/2 f K(x,x — z)(p(z) dz J Rn and the proof is complete. Remark 11.12. The proof of the Lp-boundedness of pseudo-differential operators for 1 < p < oo given in Theorem 11.7 is based on the Hormander multiplier theorem, which is formulated without proof as Theorem 11.8. All proofs of the alluded Lp-boundedness invoke some results and concepts that go beyond the elementary nature of this book. For L2-boundedness, we can provide a complete and self-contained proof. Indeed, for all positive integers N, we obtain by Lemma 11.9 a positive constant Cjv such that \^(\0\<Cn(1 + \\\)-N, A,|eRn. By (11.7) and the Plancherel theorem, we get for all positive integers N, a positice constant Cn such that ||I>||2 = ||5£(A, < CN(l + lAD-^lh = CN{ 1 + |A|)-"|M|a, which is (11.8) for p = 2. The use of Hormander’s multiplier theorem is thus avoided.
86 An Introduction to Pseudo-Differential Operators, 3rd Edition Exercises 11.1. Prove that the definition of Tau given for a tempered distribution u given in (11.4) coincides with the definition of Tau for a Schwartz function u given in (6.4). 11.2. Let a G S° and 1 < p < oo. Then we have shown in the proof of Theorem 10.7 that there is a positive constant C such that \\TMp<C\Mp, <p€S. Hence Ta can be extended to a unique bounded linear operator from Lp(Rn) into Lp(Rn). Prove that the extension coincides with Ta : S' S' re¬ stricted to Lp(Rn). 11.3. Let cr(x, ¢) G 5° be a nonzero symbol which is independent of f G I71. Prove that the bounded linear operator Ta : LP(Rn) -¥ Lp(Rn), 1 < p < oo, is not compact. 11.4. Show that the limit in operator norm of a sequence of pseudo¬ differential operators Tak : L2(Rn) L2(Rn), where crG 5°, need not be a pseudo-differential operator. 11.5. Suppose that a G 5° has compact support in x. Prove that the pseudo-differential operator Ta : Lp(Rn) —» Lp(Rn) is a bounded linear operator for 1 < p < oo.
Chapter 12 The Sobolev Spaces Hs,p, —oo < s < oo, 1 < p < oo Theorem 11.7 tells us that Ta : Lp(Rn) -¥ Lp(Rn) is a bounded linear operator for 1 < p < oo if a is a symbol in 5°. In order to find an analog of Theorem 11.7 for an arbitrary symbol in Sm, we need to introduce a family of spaces of tempered distributions. For — oo < s < oo, we denote by Js the pseudo-differential operator of which the symbol crs(£) is given by M« = (i+i£i2rs/2, £€Rn. It should be noted that the symbol of Js is in S~s. (See Example 6.4.) The operator Js is often called the Bessel potential of order s. It is an easy exercise to prove that for any u £ <S', the product asu of a3 and u defined by {<Jsu){ip) =u{a8(p), (p G«S, is also in 5'. See Exercise 12.1. The following proposition is an easy consequence of (11.3) and the def¬ inition of Js, and the proof is left as an exercise. See Exercise 12.2. Proposition 12.1. Jsu = T^OsTu, u £ S'. An easy corollary of Proposition 12.1 is the following proposition. Its proof is also left as an exercise. See Exercise 12.3. Proposition 12.2. Let u € S'. Then (i) JsJtU = Jg-j-tti, (ii) J0u = u. For —oo < s < oo and 1 < p < oo, we define Hs'p to be the set of all tempered distributions u for which J-8u is a function in Lp(Rn). It is 87
88 An Introduction to Pseudo-Differential Operators, 3rd Edition obvious that H8,p is a vector space. It can be made into a normed vector space if we equip it with the norm || ||s>p, where |HU,P = II j-su\\p, u G Hs'p. We usually call H8'p the Lp-Sobolev space of order s. It is obvious that tf°’p = Lp{ Rn). Theorem 12.3. H8iP is a Banach space with respect to the norm || ||s>p. Proof We need only prove completeness. To do this, let {u&} be a Cauchy sequence in Hs p. Then, by the definition of Hs,p, the sequence {J-sUk} is a Cauchy sequence in Lp(Rn). Since Lp(En) is complete, it follows that there exists a function u in Lp(Rn) such that J-8uk -» u (12.1) in Lp(Rn) as k -> oo. Let v = J8u. Then, by Proposition 12.2, = u. Hence J-8v is in Lp(Rn), i.e., v G H8tP. That Uk -> v in H8,p as k -4 oo is an immediate consequence of (12.1). □ Proposition 12.4. Jt is an isometry of HStP onto H8+t'p. More precisely, \\Jtu\\8+t,p = ||u||alP, U G H8'p. (12.2) Proof Let u G H8’p. Then, by Proposition 12.2, ||*7t^||s+t,p = II J-s-tJt'U’Wp = \\J— su\\p = ||^||«,p* Let v G £P+t’p. Then, by Proposition 12.2, J-tv G H8'p and JtJ-tv = v. This proves that Jt is onto. □ Theorem 12.5. Let 1 < p < oo and s <t. Then HUp C Hs'p, and IMUp — lklkp> u ^ Http, Theorem 12.5 is usually known as the Sobolev embedding theorem. To prove Theorem 12.5, we use a technical result which gives us an explicit formula for the inverse Fourier transform of the function (1 4- |£|2)_s/2, where £ G W1 and s > 0, in the distribution sense. Proposition 12.6. Let s > 0. If we define the function G, GJx) = - f e“r/2e"|x,2/(2rV~(n_5)/2 — 2*/2r (f) Jo then (i) Gs G L1(Wl)} (ii) lIG.Hx = (27r)n/2, (iii) G;(0 = (l + |^|2)“s/2, on E" by x € En, ^ € 1”.
The Sobolev Spaces H5’p, -oo < s < oo, 1 < p < oo 89 Proof By Fubini’s theorem, I \Gs(x)\dx= I Gs(x)dx JRn JRn _ (27r)n/2 f°° /2 /2 dr 2s/2r (4)/0 7 (12.3) if we recall that (12.4) (12.5) [ e-^^dx = (27rr)”/2. JRn But for any e > 0 and a > 0, we have e-°r(o)= f°° e~erra—. Jo r Putting e = \ and a = | in (12.4), we get jf°e-r/V/2y =2*/2r (I). Hence, by (12.3) and (12.5), / |G»(x)|dx = (2ir)"/2. JRn This proves parts (i) and (ii). To prove part (iii), let <p e S. Then, by Proposition 4.6, [ Gs(0<p(t)d$= f GsiOviOdt JRn jRn - I [ f r e-r/\-\i\VWr-(n-W-}md$. 2*/2r(f)yRnU ri Using Fubini’s theorem, we get «/Rn - 1 r°e-r/2 -(n~«)/2 / f (12.6) " 2*/2r (§) /„ Ur- J r But, by Proposition 4.6 again, we get J <£(£)e_l€|a/(2r)# = V»(€)^(0 <*£> (12.7)
90 An Introduction to Pseudo-Differential Operatorsf 3rd Edition where (12.8) V>(a;) = e-ll|2/(2r), x <E f. By Propositions 4.4 (iii) and 4.5, $(0 =rn/2e-r|«l2/2, £eRn. Hence, by (12.7), f 0(0e"l€l2/(2r)df = W2 / y>(0e-r|£|2/2d$. «/Rn ./Rn Therefore, by (12.6) and (12.8), L S<<M0 « = 57½) f {/,. * Using Fubini’s theorem again, we get [ GsitMOdti J Rn = 57¾ Putting e = (1 + |£|2)/2 and a = § in (12.4), we get J G.MrtOdt = f (i + KIT'/2¥>(0«- ^Rn Hence, by Lemma 6.6 and (12.9), we can conclude that Gs(0 = (1 + |£|2)-s/2, £€Kn. (12.9) The following consequence of Proposition 12.6 will be useful to us. Proposition 12.7. Let s > 0 and 1 < p < oo. Then II J.»||p < IMIr, « e Lp(Rn). (12.10)
The Sobolev Spaces if3,p, —oo <s<oo,l<p<oo 91 Proof Let ip G S. Then, by the definition of Js, (^)A(0 = (i + kl2)"s/V(a On the other hand, by Propositions 4.1 and 12.6, (g. *<pno = (2*)n/2Gs(om = (2^(1 + 1^)-^(0, C€En. Hence for all (p G 5, JS(P = (27T )~n/2(Gs*if), and, by Theorem 3.1, \\JM\p < (2^)-^110.1^ IMIP = IMIp. Since S is dense in Lp(Rn) by Remark 3.10, it follows that Js can be extended to a bounded linear operator on Lp(Rn) satisfying (12.10). □ Remark 12.8. As in Remark 11.11, we leave it as an exercise to prove that the bounded extension coincides with Js : S' -> S' restricted to the space Lp(Rn). Proof of Theorem 12.5 Let u G Ht,p. Then, by the definition of HtiP, we have J-tu G Lp(Rn). Hence, by Proposition 12.2, J-Su = Jt-8J-tu• By the definition of H9'p and Proposition 12.7, we get llwlls,p = ||«/— $u\\p = ||«/it—s«/—$u||p < ||t7_iix||p = ||u||t,p, and hence Theorem 12.5 follows. □ We can now give a more precise result generalizing Theorem 11.7. Theorem 12.9. Let a be a symbol in Sm. Then Ta : Hs'p H8~m'p is a bounded linear operator for —oo<s<oo and 1 < p < oo. Proof Since Jm-gT^Js G 5°, it follows from Theorem 11.7 that there is a positive constant C such that llTfl-uHs—jT^p = ||«7m—= \\Jm—sTffJsJ~-su\\p < C\\j-8u\\p = C\\u\\s,p, u G H8'p. □ Remark 12.10. A weaker result than the Sobolev embedding theorem formulated in Theorem 12.5 is one that asserts the existence of a positive constant C such that IMkp < CIMkp, n € H
92 An Introduction to Pseudo-Differential Operatorst 3rd Edition where 1 < p < oo. The weaker inequality has a much shorter proof. Indeed, for all u G H IMU,p = ll*^-swllp= \\Jt—sJ—t^l Ip* Since s < £, Jt-S is a pseudo-differential operator with symbol in 5°. So, by Theorem 11.7, there exists a positive constant C such that ||= IIJt-sJ-tu\\p < C\\J-M\P = CIMkp, u € H Thus, IMkp < ll«lkp, « € HtlP. Notwithstanding the apparently shorter proof, the Lp-boundedness of pseudo-differential operators is invoked. So, the proof of Theorem 12.5 as given in this chapter is self-contained and elementary, and it gives a better result. Exercises 12.1. Let s G (—oo, oo) and as be the function on Rn defined by MO = (i+iei2rs/2> £ € Mn. Prove that if u G S', then the product asu of as and u defined by (cr8u)(ip) = u(as(p), ip G <S, is also in <S'. 12.2. Prove that for all s G (—oo, oo), Jsu = T~lasTu, u G <S', where as is the function on Rn defined in Exercise 12.1. 12.3. (i) Prove that for all s,t G (—00,00), JsJt'U’ — Js-\-t'U,i w G <S . (ii) Prove that Jqu = u for all u G S'. 12.4. (i) Prove that for all s G (—00,00),
The Sobolev Spaces Hs'p, — oo <s<oo, l<p<oo 93 (ii) Prove that for all s £ (-00,00), IW* = y (1 + iei2)sl«(0l2rfe}1/2, t* € h°*. 12.5. (i) Prove that if u £ if5’2, s > § + fc, where k is a nonnegative integer, then u is equal to a Ck function on Rn almost everywhere. (ii) Prove that if u £ i/5’2, s > then u can be modified on a set of measure zero to a continuous function v on Rn such that lim v(x) = 0. |x|—>00 12.6. (Erhling’s Inequality) Prove that if s < t, then for any positive number e, there exists a positive constant C, depending on e, s and t only, such that IMI«,2 < e|Mlt,2 + CIMIo.2, € s. 12.7. Let s > 0 and 1 < p < 00. Then we have shown in the proof of Proposition 12.7 that II JsvWp < IMIp, v e 5. Hence Js can be extended to a unique bounded linear operator from Lp(Rn) into Lp(Rn). Prove that the extension coincides with J8 : S' -¥ S' restricted to Lp(Rn). 12.8. Let cr be an elliptic symbol in Sm. Let u € Lp(Rn) be a solution of the pseudo-differential equation Tau = /, where / 6 Lp( Rn). Prove that u £ Hm'p. 12.9. Let a £ Sm, m > 0, and / £ Lp(Rn), 1 < p < 00. An approximate solution of the pseudo-differential equation Tau = / on Rn is a function u in such that Tau - f modulo i.e., Tau - f £ r\seuHs'p. Prove that an approximate solution exists if a is elliptic. 12.10. Find all real numbers s such that <5 £ H8'2. 12.11. Let s £ (—00,00). Find Js5. 12.12. Let s > 0. Find a solution u in Lx(Rn) such that (/ - A)8f2u = S on Rn, where (/ - A)5/2 is the operator introduced in Example 6.4. (u is termed a fundamental solution of the operator (/ — A)5/2.) 12.13. Is DseRH8'2 = 5? Explain your answer.
94 An Introduction to Pseudo-Differential Operators, 3rd Edition 12.14. For —oo < s < oo and 1 < p < oo, let u G Hs,p and v G H~s'p'. Let {<^j} and {^j} be sequences in S such that ipj u in JTS’P and ^ t; in H~s,p' as j -¥ oo. (i) Prove that \imj^oo((pj^j) exists and the limit is independent of the choice of the sequences {ipj} and {^}. (The limit is denoted by (u,v).) (ii) Prove that \{u,v)\ < IM|s,p|M|_s,p/, u G H°>p, v G H~s’p'.
Chapter 13 Closed Linear Operators In this chapter we give a brief account of the theory of closed linear oper¬ ators on Banach spaces. The choice of topics is dictated by what we need for the theory of minimal and maximal pseudo-differential operators in the next chapter. Let X and Y be complex Banach spaces with norms denoted by || ||x and || ||y respectively. We are interested in linear operators A mapping a dense subspace of X, usually denoted by V{A), into Y. We call V(A) the domain of the operator A. Definition 13.1. The operator A is said to be closed if for any sequence {xk} of vectors in V{A) such that Xk ->■ x in X and Axk —> y in Y as k -¥ oo, we have x £ T>(A) and Ax = y. Definition 13.2. The operator A is said to be closable if for any sequence {xk} of vectors in V(A) such that Xk ->* 0 in X and Axk y in Y as k —► oo, we have y = 0. Obviously, a closed linear operator is closable. Definition 13.3. Let A and B be linear operators from X into Y with domains V(A) and V(B) respectively. We call B an extension of A if V(A) C V(B) and Bx = Ax for all x e P(A). Proposition 13.4. Let A be any linear operator from X into Y with do¬ main V{A). Then A has a closed extension if and only if A is closable. Proof Let B be a closed extension of A. Let {#&} be a sequence of vectors in V(A) such that ->* 0 in X and Axk y in Y as k -)- oo. Since B is an extension of A, it follows that Xk € V(B),Xk 0 in X and Bxk y in Y as k -¥ oo. Since B is closed, we have y — 0. Hence A is closable. 95
96 An Introduction to Pseudo-Differential Operatorsf 3rd Edition Conversely, suppose that A is closable. We define an operator Ao as follows. V(Ao) is the set of all vectors x G X such that there is a sequence {xu} of vectors in V{A) with the property that x& ->> x in X and Axk -» y in Y for some y G Y as k oo. For any x G V(Aq), we define Aox to be equal to y. We have to check that the definition of Ao does not depend on the particular choice of the sequence {x*;}. Indeed, if {zk} is another sequence of vectors in V(A) such that 2* x in X and Azk -> w for some other w eY as k -> 00, then x* — 2* ->■ 0 in X as k ->• 00, x/. - Zk G £>(A), and A(x& - -> y - w in T as A; ->• 00. Since A is closable, it follows that y - w = 0, i.e., y = w. Obviously, Ao is an extension of A. It is closed. For let {xk} be a sequence of vectors in V{Aq) such that Xk -> x in X and A0xk y in Y as k -> 00. For each fc, there is a sequence {x^ } of vectors in V(A) such that x^. ->> Xk in X and Ax^ AoXk in Y as j 00. Hence for each k, there is a 2¾ G X>(A) such that ll*fc - zfclU < ^ and - Azk\\y < Hence for each k, IN* - ®||x < IN* - + |\Xk - x\\x < ^ + ||Xfe - x||x* Therefore IN* — x||x ->■ 0 as fc 00. Similarly, ||A&* — 2/||y -f 0 as A; -* 00. This proves that x G V(Ao) and Aox = y. Hence Ao is closed. □ Let us study the operator Aq constructed in the proof of Proposition 13.4. Proposition 13.5. Aq is the smallest closed extension of A. This means that if B is any closed extension of A, then B is an extension of Ao. Proof Let x G V{Ao) and Aox = y. Then, by the definition of Aq, we can find a sequence {x*} of vectors in A) such that Xk x in X and Axk ->• y in Y as k ->* 00. Since B is an extension of A, it follows that Xk G T>(B), xu ->* x in X and Bxk ->■ y in Y as k 00. Since B is closed, we can conclude that x G T>(B) and Bx = y. This proves that B is an extension of Aq. □ Remark 13.6. In view of Proposition 13.5, we call Ao the minimal operator of A.
Closed Linear Operators 97 Let X be any complex Banach space with norm \\\\x- We denote by X' the dual space of X. Let us recall that X1 is the Banach space of all bounded conjugate linear functionals on X. The norm || \\x> in X' is given by ll/ll*'= supInP. /ex'. *ex IfIIx x^O Let X and Y be complex Banach spaces. For any linear operator A from X into Y with domain V{A) dense in X, we define an operator A1 :Y' —► X1 as follows. V{At) is the set of all functionals y' in Y* for which there is a functional x* in X* such that yf(Ax) = x'(x), x G 'D(A). (13.1) Lemma 13.7. Let y1 GY*. Then there is at most one xf G X' for which (13.1) holds. By Lemma 13.7, we can define Aty* to be equal to x* for all y' G V(At). We call A1 the true adjoint or simply the adjoint of A. Proof of Lemma 13.7 Let yf G7'. Suppose x' and z* are functionals in X' for which y\Ax) = x’(x), x G V{A), and y\Ax) - z\x), x G 'D(A). Obviously, x' = zf on V(A). Since V(A) is dense in X, a simple limiting argument will show that x1 = zf on X. □ Proposition 13.8. At is a closed linear operator from Yl into X'. Proof Linearity, as usual, is easy to check. To prove that At is closed, let Wk} l>e a sequence of functionals in 1^{A1) such that y'k -* yr in Y* and A% -»• x' in X' as k oo. Then, by (13.1) and the definition of At1 y'k(Ax) = (A* y'k)(x) for all x G V(A) and k = 1,2, Let k -¥ oo. Then y* (Ax) = x'(x), x G T>(A).
98 An Introduction to Pseudo-Differential Operators, 3rd Edition This proves that y' G V{At) and Atyf = x'. Therefore A1 is a closed operator. □ Another observation about the adjoint of a linear operator we shall use is given in the following proposition. Proposition 13.9. Let A be any linear operator from X into Y with do¬ main V(A) dense in X. Then for any extension B of A, the operator A1 is an extension of Bl, Proof Let y' G Vi^B1). Then we can find a functional x1 in X' such that y\Bx) = x,{x), xeV(B). Since B is an extension of A, we have y1 (Ax) = x;(x), x G V(A). This proves that yf G V{At) and Aty/ = x' = Bty*. □ Exercises 13.1. Prove that for -oo < s < oo and 1 < p < oo, the dual space of H8'p is H~8'p', where i + i-i. P P 13.2. Let A be a linear operator from Lp(Rn) into Lp(Rn), 1 < p < oo, with dense domain V{A). Prove that T>(At) consists of all functions u in Lp> (Rn) for which there exists a function / in Lp (Rn) such that (u,Av) = (f,v), v G V{A), where (9, h) = / g{x)h{x) dx JRn for all g G Lp (Rn) and h G Lp(Rn). 13.3. Let Z be a complex Banach space. A subset S of Z' total if {z G Z : z'(z) = 0 for all z* G S} = {0}. Let A be a linear operator from a complex Banach space complex Banach space with dense domain V(A). Prove that if and only if the domain T>(At) of At is a total set. is said to be into another A is closable
Chapter 14 Minimal and Maximal Pseudo-Differential Operators Let cr be a symbol in Sm. Then the pseudo-differential operator initially defined in Chapter 6 on the Schwartz space S, has later been extended in Chapter 11 to the space S' of all tempered distributions using the formal adjoint T*. By Theorem 12.9, Ta : HSiP -> Hs~m'p is a bounded linear operator for — oo < s < oo and 1 < p < oo. As a matter of fact, when m > 0, the operator Ta can also be considered as a linear operator from Lp(Rn) into Lp(En), 1 < p < oo, with domain S. We denote this operator simply by Ta. It is not closed in general. Fortunately, it is closable. Hence, by Proposition 13.4, it has a closed extension. Proposition 14.1. The operator Ta is closable. Proof Let {</?*} be a sequence of functions in S such that ipk ~+ 0 and Taifk -+ f in Lp(Rn) as k oo. Then for any function ^ in 5, we have (Ta^ki ip) = Tp'ip), k = 1,2,..., where T* is the formal adjoint of Ta. Let k —> oo. Then (/,^) = 0 for all functions ^ eS. Since S is dense in Lp(Rn), it follows that / = 0. Hence, by Definition 13.2, Ta is closable. □ Remark 14.2. A consequence of Proposition 14.1 is that the minimal op¬ erator Tffio of Ta exists. (See Remark 13.6.) Let us recall that the domain ViTafl) of Tv# consists of all functions u in Lp(Rn) for which a sequence {ifk} in S can be found such that tpk u in Lp(Rn) and Taipk -+ f in Lp(Rn) for some / G Lp(Rn) as k ->> oo. Moreover, Ta$u = /. Definition 14.3. Let u and / be functions in Lp(Rn), 1 < p < oo. We say that u lies in ©(T^i) and Ta^u = / if and only if {u,T*ip) = (f,<p), <p£S, (14.1) 99
100 An Introduction to Pseudo-Differential Operators, 3rd Edition where T* is the formal adjoint of Ta. Proposition 14.4. Let u G X>(T^i). Then Ta^u = Tau in the distribution sense. Proof By Definition 14.3, (7V,lU, if) = (U, T*(p), (peS. Hence, by considering u and Ta^u as tempered distributions, we have (T^u)(jp)=u(T^), (peS. (14.2) On the other hand, by (11.3), cTau){Tp) = u{T{*£), (14.3) Hence, by (14.2) and (14.3), Ta^u = Tau in the distribution sense. □ Proposition 14.5. T^i is a closed linear operator from Lp(Rn) into Lp(Rn) with domain V(T^i) containing S. Proof That S C £>(1^1) is obvious from (14.1) and the definition of the formal adjoint T*. Linearity is again easy to check. To prove that T^i is closed, let {uk} be a sequence of functions in V(Ta>i) such that uk u in Lp(Rn) and Ta^uk f in Lp(Rn) for some u and / G Lp(Rn) as k ->■ oo. Then, by (14.1), {uklT*(p) = (Ta,iuk,(p) (14.4) for all (p G S and k = 1,2, Let k -4 oo in (14.4). Then (u,T» = (/,</?), ip G S. Hence, by Definition 14.3, u G ^(T^i) and Ta^u = f. This proves that is closed by Definition 13.1. □ Proposition 14.6. S C V(T*tl), where T* ± is the adjoint ofTati. Proof Since T^i is a closed linear operator from Lp(lRn) into Lp(Rn), for 1 < p < oo, with domain containing S, it follows from Proposition 13.8 that T*x is a closed linear operator from Lp> (En) into Lp'(Rn), where pf is the conjugate index of p. Let ^ eS. Then for all functions u G V(Tati), (1>,Tatlu) = (T>,n) by Definition 14.3. Hence, by the definition of ^ £ T>(T£i) and T*Ai> = T;ip. ’ ' □ Proposition 14.7. is an extension ofT^0.
Minimal and Maximal Pseudo-Differential Operators 101 Proof Let u G V(Ta$) and T^oii = /. Then, by Remark 14.2, there is a sequence {tpk} of functions in <S for which <pk u and Ta(pk ->■ / in Lp(Rn) as k —> oo. Hence, by the definition of T*, we have for all ip G S and k = 1,2, Let k -¥ oo. Then (u,r» = So, by Definition 14.3, u G £>(1^1) and = /. □ Remark 14.8. Using Propositions 13.9 and 14.7, we see that 0 is an extension of T* v Since, by Proposition 14.6, the domain of T*x contains the space <S, it follows that the domain of T* 0 contains S as well. Proposition 14.9. is the largest closed extension of Ta in the sense that if B is any extension of Ta such that S C D(R£), then T^i is an extension of B. We first prove the following lemma. Lemma 14.10. T*(p = T*ip for all ip G S. In other words, the true and formal adjoints coincide on the space S. Proof Let (p G S. Then, by the definition of T*, we have So, by the definition of T* and the duality of Lp(En), ip G D(T*) and = □ Proof of Proposition 14.9 Let u G V(B). Then for all ^ G 5, we have ^ G V(Bt). Hence, by the definition of Bl, (V>,£u) = 0B^,u). (14.5) Since B is an extension of TV, it follows from Proposition 13.9 that T* is an extension of Bl. Hence, by (14.5), (^Bu) = (T^u). (14.6) By Lemma 14.10, T* = T* on S. Hence, by (14.6), we have (^bu) = (t;^u), ii>es. Therefore, by Definition 14.3, we have u G P(Taji) and Ta^u = Bu. □ Remark 14.11. Because of Proposition 14.9, we call Tay\ the maximal operator of Ta.
102 An Introduction to Pseudo-Differential Operators, 3rd Edition By Proposition 14.7, we know that T^i is an extension of T^o- The aim of this chapter is to prove that T^o = T^i if a is an elliptic symbol in 5m,m > 0. (See Chapter 10 for the definition of ellipticity.) We need some preparation. Theorem 14.12. Let m > 0 and a be an elliptic symbol in Sm. Then 2>(2V|0) = Hm* To prove Theorem 14.12, we use the following estimate, which is the analog of the Agmon-Douglis-Nirenberg estimate in [Agmon, Douglis and Nirenberg (1959)] for pseudo-differential operators. Proposition 14.13. Let m > 0 and a be an elliptic symbol in Sm. Then there exist positive constants C\ and C2 such that Ci||w||7Ti,p ^ (||T<7u||o,p H" |M|o,p) < u E H 'p. Proof By Theorems 12.5 and 12.9, there is a positive constant C' such that \\T*u\\o,p + ||u||o,P < C'||u||m,p, u E Hm'p. Next, by (10.1) in Theorem 10.1, we have u = TTTau -Ru, u E (14.7) where r E S~m and R is a pseudo-differential operator with symbol in Hence it follows from Theorem 12.9 and (14.7) that there is a positive constant C such that C\\u\\m,p < (|M|ofp + ||u||o,p), U E Hm'p. This proves Proposition 14.13. □ Proposition 14.14. S is dense in H8yP, -00 < s < 00, 1 <p < 00. Proof Let u E H8'p. Then, by the definition of Hs'p, J-Su E Lp(Rn). Since S is dense in Lp(Wl) by Remark 3.10, it follows that there is a sequence {ifk} of functions in S such that ipk -» J_su in Lp(W1) as k -»> 00. Let ipk = Js<Pk, fc = 1,2, — By Proposition 6.7, E S, k = 1,2, — Also, by the definition of H8'p again, W'lfik ~~ u||s,p = ||»7— s'lpk “ J-s^Wp — \\<Pk ~ J-sU\\p 0 as k ->• 00. This proves that 5 is dense in Hs'p. □ We can now prove Theorem 14.12.
Minimal and Maximal Pseudo-Differential Operators 103 Proof of Theorem 14.12 Let u E iTm,p. Then, by Proposition 14.14, we can find a sequence {</?*.} of functions in S such that <fk u in Hm'p as k -+ oo. By Propositions 14.13 and 14.14, {Ta(fk} and {<pk} are Cauchy sequences in Lp(En). Hence (pk -+ u and Ta<pk -+ f in Lp(En) for some u and / in Lp(Rn) as k -+ oo. Hence, by the definition of 7V}o, u E £>(7V50) and Tv^u = f. On the other hand, if u E V(Ta>o), then, by the definition of 7V,0 again, we can find a sequence {(} of functions in S for which (fk -+ u in Lp(Rn) and Ta(pk -+ f in Lp(Rn) for some / E Lp(Rn) as k -+ oo. Hence {(pk} and {Ta(fk} are Cauchy sequences in Lp(Rn). So, by Propositions 14.13 and 14.14, {<} is a Cauchy sequence in Hm'p. Since Hm'p is complete by Theorem 12.3, it follows that -+ v in Hm'p for some v E Hm'p as k -+ oo. Then, by Theorem 12.5, ifk —► v in Lp(Rn) as k-+ oo. Hence u = v and consequently u E Hm,p. □ Finally we come to the main result of this chapter. Theorem 14.15. Let m > 0 and a be an elliptic symbol in Sm. Then T„,0 = Proof Since is the smallest closed extension of Ta, it follows from Proposition 14.7 and Theorem 14.12 that it is sufficient to prove that ©(T^i) C Let u E Then, by (10.1) in Theorem 10.1, u = TrT'U - Ru, (14.8) where r E 5~m and jR is a pseudo-differential operator with symbol in CfeeR#*. By Proposition 14.4, Ta,\u = Tau in the distribution sense. Thus, by the definition of IV,i, Tau E Lp(Rn). Since r E 5_m, it follows from Theorem 12.9 that TTTau E Hm'p. Since u E Lp(En) and has symbol in 5”m, it follows from Theorem 12.9 again that Ru E Hm'p. Hence, by (14.8), u E ifm’p. □ Exercises 14.1. Let a be any symbol in Sm, m < 0. Prove that the minimal operator Ta>o oi Ta : S -+ S in Lp(En), 1 < p < oo, is a bounded linear operator from Lp(En) into Lp(En). 14.2. Let a be any symbol. Consider TV as a linear operator from Lp(En) into Lp(En), 1 < p < oo, with dense domain S. Prove that T* = (TV,o)*- 14.3. Let a and Ta be as in Exercise 14.2. Prove that = (T*)i, where (T*)i is the maximal operator of the pseudo-differential operator T*.
104 An Introduction to Pseudo-Differential Operators, 3rd Edition 14.4. For any closed linear operator A from a complex Banach space into itself with domain V(A), a subspace V of T>( A) is called a core of the operator A if the minimal operator of the restriction of A to V is equal to A. Prove that if a is any elliptic symbol, then Co°(Mn) 1S a core °f ^<r,o and
Chapter 15 Global Regularity of Elliptic Partial Differential Equations Let P(x,D) = £,a,<m act(x)Da be a linear partial differential operator of order m such that sup \(D^aa)(x)\ < oo, \a\ < m, (15.1) for all multi-indices /?. Then we have observed in Example 6.3 that P(x, D) is a pseudo-differential operator with symbol P(x,f) in Sm, where P(x,0= a<* (*)£“• \a\<m The purpose of this chapter is to use the theory of pseudo-differential operators we have developed to prove the following Lp analog of a result in [Hess and Kato (1970)]. Theorem 15.1. Let P(x,D) = 2|a|<m aa(x)Da be a linear partial dif¬ ferential operator of order m satisfying (15.1). Suppose that there exists a point xq G Rn such that we can find positive constants C\ and (¼ for which and M*o)C \a\=m (15.2) (a<*(x) ~ aa(xo))C \a\=m < C*\£T (15.3) for all x and $ G In, and C% < C\. If u G H8*p, P(x,D)u = f and f G H8*, then u G H8+m*. 105
106 An Introduction to Pseudo-Differential Operatorsf 3rd Edition Remark 15.2. If P(x, D) is a linear partial differential operator of order m satisfying the hypotheses of Theroem 15.1, and if / is any tempered distribution in #5,p, then Theorem 15.1 asserts that any solution u in H8,p of the partial differential equation P(x,D)u = / on ln lies in a more “selective” or “regular” space iJs+m>p. If we recall, by Theorem 12.5, that jjs+m,p £ jjs+m—ltp q ... q fls>P then the solution u can be thought of being m steps more “selective” or “regular” than the given source data / defined globally on Rn. For this reason, we call Theorem 15.1 a global regularity theorem. To prove Theorem 15.1, we use the following lemma. Lemma 15.3. Let P(x,D) = Yl\a\<ma<x(x)Da be a linear partial differ¬ ential operator of order m satisfying the hypotheses of Theorem 15.1. Then there exist positive constants C and R such that \p(x,o\>c(i+\s\r, ifi >r. Remark 15.4. Lemma 15.3 tells us that under the hypotheses of Theorem 15.1, P(x,D) is a pseudo-differential operator with symbol P(x,$) in Sm and P(x,£) satisfies the ellipticity condition defined in Chapter 10. Proof of Lemma 15.3 By (15.2) and (15.3), we obtain |a|=m Y - Mxo))fa + y a*(xo)£a | a |=77i | a |=7Ti > Y a<*(xo)fa | a | =777 >{Cx-c2mm, - Y (“oW ~ Oo(*o))f“ |a|=m x,i € K". (15.4)
Global Regularity of Elliptic Partial Differential Equations 107 Next, by (15.4), we can find positive constants C", C" and R such that Y °a(®)£“ \a\<m Y aa(x)Za+ Y ««(*)£“ \a\=m |a|<m > Y o«(*)€“ — ^ o0(x)r |o: 1=771 |a|<77l > c'(l+Kir - c"(i+Kir-1 = (1+ 1^(^-^(1 + 1^1)-1). Kl >R- Obviously, we can find another positive constant Ri > R such that c' - c"(i + Kl)_1 > y, Kl > Ri- Hence, by (15.5) and (15.6), (15.5) (15.6) Y Maor | a | <ra >y(l + kl)m, This proves Lemma 15.3. □ Proof of Theorem 15.1 By Lemma 15.3, P(x,D) is an elliptic pseudo¬ differential operator with symbol P(x, £) in 5m. Hence, by Theorem 10.1, we can find a symbol r E S~m and a pseudo-differential operator R with symbol in OkeRSk such that TrP{x,D) = I + R. (15.7) Hence, by (15.7), u = Tr/ — Ru. Since r E S~m and / E H8'p, it follows that Tr/ E Also, Ru E Hs+m'p because P has symbol in 5~m and u E Hs'p. Hence, by (15.7), u E JIa+m* □ Exercises 15.1. Let P(x, D) be a linear partial differential operator of order m satis¬ fying all the hypotheses of Theorem 15.1. Prove that if / E HSyP, then any solution u in UteRHtiP of the partial differential equation P(x, D)u = f on Mn is in Hs+m'p.
108 An Introduction to Pseudo-Differential Operators, 3rd Edition 15.2. Let m > 0 and a £ Sm be an elliptic symbol. Let / £ Lp(Rn), for 1 < p < oo and u be any solution in Ut&iHttP of the pseudo-differential equation Tau = / on Rn. (i) Prove that u £ Lp(Rn). (ii) Prove that (u,T£ip) = (/, p) for all ip £ S. (iii) Prove that there exists a sequence {(pk} 0f functions in S such that <Pk -> u in Lp(Rn) and ^ / in Lp(Rn) as Ar oo.
Chapter 16 Weak Solutions of Pseudo-Differential Equations We give in this chapter a result on the existence of weak solutions in Lp(En), 1 < p < oo, of pseudo-differential equations on En. We begin with the definition of a weak solution. Definition 16.1. Let a G Sm, m > 0, and / G Lp(En), 1 < p < oo. A function u in Lp(En) is said to be a weak solution of the pseudo-differential equation T^u — f on En if (u,t;<p) = <pes, where T* is the formal adjoint of Ta introduced in Chapter 9. From the definition of the maximal operator Tffii of Ta given in Chapter 14, it is obvious that the following proposition is true. Proposition 16.2. Let a G Sm, m > 0, and f G Lp(En), 1 < p < oo. Then a function u in Lp(En) is a weak solution of the pseudo-differential equation Tau = f on W1 if and only ifue ^(T^i) and Ta^u = /. For any a G Sm, m > 0, the following theorem characterizes the func¬ tions / in Lp(En), 1 < p < oo, for which the pseudo-differential equation Tau = / on En has a weak solution u in Lp(En). Theorem 16.3. Let a G Sm, m > 0, and f G Lp(En), 1 < p < oo. Then the pseudo-differential equation Tau — f on W1 has a weak solution u in Lp(En) if and only if there exists a positive constant C such that 1(/, ^)1 <c\\t;v\\p,, ves, (16.1) where p1 is the conjugate index of p. 109
110 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Suppose that Tau = / on ln has a weak solution u in Lp(Rn). Then, by Definition 16.1, (f,y) = (u,T;<p), <p<=S. Hence, by Holder’s inequality, 1(/,01 <IMIplir>||P-, and the inequality (16.1) holds with C = IM|P. Conversely, suppose that the inequality (16.1) is true. Let W be the subspace of Lp (W1) defined by W = {T> :<peS}. We define the linear functional F :W —» C by Fw = (<£,/), w e W, where <p is any function in <S with the property that T*tp = w. To see that the defintion of F : W —> C is independent of the function <p, let and (p2 be functions in S such that T*p>i = w and T*<^2 = w. Then, by (16.1), Hence (ip\,/) = (<^>2, /) and this proves that the choice of the function ip is irrelevant to the definition of F : W -»■ C. Since, by (16.1), l-FH = \{<P,f)\ < C\\T*M\P, = CIHIp-, wew, it follows that F : W -» C is a bounded linear functional. Hence, using the Hahn-Banach theorem and the Riesz representation theorem, we can find a function u in Lp(Rn) such that Fw = (</>, /) = (w,u), we W, (16.2) where ip is any function in S satisfying T*<p = w. Since {T*ip : ip e S} is obviously a subspace of W, it follows from (16.2) that iv>, f) = (T*ip,u), tpeS, and hence, by Definition 16.1, u is a weak solution in Lp(Rn) of the pseudo¬ differential equation Tau = f on Rn. □
Weak Solutions of Pseudo-Differential Equations 111 Exercises 16.1. Let <r G Sm, m > 0, and let u and / be in Lp(Rn), 1 < p < oo. Prove that u is a solution of Tau — f on Rn in the distribution sense if and only if u is a weak solution of Tau = / on Rn. 16.2. Let <j G Sm, m > 0, be an elliptic symbol, and let / G Lp(Mn), for 1 < p < oo. Prove that every weak solution u in Lp(Rn) of the pseudo¬ differential equation Tau — f on tn is in Hm'p. 16.3. Let s > 0 and J-s be the pseudo-differential operator defined in Chapter 12. Let q be any real-valued and nonnegative function on Rn such that sup \(Daq)(x)\ < oo xeRn for all multi-indices a. Prove that the pseudo-differential equation J-Su + qu = f on Rn has a weak solution u in L2(Rn) for every function / in L2(Rn). 16.4. Let cr(x,£) G Sm, m > 0, be such that a is independent of x G Rn. Prove that the pseudo-differential equation Tau = / on Rn has a unique weak solution u in L2(Rn) for all functions / in L2(Rn) if and only if there exists a positive constant C such that \*m >c, £ £ Rn.
This page intentionally left blank
Chapter 17 Garding’s Inequality We are interested in a subset of the set of all elliptic pseudo-differential operators introduced in Chapter 10. These operators satisfy an important inequality in the study of pseudo-differential operators. Theorem 17.1. (Garding’s Inequality) Let a € S2m be such that there exist positive constants C and R for which Re<r(*,0 > <?(1 + |£|)2"\ Id >R. Then we can find a positive constant C' and a constant Cs for every real number s > | such that R*C> c'Mtz ~ c.Mt-.* v e s. A symbol satisfying the hypothesis of the theorem is said to be strongly elliptic. In order to prove the theorem, we need two lemmas. Lemma 17.2. Let F be a C°° function on the complex plane C. Then for every a in S°, F o a £ S°. Proof We need to prove that for all multi-indices a and /3, there exists a positive constant Ca,/3 such that |(0“af(Fo<r))(x,0| <Cajj(l + |d)“l'J|, z,£eir. (17.1) (17.1) is true for all multi-indices a and /3 with \a + P\ = 0. Indeed, there exists a positive constant C such that K*.0I<C, x,£ € 1". Thus, F o a is in fact a bounded and C°° function on r xln. Hence there exists another positive constant C' such that |(F o cr)(a;,£)| < C', 1,^1". 113
114 An Introduction to Pseudo-Differential Operators, 3rd Edition Now, suppose that (17.1) is valid for all C°° functions F on C, a in S° and multi-indices a and /? with \a + 0\ = l. Let a and /3 be multi-indices with \a + /?| = / + 1. We first suppose that = for some multi-index 7 and some j = 1,2,..., n. Then, by the chain rule, (3?qf (F 0 <r))(*,0 = 0 <T)d(ja + (F2 o for all x and £ in Rn, where F\ and F2 are the partial derivatives of F with respect to the first and second variables respectively. Now, by Leibniz’s formula and the induction hypothesis, there exist positive constants Cpj and CatPntsj such that SI7 = ^.7,i(i + ieir(l7l+1)> *,£eRn, where Cp,5Ca,p,'Y,5,j- Similarly, there exists a positive constant Cfa/y j such that mdim O a)^a}(*,0l < C'a^(l + |6-(M+1\ for all x and £ in W1. Therefore m%{F 0 *))(*, 0| < (Ca,Jtj + C'a^){ 1 + K|)-I*l for all x and £ in Rn. Now, we suppose that = d2dXidl for some multi-index 7 and some j = 1,2,..., n. Then, as before, there exists a positive constant Canj such that m$(F 0 *))(*, 01 < <7)j( 1+ki)-m
Garding’s Inequality 115 for all x and £ in Mn. Thus, by the principle of mathematical induction, (17.1) follows. □ Lemma 17.3. Let a be a strongly elliptic symbol in S2m, m > 0. Then there exist positive constants 7 and k such that Rea(x,£) > 7<£>2m - € 1", where ( ) is the function on Mn defined by <£) = (i + I£I2)1/2> feir*. Proof By strong ellipticity, there exist positive constants C and R such that Re<7(x,£)>C<£>2m, Kl >R- Since a G 52m, we can find a positive constant K such that k(*,0l<*(02m. *>£eir. Therefore |Re<r(x, ^)| < K(02m < K(1 + R2)m, \£\ < R. Hence there exists a positive constant M such that Re a(x, £) > -M, |£| < R. Since ^ is continuous on the compact set {£ € W1 : |£| < R}, we can find a positive constant k such that Re<r(a;,£) ^2m-l > Kl < a Therefore Re<7(x,£) + K<02ro-1 >0, \(\<R. Since Re7^.ir~ - is a positive and continuous function on the compact set {(el": |£| < R}, there is a positive constant S such that Re<r(x,£) + «(02m-1 (f> 2m >J, |£| <R. So, the lemma is proved if we let 7 = min(C',<5). □ Proof of Theorem 17.1 Let Tt = where Jm is the Bessel potential of order m. In fact, Jm = )-*». Then, using the asymptotic
116 An Introduction to Pseudo-Differential Operators, 3rd Edition expansion for the product of two pseudo-differential operators in Theorem 8.1, where Tn-Jm — TT Similarly, and ri - ( )'ma e Sm-\ Tr = JmTTl (17.2) r-()-m7i GS"1. (17.3) Multiplying (17.2) by ( )“m and adding the result to (17.3), we get r - ()~2ma e S'1. Therefore r = ( )~2ma + r, where r € S~l. So, by Lemma 17.3, Rer = ()“2mRecr + Re r > 7 — k( )_1 + Rer > 7 — /c'( )_1, where k! is another positive constant. Therefore r satisfies the conclusion of Lemma 17.3 with m — 0. Let us suppose for a moment that Garding’s inequality is valid for m = 0. Then we can find a positive constant C’ and a positive constant C8 for every real number s > \ such that Re ip) = Re (J-mTr <p) = Re (TrJ-mtp, — & ll^-m^|lo,2 “ Cs\\J-m(P\\‘LSi2 for all ip in S. We are now ready to prove Garding’s inequality for m = 0. By Lemma 17.3, we have positive constants 7 and k, such that Re<r 4- k( )-1 > 7. Let F be a C°° function on C such that
G&rding’s Inequality 117 Let r be the function defined on Rn x Rn by t(x,0 = F(2(R.e<r(x,£) + k(£)-1 —7)), x,£ € Kn. Then, by Lemma 17.2, r € 5°, and for all x and £ in Rn, f(*,0 = +2Reff(*,f)+ 2^(0-1-27 = ^2Rfi<r(x,0 +2k(£>-1 - |t- Using the asymptotic expansion for the formal adjoint of a pseudo- differential operator in Theorem 9.1, we have where r* G 5° and r - r* G S”1. Using also the asymptotic expansion for the product in Theorem 8.1, t;tt = rA, where A-tVgS"1. If we let ri and r[ in 5_1 be such that r* = r + r\ and A = *T + rl, then, with r2 = nr + r[ G S"1, A = (r + ri)r + 7^ = 2Recr + 2k( )-1 So, if we let r3 = 2k( )_1 + r2 G S-1, then we get 3 A = 2Recr — -7 + r3. z But 3 27 + r2- 2Re 0- = 0- + 0 = 0- +0*+7*4 for some r4 in S 1. Therefore A = 0- + 0* - -7 + r5 (7 + 0-* = A + -7 - r5. for some 7*5 in S Thus,
118 An Introduction to Pseudo-Differential Operators, 3rd Edition Since (T\(p,tp) = (TT(p,TT(p) >0, <p G <S, it follows that 2Re (Tfip, (p) — (Tpipiip) + (Tpipytp) — cp) = (T\<P, <p) + ^711^110,2 “ (Trs<P,<p) > i\Mh + {%Mh - MINIMI-*,*} • Using the L2-boundedness of pseudo-differential operators in Theorem 11.7, we get a positive constant // such that 2Re (T„<p,(p) >7lMlo,2 + {|lMlo,2-A*IMllj,2}» V e S- But = J ^0^10(012 d£ = I + J, where I=[ p(Q-l\m\2dt and j=f Obviously, i <1 JRnmo\2dt=l\Mi2- To estimate J, we note that ^(0-1 > | =*• <0 < y • So, for /t(£)_1 > 2> we Set> f°r every real number s > Hit)-1 = M(O2'-1<0-a* ^rv*. Then for every real number s > j<»(-)2° 1 f <0-2*i^(0i2de=c;iMii.,2, \ 7 / JRn /o \ 2s—1 where Cfs = . Therefore 2Re (Tffip,tp) > Tlbll2i2 - CSlM|l.,a, V € S, and the proof of the theorem is complete. □
Gdrding’s Inequality 119 Exercises 17.1. Prove that a strongly elliptic pseudo-differential operator is elliptic. 17.2. Give an example of an elliptic pseudo-differential operator that is not strongly elliptic. 17.3. Let a e 52m, m > be such that a is strongly elliptic. Prove that there exists a constant C such that Re <p) > -CIMI*, V € S.
This page intentionally left blank
Chapter 18 Strong Solutions of Pseudo-Differential Equations Let a G Sm, m > 0, and let / G Lp(Mn), 1 < p < oo. From Chapter 16 we see that a function u G Lp(Mn) is a weak solution of the pseudo-differential equation Tau = / on Mn if u G V(Ta,i) and TV,iu = /. In this chapter we give another notion of a solution of the equation. A function u G Lp(Rn) is said to be a strong solution of the equation = / if u G X^T^o) and XV, = / • Remark 18.1. For the pseudo-differential equation T^u = / on Rn, it should be clear that strong solutions axe weak solutions. If a is elliptic, then it is also true that weak solutions are strong solutions. These simple facts are best left as exercises. See Exercises 18.1 and 18.2. We begin with strong solutions in L2(Rn). Theorem 18.2. Let a G S2m, m > 0, be an elliptic symbol such that there exists a positive constant C for which Re(2>,¥>)>C|M|^2, ¥>€5. (18.1) Then for every function f in L2(Rn), the pseudo-differential equation Tau = / on Rn has a unique strong solution u in L2(Rn). We need the following lemma. Lemma 18.3. Under the hypotheses of the preceding theorem, there exists a positive constant C such that Re (T„u,u) >C|H|^2, Proof Let u G Hm'2. Then, by Proposition 14.14, there exists a sequence {(fk) of functions in S such that u in Hm,2 as k oo. By Theorem 121
122 An Introduction to Pseudo-Differential Operators, 3rd Edition 12.9, Ta : Jfm>2 H~m'2 is a bounded linear operator. Hence, by Exercise 12.14, there exists a positive constant C' such that \{Ta(pk,<Pk) - {TaU,u)\ — 1CTaVkitpk) ” {TGU,(pk) + (TgU, (fk) — ti) | = |(2V(<p* - ti),^fc) + (Tau,(pk - w)| — — 11—fn,21|^/j11771,2 H" |I^V^II —m.,21— ^||m,2 ^ C7 ||^fe ^11m-,21111m-,2 “1“ II^Hra^H^fc 0 as k —> oo. Therefore Re(Tau,u) = lim Re(T^ip^tpk) > C lim \\<Pk\\m,2 = IMIm,2- k—tOO ft—HX> □ We also need the following fact, which is a generalization of the Riesz representation theorem. Theorem 18.4. (Lax—Milgram Lemma) Let X be a complex and sepa¬ rable Hilbert space with inner product and norm denoted by (,)x awd || \\x respectively. Let B be a bilinear mapping on X such that there exist positive constants C\ and (¾ for which \B(x,y)\ < Cillxllxllyllx, x,y € X, (18.2) (18.3) Then for every bounded linear functional f on X, there exists a unique vector y e X such that f(x) = B(x,y), x£X. Proof For a fixed vector y in X, we see from (18.2) that B(-,y) is a bounded linear functional on X. Using the Riesz representation theorem, there exists a unique vector z(y) G X such that B(x,y) = (x,z(y))x, x€X. We observe that the mapping XBy^ z{y) € X
Strong Solutions of Pseudo-Differential Equations 123 is linear. Indeed, let y\ and y2 be vectors in X and let ci and C2 be complex numbers. Since B is bilinear, (x, z(ciyi + c2y2))x = B(x,cxyi + c2y2) = c^B(x,yx) +c^B(x,y2) = ci(x,z(yA))x +ci(x,z(y2))x = (x^xziyi^x + (x,c2z(y2))x = (x,c1z(yi)x + c2z(y2))x, x € X. Therefore z(ciyi + c2y2) - ciz(yi) + c2z(y2) and linearity is established. Let M be the subspace of X given by M = {z{y): y £ X}. Then M is a closed subspace of X. Indeed, let {z(yk)} be a sequence in M such that z(Vk) -*• z as fc —^ oo. Then for j,k = 1,2,..., B(x,yj - Vk) = (x, z(yj) - z(yk))x, x e X. By (18.3) and the Cauchy-Schwarz inequality, we get for j,k = 1,2,..., (¼||Vj ~ Vk\\x < \B(Vj ~ Vk,Vj ~ Vk)\ = \{Vj-Vk,z{yj)-z(yk))x\ < \\Vj ~ Vk\\x \\z(yj) - z(yk)\\x- So, for j, k = 1,2,..., CzWVj - VkWx < ||z{yj) - z{yk)\\x- Thus, {yk} is a Cauchy sequence in X and hence Vk^y for some y in X as k -¥ 00. By (18.2), \B(x,yk -y)| < CiUarllxlly* -y||x -► 0 as k —¥ 00 for all x € X, which gives B(x,yk) -¥ B(x,y)
124 An Introduction to Pseudo-Differential Operators, 3rd Edition as k oo. Furthermore, for all (z>2(2/fc))x -* 0*>*)x as k ->* oo. Since B(x,yk) = (a?,z(i/*))x, x6l, for fc = 1,2,..., it follows that B(x,y) = {x,z)x, xeX. Therefore z E M and this proves that M is closed. Now, M = X. To see this, let us assume that M is a proper (and closed) subspace of X. Then there exists a nonzero vector x G X such that B(x, y) = (x, z(y))x = 0, y € X. If we let y = x, then by (18.3), x = 0 and this contradiction implies that M = X. By the Riesz representation theorem, there exists a unique vector w G X such that f(x) = (x,w)x, xeX. Since X = Af, we can find a vector y € X such that w = z(y). Therefore /Or) = (x,ti7)x = (®,3(y))x = xeX. The uniqueness of y follows again from (18.2). □ Proof of Theorem 18.2 Let B : Hm,2 x Hm>2 -> C be the bilinear mapping defined by B(u,v) = (u,Tav), u,v € if771’2. Then for all u and v in if771’2, we see by means of Exercise 12.14 that \B(U,V)\ < |M|m,211^11^,2 < |M|mf2|M|m,2 and, by Lemma 18.3, \B(u,u)\ > \(T„u,u)\ > C\\u\\2m,2, u € Hm’2. Let / € L2(En). Then we define the linear functional F : Hm’2 -¥ C by F(w) = (w,f), w e Hm’2. It is a bounded linear functional because, by Exercise 12.14 and Theorem 12.5, \F{w)\ = 1(1«,/)| < |MU,2||/ll-m,2 < ||/||2||tl>||m,2, W € Hm’2.
Strong Solutions of Pseudo-Differential Equations 125 So, by the Lax-Milgram lemma, we get a unique function u in Hm'2 such that F(w) = B(w,u), w G Hm'2, or equivalently, (w, f) = (w,Tau), w G Hm'2. So, u is a weak solution in L2 (Rn). Since a is elliptic, it follows from Exercise 18.2 that u is a strong solution in L2(Rn). □ Another Proof of Theorem 18.2 By Theorem 12.5, the inequality (18.1) and the Cauchy-Schwarz inequality, we get for all functions <p € S, M\l < IMIm.2 < £lMhl|T>||2 and hence IMk < 1||T>||2. Let f € L2(R"). Then m>p)\ < 11/H2IMI2 < 1||/ii2||i;vii2i vg<s. So, by Lemma 18.3, the pseudo-differential equation Tau = / on En has a weak solution u in L2(Rn). Since a is elliptic, it follows that u is also a strong solution in L2(Rn). Let v be another strong solution in L2(Rn). Then, by Theorem 18.3, llu - Him,2 ^ ^Re (T^(u ~ V), U - V) = 0. So, u = v and u is the unique solution. □ We can now give sufficient conditions in terms of the symbols a for the existence and uniqueness of strong solutions in L2(Rn) for pseudo- differential operators Ta. Theorem 18.5. Let a G 52m, m > 0, be a strongly elliptic symbol Then there exists a real number Ao such that for all f in L2(Rn) and A > Ao, the pseudo-differential equation (Ta -I- AI)u = / on Rn, where I is the identity operator on L2(Rn), has a unique strong solution u in L2(Rn). Proof By Garding’s inequality, there exist constants A and Ao such that A > 0 and Re (Tm p) > A|M|*,>2 - AolMli, p € S.
126 An Introduction to Pseudo-Differential Operators, 3rd Edition Then for A > A0, Re ((T„ + \I)<p, ip) > A\\<p\&a + (A - A0)|M|| > A\\<p\&ti, € <S. Thus, by Theorem 18.2, the proof is complete. □ The following theorem is a result on the existence and uniqueness of strong solutions in Lp(Rn). Theorem 18.6. Let a G 5m, m > 0, be an elliptic symbol such that a is independent of x in Rn and £git. Then for every function f G Lp(Rn), 1 < p < oo, the pseudo-differential equation Tau = / on Rn has a unique strong solution u in Lp(Rn). Proof By Exercise 18.3, it is sufficient to prove that there exists a positive constant C such that IMIp' < ^ s. Let r — 1/(7. Then for all multi-indices a, we use (1.4) to obtain daT = Y Cam ><*(*° ^+1 where C^u),...,<*(*) is a constant depending on ... and the sum is taken over all multi-indices ,..., that partition a. Thus, we can find positive constants Ca(i),..., Ca(k) such that ctt(D- -caW(i + |£l)*ro_w k(0lfc+1 for all f G Mn. Since a is elliptic, we can find positive constants C and R such that k(0l>c'(i + ior, Kl >R- Since is a continuous and positive function on the compact set {£ G Rn : |£| < i?}, it follows that there exists a positive number 6 such that k(0i>*(i + ifi)m, iei < «- Therefore there exists a positive constant C' such that Koi>c'(i+Kir, ^ g Rn.
Strong Solutions of Pseudo-Differential Equations 127 So, there exists a positive constant C" such that |(fl*T)(0| < c"( 1 + I£|)"m“w, £ € Kn. Therefore r E S~m. By Theorem 12.9 and the Sobolev embedding theorem in Chapter 12, we get a positive constant Cm such that Wvb = \\TtT*p\\p' < C"'\\T^p\\p'. V e 5. It remains to prove uniqueness. Let u and v be strong solutions in Lp(Rn). Let w = u — v. Then w E ifm,p and T^w = 0 on Rn. Thus, aw = 0 in the sense of distributions. So, w{ap) = (aw)(ip) = 0, ip G S. (1S.4) For all if) £ 5, we can find a function ip G 5 such that ap = That this can be done is Exercise 18.5. Thus, by (18.4), w(ip) = 0, ip € S. So, w = 0 and hence w = 0. Therefore u = v and uniqueness is proved. □ Exercises 18.1. Let a € Sm, m > 0. Let / € Lp(Rn) for 1 < p < oo. Prove that a strong solution u in Lp (En) of the equation Tau = / on ln is also a weak solution in Lp(Rn). 18.2. Let a € Sm, m > 0, be an elliptic symbol. Let / G Lp(Rn), where 1 < p < oo. Prove that every weak solution u in Lp(Rn) of the equation Tau = / on Mn is also a strong solution in Lp(Rn). 18.3. Let cr E Sm, m > 0, be elliptic and such that there exists a positive constant C for which M\p>< c\\t;<p\\p,, <pes. Prove that the pseudo-differential equation Tau = / on Rn has a strong solution u in Lp(Rn) for every function / € Lp(Mn), 1 < p < oo. 18.4. Let a be as in Exercise 18.3 and such that {T*cp : ip € S} is dense in Lp' (Rn). Prove that the equation Tau = / on Rn has a unqiue strong solution u in Lp(Rn) for every function / G Lp(Rn), 1 < p < oo. 18.5. Let a G Sm, m > 0, be an elliptic symbol such that a is independent of x in Rn and *(0^0, £ E Rn. Prove that for all functions xp G <S, there exists a function ip G 5 such that CT(£ = Xp.
This page intentionally left blank
Chapter 19 One-Parameter Semigroups Generated by Pseudo-Differential Operators Let A be a closed linear operator from a complex Banach space X into X with dense domain V(A). Let / € X. Then we are interested in the initial value problem for the heat equation governed by A given by f u'(£) = A(u(t)), t > 0, l u(0) = /, where u : [0, oo) —> X and u'(t) = lim v ' h->o u(t + h) — u(t) h if the limit in X exists. The questions to be answered are related to the existence and uniqueness of a global solution u : [0, oo) -> X. It is intuitively clear that the solution u : [0, oo) -* X is given formally by u(t) = etAf, t > 0. What is etA when A is a closed linear operator densely defined on a complex Banach space X? The aim of this chapter is to answer this question. A family {T(t) : t > 0} of bounded linear operators on X is said to be a one-parameter semigroup if (i) T(0) = /, where I is the identity operator on X, (ii) T(s)T(t) = T(s + £), s, t > 0, (iii) T(t)x —> x in X as t -¥ 0+ for every x in X. Let {T(t) : t > 0} be a one-parameter semigroup on X. We denote by T>(A) the set of all elements x in X such that lim*_*.o+ T^*~x exists in X. Proposition 19.1. V(A) is a dense subspace of X. 129
130 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof Let W be the subspace of X given by W= jxs = i^ T(S)xdS : x 6 X, s > o|. Then W is dense in X simply because xs -+ x in X for all x € X as s -+ 0+. For t > 0, let Let s and t be positive numbers. Then for t < s, = T(t) fg T(6)xdS — J0a T(S)xd6 1 s st _ j‘ T(t + 5)x d6 - /0® T(8)x d6 st f"+t T(5)x d5 - Jq T(S)xdS st _ Ss+tT{S)xd8 - /q T(5)xdS st = Asxt. Thus, Atx8 -+ Asx in X as t -+ 0 +. Therefore W C V(A) and hence V(A) is dense in X. □ The linear operator A from X into X with domain V(A) defined by Ax = lim —x ? x e 'D(A), t—>o+ t where the limit is again understood to take place in X, is called the in¬ finitesimal generator of the one-parameter semigroup {T(t) : t> 0}. Proposition 19.2. The infinitesimal generator A of a one-parameter semi¬ group {T(t) : t > 0} on a complex Banach space X is a closed linear oper¬ ator. We need two lemmas to prove Proposition 19.2. Lemma 19.3. Let A be the infinitesimal generator of a one-parameter semigroup {T(t) : t > 0} of bounded linear operators on a complex Ba¬ nach space X. Then for all t € [0, oo) and x G T)(A), T(t)x € T>(A) and AT(t)x = T{t)Ax, x e V(A).
One-Parameter Semigroups Generated by Pseudo-Differential Operators 131 Proof Let x G V(A). Then for all t G [0, oo), T(t)AhX = AhT(t)x, h > 0. Therefore lim AhT(t)x = lim T(t)AhX = T(t) lim AhX = T(t)Ax. h—>-0 h—>0 h—>0 Thus, T(t)x G V(A) and AT(t)x = T{t)Ax, as asserted. □ Lemma 19.4. Let x G V(A), where A is the infinitesimal generator of a one-parameter semigroup {T(^) : t > 0} on a complex Banach space X. Then for all t > 0, T(t)x-x= f T(s)Axds. Jo Proof Let / be a bounded linear functional on X. Then we define the function F : [0, oo) -> C by F(t) = f (r(t)x - x - j T{s)Axds^ , t > 0. So, for t > 0, the right-hand derivative F'(t-h) of F at t is given by F'(t+) = f(AT(t)x-T(t)Ax) in view of the definition of the infinitesimal generator and the fundamental theorem of calculus. By Lemma 19.3, F'(t+) = 0, t G [0, oo). Thus, F(t) is a constant for all t G [0, oo). Since F(0) = 0, it follows that F(t) = 0, t > 0. Thus, using the Hahn-Banach theorem, we see that T(t)x-x= f T(s)Axds, £ > 0. Jo □ Proof of Proposition 19.2 Let {Xj} be a sequence in V(A) such that Xj x
132 An Introduction to Pseudo-Differential Operators, 3rd Edition T(t)x — x = lim (T(t)xj - Xj) = lim / T(s)Axj ds =/ Tsyds. Therefore x € T>(A) and Ax = y. Thus, A is a closed linear operator. □ When is a closed linear operator A from a complex Banach space X into X with dense domain V(A) the infinitesimal generator of a one-parameter semigroup on X? The answer is given as the content of the celebrated Hille-Yosida-Phillips theorem. We first introduce some notation and terminology. Let A be a closed linear operator from a complex Banach space X into X with dense domain V{A). We define the resolvent set p(A) of A to be the set of all complex numbers A such that XI — A : T>(A) X is bijective, where I is the identity operator on X. If A £ p(A), then the bounded linear operator (AI — A)~l : X X is called the resolvent of A at A and is often denoted by R(A; A). The norm a bounded linear operator A from a complex Banach space X into another complex Banach space Y is denoted by ||A||. Theorem 19.5. (Hille-Yosida-Phillips Theorem) Let A be a closed linear operator from a complex Banach space X into X with dense domain V(A). Then A is the infinitesimal generator of a one-parameter semigroup {T(t) : t>0} on X if and only if we can find a positive number M and a real number u such that for all A in (u, oo) and all positive integers n. Proof Let us first prove the sufficiency. For A > a;, we define the bounded linear operator B\ : X -* X by j—+oo Hence {A 6 1: A > Go*} C p(A) and Bx = -X(I-XR(X;A)).
One-Parameter Semigroups Generated by Pseudo-Differential Operators 133 Then for alH G ®, etB, =e-Xtf^&lR{X.A) n\ n=0 and therefore So, for every in (a;, oo), ||etBx|| < Met(Jl for all sufficiently large A. Now, we claim that lim B\x = Ax, x G T>(A). (19.2) (19.3) Indeed, let x G T>(A). Then XR(X; A)x - x = XR(X; A)x - R(A; A) Ax 4 R(A; A)i4x - x Since V(A) is dense in X, it follows from a standard limiting argument that for every x in X, XR(X; A)x -4 x in X as A -4 oo. But for all x in V(A), 5ax = -Ax 4* X2R(X; A)x = -Ax + A#(A; i4)Ax - XR(X; A)Ax 4 XR(X; A)Ax = XR(X-,A)Ax and hence we get = R(A; A)(XI - A)x 4 #(A; A) Ax - x = x 4 #(A; A)Ax — x = R( A; A)4x and hence M \\XR(X;A)x-x\\x = \\R(X;A)Ax\\x < t \\Ax\\x -4 0 A — CO as A -4 oo. Moreover, for all sufficiently large A, ||Ai?(A; A)|| < M-r-^— < 2M. X — oj B\x = A.R(A; A) Aa: -»■ Aa:
134 An Introduction to Pseudo-Differential Operators, 3rc^ Edition as A -* oo. For A > cj, let S\(t) = etB\ t € R. Using the first resolvent formula in Exercise 19.4, we know that R(A; A) = R(»; A)R(A; 4), A, /x > w. Thus, = B^Bx and we get = SA(t)BM for all A and fi in (w, oo), and all t in M. Hence for all x G B[A) and t € [0, oo), Sx(t)x - SM(t)s = J ^[SM(< “ *)$*(*)*] ds = [* 5„(* - s)Sx(s)(Bx - B^x ds. (19.4) Jo By (19.2) and (19.4), there exists a positive constant A0 such that ||Sx(t)x - SM(t)x||x < M2tetu'\\Bxx - B»x\\x, t > 0, (19.5) for all A and p in (A0,oo). By (19.3) and (19.5), we see that for all x in V(A) and all t in [0, oo), ||Sx(t)x - S„(f)x||x -> 0 as A and n tend to oo. So, in view of (19.2), we see that for all x in X and all t in [0, oo), ||S\(t)x-S/t(x)||x as A and n tend to oo. Therefore for all x in X and all t in [0, oo), we define T(t)x by T(t)x = lim Sx(t)x. A—yoo It is then easy to see that T(t) is a bounded linear operator on X and ||T(t)|| < MetUl (19.6) for all t in [0, oo). Let x € V{A). Then, by (19.5), Sx{t)x -> T{t)x uniformly with respect to t on compact subsets of [0, oo) as A —> oo. Therefore for all x in X, the mapping [0, oo) T(t)x G X is continuous. To check the semigroup property, we first note that T(0) = lim 5A (0) = /. A-»oo
One-Parameter Semigroups Generated by Pseudo-Differential Operators 135 Moreover, for 0 < s, t < 00 and all x in X, T(s + t)x = lim Sx(s + t)x = lira Sx(s)Sx(t)x \—>00 A—>00 and, by (19.2), \\Sx(s)Sx(t)x -T(s)T(t)x\\x = ||5a(«)5a(*)* - Sx(s)T(t)x + Sx(s)T(t)x - T(s)T(t)x\\x < ||SA(s)|| ||5A(t)* - T{t)x\\x + ||(5a(«) - T(s))T(t)x\\x < Me°“'\\Sx(t)x - T(t)x\\x + ||(5a(«) - T(*))r(t)*||x -* 0 as A oo. It remains to prove that A is equal to the infinitesimal generator B of the one-parameter semigroup {T(t) : t > 0}. Let x 6 X. Then S\(t)x - x — f 4~S\(s)xds = f S\(s)B\xds. (19-7) Jo as Jo For all x in T>(A) and all s in [0, oo), \\Sx(s)Bxx-T(s)Ax\\x = \\S\(s)B\x - S\(s)Ax + S\(s)Ax - T(s)Ax\\x < IISawil \\Bxx - Ax\\x + \\(Sx(s) - T(s))Ax\\x -> 0 uniformly with respect to s on compact subsets of [0, oo) as A —> oo. Thus, by (19.7), T(t)x — x= f T(s)Axds, x G 'D(A). Jo Therefore Bx = lim —- = lim f T(s)Axds = Ax, xeV(A). t—>o+ t t-^o+ J0 v 7 Hence B is an extension of A. So, we only need to prove that V(A) = V(B). To this end, we claim that there exists a real number Ai such that \ep(B), A>Ai. (19.8) Assuming the claim for a moment, then there exists a real number A such that Thus, \ep(A)nP(B). (AI - B)V(A) = (AI - A)V(A) = X
136 An Introduction to Pseudo-Differential Operators, Zrd Edition and (AI - B)V(B) = X. So, if x G V(B), then there exists an element z in V(A) such that (AI - B)z = (AI - B)x. Since A/ — B is one-to-one, z = x € 'D(A). It remains to prove the claim, i.e., (19.8). For A > cji, we define R(X)x for every x in X by That the integral exists is a consequence of (19.6). It also follows from (19.6) that R(A) is a bounded linear operator on X. Now, for A > ui and all x in X, BhR{X)x T(h)R(X)x - R(X)x h -» XR(X)x - x as h ->• 0+. Thus, for A > oji and all x in X. R(X)x e V(B) and BR(X)x = XR(X)x - x, which is the same as (XI - B)R(X)x = x. So, if A > ui and x e V(B), then R(X)x G V(B) (19.9)
One-Parameter Semigroups Generated by Pseudo-Differential Operators 137 and ‘OO BR(X)x = B e~xtT(t)xdt Jo *oo e~xtBT(t)xdt 0 rOO e~uT{t)Bxdt R(X)Bx. (19.10) By (19.9) and (19.10), R(X)(XI - B)x = x. So, for A > wi, (XI - B)_1 = R(X) and the proof of the sufficiency part of the theorem is complete. As for the necessity, we let g : [0, oo) —> R be the function defined by g(t) = In ||T(t)||, t 6 [0, oo). Then g is subadditive. Indeed, for all s and t in [0, oo), g(s +t) = In ||T(s + t)\\ = In (||T(s)T(t)||) < In (||T(s)|| ||T(t)||) = In ||T(s)|| + In ||T(f)|| = g(s) + g(t). Let t0 > 0. Then for all t in [0, oo), we write t = nt0 + s, where n is an integer depending on t and 0 < s < t0. By the subadditivity of g, we get Let S > infto>0 Then there exists a positive number R such that g(t) < ng(t0) g(s) g(t0) t ~ t t to as t -¥ oo. So, and hence t>R t g(t) c sup < 8. Thus, ln||T(0ll<«. *>R-
138 An Introduction to Pseudo-Differential Operators, 3rd Edition or l|T(t)||<e« t>R. Since is a continuous function of t on the interval [0, iJ], there exists a positive constant C for which \\T(t)\\ < Cest, t € [0, i?]. Hence there is a positive constant M such that ||T(i)|| < Mest, t € [0, oo). (19.11) Now, for A > S and for all x in X, we define R(\)x by /»oo R(X)x = / e~xtT(t)xdt. Jo Since A is the infinitesimal generator of the one-parameter semigroup {T(t) :t> 0}, we can repeat the analysis given above with A for B and S for to conclude that (J, oo) C p(A) and pOO R(X; A) = / e~xtT(t)xdt (19.12) Jo for A > S and for all x in X. For all A and p in (5, oo), the first resolvent formula in Exercise 19.4 gives R(A; A) - A) = (p - A)JI(A; A)R(p; A). Thus, ±R(X;A) = -R(X;A)2 and, by induction, we get ^R(\;A) = (-irn\R(X;A)n+1 for all A in (5, oo). Differentiating both sides of (19.12) n — 1 times with respect to A and using the preceding formula for ^rR(A; A), we get R(A; A)n = / e~xttn~1T(t)x dt (19.13) (n - 1)! J0 for A > S and for all x in X. So, by (19.11) and (19.13), »*<* ^11s = for all positive integers n. □ Remark 19.6. We leave it as Exercise 19.6 to prove that the function u : [0, oo) -> X defined by u{t) = T(t)/, t > 0, is the unique solution of the initial value problem (19.1). Thus, etA is the natural notation for T(t) for t > 0.
One-Parameter Semigroups Generated by Pseudo-Differential Operators 139 As an application to pseudo-differential operators, we give the following theorem. Theorem 19-7. Let a G S2m, m > 0, be a symbol such that we can find a positive constant C and a constant Ao for which Re <p) > C\M|^>2 - Ao|M|i VeS. (19.14) Then the operator Tc$ is the infinitesimal generator of a one-parameter semigroup of bounded linear operators on L2(En). For a proof of Theorem 19.7, we need the following lemma. Lemma 19.8. Let A > A0. Then for every f G L2(Rn), there exists a unique solution u G if2m»2 of the equation (AI — T<r$)u = /, where I is the identity operator on L2(Rn). Moreover, ||(A/ - T,,o)u||a > (A - Ao)|M|2, u € H2m'2. (19.15) Proof Let A > Ao- Then, by (19.14), Re ((AI - Ta)(p,ip) = Re((A0J - Ta)<p,<p) + (A - A0)|M|\ >C||*2 + (A-Ao)IMIi > (A — A0)||<^||i, ¥>€5. So, by a limiting argument, Re ((AI - 7Vto)u, u) > (A - A0)||«||i, « € H2m'2. Thus, for every / G L2(Rn), we can use Theorem 18.2 to obtain a unique solution u G #2m’2 of the equation (AJ-^,0)14 =/■ Moreover, IKAi-r^oMil = (A — Ao)2H^lll + 2(A — Ao)Re((AoI — Ta$)(p,(p) -I- ||Ao/ — > (A — Ao)2|Ml2? <p € S. Thus, by a standard limiting argument again, ll(AI - T,to)«||a > (A - A0)|M|2, u 6 H2m’2. □ Proof of Theorem 19.7 Ta,o is a closed and densely defined linear oper¬ ator from L2(Rn) into L2(IRn). By Lemma 19.8, (XI - To)”1 exists for A > Ao* Let A > Ao* Then, by (19.15), ll(AJ - T\,o)_1|| < (A - Ao)-1-
140 An Introduction to Pseudo-Differential Operators, 3r<* Edition Thus, ||(AI - Ta%0)-n|| < ||(A7 - TA,o)-1||n < (A - A0)-n, n = 1,2,.... Hence, using the Hille-Yosida-Phillips theorem, the proof is complete. □ The following theorem gives a familiar class of symbols a for which —Ta is the infinitesimal generator of a one-parameter semigroup of bounded linear operators on L2(Rn). Theorem 19.9. Let a G S2m, m > 0, be a strongly elliptic symbol Then —Tato is the infinitesimal generator of a one-parameter semigroup of bounded linear operators on L2(Rn). Proof By Garding’s inequality, i.e., Theorem 17.1, we can find a positive constant C and a constant Cs for every real number s > \ such that Re p) > C\\p\\2m>2 - Cs\\p\\h-,,2, p€S. (19.16) If Cs < 0, then Re(T^,<p) > CWpW^, p&S. Now, suppose that Cs > 0. Let e G (0, C/Cs). Since m - s < m, it fol¬ lows from Erhling’s inequality in Exercise 12.6 that there exists a positive constant C£ for which + C'M\l p€S. (19.17) Thus, by (19.16) and (19.17), Re (Trp, p)>(C- Cse)||vC,2 - C.C,M\\, p € 5. So, by Theorem 19.7, the operator Tc$ is the infinitesimal generator of a one-parameter semigroup of bounded linear operators on L2(Rn). □ Exercises 19.1. Let A be a bounded linear operator from a complex Banach space X into X. (i) Prove that for t > 0, the series YlkLo *~W~ converges absolutely. (We denote the sum by etA.) (ii) Prove that for all x G X, e(t+h)Ax __ etAx lim : h—¥ o+ h AetAx, t > 0.
One-Parameter Semigroups Generated by Pseudo-Differential Operators 141 (iii) Prove that for all s, t € [0, oo), esAetA = e(s+t)A' (iv) Prove that e0A = J, where I is the identity operator on X. (v) Prove that for all t € [0, oo), AetA = etAA. 19.2. Let {T(t) : t > 0} be a one-parameter semigroup on a complex Banach space X. Prove that the subspace W of X defined by is dense in X. 19.3. Let A be a closed linear operator from a complex Banach space X into X with dense domain. Prove that p(A) is an open subset of the complex plane C. 19.4. Let A be a closed linear operator from a complex Banach space X into X with dense domain. Prove that for all Ai and A2 in p(A), fl(Ax; A) - R(\2;A) = (A2 - X1)R(X1;A)R(X2;A). (This is known as the first resolvent formula.) 19.5. Prove (19.4) in the proof of the Hille-Yosida-Phillips theorem. 19.6. Prove that the function u : [0,00) ->• X defined by is the unique solution of the initial value problem (19.1). 19.7. Let X be a complex Banach space. Let A be a closed linear operator from X into X with dense domain P(A). Prove that A is the infinitesimal generator of a one-parameter semigroup {T(t) : t > 0} on X with u(t) = T(t)f, t > 0, I|T(*)||<1, t>0, if and only if (0,00) C p(A) and ll*(A,A)||<i
142 An Introduction to Pseudo-Differential Operators, 3rd Edition for all A G (0, oo). (A one-parameter semigroup {T(t) : t > 0} on a complex Banach space X with the property that I|T(*)II<1, t>0, is called a one-parameter contraction semigroup.) 19.8. Let X be a complex and separable Hilbert space in which the inner product is denoted by (, )x- A closed linear operator from X into X with dense domain T>{A) is said to be dissipative if Re (Ax, x)x <0, x e V(A). Prove that A is the infinitesimal generator of a one-parameter contraction semigroup if and only if A is dissipative and the operator XI- A is surjective for all A G (0, oo).
Chapter 20 Fredholm Operators Among the basic questions in the study of equations are the existence and uniqueness of solutions. However, it is in the nature of the subject that we cannot always expect to have both for a given equation. So, we have to seek the second best. To see what is sensible to ask for, let us note that existence and uniqueness are intimately related to, respectively, the range and the null space of the operator associated with the equation under investigation. Intuitively, we naturally want the range to be big and the null space to be small To make these ideas precise, we introduce Fredholm operators in this chapter. Since there are very lucid accounts of Fredholm operators in the literature, we give only the results that we need about these operators. Details and proofs can be found in [Schechter (2002)]. Definition 20.1. Let A be a bounded linear operator from a complex Banach space X into a complex Banach space Y. Suppose that (i) the range R(A) of A is a closed subspace of Y, (ii) the null space N(A) of A is finite-dimensional, (iii) the null space N(A1) of the adjoint A£ of A is finite-dimensional. Then we call A a Fredholm operator. Definition 20.2. Let A be a Fredholm operator from X into Y as in Definition 20.1. Then the index i(A) of A is defined by i(A) = dim N(A) - dim iV(A£), where dim N(A) and dim N(At) are the dimension of N(A) and the dimen¬ sion of AT(A£) respectively. Remark 20.3. The requirement that N(A) be finite-dimensional is what we mean by saying that the null space of A is small. It is well known that the quotient space Y/R(A) can be identified with N(At). Thus, the condition 143
144 An Introduction to Pseudo-Differential Operators, 3rd Edition that N(At) is finite-dimensional can be translated into the requirement that the range R(A) of A is big in Y. Remark 20.4. Let K be a compact operator on X. Then it is well known that I — K is a Fredholm operator on X and %{I - K) = 0. This fact is the Riesz theory of compact operators on Banach spaces. The following theorem is the main tool that we use to study Fredholm pseudo-differential operators. It is known as Atkinson’s theorem. Theorem 20.5. (Atkinson’s Theorem) Let A be a bounded linear op¬ erator from X into Y. Then A is Fredholm if and only if we can find a bounded linear operator from Y into X, a compact operator K\ on X and a compact operator K<i on Y such that BA = I - Ki on X, where I is the identity operator on X, and AB = I -K2 on Y, where I is the identity operator on Y. We give a necessary and sufficient condition for a bounded linear oper¬ ator to have a closed range. Theorem 20.6. Let A be a bounded linear operator from a complex Banach space X into a complex Banach space Y. Suppose that A is injective. Then the range R(A) of A is closed in Y if and only if there exists a positive constant C such that Mx < C\\Ax\\y, xeX. Proof Suppose that R(A) is closed in Y. Then R(A) is a Banach space with norm inherited from Y. So, A : X —> R(A) is a bijective and bounded linear operator. Thus, by the bounded inverse theorem, A~l : R(A) X is a bounded linear operator. So, there exists a positive constant C such that p-'yllx < C’llvllr, y € R(A). This means that WA-'AxWx < C\\Ax\\y, xgX.
Fredholm Operators 145 Thus, INIx < C\\Ax\W, Conversely, suppose that there exists a positive constant C such that \\x\\x<C\\Ax\\y, xeX. Let {yk} be a sequence in R{A) such that Vk^y in Y as k —► oo. For k = 1,2,..., let Xk G X be such that Axk — yk- Then 11¾ - Xk\\x < C\\Axj - Axk\\y = 11¾ " Vk\W ->• 0 as j, k -t oo. So, {xk} is a Cauchy sequence in X. Since X is complete, it follows that Xk -> x for some x in X as oo. But as fc —>• oo. Therefore -> y Ax = y. Hence y £ R(A) and this proves that R(A) is a closed subspace of Y. □ That the index behaves like the logarithm converting multipication into addition is made precise in the following theorem. Since the index is com¬ puted only for Fredholm operators on Hilbert spaces in this book, the the¬ orem is stated and proved in a Hilbert space setting. Theorem 20.7. Let X, Y and Z be complex, separable and infinite¬ dimensional Hilbert spaces. Let A\ : X ->• Y and A2 :Y Z be Fredholm operators. Then A2A\ :X ->• Z is a Fredholm operator and i(A2Ai) = i(A2) + i(A\). Before giving a proof of Theorem 20.7, it is helpful to give a formula for the null space N {A1) of the adjoint A1 of a Fredholm operator A from X into Y, where X and Y are complex and separable Hilbert spaces. Proposition 20.8. Let X and Y be complex and separable Hilbert spaces. Let A be a Fredholm operator from X into Y. Then Y = N(At)®R(A).
146 An Introduction to Pseudo-Differential Operators, 3rd Edition The proof of Proposition 20.8 is left as an exercise. We note here that each of N(At) and R(A) is an orthogonal complement to the other. Of particular importance here is the fact that N{At)=R(A)±, which we also write as N(At)=YeR(A). Proof of Theorem 20.7 That A2 Ai : X Z is a bounded linear operator is trivial. By Atkinson’s Theorem 20.5, there exist bounded linear operators B\ : Y X and B2 : Z Y, and compact operators K\ on X, K2 and Kz on Y and K4 on Z such that B1 Ai = I-K 1 on X, where I is the identity operator on X, AlBl=I-K2 on y, where I is the identity on Y, B2A2 = I — Kz on y, where I is the identity operator on Y and A2B2 = i-k4 on Z, where I is the identity operator on Z. Thus, BxB2A2Ax = £1 (I - Ks)Al = BxAi - B\KzA\ BxKzAi on X and A2A\B\B2 — A2(I — K2)B2 = A2B2 — A2K2B2 — I — K4 — A2K2B2 on Z. Thus, by Atkinson’s Theorem 20.5 again, A2 A\ : X —> Z is Fredholm. Now, let Mi be the finite-dimensional subspace of Y given by Mi = R(Ai) fl N(A2), (20.1) and we write R{A\) = Mi 0 M2, N(A2) = Mi 0 Mz (20.2) and y = R(Ai) 0 Mz 0 M4, (20.3)
Fredholm Operators 147 where M3 and M4 are finite-dimensional subspaces of Y and M2 is a closed subspace of Y. Now, let X\ be the subspace of X defined by Xi = N{A2A1) © N{Ai). (20.4) Then Ai(X1) = M1. (20.5) Indeed, let xi G X\. Then Axx\ G R(A\). Also, by (20.4), A2A\X\ = 0. Therefore, by (20.1), A\(X\) C Mi. Conversely, let m 1 E Mi. Then, by (20.1), mi = A\x for some x E X and, by (20.1) again, Azmi = A2A\x = 0. Therefore x E N{A2A\). If we write x = xo - w, where xo E N(A2Ai) and w G N(A\), then Aix = Aix0 - Aiw = Aix = mi. Thus, Mi C Ai(Xi). Let Z4 be the subspace of Z given by Z± = R(A2)eR{A2Ai). (20.6) Then Z4 = A2(M4). (20.7) To prove (20.7), we just note that for all y G T, by (20.3), A2y = A2Aix0 + A2m3 + A2m4, where xq G X, m3 G M3 and 7714 G M4. Thus, A2y = A2Aix0 + A2m4, and the proof of (20.7) is complete. Let dj = dim Mj, j = 1,3,4. Since Ai : Xi Mi and A2 : M4 Z4 are bijective, it follows from (20.5) and (20.7) that dim Xi = d\ (20.8) and dim Z4 = d4 (20.9) respectively. Thus, by (20.4)-(20.6), i(A2Ai) = dimiV(A2Ai) - dim(Z © R{A2A\)) = dimiV(Ai) + d\ — dim (Z © R(A2) 0 Z4) = dimiV(Ai) + di — dim (Z © R{A2)) — d*.
148 An Introduction to Pseudo-Differential Operators, 3rd Edition On the other hand, by (20.2) and (20.3), i(A2)+i{Ai) = dimN{A2) - dim (Z © R(A2)) + dimN(AX) - dim (Y © R(Ai)) = d\ + c?3 — dim (Z © R(A2)) + dim iV(Ai) — ds — = d\ - dim (Z © R(A2)) + dim N(Ai) - d4. Thus, i(A2Ai) = i(A2) + i(A\), as claimed. □ Given a Fredholm operator, the problem of computing its index is usu¬ ally an important but difficult problem. We give in this chapter a formula for computing the index of a Fredholm operator from a Hilbert space X into a Hilbert space Y. As the formula is expressed in terms of traces of trace class operators on Hilbert spaces, we first recall without proofs the basic facts on trace class operators and traces, which can be found in, for instance, [Lax (2002)] and [Reed and Simon (1980)]. Let A be a compact operator on a complex and separable Hilbert space X. Then A1 A is a compact operator on X. Moreover, A1 A is a positive operator on X in the sense that (AlAx,x)> 0, x € X. Then it is well known that the operator A1 A has a positive and compact square root, which we denote by y/AlA. This simply means that y/A*A is a positive and compact operator on X such that its square is the same as A1 A. By the spectral theorem, there exists an orthonormal basis {<pk} for X consisting of eigenvectors of y/AtA. For k = 1,2,..., let Sk be the eigenvalue of y/AlA corresponding to the eigenvector <pk. Then we say that A is a trace class operator on X if oo Sk < 00 k=i and we say that A is a Hilbert-Schmidt operator on X if oo 5^4 <00. k=1 Remark 20.9. Let Si be the set of all trace class operators on X and let S2 be the set of all Hilbert-Schmidt operators on X. Then it is well known that S\ and S2 are two-sided ideals in the set B(X) of all bounded linear operators on X.
Fredholm Operators 149 Let A : X X be a bounded linear operator. Suppose that oo $3l(A¥’fc>Vfc)x| <00 k=l for all orthonormal bases {y>k} for X. Then it can be proved that A is a trace class operator. Let A be a trace class operator on X. Then it can also be proved that YlkLi(A<Pki<Pk)x is absolutely convergent for all orthonormal bases {<£&} for X and the sum is independent of the choice of the orthonormal basis. This sum is termed the trace of the trace class operator A on X and is denoted by tr(A). It is easy to prove that if A and B are trace class operators on X and c is any complex number, then A + B and cA are also trace class operators on X. It is also easy to prove that We need the following property of the trace. Theorem 20.10. Let A be a trace class operator on a complex and sepa¬ rable Hilbert space X and let B be a bounded linear operator on X. Then AB and BA are also trace class operators on X. Moreover, A proof of Theorem 20.10 can be found on page 334 of [Lax (2002)]. As examples of Hilbert-Schmidt and trace class operators on L2(Rn), we give one for Hilbert-Schmidt operators and one for trace class operators. They are useful to us for studying Fredholm pseudo-differential operators on L2(Rn). Theorem 20.11. Let A : L2(Rn) L2(Rn) be a bounded linear operator. Then A is a Hilbert-Schmidt operator if and only if there exists a function K E L2(Rn x Rn) such that for all f £ L2(Rn). The function K is usually called the kernel of the Hilbert-Schmidt op¬ erator A. tr (A + B) = tr(A) + tr(J5) and tr(cA) = ctr(A). tr (AB) = tr (BA). (Af)(x)= K(x, y) f(y) dy, xeRn,
150 An Introduction to Pseudo-Differential Operators, 3rd Edition Theorem 20.12. Let A be a bounded linear operator on L2(Rn) given by for all f € L2(Rn), where K is a Schwartz function on W1 x Rn. Then A is a trace class operator on L2(Rn) and Theorem 20.12 is a special case of the more general results in [Brislawn (1988)]. We can now give the formula in Section 30.7 of [Lax (2002)] for the index of a Fredholm operator from one Hilbert space into another Hilbert space. Theorem 20.13. Let X and Y be complex, separable and infinite¬ dimensional Hilbert spaces. Let A : X Y be a Fredholm operator. Sup¬ pose that we can find a bounded linear operator B :Y X, a trace class operator T\ on X and a trace class operator T2 on Y such that where I is the identity operator on Y. Then the index i(A) of A is given by (Af)(x) = K(x,y)f(y)dy, iGP, BA = I-TU where I is the identity operator on X, and AB = /- T2, (20.10) (20.11) (20.12) and Y = R(A)®Y0,
Fredholm Operators 151 where Xo and Yq are the orthogonal complements of N(A) in X and R(A) in Y respectively. Let P be the projection of X onto Xo along N(A). Then AP = A. (20.13) Indeed, let x G X. Then we can write x = z + xo, where z G N(A) and xo G Xq. So, APx = Axq and Ax — Ax o. Thus, APx = Ax, which gives (20.13). By (20.12) and (20.13), we get APTi = T2A. (20.14) It is obvious that A : Xo R(A) is a bijection. Since P : X Xo is a projection, it follows that PT\ maps Xo into X0. By (20.12), T2 maps R(A) into R(A). Now, tr(FTilx„)=tr(T2|fl(jl)). (20.15) Indeed, by Exercise 20.11, there exists a surjective isometry U of R(A) onto X0. Then, by (20.14), UAPTX = UT2A = UT2U~lUA and hence (U A)(PTi)(U A)~x = UT2U~x. Since each factor on the left hand side and the operator on the right hand side of the preceding equation maps Xo into Xo, we can invoke Theorem 20.10 to conclude that tr(PT1|Xo)=tr([/r2t/-1|Xo). (20.16) Now, we note that tr (20.17) Indeed, let {xj-} be an orthonormal basis for Xo. Then {U~lx*•} is an orthonormal basis for i?(A). Thus, oo tr(i/ratr1|jro) 3=1 = ^T2U-lx%U-lx])Y = tr (t2|r(/4)) .
152 An Introduction to Pseudo-Differential Operators, 3rd Edition So, (20.17) is proved and (20.15) follows from (20.16) and (20.17). Let be an orthonormal basis for N(A) and let {#*•} be an or¬ thonormal basis for Xq. By (20.10), T\ — I on N(A). It follows that dim N(A) oo tr(2i) = 53 (TlZj,Zj)x + Y,(TiX0j,x°j)x j= 1 3=1 dim N(A) oo = E (zi>zi)*+ E(ri xj>Pxj)x j=l j=1 dim N(A) oo = E i=i i=i = dim N(A) + tr (PTi |Xo). (20.18) Let {j/jA*4 ' be an orthonormal basis for lo and let {wj} be an or- thonormal basis for R(A). Then dim N(At) oo tr(T2) = 53 (r2j,9, y0)Y + 53(^-,Wj)y j=i i=i dim N(At) = 53 (r22/o,2/°)y+tr(T2|iJ{A)). (20.19) 3=1 Since #(/ - 7¼) C -R(A), we see that i?(/ - t2) _l y0. Therefore ((/-7^,^=0, j = 1)2, So, by (20.19), dim N(Al) tr(T2) = 5Z (y%Vj)Y + *r (T2ln(A)) J =1 dim JV(j4*) = E ^(¾)) j=1 = dim JV(A*) + tr (r2|fl(jl)) . (20.20) Thus, by (20.15), (20.18) and (20.20), i(A) = dim N(A) — dim N(At) = tr(Ti) - tr(T2).
Fredholm Operators 153 Exercises 20.1. Can a compact operator from a complex Banach space X into a complex Banach space Y be Fredholm? Explain your answer. 20.2. Can a Fredholm operator from a complex Banach space X into a complex Banach space Y be compact? Explain your answer. 20.3. Let X and Y be complex, separable and infinite-dimensional Hilbert spaces. Let A : X -+ Y be a Fredholm operator and let K : X -+ Y be a compact operator. Prove that A + K : X -* Y is Fredholm and i(A + K)=i(A). 20.4. Let A be a bounded linear operator on a complex and separable Hilbert space X. Prove that A* A is a positive operator on X. 20.5. Let A be a bounded and positive operator on a complex and separable Hilbert space X. Prove that A is self-adjoint, i.e., A1 = A. 20.6. Let A : L2(Rn) -+ L2(Rn) be a Hilbert-Schimdt operator given by the kernel K. Prove that the kernel Kl of Af : L2 (Rn) -+ L2(Rn) is given by Kt{x,y) = K{y,x), x,y€Rn. 20.7. Prove that Si is a subspace of B(X). 20.8. Prove that tr : Si -+ C is a linear functional. 20.9. Prove that if A is a Fredholm operator on a complex ansd separable Hilbert space X, then X = NiA1) © R(A). (This exercise says that N(At) and R(A) are orthogonal to each other.) 20.10. Let M be a closed subspace of a complex and separable Hilbert space X. If we write X = M 0 M1- and let P be the projection of X onto M along M-1. Find Pt. 20.11. Find a surjective isometry U : R(Ai) -+ Xo used in the proof of Theorem 20.13
This page intentionally left blank
Chapter 21 Fredholm Pseudo-Differential Operators We begin with proving that Fredholm pseudo-differential operators on Lp(Mn), 1 < p < oo, with symbols in Sm, —oo < m < oo, are elliptic. To this end, we need some technical preparations. Definition 21.1. For A > 0, r > 0 and #o,£o Gln,we define the operator R\Axo,£o) : Lp(Rn) Lp(Rn), 1 < p < oo, by («A,r(®b,&)«)(*) = XTn/peiXx<au(\T(x - x0)), X € ®n, for all u € Lp(»n). Proposition 21.2. The operator R\^(xo,£q) : Lp(Rn) -> Lp(Rn) *5 a surjective isometry and the inverse is given by (®A,r(xo,£o)-1u)(x) = A-Tn/pe-a(*0+A'T:c)-«0w(xo + A~Tx), x € IT, /or aM u € Lp(Rn). Proof We get This proves that -Ra,t(#o,£o) is an isometry. Now, let v € Lp(Rn). We need to find a function u in Lp(Rn) such that ||#a,t(zo,£oHI£ f |(ftA,r(#0,fo)u)(z)|Pdz Xrn\u(XT(x - x0))\pdx, ti e Tp(Mn). (21.1) Let y = Ar(# — #o) in (21.1). Then we have ||flAfr(®o,&)u||p = ||u||p, u € Lp(Rn). R\,t(xo,£o)u = v. 155
156 An Introduction to Pseudo-Differential Operators, 3rd Edition Let u be the function in Lp(E71) given by u(x) = + \~TX), x € Rn. Then #A,r(tfo,fo)u = v. □ Proposition 21.3. For all u E Lp(Rn) and v E Lp' (E71), where ^ 4- y = 1, 1 < p < oo, (R\,r(Xo^0)u,v) -+ 0 as A -» oo. Proof Let (^(E71). Then we have |(tfA,r(£0,fo)u,v)| < Arn/p [ |i/(Ar(x-aro))||v(a:)|dar J Rn = A™tp f A”rn|u(2/)||v(ar0 + A“Ty)| dy jRn = X~Tn/p' f Kx0 + A-Tj/)|Ky)|dy J Rn < X~Tn/p' sup \v(y)\ f (21.2) y£Rn JRn So, by (21.2), (R\^t(xo£o)u’v) 0 35 00 f°r u>v € Co^E71). Let u E Lp(Wl) and v E Lp>(E71). By density, we can find sequences {<pj} and {'ipj} in Co°(En) such that and (pj -+ u ifrj -+ v in Lp(En) and Lp (Mn) as j -+ oo respectively. Therefore for every positive number e, there exists a positive integer J such that \(R\,t(x0,£o)u,v) - (R\,T(xO,&>)Uj,Vj)\ < £ for all j > J. Then, again using the fact that (R\tT(xo,£o)uj,Vj) -t 0 as A -+ oo for Uj, Vj E Co°(En), the result follows immediately. □ Proposition 21.4. Let a E Sm. Then R\,t(xqi£o) TaR\yT{xo-,Co) = Tax r, (21.3)
Fredholm Pseudo-Differential Operators 157 where <J\,T(x,rj) = a(x0 + A"rx,Afo + ATrj), x,rj G Mn. (21.4) Moreover, if a G 5°, A > 1, 0 < r < 1/2 and fo i= 0, then for all multi- indices a and /3, there exists a positive constant Cp such that l(3?3S,r)MI < C'^aJ,(a)^A-rl“lA-<1-aT)l',l, x,V € R», (21.5) where pa,p denotes the corresponding norm in S°. An ingredient in the proof of Proposition 21.4 is the following Peetre’s inequality. Lemma 21.5. (Peetre’s Inequality) For all t G (-00,00) and for all x,y G Mn, (Tw)'£2l‘l(1 + |l','|i)'‘1' Proof The inequality is obviously true if t = 0. For all y and z in Rn, 1 + \y - *\2 = 1 + (y - z) ■ (y - z) = l + \y\2-2yz + \z\2 <l + \y\2 + 2\y\\z\ + \z\2. Since 2\y\\z\<\y\2 + \z\2, it follows that l + \y-z\2<l + 2\y\2 + 2\z\2 < 2(1 + |y|2)(l + \z\2) for all y and z in Rn. Replacing z by y — x, we get 1 + \x\2 < 2(1 + |j/|2)(l + \x - y\2). If £ > 0, then (1 + |ar|2)* < 2*(1 + |»|2)*(1 + \x-y\2Y, as asserted. If t < 0, then — t > 0 and by what we have proved, (1 + |y|2)-‘ < 2-*(l + |*| V(1 + I* - y\2)-\ which is the same as (l +1*!2)* < 2l*l(l + |*r|a)*(l +1* -
158 An Introduction to Pseudo-Differential Operators, 3rd Edition as required. □ Proof of Proposition 21.4 We first note that for all ip € <S, (!»(*) = (2tt)-"/2 f eix'^a(x, ¢)0(0 ^ = (2n)~n f f et(-x~yH<T(x,£)u(y)dyd£, J]Rn «/Rn where the integral JRn JjRn is understood to be an iterated integral in which the integration is first performed with respect to y and then £. Let u e S. Then for all x £ W1, (R\,T(xo,£o)~lTaR\,r(xo, £o)«) (x) = e-iA(*o+A-^Ko(2,r)-« f f e<A(*o+A-’-*-»)-€<T(aro + A_Tar,0 «/Rn JR" eiXy^u(\T(y-xo))dyd^ = \-Tn(2ir)~n f [ eiX~T^-^x-^a(x0 + A-r*,0«(z)dz<% JR» JR» = (27r)_n f j e^x~z^Tla(xo + A_Tar, A£0 + ATri)u{z) dz dq. JRn «/Rn Thus, we get (21.3) and (21.4), as asserted. Now, using (21.4), the chain rule and Peetre’s inequality, m^x,r)M\ = +A-r®, a&+\rn)\-T'a'\rm\ < paj,{<r)( 1 + |A£0 + Ar.f|)-MA-T|o|ATM < C0pa,0(o)(Ko + <c?pa,?{<!){ Ae0)-|/J|(Ar»?)l's|A-T|a| < C0pa,0(<r)\$>\-m{v)WaP’-WA-’-W (21.6) Hence (21.5) is proved. □ The following theorem is one of the main theorems of this chapter. Theorem 21.6. Let a £ S° be such that Ta : Lp(Rn) ->* Lp(Wl) is a Fredholm operator for 1 < p < oo. Then there are positive constants C and R such that k(*,£)l > c for all x e W1 and |£| > R, i.e., a is elliptic.
Fredholm Pseudo-Differential Operators 159 Proof Since Ta is a Fredholm operator, it follows that we can find a bounded linear operator S on Lp(Rn) and a compact operator K on Lp(Rn) such that STa = I + K. Let M be the set of all points £ in Mn such that there exists a point x in W1 for which W*'01 S 2pii' Now, if M is bounded, then there exists a positive number R such that Thus, for each point £ G Rn with |£| > R, we get, for all iGln, |<r(ae,OI > 2jjsj|> which is the same as saying that a is elliptic. So, suppose that M is not bounded. Then there exists a sequence {(#&,£&)} m x such that l&l -> 00 as k oo and |£*)| 2\\S\\ ’ ^ 1,2,.... Thus, there exists a subsequence of {(#&, £&)}, again denoted by {(#&, £&)}> such that ^ ^oo for some complex number (Too as k -» oo. Therefore l<To01 - 2pii- For k = 1,2,..., let Afc = |£*|. Then, by Proposition 21.4, we have where (21.7) <?\= <r{xk + Aferx, £^ + A*//), x,rj G Rn. Let a and /3 be multi-indices. Then, in view of (21.5), there exists a positive constant Cp such that (21.8)
160 An Introduction to Pseudo-Differential Operators, 3rd Edition For k = 1,2,..., let <t£° be given by <r*° = cta*,t(0,0) = a{xk,ik)- Using Taylor’s formula given in Theorem 7.3 and the estimate (21.8), we get W\„AX,V) ~0fc°| = kA»,r(*,»?) -^Afc,r(0,0)| E f maXk,r){6xMdB +#*l=i = E w|7|nImI/1 ^,#.k)^)l/i|Arw^(1-2T)M^ 17+/.1=1 Jo (21.9) uniformly for (x,rj) on every compact subset K of Rn xln as k —» oo. Let u £ S. Then (7VX ,«)(*) - ^u(ar) = (2#r)-n/2 / eixt,(axk<r(x,v) ~ dV k' jRn for all x € Rn. By (21.9), the assumption that cr € S° and Lebesgue’s dominated convergence theorem, (T,Xk.ru)(*)->o?u(x) (21.10) for all x £ Rn as fc oo. Moreover, for all l € N, using (21.8) and an integration by parts, we can find a positive constant C^ for each p, with | //1 < 21 such that \{x)2l(TaXk,ru)(x)\ (x)21 (2n)~n/2 [ eix T><jXk<T(x,V)u(v)dv J Rn (2ir)-”/2 f ((/ - £)«(£) <*£ Jr* = (2ir)~n/2 f E /r» , .T7l, M! l#»l<a for all x € K", where />(£>) = (/- A)1. So, there exists a positive constant C such that 1(^.^)(^)1^(7^)-^, *€Mn.
Fredholm Pseudo-Differential Operators 161 Now, if 2Ip > n, then (x) 2lp € L1(Rn). So, there exists a positive constant Ci such that l(^fc.r - <r?)«)(*)| < x e Rn. (21.11) Thus, T<ry. U (TonU in Lp(Wl) as k -» oo. Let u be a nonzero function Lp(Rn). Since R\kyT ||jj^ is an isometry, it follows that 0 < Hip = RXk,r (**>|^|)W {STa-K)RXk<r (^>|||) (**’ i&t) STaR\ + \\KRx, i|i) 1^,,(..,,1,).1 Ml Now, using the fact that K is a compact operator and Proposition 21.3, it follows that *Kw(*‘’iti)“ll^0 (21.13) as k -¥ oo. Then by (21.12), IHIP< 11-511 KolNIp. Then (21.7) and (21.13) give the contradiction that 1 I l 1 M -l<To01 - 2pi’ which completes the proof. C The preceding theorem can be generalized to the following theorem. Theorem 21.7. Let a € 5m, -oo < m < oo, and let Ta : H8iP H8~m'p be a Fredholm operator for some s € (—00,00). Then Ta is an elliptic operator.
162 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof The operators : H8* -> H8~m>p, J_s : Hs* Lp(Rn) and Jm-s ' H8~m'p Lp(Rn) are bounded linear operators. Here Js, the Bessel potential, is a pseudo-differential operator with symbol as £ S“s, where MO = £^n- Let Jm—sTaJs = Tr. Then Tt : Lp(En) -> Lp{Rn), where r £ 5°. Since Js is bijective, it follows that Js is Fredholm and elliptic for all s £ (-oo, oo). So, Tr is elliptic. By the fact that Js, s £ Mn, is bijective, it follows immediately that Ta is elliptic. □ For pseudo-differential operators with symbols in 5m, -oo < m < oo, that ellipticity does not imply Fredholmness can be seen from the following example. Example 21.8. Let P(D) — £|a|<m aa Da be a linear partial differential operator with constant coefficients on Mn such that P(D) : Hm'p -+ Lp(Rn) is injective and there exists a sequence {£&} in Rn such that P(Sk) -»• o as k —y oo. Then P(D) : Hm,p —> Lp(Rn) is not Fredholm. More precisely, the range R{P) of the operator P(D) : Hm'p -+ Lp(Rn) is not a closed subspace of Lp(Rn). To this end, let <p be a function in S such that ||^||p = 1. Let {ek} be a sequence of positive numbers such that 4^(6)^0 for 0 < \ii\ < m as k -+ oo. Let {ipk} be the sequence of functions in S defined by <Pk(x) = e^n/pip(skx)etx'ik, x € M", for k = 1,2, — Then ll^fcllp — k — 1,2,....
Fredholm Pseudo-Differential Operators 163 By Leibniz’s formula, we get for fc = 1,2,..., (P(D)<pk)( x) = e~knlp E -i{P(*\Dy**'‘)(.D»v){ekx)£M M<m ^* = Hn/P E -}P{,i)(^)eixik(D^)(skx)£[ri \p\<m = er/pP(^)eix^<p(ekx) + er/p E ^P{l,)(^)eix^(D^)(ekx)e^ l<\p\<m for all x € Rn. So, for fc = 1,2,..., \\P(D)<fk\\P < |P(&)| + E 7^1^(6)1 WD^Wp -+ 0 l<|^|<m as fc -» oo. Thus, we cannot find a positive constant C such that \\Vk\\p < C\\P(D)<pk\\p, & = 1,2,— So, the range R(P) is not closed in Lp(Rn) by Theorem 20.6. For a positive result, we have the following proposition. Proposition 21.9. Let P(D) be a linear partial differential operator with constant coefficients and of order m onln. Then P(D) : ifm’2 ->• L2(Rn) is Fredholm with zero index if and only if 0£{P(0:£eR”}, where {• • • } is the closure in W1 of the set {•••}. Before giving a proof of Proposition 21.9, we need two lemmas. Lemma 21.10. Let P be a polynomial on W1 given by p(o=Ea^°’ £eRn- \a\<m Let Z(P) be the zero set of P, i.e., Z(P) = {£ £ Rn : P(0 = 0}. Then m(Z(P)) = 0, where m is the Lebesgue measure on Rn.
164 An Introduction to Pseudo-Differential Operators, Zrd Edition Proof The lemma is certainly true for n = 1. Suppose that the lemma is true for all polynomials on Rn_1. Write En = Rn_1 x E and any point in En is of the form (£', £n), where £' £ Rn 1. Any multi¬ index a = (ai, a2,..., an) can be written as (o', an), where ol — (ai,a2,...,an-i). Now, P(0=P(Z'An)= £ \a'\<m-an t/oc Thus, Z(P) = j(£',£«)€lT ^ ( y aa',an^nn |a' |<m—\an=0 Let Xz(P) be the characteristic function of Z(P). Then, by the induction hypothesis, we get m(Z(P))=r f XZ(P)(£',£n)«n J—oo iR"-1 = /°° (J Xzwti,U)#)dL = 0. □ Lemma 21.11. Lei P(D) = 6e a /mear partial differential operator with constant coefficients and of order m on En. Then the op¬ erators P(D) : Hm>2 -> L2(En) and P(D)* : L2(En) if"™’2 are both injective. Proof Let u £ Hm,2 be such that P(D)u = 0. Then u £ L2(Rn) by the Sobolev embedding theorem in Chapter 12. Taking the Fourier transform on both sides, we get W = 0 for almost all f £ En. By Lemma 21.10, u = 0 a.e. on En. Therefore u = 0. This proves that P(D) : Hm,2 ->• L2(En) is injective. Now, let u £ L2(Mn) be such that P{D)lu = 0.
Fredholm Pseudo-Differential Operators 165 Then, by Exercise 13.1, Exercise 13.2 and Plancherel’s theorem, (P(D)tu,v) = (u, P(D)v) = (ii,Pv) = (Pu,v) = 0, v € Hm’2. Therefore (Pu, ip)= 0, <p £ S. Since S is dense in L2(Rn), it follows that Pu = 0. By Lemma 21.10, u = 0 a.e. on Rn. Therefore u — 0. So, P(D)1 : L2(Rn) ->> #“m’2 is injective. □ Proof of Proposition 21.9 Suppose that P{D) : Hm’2 -> L2(Rn) is Fredholm with zero index. Then the range R{P) of P(D) : Hm,2 —>• L2(Rn) is closed in L2(Rn). By Theorem 20.6, there exists a positive constant C such that IMka < C||P(Z?)tt||a, u € Hm'\ So, 0¢{P(¢) :€€«»}. For if this were not true, then there would be a sequence {&} in Rn such that P(£k) -> 0 as k oo, and the analysis in Example 21.8 would apply to give a contradiction. Conversely, suppose that 0 i {P(0 :*€**}. Then there exists a positive constant C such that ^ € Bn. So, for all u € Hm,2, by Plancherel’s theorem, ||P(D)w||2 = HF&lla > C||fi||2 = C||ti||2, t« 6 Therefore the range R(P) is closed in L2(Rn). The rest of the proof follows from Lemma 21.11. □ The following result follows from Proposition 21.9. Theorem 21.12. Let P(D) be a linear partial differential operator with constant coefficients and of order m on Mn. Then, for —oo < s < oo, the operator P(D) : Hs’2 -» Hs~m,2 is Fredholm with zero index if and only if 0
166 An Introduction to Pseudo-Differential Operators, 3rd Edition Proof We only need to prove that P(D) : H8'2 -+ H8~~m,2 is Fredholm with zero index if and only if P(D) : Hm'2 -+ L2(En) is Fredholm with zero index. Suppose that P(D) : H5,2 -+ Hs~m'2 is Fredholm with zero index. By Exercise 21.5, P(D) = Jm-SP(D)JS-Tn : Hm'2 -+ L2(En) is Fredholm with zero index. Similarly, the converse is true. □ Remark 21.13. In fact, in Theorem 21.12, if 0 ^ {P(£) : £ E E71}, then P(D) : H5,2 -+ is a bijection for — oo < s < oo. Exercises 21.1. Give an example of a linear partial differential operator P(D) on En such that P(D) is elliptic, injective and there exists a sequence {£*} in En such that ^(6)^0 as k -* oo. 21.2. Prove that a linear partial differential operator P(D) with constant coefficients on En is elliptic if 0£{P(ChiTE^}. Is the converse true? 21.3. Use the definition of 5° to give another formula for pa,p((r) in Propo¬ sition 21.4. 21.4. Use the definition of S° and Peetre’s inequality to fill in the details in deriving (21.6). 21.5. Let P(D) be a linear partial differential operator with constant coefficients and of order m on En. Prove that for all s E (-00,00), P(D) and Js commute in the sense that P(D)JS : H*’2 -+ and JSP{D) : tf*’2 -> 2 are equal for all t € (—00,00).
Chapter 22 Symmetrically Global Pseudo-Differential Operators The pseudo-differential operators studied in this book are global operators on Rn. The decay estimates on the associated symbols due to differentia¬ tions have hitherto been imposed only on the ^-variable. We introduce in this chapter another class of pseudo-differential operators on Mn of which the symbols satisfy similar decay estimates due to differentiations with re¬ spect to the ^-variable and the ^-variable. They are appropriately called symmetrically global pseudo-differential operators. These operators are also known in the literature as SG operators with SG standing for symbol-global, scattering operators and operators with exit at infinity. For the sake of con¬ venience and the close match with the title of this chapter, we simply call these operators SG operators and their symbols SG symbols. Let mi, m2 € (—00,00). Then we let Smi,m2 be the set of all functions in C00 (Rn x En) such that for all multi-indices a and /3, there exists a positive constant Ca,p for which \(D%Dl<T)(x,t)\ < Ca,^)m2~laiiOmi~W, € Rn. A function in Smi ,m2 is said to be a SG symbol of order (mi, m2). It is clear that if a G Sm 1,7712 and m2 < 0, then a G Smi, where Smi is the class of symbols of pseudo-differential operators studied before this chapter in the book. Let a G 5mi,m2. Then we define the SG pseudo-differential operator Ta with symbol a by (7»(z) = (2tt)-"/2 / eix<a(x,0<p(0dti, * € M", (22.1) JRn for all functions (p in the Schwartz space S. It can be proved easily that Ta : S -» S is a continuous linear mapping. (See Propositions 6.7 and li.i.) We begin with the product formula. 167
168 An Introduction to Pseudo-Differential Operators, 3r<i Edition Theorem 22.1. Let a € Sm'-m2 and r € S'11-"2. Then TaTT = Tx, where A € Smy+<*i ."»*+<** and M **' iTere the asymptotic expansion means that for every positive integer M, there exists a positive integer N such that A- ^ 6 5"»1+Mi-M,m2+M2-M ImI <n The next important result is the formula for the formal adjoint. Theorem 22.2. Let a G Smi,m2. Then the formal adjoint T* of Ta is a SG pseudo-differential operator Tt, where r G Smi,m2 and T ~ E (—*)|M| d$d»a. Here the asymptotic expansion means that for every positive integer M, there exists a positive integer N such that ■ _ ( QV’QV'tf q ^mi-M,rri2-M ^ n\ x S As in Chapter 11, we can extend the definition of a SG pseudo¬ differential operator from the Schwartz space S to the space <S' of all tem¬ pered distributions. Let a G Smi,m2. Then for all u in S', we define the linear functional Ta : S C by (Tau)(<p) = u(T*Jp), ipeS. As an analog of Proposition 11.6, Ta maps S' into S' continuously. The more interesting result is the following L^-boundedness theorem. It follows from the fact that each symbol in 50,0 is in 5° and Theorem 11.7. Theorem 22.3. Let a G 50,0. Then Ta : Lp(Rn) -» Lp(Rn) is a bounded linear operator for 1 < p < oo. We now come to the ellipticity of SG pseudo-differential operators. Let a G Smi,m2, -oo < mi,m2 < oo. Then cr is said to be SG elliptic if there exist positive constants C and R such that k(*,0| > C'<*>TOa(0m‘, l*l2 + Kl2 > R2-
Symmetrically Global Pseudo-Differential Operators 169 The following theorem tells us that SG elliptic operators have SG para- metrices. Theorem 22.4. Let a G 5mi,m2, —oo < mi,m2 < 00, be SG elliptic. Then there exists a symbol r in S~mi’~m2 such that TTTa = I + R and TaTr = / + S, where R and S are infinitely smoothing in the sense that they are SG pseudo-differential operators with symbols in ' The SG pseudo-differential operator Tr in the preceding theorem is known as a SG parametrix of Ta. The next development of the theory is to determine the domain of a SG elliptic pseudo-differential operator. We need Sobolev spaces for SG pseudo-differential operators for this task. For Si,$2 G (—00,00), we let Jsi,s2 be the Bessel potential of order (si,S2) defined by J$1,S2 = 31,82 ’ where ctsus2M = (x)-82(0'91^ Obviously, cr8uS2 G S~Sl~82. It is left as Exercise 22.6 to show that the mapping J5l,52 : S' S' is a bijection for —00 < «i, $2 < 00, and J.-U = (22-2) and hence, by Theorem 22.1, J~\St is a SG pseudo-differential operator of order (-^1,-^2)- For 1 < p < 00, and -00 < si,s2 < 00, the U’-Sobolev space HSl'S2’p of order (*’i, s2) is defined by tf81'*2’3’ = {u € S' : J-Sl,-s2n <E Lp(Rn)}. Then H8uS2’p is a Banach space in which the norm || ||3l,S2lP is given by IMI»1,*2,P = ^ H ’ 2’P- Obviously, Ap = Lp(En). Theorem 22.3 can now be improved to the following result. Its proof is left as an exercise. Theorem 22.5. Let a € Smi’m2, -oo < mi,m2 < oo, be a SG symbol. Then T * HSliS2iP ff8^~~Tn^*82~rn2,p is a bounded linear operator for 1 < p < oo and -oo < s\, $2 < oo.
170 An Introduction to Pseudo-Differential Operators, 3rd Edition We have the following Sobolev embedding theorem for SG pseudo- differential operators. (See Theorem 12.5 and Remark 12.10.) In Theorem 12.5, we have a bounded linear operator of which the norm is at most one. There is a short proof if we relax the requirement of the norm. The proof of the following theorem is also not difficult if we just need the inclusion to be a bounded linear operator. This is a good exercise. Theorem 22.6. Let si,S2,ti,t2 £ (—00,00) be such that si < t\ and S2 < Then Htl,t2jP C H81,S2,P and the inclusion i : Htl,t2iP c—HSliS2,p is a bounded linear operator. The most important feature of the Sobolev spaces for SG pseudo¬ differential operators is the following compact embedding theorem, which is not true for the Sobolev spaces in Chapter 12. Theorem 22.7. Let Si,S2,h,t2 £ (-00,00) be such that Si < t\ and S2 < Then the inclusion i : Htl,t2'p Hs 1,S2,P is a compact operator. A proof of Theorem 22.7 depends on the following result in [Wong (1983, 1994)]. It is a result in [Grushin (1970)] at least for the L2-case and we skip the proof. Theorem 22.8. Let a G Sm, m G (-00,00), be such that lim Cad3(3)=0 | x | —^00 for all multi-indices a and /3. Then for every positive numbers, the operator Ta : Hs+m'p —► Hs~£,p is compact for — 00 < s < 00 and 1 < p < 00. The following corollary follows immediately. We leave it as an exercise. Corollary 22.9. For every positive number e, Je,£ : Lp(Rn) —► Lp(Rn) is a compact operator for 1 < p < 00. Proof of Theorem 22.7 Let e be a positive number such that t\ — si — e > 0 and *2 - 82 - e > 0. Then, by Theorem 22.1 and (22.2), J~£j-Sli-S2 is a SG pseudo-differential operator of order (51 + e, S2 + ¢). So, j—l j . jjtifaiP jjti—si—e,t2—S2—e
Symmetrically Global Pseudo-Differential Operators 171 is a bounded linear operator, the inclusion i: Hl Lp(Rn) is a bounded linear operator, the operator Je,e : Lp(Rn) -+ Lp(En) is compact, and the operator JSu82 : Lp(En) -+ H8"82'p is a bounded linear operator. Thus, i = : Htl,t2,p -» HS1'S2’P is a compact operator. The first main result in this chapter is the following theorem. Theorem 22.10. Let a € Smi’m2,-oo < mi,m2 < oo, be SG elliptic. Then for all sus2 € (-00,00), T<, : H*l'32'p ->■ HSl~mi'32~m2’p is a Fred- holm operator for 1 < p < oo. Proof Since a is SG elliptic, it follows from Theorem 22.4 that there exists a symbol r in S~~m 1,-m2 such that TrTCT = I + # and 71* Tr = / + 5, where and 5 are infinitely smoothing in the sense that they are SG pseudo-differential operators with symbols in ^or positive numbers t\ and t2, the linear operator R : Lp(Rn) -+ Lp(En) is the same as the composition of the linear operators R : Lp(En) -+ 7ftl,t2,p and i. jjtut2,p £,p(En). Since 1? : Lp(En) -+ Htut2'p is bounded by Theorem 22.5 and i: HUM'P *-+ Lp(Rn) is compact by Theorem 22.7, it follows that R : Lp(Rn) -+ Lp(Rn) is compact. Similarly, 5 : Lp(En) -► Lp(En) is compact. So, by Theorem 20.5 of Atkinson, TV,o is Fredholm. □ We give an analog of Theorem 21.7 for SG pseudo-differential operators. Theorem 22.11. Let a G 5mi’m2, mi,m2 G (-00,00), be such that the operator TV : HSl'82'p -+ Hs^ Fredholm for si,s2 G (-00,00) and 1 < p < 00. 77&en o* «5 SG elliptic.
172 An Introduction to Pseudo-Differential Operators, 3rd Edition We first establish an analog of Theorem 21.6 for SG pseudo-differential operators. Theorem 22.12. Let a G 50,0 be such that Ta : Lp(Wl) —> Lp(Rn) is Fredholm for 1 < p < oo. Then we can find positive constants C and R such that for all x and f in W1 with K*iOI > o \x\2 + |£|2 > R2 Proof By Atkinson’s theorem, i.e., Theorem 20.5, we can find a bounded linear operator S on Lp(Wl) and a compact operator K on Lp(Rn) such that STa = I + K. Let M be the subset of W1 x Rn given by {(*>£) e En : |ff(ar,OI < 2p||} ' We first suppose that M is bounded, i.e., there exists a positive number R such that So, \(x,0\<R, (x,0eM. X2 +£2 > R2 => (x,£) ¢ M => \cr(x, ^)| > 2iisir Thus, the theorem is proved. If M is unbounded, then there exists a se¬ quence {(xfe,^)} in M such that |(Xfc,6)l -t oo as k -»• oo and 1 2||5|| , A: = 1,2, — So, there exists a subsequence of {(xft,£jt)}, again denoted by {(xfc,£*)}, such that k(x*,&)| -*■ <7oo as fc -t oo, where <Too is some complex number. Thus, 1
Symmetrically Global Pseudo-Differential Operators 173 Now, for k = 1,2,..., let A& = |(^A5?Cfe)|- Since (21.3) is equally valid for S0,0, it follows that RXk>T (**• ill) T°Rxk'T (**’ ill)= T°k'r’ where (*>»?) = ° (** + AfcTl’Aft]|j + A*»l) ’ € By Exercise 22.13, (21.5) is valid also for 50,0 if we replace pa,p by the corresponding norm qa^ in 50,0. So, for all multi-indices a and /?, there exists a positive constant Cp such that m^aXk,T){x,v)\ < C0qa,0{ri)W^TlaXil'2Tm^ and the rest of the proof is exactly the same as that of Theorem 21.6. □ Using Theorem 22.12, the proof of Theorem 22.11 is similar to that of Theorem 21.7 and is best left as an exercise. We end this chapter with an index formula for SG pseudo-differential operators. We need a lemma. Lemma 22.13. f\kltk2eRSkl'k2 = <S. Proof It is easy to check that S C n&^eRS*1’*2 and is left as an exercise. Let a G nkuk2&tSkuk2. Then for all multi-indices a, /?, 7 and S, let k\ and /¾ be real numbers such that M + k2 - H < 0 and 101+*i-i*i <0. Since a G Skuk2, there exists a positive constant such that sup \xat?(D2D%<r)(x,0\ ®,€€Rn < sup z,£6Rn <CklM,^5 sup ((*)H+**-^<Oi<5,+*1_|4|)<oo. z,£€Rn Therefore a G S. Let a € 5°’° be a SG elliptic symbol. Let r € S°'° be such that TtT„ = I — Ti □
174 An Introduction to Pseudo-Differential Operators, 3rd Edition and T,Tt=I-T2j where Tj is a SG pseudo-differential operator with symbol tj in j = 1,2. Therefore for j = 1,2, Tj e S by Lemma 22.13. Let <p e S. Then for j = 1,2, (Tj<fi)(x) = (2tt)-"/2 f = {2ir)~n/2 f T-x[T2Tj)(x, y) <p{y) dy J Rn for all x £ Kn, where denotes the Fourier transform with respect to the second variable. So, for j = 1,2, (Tjtp)(x) = (27r)“n/2 f {?2tj)(x,y-x) <p(y) dy Jrn = (27r)“n/2 f {T2lTj){x, x-y) cp{y) dy Jr» for all ar € Mn. For j = 1,2, (^_1'rJ)(a;, ar - y) is a Schwartz function of (*,y) in Kn x IT. So, by Theorem 20.12, Tj : L2(Rn) L2(Rn) is a trace class operator with tr(Tj) = (27r)-n/2 / (^^)(^0) <fc «/Rn = (2tt)-" / f e^rjfaQdxdt Jrn Jr* = (2?r)"n / f Tj{x,t)dxd¢, «/Rn ,/r» for j = 1,2. By Theorem 20.13, i(7V) = tr(Tj) - tr(T2) = (27r)“n f f (n(*»0 ~ r2(x,0)dxd^. Jr* «/Rn Finally, let cr € S7"1'™2 be such that the corresponding SG pseudo¬ differential operator T„ : H3^3^2 -+ H*i-™-82'7"2’2 is a Fredholm operator for -oo <mi,m2 < oo and —oo < «i, «2 < °°- Then, by Theorem 22.11, a is elliptic. So, there exists a symbol r € 5'7”1’ m2 suc^ TtT„ = I-T{ (22-3) TaTr = I-T3, and (22.4)
Symmetrically Global Pseudo-Differential Operators 175 where T[ and tively, symbols Jm\—s\tms “^2 n axe SG pseudo-differential operators with, respec- and t'2 in f\uk2&tSkl>k2. Now, the operator ,/ is a SG pseudo-differential operator with symbol in 50,0 _ By the analysis carried out in the preceding paragraph, Theorem 20.7 and the fact that JS1)S2 is a surjective isometry for all «i and s2 in K, L2(Rn) is a Fredholm operator such that )=i{Ta). (22.5) Jmi-sum2-S2T<rJsi,s2 • L ) ,m2— By (22.3), (J_Sl,_S2Tr JSl-mi,«2-m2)(^mi-si,m2-S2^T^si,S2) = J— Sis — 32^^0 JsiyS2 = J_Sl}_S2(I — T[)JSl,S2 = I — J-81,-82^1 J81,82* Moreover, by (22.4), (Jmi-si,m2-S2^crJsuS2)(J-si-82,^TJsi-muS2-m2) = Jmi—si,m2-82^'^'T^8i-mi,S2-fn2 — Jfji\—S\,m2 — S2 -^2 ) ^81 7711, S2 m2 — / Jm\— s\,m2—828\—m\,82—^2 - By Theorem 20.12, the equation (22.5) and Theorem 20.7, we get i(T<r) = tr(J_Sl,_S2T/JSl>S2) — tr(Jmi-si,m2-«2^2^»i-mi,S2-m2) = tr(I?)-tr(22). Using the same analysis as in the preceding paragraph, we conclude that i(Ta) = (2<Tn [ f (T[(z,t)-TiM)dxdt. Exercises 22.1. Prove that every linear partial differential operator with constant coefficients on W1 is a SG pseudo-differential operator. 22.2 What is the SG order of the linear partial differential operator 2|a|<ma<*Da with constant coefficients on Rn? 22.3. Prove that every linear partial differential operator P(D) on Rn is SG elliptic if and only if 0*{P(fl :*€»>}•
176 An Introduction to Pseudo-Differential Operators, 3rd Edition 22.4. Let a G Sm 1,m2, -oo < mi,m2 < 00. Prove that Ta : -> «5 is a continuous linear mapping. 22.5. For -00 <mi,m2 <00, prove that if cr G Smi,m2 and m2 < 0» then a G Smi. 22.6. Prove that for -00 < si,s2 < 00, the mapping J8liS2 : 5' -> 5' is a bijection and 22.7 Prove that for 1 < p < 00 and -00 < s < 00, #s’0,p = 22.8. Prove that for 1 < p < 00 and —00 < si,$2 < 00, the mapping J-Sl-S2 : H8l'S2'p -» Lp(Rn) is a surjective isometry. 22.9. Prove Theorem 22.5. 22.10. Let Si,S2,ti,t2 £ (-00,00) be such that si < t\ and s2 < t2. Prove that Htlit2tP C HSl'S2'p and the inclusion i : Htlit2iP ^ H81 yS2'p is a bounded linear operator. 22.11. Prove Corollary 22.9. 22.12. Prove that, for —00 < si,s2 < and 1 < p < 00, the Schwartz space S is dense in i/5l,S2,p. 22.13. Prove (21.5) for symbols in S0,0, where pa$ is now replaced by the corresponding norm qa,p in 50,0. 22.14. Prove that if Ta and Tr are SG elliptic pseudo-differential operators, then TaTr is a SG pseudo-differential operator. 22.15. Prove that if Ta is a SG elliptic pseudo-differential operator, then T* is also a SG elliptic pseudo-differential operator. 22.16. Prove Theorem 22.11. 22.17. Prove that S C ,k26RSkl'k2. 22.18. Let A : L2(Rn) -> L2(Rn) be an operator of finite rank such that the range R{A) of A is contained in S. Prove that A is a SG pseudo-differential operator with symbol in n^^RS*1’*52.
Chapter 23 Spectral Invariance of Symmetrically Global Pseudo-Differential Operators This chapter contains an application of the equivalence of the ellipticity and Fredholmness of SG pseudo-differential operators on Lp(Mn), 1 < p < oo. To motivate the topic, let us first observe that, in view of Theorems 22.2 and 22.5, the set of SG pseudo-differential operators with symbols in 50,0 is an algebra of bounded linear operators on Lp(Rn) for 1 < p < oo. In the case when p = 2, the algebra is in fact a *-algebra. Suppose that a E 50,0 is such that the corresponding SG pseudo-differential operator Ta : Lp(Rn) Lp(Rn) is bijective, where 1 < p < oo. The problem is to determine whether or not the inverse is also a SG pseudo-differential operator with symbol in 50,0. This is known as the spectral invariance problem. See Exercise 23.4 for an explanation of the terminology. We begin with the L2-case. Theorem 23.1. Let a E 50,0 be such that the corresponding SG pseudo¬ differential operator Ta : L2(Rn) —>• L2(Rn) is bijective. Then its inverse T~l : L2(Rn) -> L2(Rn) is also a SG pseudo-differential operator with symbol in 50,0. Proof By bijectivity, Ta : L2(Rn) —» L2(Rn) is Fredholm and i{Ta)= 0. By Theorem 22.12, Ta is SG elliptic. By Theorem 22.4, there exists a symbol r in 50,0 such that TaTr = / + 5, where R and S are infinitely smoothing in the sense that they are SG pseudo-differential operators with symbols in C\kiMeRSkl,k2. By Atkinson’s 177
178 An Introduction to Pseudo-Differential Operators, 3rd Edition Theorem 20.5, Tr is Fredholm and hence by Theorem 22.4, Tr is elliptic. Also, by Remark 20.4 and Theorem 20.7, So, i(Tr) + m = i{TrTa) = i(I + R)= 0. i(Tr) = 0. Now, let u e N(Tr). Then Tru = 0 => TaTTu = 0 (J + S)u = 0 => u = —Su. Since n*lj*2eRfr*1,tf2'2 = <S, it follows that u = —Su 6 S. Similarly, the null space N(T'*) of the true adjoint T* : L2(En) ->• L2(En) of Tt : L2(En) -A L2(En) is also a subspace of S. Now, we write L2(Rn) = N(TT)®N{TT)± and L2(En) = N(T*) 0 -Rx,2(Tr), where Rl^{Tt) is the range of Tr : L2(En) -+ L2(En). Let P = iFn, where 7r is the projection of L2(En) onto N(Tr), F is an isomorphism of N(Tt) onto N(T*) and i is the inclusion of AT(T*) into L2(En). Then the operator Tt + P : L2(En) -+ L2(En) is a bijective parametrix of Ta. Therefore, without loss of generality, we may assume that the parametrix Tt : L2(En) ->■ L2(En) is bijective. So, I + R is bijective. In fact, (J + P)-1 =I + K, where K maps L2(En) into S. Indeed, there exists a bounded linear oper¬ ator K : L2(Rn) -+ Z,2(En) such that (I + R)(I + K)=I. So, K = -R - RK and K* = -R* - K*R*. It is obvious that K and K* map L2(En) into S. Thus, by Exercise 23.2, the kernel of K : L2(En) -+ L2(En) is a Schwartz function on En x En,
Spectral Invariance of Symmetrically Global Pseudo-Differential Operators 179 and hence, by Exercise 23.3, K is a SG pseudo-differential operator with symbol in C\kltk2euSklM- Thus, T~1T~l=I + K or equivalently T-1 = (I+ K)Tr and this completes the proof. □ We can now give the Inversion of the spectral invariance of SG pseudo¬ differential operators. Theorem 23.2. Let a G 50,0 be such that the pseudo-differential operator Ta : Lp(En) ->• Lp(En) is invertible, where 1 < p < oo. Then its inverse T~l : Lp(Rn) ->• Lp(Rn) is also a pseudo-differential operator with symbol in 50,0. Proof Since Ta : Lp(Rn) ->■ Lp(En) is bijective, it follows that it is Fred¬ holm. So, by Theorem 22.4, a is an elliptic symbol in 50,0. Thus, by Theorem 22.4, Ta : L2(Rn) —»• L2(En) is Fredholm. So, there exists a symbol r in 50,0 such that TaTr = / + 5, where R and 5 are infinitely smoothing. Let u G L2(En) be such that Tau = 0. Then TrTau = 0 => (/ + R)u = 0 =½ u = —Ru G 5. (23.1) So, u is also in the null space of Ta : Lp(Rn) -> Lp(Rn). Since Ta : Lp(En) —> Lp(En) is injective, it follows that u = 0. Therefore Ta : L2(En) -> L2(En) is injective. To show that Ta : L2(En) -> L2(En) is surjective, let u be a function in the null space N(T'*) of the adjoint Tl : L2(En) -> L2(En) of Ta : L2(En) -> L2(En). Then T^u = 0 => Ta*u = 0, where a* in 50,0 is the symbol of the formal adjoint of Ta. Since a* is elliptic, we can use a left parametrix of Ta* as in (23.1) to conclude that u = 0. This proves that Ta : L2(En) L2(En) is surjective and hence bijective. So, by Theorem 23.1, the inverse T”1 : L2(En) -t L2(En) is a SG pseudo-differential operator Tt with symbol r in 50,0. Since 77 V = I>, f €5, and S is dense in Lp(En), it follows that T”1 = Tr on Lp(En). □
180 An Introduction to Pseudo-Differential Operators, 3rd Edition Exercises 23.1. Prove that DSl>S2eRHSl’S2’2 = <S- (Compare this exercise with Exer¬ cise 12.13.) 23.2. Prove that if A : L2(Rn) L2(Mn) and A1 : L2(Rn) -> L2(Mn) are bounded linear operators such that A and A* map L2(Rn) into <S, then the kernel of A is a Schwartz function on W1 xln. (Hint: Use Exercise 20.6.) 23.3. Let A : L2(Rn) L2(Rn) be a Hilbert-Schmidt operator such that its kernel is a Schwartz function on Mn xln. Prove that A is a SG pseudo- differential operator with symbol in CikiMeR*^1’*2- 23.4. Let B(X) be the Banach algebra of all bounded linear operators on a complex Banach space X. Let A be a sub-algebra of B(X) containing the identity operator I on X. Then for every element A in A, we define the resolvent set Pa(A) of A with respect to A to be the set of all complex numbers A such that A — XI has an inverse, which also lies in A. The spectrum E^(A) of A with respect to A is defined to be the complement in C of the resolvent set pa(A). (i) Prove that Eb(x){A) C E^(A), A e A. (ii) A sub-algebra A of B(X) containing the identity operator I on X is said to be nontrivial if A ± B(X). Give an example to show that there exist complex Banach spaces X and nontrivial sub-algebras A of B{X) for which %)(A) = E^(A), A€ A. (23.2) (A nontrivial sub-algebra A of B(X) containing the identity operator I on X and satisfying (23.2) is said to be spectrally invariant)
Bibliography Agmon, S., Douglis, A. and Nirenberg, L. (1959) Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions, I, Comm. Pure Appl. Math. 12, pp.623-727. Arveson, W. (2002) A Short Course on Spectral Theory (Springer). Atiyah, M. F. (1997) The index of elliptic operators, in Fields Medallists; Lectures, World Scientific and Singapore University Press, pp. 115-127. Atkinson, F. V. (1951) The normal solvability of linear equations in normed spaces, Mat. Sbornik N.S. 28, pp.3-14. Brislawn, C. (1988) Kernels of trace class operators, Proc. Amer. Math. Soc. 104, pp.1181-1190. Dasgupta, A. (2008) The Twisted Laplacian, the Laplacians on the Heisenberg Group and SG Pseudo-Differential Operators, Ph.D. Dissertation, York University. Dasgupta, A. (2008) Ellipticity of Fredholm pseudo-differential operators on Lp(En), in New Developments in Pseudo-Differential Operators, Operator Theory: Advances and Applications 189, Birkhauser, pp. 107-116. Dasgupta, A. and Wong, M. W. (2008) Spectral theory of SG pseudo-differential operators, Studia Math. 187, pp. 186-197. Dasgupta, A. and Wong, M. W. (2010) Spectral invariance of SG pseudo¬ differential operators on Lp(Rn), in Pseudo-Differential Operators: Com¬ plex Analysis and Partial Differential Equations, Operator Theory: Ad¬ vances and Applications 205, Birkhauser, pp. 51-57. Davies, E. B. (1980) One-Parameter Semigroups (Oxford University Press). Davies, E. B. (2007) Linear Operators and their Spectra (Cambridge University Press). Douglas, R. G. (1998) Banach Algebra Techniques in Operator Theory, 2nd edn. (Springer). Fan, Q. and Wong, M. W. (1997) A characterization of Fredholm pseudo¬ differential operators, J. London Math. Soc. (2) 55, pp.139-145. Fan, Q. and Wong, M. W. (1997) A characterization of some non-elliptic pseudo¬ differential operators as Fredholm operators, Forum Math. 9, pp. 17-28. Friedman, A. (1969) Partial Differential Equations (Holt, Rinehart and Winston). 181
182 An Introduction to Pseudo-Differential Operators, 3rd Edition Friedman, A. (1971) Advanced Calculus (Holt, Rinehart and Winston). Garding, L. (1953) Dirichlet’s problem for linear partial differential equations, Math. Scand. 1, pp.55-72. Gel’fand, I. M. (1960) On elliptic equations, Russian Math. Surveys 15, pp.113- 123. Grushin, V. V. (1970) Pseudo-differential operators on Rn with bounded symbols, Fund. Anal. Appl. 4, pp.202-212. Harutyunyan, G. and Schulze, B.-W. (2008) Elliptic Mixed, Transmission and Singular Crack Problems (European Mathematical Society). Hess, P. and Kato, T. (1970) Peturbation of closed operators and their adjoints, Comment. Math. Helv. 45, pp.524-529. Hille, E. and Phillips, R. S. (1974) Functional Analysis and Semi-Groups, 3rd printing of revised (1957) edn. (American Mathematical Society). Hormander, L. (1960) Estimates for translation invariant operators in Lp spaces, Acta Math. 104, pp.93-140. Iancu, G. M (1999) Dynamical Systems Generated by Pseudo-Differential Opera¬ tors, Ph.D. Dissertation, York University. Kumano-go, H. (1981) Pseudo-Differential Operators (MIT Press). Lax, P. D. (2002) Functional Analysis (Wiley-Interscience). Reed, M. and Simon, B. (1980) Functional Analysis, revised and enlarged edn. (Academic Press). Royden, H. L. (1988) Real Analysis, 3rd edn. (Prentice-Hall). Rudin, W. (1987) Real and Complex Analysis, 3rd edn. (McGraw-Hill). Saint Raymond, X. (1991) Elementary Introduction to the Theory of Pseudodif¬ ferential Operators (CRC Press). Schechter, M. (1977) Modem Methods in Partial Differential Equations (McGraw- Hill). Schechter, M. (1986) Spectra of Partial Differential Operators, 2nd edn. (North- Holland). Schechter, M. (2002) Principles of Functional Analysis, 2nd edn. (American Mathematical Society). Schulze, B.-W. (1998) Boundary Value Problems and Singular Pseudo-Differential Operators (Wiley). Wheeden, R. L. and Zygmund A. (1977) Measure and Integral (Marcel Dekker). Wong, M. W. (1983) Fredholm pseudo-differential operators on weighted Sobolev spaces, Ark. Mat. 21, pp.271-282. Wong, M. W. (1987) On some spectral properties of elliptic pseudo-differential operators, Proc. Amer. Math. Soc. 99, pp.683-689. Wong, M. W. (1988) Essential spectra of elliptic pseudo-differential operators, Comm. Partial Differential Equations 13, pp.1209-1221. Wong, M. W. (1991) Minimal and maximal operator theory with applications, Canad. J. Math. 43, pp.617-627. Wong, M. W. (1994) Spectral theory of pseudo-differential operators, Adv. in Appl. Math. 15, pp.437-451. Yosida, K. (1995) Functional Analysis, reprint of the 6th (1980) edn. (Springer- Verlag).
Index adjoint, 97 adjoint formula, 21 Agmon-Douglis-Nirenberg estimate, 102 approximate solution, 93 asymptotic expansion, 36, 55, 61, 168 Atkinson’s theorem, 144 Bessel potential, 87, 169 boundedness in L2(Rn), 85 boundedness in Lp(Rn ), 77, 85, 91, 168 closable linear operator, 95, 99 closed linear operator, 95 continuity of translations in Lp (Rn), 11 convolution, 9 core, 104 dissipative operator, 142 distribution solution, 111 elliptic, 69, 73 Erhling’s inequality, 93 extension, 95 first resolvent formula, 141 formal adjoint, 61 Fourier inversion formula, 22, 29 Fourier transform, 17, 28 Fredholm operator, 143 fundamental solution, 93 Garding’s inequality, 113 global regularity, 106 Hormander’s multiplier theorem, 85 heat equation, 129 Hermite function of order n, 25 Hilbert-Schmidt operator, 148 Hille-Yosida-Phillips theorem, 132 index, 143 infinitely smoothing, 69, 169 infinitesimal generator, 130 initial value problem, 129 kernel, 149 Lax-Milgram lemma, 122 left parametrix, 72 Leibniz’s formula, 3 length, 2 maximal operator, 101 minimal operator, 96, 99 Minkowski’s inequality in integral form, 3 multi-index, 2 one-parameter contraction semigroup, 142 one-parameter semigroup, 129 183
184 An Introduction to Pseudo-Differential Operators, 3rd Edition order, 167 parametrix, 69 partial differential operator, 1 partition of unity, 43 Peetre’s inequality, 157 Plancherel’s theorem, 23, 24 positive operator, 148 principal symbol, 73 product, 55 pseudo-differential operator, 32 resolvent, 132 resolvent set, 132, 180 Riemann-Lebesgue lemma, 19 right parametrix, 72 Schwartz space, 13 SG elliptic, 168 SG parametrix, 169 SG pseudo-differential operator, 167 SG symbol, 167 singular integral operator, 79 Sobolev embedding theorem, 88, 170 Sobolev space, 88, 169 spectral invariance, 177 spectrum, 180 strong solution, 121 strongly elliptic, 113 support, 11 symbol, 2, 32 Taylor’s formula with integral remainder, 49 tempered distribution, 27 tempered function, 27 total set, 98 trace, 149 trace class operator, 148 weak solution, 109 Young’s inequality, 9