Second Order Equations of Elliptic and Parabolic Type

477 84 24MB

English Pages [217] Year 1998

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Second Order Equations of Elliptic and Parabolic Type

  • Commentary
  • decrypted from B85D0099C41BC26E734A53CB56BE5E70 source file

Table of contents :
Cover
Title page
Contents
Preface to the English edition
Preface
Elliptic equations in nondivergence form
Elliptic equations in divergence form
Parabolic equations
Appendix
Bibliography
Index
Back Cover

Citation preview

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

Selected Title s i n Thi s Serie s 171 E . M . Landis , Secon d orde r equation s o f ellipti c an d paraboli c type , 1 99 8 170 V i k t o r Prasolo v an d Yur i Solovyev , Ellipti c function s an d ellipti c integrals , 1 99 7 169 S . K . G o d u n o v , Ordinar y differentia l equation s wit h constan t coefficient , 1 99 7 168 Junjir o N o g u c h i , Introductio n t o comple x analysis , 1 99 7 167 M a s a y a Y a m a g u t i , Masayosh i Hata , an d J u n Kigami , Mathematic s o f fractals , 1 99 7 166 Kenj i U e n o , A n introductio n t o algebrai c geometry , 1 99 7 165 V . V . Ishkhanov , B . B . Lur'e , an d D . K . Faddeev , Th e embeddin g proble m i n Galois theory , 1 99 7 164 E . I . G o r d o n , Nonstandar d method s i n commutativ e harmoni c analysis , 1 99 7 163 A . Ya . D o r o g o v t s e v , D . S . Silvestrov , A . V . Skorokhod , an d M . I . Yadrenko , Probability theory : Collectio n o f problems , 1 99 7 162 M . V . B o l d i n , G . I . Simonova , an d Yu . N . Tyurin , Sign-base d method s i n linea r statistical models , 1 99 7 161 Michae l Blank , Discretenes s an d continuit y i n problem s o f chaoti c dynamics , 1 99 7 160 V . G . Osmolovskii , Linea r an d nonlinea r perturbation s o f th e operato r div , 1 99 7 159 S . Ya . K h a v i n s o n , Bes t approximatio n b y linea r superposition s (approximat e nomography), 1 99 7 158 Hidek i Omori , Infinite-dimensiona l Li e groups , 1 99 7 157 V . B . Kolmanovski i an d L . E . Shatkhet , Contro l o f system s wit h aftereffect , 1 99 6 156 V . N . Shevchenko , Qualitativ e topic s i n intege r linea r programming , 1 99 7 155 Yu . Safaro v an d D . Vassiliev , Th e asymptoti c distributio n o f eigenvalue s o f partia l differential operators , 1 99 7 154 V . V . Prasolo v an d A . B . Sossinsky , Knots , links , braid s an d 3-manifolds . A n introduction t o th e ne w invariant s i n low-dimensiona l topology , 1 99 7 153 S . K h . A r a n s o n , G . R . B e l i t s k y , an d E . V . Zhuzhoma , Introductio n t o th e qualitative theor y o f dynamica l system s o n surfaces , 1 99 6 152 R . S . Ismagilov , Representation s o f infinite-dimensiona l groups , 1 99 6 151 S . Yu . Slavyanov , Asymptoti c solution s o f th e one-dimensiona l Schrodinge r equation , 1996 150 B . Ya . Levin , Lecture s o n entir e functions , 1 99 6 149 Takash i Sakai , Riemannia n geometry , 1 99 6 148 Vladimi r I . Piterbarg , Asymptoti c method s i n th e theor y o f Gaussia n processe s an d fields, 1 99 6 147 S . G . Gindiki n an d L . R . Volevich , Mixe d proble m fo r partia l differentia l equation s with quasihomogeneou s principa l part , 1 99 6 146 L . Ya . Adrianova , Introductio n t o linea r system s o f differentia l equations , 1 99 5 145 A . N . A n d r i a n o v an d V . G . Zhuravlev , Modula r form s an d Heck e operators , 1 99 5 144 O . V . Troshkin , Nontraditiona l method s i n mathematica l hydrodynamics , 1 99 5 143 V . A . M a l y s h e v an d R . A . Minlos , Linea r infinite-particl e operators , 1 99 5 142 N . V . Krylov , Introductio n t o th e theor y o f diffusio n processes , 1 99 5 141 A . A . D a v y d o v , Qualitativ e theor y o f contro l systems , 1 99 4 140 Aizi k I . Volpert , V i t a l y A . Volpert , an d V l a d i m i r A . Volpert , Travelin g wav e solutions o f paraboli c systems , 1 99 4 139 I . V . Skrypnik , Method s fo r analysi s o f nonlinea r ellipti c boundar y valu e problems , 1 99 4 138 Yu . P . R a z m y s l o v , Identitie s o f algebra s an d thei r representations , 1 99 4 137 F . I . K a r p e l e v i c h an d A . Ya . Kreinin , Heav y traffi c limit s fo r multiphas e queues , 1 99 4 136 Masayosh i Miyanishi , Algebrai c geometry , 1 99 4 135 M a s a r u Takeuchi , Moder n spherica l functions , 1 99 4 134 V . V . Prasolov , Problem s an d theorem s i n linea r algebra , 1 994 133 P . I . N a u m k i n an d I . A . Shishmarev , Nonlinea r nonloca l equation s i n th e theor y o f waves, 1 99 4 (See th e AM S catalo g fo r earlie r titles ) Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

This page intentionally left blank

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

Second Orde r Equations o f Ellipti c and Paraboli c Typ e

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

This page intentionally left blank

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

10.1090/mmono/171

Translations o f

MATHEMATICAL MONOGRAPHS Volume 1 7 1

Second Orde r Equations o f Ellipti c and Paraboli c Typ e E. M . Landi s

> ^ % ,

American Mathematical Societ y Providence, Rhode Island Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

EDITORIAL COMMITTE E AMS Subcommitte e Robert D . MacPherso n Grigorii A . Marguli s J a m e s D . Stashef f (Chair ) A S L S u b c o m m i t t e e Steffe n Lemp p (Chair ) I M S S u b c o m m i t t e e Mar k I . Freidli n (Chair ) E. M . J I A H U H C yPABHEHMH B T O P O r O nOPtfUK 3JIJmnTHHECKOrOH

A

nAPABOJIMMECKOr O TMnO

B

«HAYKA», MOCKBA , 1 97 1 Translated fro m th e Russia n b y Tamar a Rozhkovskay a with th e participatio n o f Scientifi c Book s (RIMIB E NSU) , Novosibirsk , Russi a 1991 Mathematics Subject

Classification. Primar y 35J1 5 , 35K1 0 .

ABSTRACT. Th e book introduce s th e reader t o the theory o f linear ellipti c an d parabolic equation s of th e secon d order . I n additio n t o th e discussio n o f classica l result s fo r equation s wit h smoot h coefficients (Schaude r estimate s an d the solvability o f the Dirichlet proble m fo r elliptic equations ; the Dirichle t proble m fo r the heat equation) , th e book describe s propertie s o f solutions t o secon d order ellipti c an d paraboli c equation s wit h measurabl e coefficient s nea r th e boundar y an d a t infinity. The boo k present s a fine elementar y introductio n t o th e theory o f elliptic an d parabolic equa tions o f the secon d order . Th e precise an d clea r expositio n o f the materia l make s i t suitabl e fo r graduate student s a s wel l a s fo r researc h mathematician s wh o want t o ge t acquainte d wit h thi s area o f the theory o f partial differentia l equations .

Library o f Congres s Cataloging-in-Publicatio n D a t a Landis, E . M. (Evgen h MikhaTlovich ) [Uravneniia vtorog o poriadk a ellipticheskog o i parabolicheskogo tipov . English ] Second orde r equation s o f elliptic an d parabolic typ e / E . M . Landi s ; [translate d b y Tamar a Rozhkovskaya]. p. cm . — (Translation s o f mathematical monographs , ISS N 0065-928 2 ; v. 171) Includes bibliographica l reference s an d index . ISBN 0-821 8-0857- 5 (acid-fre e paper ) 1. Differentia l equations , Elliptic . 2 . Differential equations , Parabolic . I . Title . II . Series . QA377.L35131 99 7 97-3467 0 CIP

C o p y i n g an d reprinting . Individua l reader s o f this publication , an d nonprofit librarie s actin g for them , ar e permitte d t o mak e fai r us e o f th e material , suc h a s t o cop y a chapte r fo r us e in teachin g o r research . Permissio n i s grante d t o quot e brie f passage s fro m thi s publicatio n i n reviews, provide d th e customary acknowledgmen t o f the source i s given . Republication, systemati c copying , o r multiple reproductio n o f any materia l i n this publicatio n (including abstracts ) i s permitte d onl y unde r licens e fro m th e America n Mathematica l Society . Requests fo r suc h permissio n shoul d b e addresse d t o th e Assistan t t o th e Publisher , America n Mathematical Society , P . O. Bo x 6248, Providence , Rhod e Islan d 02940-6248 . Request s ca n als o be mad e b y e-mail t o [email protected] . © 1 99 8 by the American Mathematica l Society . Al l rights reserved . The America n Mathematica l Societ y retain s al l right s except thos e grante d t o the United State s Government . Printed i n the United State s o f America . @ Th e paper use d i n thi s boo k i s acid-fre e an d falls withi n th e guideline s established t o ensur e permanenc e an d durability . Visit th e AMS home pag e a t URL : http://www.ams.org / 10 9 8 7 6 5 4 3 2 1 0

3 02 01 0 0 99 9 8

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

Contents Preface t o th e Englis h Editio n i

x

Preface x

i

Chapter 1 . Ellipti c Equation1 s i n Nondivergenc e For m Introduction §1. Th e maximu m principl e 2 §2. 5-capacit y 9 §3. Th e lemm a o n th e norma l derivativ e an d th e stron g maximu m principle §4. Th e growt h lemm a §5. Th e behavio r o f solution s i n a neighborhoo d o f a boundar y poin 1t §6. Th e behavio r o f solution s a t infinit y 3 §7. Equation s o f Corde s type . A prior i estimate s fo r Holde r norm s 3 §8. Th e existenc e o f a solutio n t o th e Dirichle t proble m fo r a linea r equation 4 §9. Th e existenc e o f a solutio n t o th e Dirichle t proble m fo r a quasilin ear equatio n 5 §10. Th e Harnac k inequalit y an d th e Liouvill e theore m fo r equation s o f Cordes typ e 5

4 5 8 0 7 4 0 5

Chapter 2 . Ellipti c Equation s i n Divergenc e For m 6 9 §1. Th e existenc e an d uniquenes s o f a wea k solutio n t o th e Dirichle t problem 6 9 §2. Som e fact s abou t function s o f severa l rea l variable s 8 1 §3. A prior i estimate s fo r th e Holde r norm s o f solution s t o equation s in divergenc e for m 9 1 §4. A priori estimate s fo r th e Holde r norm s (continued ) 9 9 1 1 Chapter 3 . Paraboli c Equation s 1 1 §1. Definition s an d notatio n 1 1 §2. Th e maximu m principl e §3. Superparaboli c an d subparaboli c potential-typ e1 function s 2 §4. Th e uniquenes s o f a solutio n t o th e Cauch y problem , an d th e sta bilization o f a solutio n t o th e Cauch y proble m a1 s t— > o o 2 1 §5. Paraboli c s, /^-capacity 2 §6. Th e growt h lemm a 2 §7. Th e behavio r o f a solutio n i n a neighborhoo d o f a boundar y poin1 t 3 §8. Equation s o f Corde s type . Th e oscillatio n theore m an d corollarie1 s 4 §9. Equation s o f Corde s type . Th e Harnac k inequalit y an d corollarie1 s 4

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

3 3 5 2 4 7 9 3 0 4

viii C O N T E N T

S

§10. Hea t potential s 5 §11. Th e existenc e o f a solutio n t o th e firs t boundar y valu e proble m i n a cylindrica l domain . Estimate s fo r th e derivative s o f a solutio n and th e compactnes 1 s theore m fo r solution s 5 §12. Constructio n o f the generalized solutio n t o the first boundar y valu e problem i n a bounde d domai n i n IR n+1 . Th e behavio r o f th e gen eralized 1 solutio n a t boundar y point s 6 Appendix 7 §1. Th e proo f o f Lemm 1 a 5. 1 fro m Chapte r 1 7 §2. Th e proo f o f th e Schaude r1 fixe d poin t theore m 8 1 §3. Isoperimetri c inequalit y 8 1 §4. Th e Schaude r estimate s 8 Bibliography 9 Index

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

3

9

5 5 5 0 2 3 9 203

Preface t o th e Englis h Editio n The boo k introduce s th e reade r t o th e theor y o f linea r ellipti c an d paraboli c equations o f secon d order . Althoug h it s mai n goa l i s t o stud y equation s wit h bounded measurabl e coefficients, th e book also contains a number o f classical result s concerning equations with smooth coefficients. Fo r example, the Schauder estimate s and th e solvabilit y o f th e Dirichle t proble m fo r ellipti c equation s i n Holde r spaces , the Wiener criterion for the regularity of a boundary poin t i n the case of the Laplac e equation, th e solvailit y o f th e Dirichle t proble m fo r th e hea t equatio n b y mean s o f heat potentials , an d som e othe r result s ar e discusse d i n detail . For equation s i n nondivergenc e for m wit h discontinuou s coefficient s th e proof s of result s presente d i n th e boo k ar e base d entirel y o n th e classica l maximu m prin ciple. To construc t th e correspondin g subellipti c an d superellipti c functions , th e au thor use s th e Ries z potentials , whic h lead s t o a natural characteristi c o f set s unde r consideration i n term s o f s-capacity , wher e s depend s o n th e ellipticit y constan t e. Th e essenc e o f th e approac h ca n b e observe d i n th e growt h lemm a (Chapte r 1 , Lemma 4.1 ) , whic h i s permanently use d throughou t Chapte r 1 . In particular , wit h th e hel p o f the growt h lemma , th e autho r deduce s sufficien t conditions fo r th e regularit y o f a boundar y poin t i n th e cas e o f ellipti c operator s with measurabl e coefficient s (a n analo g o f the Wiene r criterion ) an d prove s a serie s of theorem s o f Phragmen-Lindelo f typ e o n th e behavio r o f solution s a t infinity . A simila r approac h i s applie d t o paraboli c equation s wit h mesuarabl e coeffi cients i n Chapte r 3 . Th e autho r introduce s specia l familie s o f subparaboli c func tions and , usin g th e notio n o f paraboli c capacity , establishe s a paraboli c analo g o f the growt h lemm a an d othe r result s simila r t o thos e i n th e ellipti c case . The abov e approac h i s als o use d t o obtai n a n estimat e fo r th e Holde r nor m o f a solutio n an d t o prove th e Harnac k inequality . However , i n this cas e the us e of th e growth lemm a require s a lowe r estimat e o f th e s-capacit y o f a se t i n term s o f it s measure, whic h lead s t o a stron g restrictio n o n th e ellipticit y constan t e . Namely , it mus t satisf y th e inequalitie s n < e < n + 2 , wher e n i s th e dimensio n o f th e space. Suc h restriction s o n th e sprea d o f eigenvalue s o f th e matri x o f coefficient s are usuall y calle d th e Corde s condition , an d th e correspondin g equation s ar e calle d the equation s o f Corde s type . In th e earl y seventies , whe n th e Russia n editio n o f th e boo k wa s published , i t seemed unlikel y tha t th e Corde s conditio n coul d b e avoided . I n th e lat e seventies , the Harnack inequality an d the Holder estimates for solutions were proved by Krylov and Safonov . Thei r result s ar e o f significan t importanc e i n th e theor y o f nonlinea r elliptic an d paraboli c equations . Th e proof s o f thes e result s requir e mor e powefu l analytic method s tha n th e method s presente d i n thi s book . I n particular , th e crucial ste p consist s i n usin g the estimate s foun d b y A. D. Aleksandro v an d Krylo v ix

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

x PREFAC

E T O T H E ENGLIS H E D I T I O N

for th e maximu m valu e o f a solution i n terms o f the integra l nor m o f the right-han d side o f th e equation . Th e correspondin g bibliographi c remark s an d reference s t o Krylov-Safonov's work s are added b y the autho r t o the Englis h editio n o f the book . For detaile d proof s i n th e ellipti c case , w e als o mentio n th e secon d editio n o f th e book b y D . Gilbar g an d N . S . Trudinger , "Ellipti c partia l differentia l equation s o f second order " (Springer-Verlag , 1 983) . Chapter 2 i s devote d t o ellipti c equation s i n diveregenc e form . T o stud y suc h equations, i t i s necessar y t o appl y essentiall y differen t methods . I n th e cas e o f simplest ellipti c equation s i n divergenc e form , th e estimat e fo r th e Holde r nor m o f a solutio n wa s first establishe d b y D e Giorg i i n 1 957 . Anothe r proo f wa s suggeste d by Moser i n 1 960 . I n Chapte r 2 , the estimates fo r th e Holde r norm s o f solutions ar e derived wit h th e hel p o f som e technica l tool s (cf . Chapte r 2 , Theore m 2.2) , whic h are differen t fro m thos e use d b y D e Giorg i an d Moser . Although th e Russia n editio n o f thi s boo k wa s publishe d 2 5 year s ago , th e book remain s attractiv e fo r th e reader . I t i s a fine elementar y introductio n t o th e theory o f ellipti c an d paraboli c equation s o f secon d order . Th e precis e an d clea r exposition o f the materia l make s the boo k understandabl e fo r graduat e student s a s well a s mathematician s wh o wan t t o ge t acquainte d wit h thi s are a o f the theor y o f partial differentia l equations . To illustrat e mai n idea s an d methods , som e result s ar e describe d i n th e boo k for th e simples t equation s only . Fo r mor e complet e an d detaile d description , th e reader ca n b e referre d t o th e above-mentione d boo k b y Gilbar g an d Trudinge r an d the recen t boo k b y N . V . Krylov , "Lecture s o n ellipti c an d paraboli c equation s in Holde r spaces " (America n Mathematica l Society , 1 996) . However , significan t part o f th e boo k b y Landi s (i n particular , th e constructio n o f barrie r function s o f potential type , an d th e us e o f capacit y i n th e stud y o f equation s i n nondivergenc e form wit h discontinuou s coefficients ) present s a nontraditiona l approac h whic h i s of interes t fo r specialist s a t presen t an d canno t b e foun d i n othe r monograph s an d textbooks. Nina Ural'tsev a July 1 99 7

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

Preface This boo k i s based o n lecture s give n b y th e autho r a t Mosco w Stat e Universit y in 1 967-1 968 . The reade r mus t b e familia r onl y wit h basi c notion s an d fact s fro m th e theor y of partia l differentia l equation s (see , e.g. , [Pel]) . Th e goa l o f th e boo k i s t o intro duce th e reade r t o som e recen t problem s i n th e theor y o f secon d orde r ellipti c an d parabolic equations . The autho r tried , o n on e hand , t o mak e th e expositio n self-containe d and , o n the othe r hand , t o kee p th e siz e o f th e boo k i n reasonabl e limits . Therefore , a limited numbe r o f topic s (a s ca n b e see n fro m th e tabl e o f contents ) hav e bee n chosen t o b e presente d systematicall y an d i n detail . The mai n too l i n the stud y o f solutions i n Chapte r 1 (nonselfadjoint equations ) and i n Chapte r 3 (paraboli c equations ) i s th e us e o f subfundamenta l an d super fundamental solutions , whic h ar e constructe d wit h th e hel p o f th e Ries z potential . Such a n approac h facilitate s th e qualitativ e stud y o f the behavio r o f solutions nea r boundary point s (theorem s of Wiener type ) an d a t infinit y (theorem s of Phragmen Lindelof typ e an d Liouville' s theorems) . Th e sam e metho d allow s u s t o obtai n a priori estimate s fo r th e Holde r norm s o f solution s provide d tha t th e root s o f th e characteristic equatio n ar e no t sprea d to o fa r apart . Thes e estimate s ar e use d i n the proo f o f the existenc e o f a solution t o a boundary valu e proble m fo r quasilinea r equations. Chapter 2 (selfadjoint ellipti c operators ) present s ne w and , i n author's opinion , promising method s base d o n theorem s fro m th e theor y o f function s o f severa l rea l variables. These main topics of the book are not discusse d i n other monographs . However , our intentio n t o presen t informatio n withou t referrin g th e reade r t o othe r book s made i t necessar y t o includ e muc h o f traditiona l materia l (th e Leray-Schaude r method, variationa l method s o f solutio n o f Dirichle t problem s i n energ y spaces , heat potentials , etc.) . Almost ever y sectio n conclude s wit h bibliographi c remark s an d comments. 1 Material tha t i s not difficul t i n principle, bu t involve s cumbersome calculations , which woul d interrup t th e exposition , i s delegate d t o th e Appendix . I n particular , the fixed poin t theore m w e us e i s proved i n § 2 of the Appendix . The autho r thank s Yu . S . Il'yashenk o wh o helpe d a t al l stage s o f preparin g the lecture s an d th e book . Hi s assistanc e allowe d u s t o avoi d certai n error s an d t o improve som e proofs .

1 Translator's note. Som e section s ar e provide d wit h additiona l bibliographi c remark s writte n by th e autho r fo r th e Englis h editio n i n 1 996 . I n th e lis t o f references , th e ne w reference s ar e marked b y th e asterisk .

xi

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

xii P R E F A C

E

The autho r i s grateful t o M. I. Vishik who read th e manuscrip t an d mad e man y useful remarks . E. M . Landi s 1970

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

10.1090/mmono/171/01

CHAPTER 1

Elliptic Equation s i n Nondivergenc e For m Introduction A linea r secon d orde r differentia l operato r o f th e for m

=£aaix)^+£bi{x)£+c{x),

a) L i,k=l i—\

X = \X\ , . . . , X

n),

is said t o b e elliptic i f th e quadrati c for m n

/ , dikiiik is positive definite . W e alway s assum e tha t a ^ = dkiThe operato r (1 ) i s sai d t o b e uniformly elliptic i n a domai n D i f ther e exis t two positiv e constant s C\ an d C 0 and M > 0 are constants calle d the ellipticity constants. Hereinafter, w e use th e followin g notation : R n i s th e n-dimensiona l Euclidea n space , QXR i s a n ope n bal l i n W 1 o f radiu s R wit h cente r x° , Sft i s the spher e \x — x°\ — R, Q^ R i s the spherica l laye r define d b y th e inequalitie s R\ ^ \x — x°\ < it^ , O i s th e origin . In th e abov e notation , Q^ i s the bal l o f radiu s R an d cente r a t th e origin . E i s th e closur e o f a se t E an d dE i s the boundar y o f E. In Chapte r 1 , we will usuall y conside r operator s L withou t lowe r terms :

i,fc=l

Throughout Chapte r 1 , unles s otherwis e stated , b y L w e mea n th e operato r {!'). For suc h operator s onl y th e rati o o f th e constant s a an d M i s important , bu t no t the constant s themselves . It i s mor e convenien t t o introduc e int o consideratio n th e quantit y Tii=laii(x) , n.

e — su p „

which i s calle d th e ellipticit y constan t o f th e operato r L. In othe r words , fo r an y x G D w e tak e th e rati o o f th e su m o f al l eigenvalue s of th e matri x ||a;/c(:r)| | t o th e smalles t eigenvalue , an d th e ellipticit y constan t e i s the supremu m o f thi s rati o ove r al l x £ D. We not e tha t e ^ n , an d i f e = n , the n th e operato r L become s th e Laplac e operator afte r th e multiplicatio n b y som e positiv e function . §1. Th e maximu m principl e 1 .1 . Let D C R n be a bounded domain, L the elliptic operator (V) in D, and u a superelliptic (subelliptic) continuous function in D. Then THEOREM

u(x) ^ mini / (u(x) vJ

dD

v

v

J

^ maxw) , x J

dD

G D.

It suffice s t o conside r th e case o f a superellipti c functio n u. W e nee d th e following lemma . LEMMA 1 .1 . /

(6) Lu

/

< 0 in

D,

then u does not attain the minimum value at any interior point of D.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

3

§1. TH E MAXIMU M PRINCIPL E

P R O O F . W e assum e th e contrary . Le t th e minimu m valu e o f u b e attaine d a t an interio r poin t x° o f D. W e mak e th e chang e o f variables x y unde r whic h th e operator L take s th e canonica l for m a t x°. W e obtai n

_ \ ^/

Q\

u

u

I

2u

_ST^d

>0>

which contradict s (6) . D P R O O F O F T H E O R E M 1 .1 . W e note tha t th e ellipticit y o f the operato r implie s the inequalit y an > 0 . Conside r a n auxiliar y functio n

> 0.

ve — u — sx\, e Since

Lv£ = Lu — ean < 0, we hav e vAx) ^ mirivJx) 3D

in vie w o f Lemm a 1 .1 . Sinc e

u(x) ^ v £(x) and v£(x) — • u(x) a

s e —> 0 ,

we hav e u(x) ^ minu(x) . D V

' 3D

K

J

EXERCISE 1 .1 . Prov e th e followin g version s o f th e maximu m principl e fo r th e operator (1 ) . (a) I f c(x) ^ 0 , the n a superellipti c (subelliptic ) functio n attain s th e negativ e minimum valu e (th e positiv e maximu m value ) o n th e boundar y o f a domain . (b) Le t c(x) ^ 0 and le t u(x) b e a solution t o th e equatio n Lu — 0 in a domai n D, wher e L i s the operato r (1 ) . The n u(x) doe s no t attai n th e zer o extrema l valu e inside th e domai n D (i.e. , u^0). (c) Le t L b e a uniforml y ellipti c operato r (1 ) , wit h th e coefficient s bi bounde d and th e coefficien t c(x) bounde d fro m above . The n ther e exist s a constan t do > 0 such tha t do depend s o n th e constant s a , M fro m th e inequalit y (2 ;) an d o n th e bounds o f the coefficient s bi and c . Fo r an y domai n D o f diameter les s than do an d any functio n u tha t i s a solutio n t o th e equatio n Lu = 0 in D an d i s continuou s i n D, th e followin g inequalit y holds :

max \u\ ^ 2 max \u\. D dD

We tur n t o a n operato r L o f th e for m (l 7 ). Fo r n = 2 we ca n mak e som e con clusions abou t th e globa l behavio r o f solution s usin g onl y th e maximu m principle . Let n = 2 and le t u(x,y) b e a solutio n t o th e equatio n (60 Lu

= 0

in a plane domai n D. W e note that an y linea r functio n ax + by + c satisfies equatio n (&). W e conside r th e se t o f point s (x,y) a t whic h u(x, y) > ax + by + c.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

1 4

. ELLIPTI C EQUATION S I N NONDIVERGENC E FOR M

FIGURE 1

Any connecte d componen t D f o f this set must reac h th e boundary o f D; otherwise , applying th e maximum principl e t o the functio n v(x, y) = u(x, y) — {ax + by + c) in th e domain D 1 ', we arrive a t a contradiction . The sam e assertio n i s true fo r the set of points a t whic h u(x, y) < ax + by + c. This fac t ca n be geometrically expresse d a s follows: I t is impossible t o cut a "cap " off th e grap h o f u by any plane (Figur e 1). We sa y that th e graph o f a continuou s functio n f(x,y) i s of generalized nonpositive curvatur e i f it is impossible to cut a "cap " of f the grap h o f / b y any plane. Thus, the graph of a solution to equation (6 ) is of generalized nonpositiv e curvature . Let a functio n f(x,y) b e continuou s i n th e entir e plan e an d le t th e grap h of / ( # , y) b e of generalize d nonpositiv e curvature . The n th e followin g Liouvill e theorem i s valid. LIOUVILLE THEOREM . Let

M(r)= ma

x \u(x,y)\.

Then either the inequality lim — ^ > 0 1—>oo T

holds or the graph of f is a cylinder whose generator is parallel to the plane xy. If th e graph o f a solutio n t o equation (6 ) is such a cylinde r an d the solutio n increases mor e slowly than a linear function , the n the solution is a constant (why?) . This theore m i s due to Adel'son-Vel'skii wh o gav e a rather complicate d proof . We do not prove this theorem, but establish another relate d property of a continuous function wit h th e graph o f generalized nonpositiv e curvature . W e mean a theore m of Phragmen-Lindelo f type . W e first recal l th e Phragmen-Lindelo f theore m t o clarify connection s betwee n thi s classica l theore m an d the theorem w e will prove . Furthermore, w e will conside r variou s generalization s o f the classical Phragmen Lindelof theorem .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§1. T H E M A X I M U M P R I N C I P L E 5

P H R A G M E N - L I N D E L O F THEOREM . Let an analytic function f(z) be defined in the upper complex half-plane z and let the following inequality hold for any point x on the real axis:

IH5|/(z)|x

Then either \f(z)\ ^ 1 in the entire upper half-plane or (7) li

— lnM(r ) m^ > i—>oo r

0,

where M(r) = su p \f(z)\. \=r \z\=r

We not e tha t th e uppe r limi t ca n b e replace d b y th e lowe r limi t i n (7) , but w e will generaliz e th e weake r inequalit y (7 ) here . The classica l Phragmen-Lindelo f theore m ca n b e reformulate d a s a theore m about harmoni c functions . P H R A G M E N - L I N D E L O F THEORE M (fo r a harmoni c function) . Let u(x,y) be harmonic function in the upper half-plane y > 0 and let

lim u(x,y)

a

< 0

(x,2/)—>(x 0 ,0)

for any x$. Then either u(x,y) ^ 0 everywhere in the upper half-plane or M r *— lim —( )—„ > 0 ,

r—>oo r

where M(r) = su

p u(x,y).

x2-\-y2=r2

PROOF. Le t th e harmoni c functio n v(x, y) b e conjugate t o the function u(x, y). We se t f(z) = e u+lv and appl y th e classica l Phragmen-Lindelo f theore m t o th e functio n / . • This theore m (wit h som e modifications ) remain s vali d fo r a n arbitrar y contin uous functio n wit h th e grap h o f generalize d nonpositiv e curvature . W e formulat e the correspondin g assertion . T H E O R E M 1 .2 . Let f(x,y) be a continuous function in the half-plane y ^ 0 with the graph of generalized nonpositive curvature. If

(8) / ( x , 0 X

0

for all x, then one of the following assertions holds: (a) f(x,y)^0fory>0; (b) -,— M(r) n lim — ^ > 0 , i—>oo

r

where M(r)= su

p |/(x,2/)|

;

x2_|_y2_r2

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

6

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M

FIGURE 2

(c) there exists Y such that for y ^ Y the graph of f(x,y) is generator is parallel to the x-axis.

a cylinder whose

P R O O F . Instea d o f (8), we can write f(x, 0 ) < 0 since we can replace / b y / — e with arbitrar y e > 0 . Assuming tha t assertion s (a ) and (b) fail, w e show the validity o f assertion (c). Let f{x0,y0) =a > 0, y 0 > 0. We conside r th e functio n

0 , 77(0:3) > 0 and the point (3:2,77(0:2) ) ^ e s o n ^ n e plan e (x , z) abov e the segment joinin g th e points (xi , 77(0:1)) an d (#3,77(23)) . The n thi s segmen t cut s a "cap" of f the graph o f 77(0:) so that th e "cap " i s directed upwar d an d is located in the uppe r half-plan e (x,z) (Figur e 3). But i n this case , the plane i n the space (x , y, z) tha t contain s thi s segmen t and is paralle l t o the y-axis cut s a "cap " of f the graph o f Lp(x,y), whic h i s impossible. Consequently, rj(x) = cons t = A > 0 . We consider th e level set E = {(x, y) \ ip(x,y) = A}. Th e set E, bein g the level set o f a continuous function , i s closed. I t intersects eac h lin e x = const . O n each of these lines , w e take th e lowest poin t o f E an d denote th e obtained se t by EQ. We prove tha t EQ i s a line paralle l t o the x-axis.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

7

§1. T H E M A X I M U M P R I N C I P L E

Z

k

(X3,7/(X3)>

(x2,J7(x2))

(*i,?7(Xi))

FIGURE 3 (x 3 ,y 3 )

^ —

&\.y[)

^^^^^^VlW ^^^"""^

I

*2

*3

FIGURE 4

We assum e th e contrary . Le t ther e exis t tw o point s (#1 ,2/1 ) an d (#2,2/2 ) ° f ^ 0 such tha t 2/ 1 7 ^ 2/2- Fo r defmiteness , le t X2 < x\ an d 2/ 2 < 2/i - Ther e i s a poin t (X3,2/3) £ £? o that i s located t o th e lef t o f (#2,2/2 ) an d satisfie s th e relatio n 2/i 2/2 2/ 1 #1 xi - # 2

2/3 - #

3

We conside r th e trapezoi d wit h vertice s (#1 ,2/1 ) , (#i,0) , (#3,0) , an d (#3,2/3) . On th e lowe r bas e an d latera l sides , th e inequalit y ip(x,y) < A hold s everywher e except th e point s (#1 ,2/1 ) an d (#3,2/3) . O n the plan e (#,2/) , we draw a line I between the poin t (#2,2/2 ) an d th e segmen t [(#1 ,2/1 ) , (#3,2/3)] s o tha t th e lin e i s paralle l t o the segment an d intersect s bot h latera l side s of the trapezoid . Le t thi s lin e be give n by th e equatio n 2/ = kx + b and le t (xi,y[) an d (#3,2/3 ) be - the intersectio n point s o f th e lin e wit h th e latera l sides o f th e trapezoi d (Figur e 4) . On the segments [(#1 , 0), (#1,2/i)], [(#3, 0), (#3,2/3)], and [(#3 , 0), (#1, 0)] we have ip(x,y) < A — 6, where 6 is a positiv e number . W e choos e e > 0 so smal l tha t e\y - kx - b\ < 6

on th e sam e segments . I n th e spac e (#,2/ , 2), th e plan e z — A + e{y — kx — b)

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

1 8

. ELLIPTI C EQUATION S I N NONDIVERGENC E FOR M

cuts a "cap " of f the graph o f ip(x,y). Indeed , 0 and (p(x, y) - [A + e(y - kx - b)] < 0 on al l sides of the trapezoid. Thus, th e line E$ is parallel t o the x-axis. Denotin g b y Y it s ordinate, w e se e that th e graph o f / ( x, y) contain s th e line y = Y, z — A\ — A-\- if^-Y. Settin g

fi(x,y) = f(x,y-Y)-A

1

and applyin g the obtained resul t t o f\, w e conclude that i f the grap h of the functio n f(x,y) contain s a line paralle l t o the x-axis, the n i t als o contain s anothe r suc h lin e with large r ordinate . T o complete th e proof, i t remain s t o note tha t i f the grap h of / contain s tw o lines paralle l t o th e x-axis , the n ther e i s on e more suc h lin e between them . T o prove thi s assertion , w e consider a plane containin g thes e line s and subtrac t fro m / th e linea r functio n whos e grap h i s thi s plan e (perhaps , w e must chang e th e sign o f the difference). The n th e projections o f these line s ont o the plan e (x , y) can be regarded a s the boundary o f the strip II , which mean s tha t the require d lin e exists . • T E S T QUESTION . I n which par t o f the proo f di d we use the fact tha t M(r) i s an uppe r boun d o f the module o f u (bu t not of u itself ) fo r x 2 + y 2 — r2l Wh y i s the conditio n o n the module necessar y here ?

Bibliographic remarks . Th e main goa l of this sectio n is to present th e properties o f a functio n tha t ca n be derived fro m th e maximum principl e regarde d a s a geometrica l propert y independentl y o f an equation satisfie d b y the function. I n this section , no t only th e initial function , bu t any function tha t differ s b y a linea r function i s assumed t o satisfy th e maximum principle . For a function define d i n the entire plan e an d possessing suc h propertie s Bern shtein [Bel ] proved th e following so-calle d "two-sided " Liouvill e theorem : a func tion i s a constan t i f i t i s bounde d fro m below an d fro m above . H e als o prove d that th e ordinary "one-sided " Liouvill e theore m fail s i n this case . I n the proof of Bernshtein, th e function i s assume d t o b e of certai n smoothness . Gelfan d mad e the proposa l tha t th e smoothness i s not essential; i t suffice s t o assum e continuity . This hypothesi s turn s ou t to be true. Th e corresponding resul t wa s established by Adel'son-Vel'skii [Ad]. The principa l ide a o f the proo f o f the Phragmen-Lindelof theore m fo r surface s of generalize d nonpositiv e curvatur e (cf . §1 ) is due to Adel'son-Vel'skii. Th e proof itself wa s obtaine d b y Gerasimov [Ge]. For n > 2 the Bernshtein theore m (henc e als o th e Adel'son-Vel'skii theorem ) fails. Th e corresponding exampl e wa s constructed b y Hopf [Hoi] . A s was noted by Kosmodem'yanskii , a sligh t modificatio n o f the Hopf exampl e lead s t o a coun terexample t o the theorem o f Phragmen-Lindelof typ e fo r surfaces o f generalize d nonpositive curvature . I n this section , thi s theore m i s formulated fo r n — 2 . However, in the case n > 2 it is also possible to make some conclusions about the global behavio r o f a function wit h th e graph o f generalized nonpositiv e curvature .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§2. s-CAPACIT Y

9

The followin g theore m i s due to Kosmodem'yanskii . Le t f(x) b e a functio n define d in th e cylinde r n i=2

and havin g th e grap h o f generalize d nonpositiv e curvature . Suppos e tha t f(x) i s nonpositive o n th e latera l surfac e o f th e cylinder . Introduc e th e notatio n M(t) = m a x / ( x ) . X\=t

One o f th e followin g case s take s place : Case 1 . f(x) i s nonpositiv e everywher e i n th e cylinder . Case 2 . lirn| rri|_).00 M(x 1 1 )/\x \>0. Case 3 . M(x\) = cons t > 0 . In Cas e 3, the function f(x) attain s the maximum valu e on a line that i s parallel to th e xi-axis . Not e tha t th e rat e o f growt h i n Cas e 2 canno t b e improved . Th e fact tha t th e grap h ha s generalize d nonpositiv e curvatur e mean s tha t i t i s a saddl e in a sense . W e see that som e conclusions, especiall y i n the cas e n — 2 , can b e mad e from th e saddl e propert y alone . Bu t thi s propert y itsel f i s no t sufficien t fo r mor e serious assertions . W e nee d a quantitativ e estimat e fo r th e "saddle-property" . I n the followin g sections , w e will obtai n i t i n term s o f th e s-capacity . §2. s-capacit y Let s b e a positiv e number . Le t E an d T b e B-sets i n W 1 . W e conside r al l possible measure s / x on E, i.e. , completel y additiv e nonnegativ e function s tha t ar e defined o n a cr-algebra consistin g o f subsets o f E an d containin g B-sets (cf . Halmo s [Ha]). A measure \i i s sai d t o b e admissible i f th e followin g conditio n holds :

We se t sup/xE = Cj(E), where th e supremu m i s take n ove r al l admissibl e measures . The numbe r Cj(E) i s calle d th e relative s-capacity o f th e se t E wit h respec t to th e se t T . The relativ e capacit y o f a se t E wit h respec t t o it s complemen t wil l b e calle d simply th e s-capacity an d denote d b y C S(E). Thus , Cf\E(E) =

C

B(E).

We nee d som e propertie s o f th e s-capacity . THEOREM

2.1 . LetE l \, let Q p^ be the cylinder defined by the conditions n-l

2=1

and let p , h — p] o f th e x naxis. W e suppose tha t p i s distributed uniforml y o n [/) , h — p] wit h th e linea r densit y v suc h tha t i f x doe s no t belon g t o th e cylinder , the n

(11) v

t~P-t—;— w^ E ^ * ? + (*„-0 1

L s

In thi s case , w e hav e (12) C

vh

s(Qp,h)^v(h-2p)^

We estimat e th e integra l o n th e left-han d sid e o f the inequalit y (1 1 ) . Le t a poin t x lie o n th e latera l surfac e o f the cylinder . W e eliminat e th e segmen t [x n — p, xn + p] from th e segmen t [p, h — p] of the x n -axis. Th e remainin g se t (i t ma y coincid e wit h the entir e segmen t [/? , h — p]) will b e denote d b y H. W e hav e rp ~pdAt r /r i Jp h

ir


s

Y.^kiak[(s +

2)-e] + ]T>*7 i + c_

Therefore, fo r s > e — 2 and sufficientl y smal l r (r < ro, where r o depend s o n s and the operator , L) the functio n l/\x — x°\ s is subelliptic i n (D D Qf0) \ i x°), wher e D i s the domai n o f th e operator . Bibliographic remarks . Th e excellen t surve y concernin g th e notion s o f a capacity an d a potential a s well a s a detailed bibliograph y ca n b e foun d i n [Land]. We briefly discus s the difference betwee n the approac h i n this book an d th e tra ditional approac h t o th e us e o f capacities an d potentials . Th e traditiona l approac h consists i n considering th e potentia l (generate d b y th e Ries z potential 1 ) y

'J

E\x-y\

s

in th e entir e space , whic h require s th e conditio n s < n on s. I f we ar e intereste d in th e behavio r o f U(x) o n a set T a t a positive distanc e fro m £" , th e valu e o f s is no t essential . Fo r an y linea r uniforml y ellipti c operato r an d sufficientl y larg e 5 (perhaps, greate r tha n n) the functio n U(x) i s subelliptic. I f we ar e intereste d i n a solution t o a linear equatio n withou t th e right-han d side , then it is useful t o regar d U(x) a s a barrier; moreover , E is placed outsid e th e domai n wher e th e equatio n is considered, wherea s U(x) i s studied insid e thi s domain . For the classica l Wiener capacit y w e can give an equivalen t definitio n b y mean s of the Dirichlet integra l for a n external problem for the set E with the unit boundar y condition. A number o f authors prefe r t o work i n terms o f this definitio n an d its generalizations. Fo r equation s wit h selfadjoin t principa l par t thi s approac h i s mor e suitable becaus e i t allow s u s to deal wit h th e principa l par t o f the fundamenta l solution o r eve n the fundamenta l solutio n itsel f instea d o f its roug h approximation . Some work s usin g thes e method s wil l b e discusse d later . However, i n the author' s opinion , th e elementar y approac h presente d i n this book ha s som e advantages . First , i t is simple. Secon d (mor e serious ) i s that in considering nonselfadjoin t equatio n wit h discontinuou s coefficients , th e bes t wa y t o find charasteristi c feature s o f a solution i s to study th e spreadin g o f roots o f the characteristic equatio n a t a point. W e not e tha t th e spreadin g canno t b e remove d since the coefficients ar e discontinuous. Therefore , a constant appears , referre d t o as the ellipticity constant an d denoted b y e, i.e., the ratio of the trace of a matrix to th e smallest eigenvalu e (also , i t should b e minimize d ove r al l linea r transformations) . The behavio r o f a solution depend s o n this constant . I n a sense, it cannot b e bette r than i t is possible t o describe i n terms o f the potentia l wit h kerne l l / r e _ 2 . W e note tha t th e numbe r e plays th e rol e o f dimension i n a series o f exact theorem s (see, fo r example , Chapte r 1 , Theore m 5.3 , simila r t o the Uryso n theore m [Ur] , i n which th e exponen t l/( e — 3), similar t o th e Uryso n exponen t l / ( n — 3), cannot b e replaced b y -^K$ + e). In thi s section , w e also introduce d th e notion s o f s-capacity an d relativ e scapacity. However , onl y the s-capacit y i s essentially use d i n this book . Th e relativ e 1 For th e Riesz potentia l w e recommend [Land] , wher e al l necessary reference s t o origina l works ar e als o presented .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

14

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M

s-capacity, introduce d i n order t o obtai n semi-additivity , wa s used elsewher e (we refer th e reader t o [Lai]) . §3. Th e lemma o n the normal derivativ e and th e strong maximu m principl e LEMMA 3. 1 (o n the normal derivative) . Let L be the operator (1 ) in the ball QQR and let u(x) be a function subelliptic (superelliptic) in Q° R and continuous in

QR. Suppose that x° G 5^ and > u(x 0))

u(x) < u(x°) (u(x)

for all x G Q°R. If there exists the inner normal derivative f ^ at the point x°, then

0

)•

P R O O F . Fo r definiteness , w e assum e tha t u(x) i s subelliptic . Le t e b e the ellipticity constan t o f the operato r L. W e set u(x°) = m. The n

max u(x) = m — a. a

> 0.

\x\ = R/2

We conside r th e auxiliary functio n v[x) = m — |-— —^ + e|e-2 x

Re

-2>

where e > 0 is so small tha t v(x) > m — a for \x\ = R/2, an d the functions u(x) and v(x) ar e defined i n the spherical laye r

By Lemm a 2.1 , Lv < 0, v\

^

mo UJ L o

* R/2,R U

^

u\\80,0

*LR/2)R

By Theore m 1.1, v(x) ^

u(x) fo

rx

G Q°R/2,R-

Since v(x°) — u(x°), w e have du dn

dv ^ dn X = XQ

M l-l 0 (althoug h w e hav e v > Cs{H) ^P > 0 in D) becaus e w e d o no t kno w i f ther e exist s a limi t o f U(x) a s x G D tend s t o I \ However, i n an y case , w e can asser t tha t lim v(x) xeD,x->r

> 0,

which i s sufficient fo r application s o f th e maximu m principle . Thus, n < v. W e conclud e tha t sup u(x)

^ su

+

^ > )

v^Qi

i

which complete s th e proo f o f th e lemm a sinc e e i s arbitrary .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms



§4. T H E G R O W T H LEMM A

17

If s < n an d R < 1 in the assumption s o f Lemma 4.1 , then , usin g the inequalit y

cm > ^ which i s valid b y Theore m 2.6 , w e obtai n /x / (16) supw(x

x f £ measiJ \ , ) ^1 + — — ^ — su

x

p u(x)

.

However, th e conditio n J R < 1 is undesirable. Furthermore , th e inequalit y (1 6 ) ca n be improve d b y replacin g R s wit h R n o n th e righ t hand-sid e o f (1 6) . Now w e prov e th e followin g lemma . LEMMA

4.2 . If s [ 1 + r] ——— ] su p u{x), xeD \ R j T ,DnOx0

where rj > 0 is a constant depending on s and n. PROOF. W e first mak e th e similarit y transformatio n whic h transform s Q\ R t o Q% . A s wa s mentioned , th e operato r i s only multiplie d b y a positiv e number . Le t H g o to H' unde r thi s transformation . The n __, mea s H meas H = — . n R Consequently, meas# which complete s th e proo f o f th e lemma . • REMARK 4.1 . Remar k 2. 1 and Exercis e 1 .1 (a ) imply that i f the coefficien t c(x) of th e operato r (1 ) satisfie s th e conditio n c(x) ^ 0 and M/a < n + 2 , then Lemm a 4.2 remain s vali d fo r R < i?o , wher e Ro depend s o n M , c , an d th e bound s o f th e coefficients bi and c ^ 0 . EXERCISE

4.1 . Prov e th e sam e assertio n withou t th e assumptio n c ^ 0 .

Bibliographic remarks . W e consider a solution t o the Dirichle t proble m i n a domain D. Le t th e zer o boundary conditio n b e impose d o n a part o f the boundary . Suppose tha t th e solutio n i s bounded , say , b y 1 . I t i s clear tha t th e zer o boundar y condition o n a par t o f th e boundar y o f D affect s th e solutio n i n som e way ; th e solution i s clos e t o zer o nea r thi s par t o f th e boundary , bu t als o i t canno t b e clos e to 1 a t som e distanc e fro m thi s part . Fo r applications , i t i s importan t t o hav e a quantitative estimat e fo r thi s behavio r dependin g o n th e structur e o f th e par t o f the boundar y wit h zer o boundar y condition . Th e growt h lemm a turne d ou t t o b e a convenien t too l fo r this . O n on e hand , it s proo f i s elementary , an d o n th e othe r hand, i t i s convenien t fo r application s (cf . §§5-7 , 1 0) . Perhaps, th e autho r treat s th e lemm a wit h partiality , becaus e severa l theorem s in the qualitativ e theor y o f second orde r ellipti c equation s hav e bee n prove d b y th e author b y usin g version s o f thi s lemm a (cf. , fo r example , [La2]) . In this lemma , th e "capacity " o f the par t o f th e boundar y wit h zer o boundar y condition i s estimate d i n term s o f th e siz e o f a portio n o f th e complemen t o f th e

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M

18

domain i n which the solutio n i s considered. Thi s portio n i s located i n some ball. I n [La2] an d othe r work s b y th e author , th e siz e o f th e portio n wa s characterize d b y its measure . I t i s clear tha t measur e provide s onl y a rough sufficien t characteristic , which does not relat e to the essenc e of the matter . Th e capacity , o n the othe r hand , yields a n adequat e characteristic . I n term s o f capacity , th e lemm a wa s first prove d by Blokhin a [Bll ] (se e detail s i n [B1 2]) . §5. Th e behavio r o f solution s i n a neighborhoo d o f a boundar y poin t We consider a domain D' C D wit h boundar y Y. Le t e be a positive number. A point x° G T i s calle d a n e-regular boundary point i f th e followin g condition s hold . For ever y pai r e\ > 0 an d £ 2 > 0 there i s 6 > 0 suc h tha t fo r an y domai n D' c D with boundar y T' , a uniforml y ellipti c operato r L define d i n D' wit h th e ellipticit y constant e' ^ e , an d a functio n u(x) < 1 , subellipti c fo r L an d continuou s i n D , the inequalit y u\ n ^ 0 0 Ir'nQ^ ^ implies th e inequalit y u\ 0

THEOREM 5.1 . Let x° eY and

< £2 -

\D'nQf e ^ n . We set s = e - 2 and denote

Ca{Q{-m\D)=7m. The point x° is e-regular if the series 00

V 4™ 7r> 771=1

is divergent. PROOF. Le t e\ > 0 an d 82 > 0 b e given . W e conside r a subdomai n D' o f D with boundar y r x , a uniforml y ellipti c operato r L define d i n D' wit h th e ellipticit y constant e' suc h tha t e' ^ e , an d a subellipti c functio n u(x) tha t i s continuou s i n D an d satisfie s th e inequalitie s u(x) < 1 , u(x)\

n

< 0.

It i s require d t o sho w tha t ther e exist s 6 > 0 depending o n e , £1 , and £ 2 such tha t u(x) < 82 at point s x e D' insid e th e ^-neighborhoo d o f th e poin t x° (Figur e 7) . Denote b y m o th e smalles t natura l numbe r suc h tha t 4-mo rao le t ther e exis t a poin t x' G D' suc h tha t \x' -x*\ ^ 4 ~

m

, u{x')

^e

2.

Our goa l i s to sho w tha t th e numbe r m i s less tha n a constan t dependin g o n e , £1, and £2 . Fo r eac h i = m o , . . . , m w e introduc e th e notatio n M; = su

p i^(x)

.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§5. TH E BEHAVIO R O F SOLUTION S NEA R A BOUNDAR Y POIN T

19

FIGURE 7

For eac h i — mo + l , . . . , r a w e conside r th e ball s Q\- x an d Q*_ (i_iy W e als o consider th e se t o f point s x G D' D Q^_(i_1) a t whic h u(x) > 0 . Denot e b y Di the componen t o f thi s se t tha t contain s a poin t o n th e spher e S%-i a t whic h th e function u take s th e valu e Mi. W e hav e

Cs{Qf-,\Di)

^ Cs{Qf-,\D) = lt.

Applying Lemm a 4. 1 t o th e ball s Q^-z an d Q^_ (i_1 )5 th e domai n Di, an d th e function u, w e obtai n M,_1^(l+ ^-4^)M, . Consequently,

hence

m 1

n

(i+€-4"7i)

V

m)'

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§5. T H E BEHAVIO R O F S O L U T I O N S N E A R A BOUNDAR Y P O I N T

23

Therefore, th e sequenc e {vm}, ra =

fc,fc + 1, . .. ({v

m},

rn

= k, k + 1 ,... )

is monotone i n every D^ ; moreover, i t i s bounded i n view of the maximum principl e and, consequently , i t converges . We not e tha t fo r solution s u m t o th e Dirichle t problem s wit h th e boundar y conditions Um\ =

F

we hav e m a x | u m -Vm\ < Dm

£ ,

where Vm =V

m+Vm.

Hence {u m} converges . 3. Part s 1 and 2 mean tha t fo r a monoton e sequenc e o f domain s th e sequenc e {um} converge s t o a functio n uj whic h i s independen t o f th e metho d o f extensio n of / . Henc e i t converge s fo r an y sequenc e o f domain s {Dm}. Indeed , i f fo r som e sequence o f domain s {D m} th e sequenc e {u m} doe s no t converge , the n w e ca n choose a subsequenc e {D mk} s o tha t D mk C D mk+1 an d th e sequenc e {u m} doe s not converg e either . 4. Th e limi t functio n Uf i s harmonic . Thi s fac t follows , e.g. , fro m th e com pactness theore m whic h assert s tha t an y sequenc e o f function s tha t ar e harmoni c and uniforml y bounde d i n D , i s compact i n an y subdomai n lyin g strictl y insid e D. REMARK 5.1 . Th e argument s i n 1 - 3 ca n b e applie d t o an y uniforml y ellipti c operator L provide d tha t th e Dirichle t proble m fo r L i s solvable i n an y subdomai n that lie s strictl y insid e D an d ha s sufficientl y smoot h boundary . Thus, th e abov e notio n o f a generalize d solutio n t o th e Dirichle t proble m fo r the Laplac e equatio n i s well defined . DEFINITION 5.1 . A poin t x° o n th e boundar y dD o f a domai n D i s sai d t o b e regular if fo r an y continuou s functio n / define d o n dD th e generalize d solutio n Uf to th e Dirichle t proble m wit h th e boundar y conditio n \dD

J

satisfies th e conditio n lim u f{x) = f(x°). xeD As above , w e se t -ym = C

x n-2{Q 4-m\D).

THEOREM 5. 4 (Wiene r criterion) . A point x° is regular if and only if the series oo

(21) £

A

m{n 2

~ hm

is divergent.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

24

1. ELLIPTI C EQUATION S I N NONDIVERGENC E FOR M

PROOF. Sufficiency. W e extend th e functio n b y continuit y t o th e domai n D and denot e th e extended functio n b y F. Fo r e > 0 we find £\ > 0 such tha t

\F(x)-f(x°)\ 0 such tha t

Vm(x) < ^ xeD

mnQf.

Hence ttmW-/(a50) n Q f °

.

Therefore,

uf(x)-f(x°)^e, xeDHQf. The invers e inequalit y i s obtained i n a simila r way . Consequently , u(x) converge s to f(x°) a s x— > x°. REMARK 5.2 . Instea d o f the Laplace operator , w e consider a uniformly ellipti c operator L with th e ellipticity constan t e' ^ e. Le t izm be a solution t o the problem Lum=0, w

mLn =

^

Let Um converge t o a solutio n u(x) o f the equation Lu — 0 . Pro m th e above proo f it i s clear tha t i f a point x° G 3D i s e-regular, the n Uf(x)— > /(x° ) a s x —> x°, x G D . Necessity. T o prov e th e necessit y w e nee d som e propertie s o f th e s-capacit y and especiall y o f the Wiener capacity . Thes e propertie s ar e established i n Lemma s 5.1 and 5.2 below . LEMMA 5.1 . Let E be a bounded closed set, G D E an open set, and e > 0 an arbitrary number. There exists an open set G, E C G C G f, with twice continuously differentiable boundary and a measure ji on G so that for

JG\v-An-2 we have U\ = 1 \dG

and (22) j[_d/i7m < «,

771=7710 — 1

where a wil l b e chosen later . Le t the boundary functio n / b e such tha t /(s°) = l , / ( * ) < !

, f(x)=0,

\x

- x°\ > 4~ m°.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

26

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M

We exten d / b y continuit y t o th e entir e spac e s o tha t th e abov e condition s ar e preserved fo r th e extende d functio n F(x), i.e. , F(x) ^ 1 , F(x)

= 0 , \x-

x°\ ^ 4

_m

°.

We sho w tha t fo r 6 > 0 there i s a poin t x e D , \X — x°\ < oo

Let Uk\

Dk ~

Uf = li i

'

k—

Introducing th e notatio n Em =

--Qf-m\D,

we ca n writ e C n - 2 y^m) 7m

*

Hm = Em \ E

m+l

Let Then Hm C E

m-\.

:

n-2{Hrn)

In th e notatio n im

C

we hav e oo

m—rriQ

By Lemm a 5.1 , every Hm ca n b e embedde d i n a n ope n se t G m wit h twic e contin uously differentiabl e boundar y suc h tha t On Gmi w e introduce a measur e \im suc h tha t

tem \ x - y\ implies Um\ = and (23) /

1

l 0 , ! ^ i < q, 0 < q < 1 , we se t A measur e /i m o n £? m i s sai d t o b e admissibl e i f

^ ( | L < X fo

rX

£Em

J'EE,m \*-y We se t

where th e supremu m i s take n ove r al l admissibl e measure s o n E m. Prov e tha t th e point x° i s regula r i f an d onl y i f J2 Um(x°) = CXD . ra=l

Bibliographic remarks . W e hav e see n tha t sufficien t an d necessar y condi tions fo r th e regularit y o f a boundar y poin t i n th e cas e o f th e Dirichle t proble m for th e Laplac e equatio n ar e give n b y th e Wiene r criterio n (anothe r proo f ca n b e found i n [Ke ] an d [Lai]) . In 1 949 , Oleini k [01 2 ] prove d tha t th e criterio n fo r th e regularit y o f a bound ary poin t fo r a linea r secon d orde r equatio n wit h sufficientl y smoot h coefficient s coincides wit h a similar criterio n fo r th e Laplac e equation . Later , th e condition s o n coefficients wer e weakened . I n 1 962 , Herv e [He ] prove d tha t i t suffice s t o assum e that th e coefficient s satisf y th e Holde r condition . Recently , Krylo v [Kr ] prove d that instea d o f th e smoothnes s o f coefficient s on e ca n impos e a weake r condition ; namely, th e continuit y o f coefficient s provide d tha t th e modul e o f continuit y uni formly satisfie s th e Din i condition . However, i t turn s ou t tha t a simila r conditio n (perhaps , th e Din i conditio n itself) represent s a border case . I t is possible to construct a n example of an equatio n that ha s discontinuous coefficient s an d i s as close to the Laplace equation a s desired, but a poin t tha t i s regula r fo r th e Laplac e equatio n canno t b e regula r fo r th e constructed equatio n (se e [Lai , M]) . Moreover , a s wa s show n b y Zogra f [Zo] , coefficients ca n eve n b e mad e continuou s i n thi s example . The abov e i s tru e fo r a n equatio n i n nondivergenc e form . I n [LSW ] i t wa s shown tha t th e regularit y conditio n fo r a uniforml y ellipti c equatio n i n divergenc e form wit h arbitrar y measurabl e an d bounde d coefficient s coincide s wit h tha t fo r the Laplac e equation . If coefficient s ar e goo d (i n som e sense ) u p t o a neighborhoo d o f a boundar y point (fo r example , satisf y th e Holde r conditio n i n a neighborhoo d o f a boundar y point), the n th e Wiener conditio n is necessary an d sufficien t fo r the regularity o f the boundary point . Suppos e tha t th e Wiene r conditio n holds , i.e. , th e correspondin g series i s divergent , an d th e poin t i s regular , whic h mean s tha t th e generalize d solution i s continuous a t thi s point . Studyin g th e rat e o f divergenc e o f th e Wiene r series, w e ca n sa y mor e abou t th e behavio r o f th e solutio n a t thi s point . W e ca n indicate th e exac t boun d fo r th e modul e o f continuit y o f th e solutio n a t thi s point .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

30

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M

This stud y wa s firs t develope d b y Maz'ya [Mai ] wh o use d a different approac h t o the definitio n o f capacity. The majo r par t o f thi s sectio n i s devote d t o th e proo f o f th e necessit y an d sufficiency o f the Wiener criterio n fo r the regularity o f a boundar y poin t fo r the Laplace equation . Th e sufficiency i s established i n a simple way , wherea s th e proof of the necessity occupie s th e rest o f the section (an d Appendix, §1 ) . It seem s tha t the situatio n shoul d b e th e opposite , i.e. , th e necessit y o f th e Wiene r criterio n should b e "almos t obvious " wherea s th e sufficiency "mus t b e proved." Th e matter is tha t introducin g th e notio n o f capacity , w e did no t prov e tha t th e extrema l measure exists . Indeed , w e had no need o f this fac t anywher e els e in the book. Bu t here, w e had to construct (cf . Appendix , §1 ) a measure tha t realize s the extremum. The constructio n itsel f i s not too complicated, bu t looks heavy in contrast wit h the simplicity o f other construction s i n this chapter . Additional bibliographi c remarks . Th e proble m o f th e regularit y o f a boundary poin t i n the case o f a nonsmoot h boundar y an d a secon d orde r linea r equation i n nondivergence for m ha s not been studie d lately . Perhaps , th e most interesting result s wer e obtaine d b y Yu. A . Alkhutov wh o suggeste d som e regularit y conditions i n the case o f a discontinuous boundary . §6. Th e behavior o f solutions a t infinit y We retur n t o an arbitrary uniforml y ellipti c equatio n o f the form (1 /) . I n this section, w e use the growth lemm a t o prove theorem s o f Phragmen-Lindelof type . By this , w e mean theorem s tha t dea l wit h subellipti c function s tha t ar e define d in a n unbounded domai n an d are nonpositive o n the boundary. Dependin g o n th e shape o f the domain , fo r such function s th e minimum rat e of growth a t infinity can be foun d i f there i s at leas t on e point o f the domain a t whic h th e function unde r consideration i s positive. THEOREM

6.1 . Let an unbounded domain D lie in the layer \xn\ < h, —o c < x % < oo , i

= 1 , . . . ,n — 1 .

Let L be a uniformly elliptic operator in D and let u be a subsolution that is continuous in the closure of D and nonpositive on the boundary of D. Then one of the following assertions holds: (a) u ^ 0 in D; (b) ,. M(r) lim — ^ n> 0 , r—>oo e

h

where M(r) = ma

x u(x)

and c > 0 is a constant depending on n and the ellipticity constant e of the operator L. P R O O F . B y using a similarit y transformation , i t i s possible t o reduc e th e re quired assertio n t o the case of some fixed h. I t is convenient t o take h = 1/2. Suppose tha t assertio n (a ) i s false , i.e. , u(x°) = a > 0 fo r a poin t x° = (x°1,x^...,xl) e D. W e set r 0 = \x°\ and r k = r 0 + 4/c , k = 0 , 1 , 2 , . . . . Le t k y b e a point o n the sphere S® wher e u(y k) = M(rk). W e consider th e balls Q\

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§6. T H E BEHAVIO R O F S O L U T I O N S A T I N F I N I T Y

31

and Q\ . Th e differenc e Q\ \D contain s a bal l o f radiu s 1 /4 . Settin g s — e — 2 , from Lemm a 4. 1 w e obtain th e relatio n sup u(x)

> ( l + £,C s(Q°1 /4)) su

xeDnQlk

p u(x)

= ( 1 + 7 ) su

xeDnQ\k

p u(x),

xeDnQ\k

where 7 > 0 depends o n e and n. B y th e maximu m principle , M(r f e + i) ^ su

p u(x).

xeDnQf

Therefore, M(rfc+i)>(l+7)Af(rfc). Hence M(r0 + 4k ) > a( l + 7) * = ae

kln1 +

< ^

and M(r 0 + 4 f c ) > a e

M

^[''0+4('=+1 )]

for sufficientl y larg e /c , which prove s th e theore m i n vie w o f th e maximu m princi ple. • Let Bh denot e th e exterio r o f th e doubl e circula r con e

i=l

We conside r a n unbounde d domai n suc h tha t a poin t x o f th e domai n i s locate d inside Bh i f \x\ i s sufficientl y large . W e wil l se e (cf . Theore m 6.2 ) ho w a func tion subellipti c i n thi s domai n an d nonpositiv e o n it s boundar y increase s whe n h increases, provide d tha t th e functio n i s positive somewher e insid e th e domain . Moreover, w e will assum e tha t h i s smal l enough : h < 1 (instead o f 1 , anothe r constant ca n b e take n here) . I f h i s large , the n th e fac t tha t th e con e i s doubl e i s not essential . Thi s cas e wil l b e discusse d i n Theore m 6.3 . T H E O R E M 6.2 . Let 0 < h < 1 and let D be an unbounded domain. Suppose that there exists R > 0 such that D \ Q° R C Bh- Let L be a uniformly elliptic operator in D and let a subelliptic function u(x) be continuous in the closure of D and nonpositive on the boundary of D. Then one of the following assertions holds: (a) u(x) ^ 0 everywhere in D; (b)

(25) li

m M(r)/r c/h >

0,

r—->oo

where M{r) = max u \x\=r

and c> 0 is a constant depending on e.

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

32

1. E L L I P T I C E Q U A T I O N S I N N O N D I V E R G E N C E F O R M x

nl

\

^

\

k

^ y * ] ^

^ 0

c^

A

^

^

^

F I G U R E1 PROOF.

0

W e assum e tha t assertio n (a ) is false, i.e. , ther e exist s a point x° £ D

such tha t u(x°) = a > 0 . Set (26) |x°

| = r 0, r

k

= r 0 ( l + 8ft) fc, k

= 1 ,2,.. . .

Let i t attain th e maximum valu e M{rk) a t a point 7/ fc £ 5 ^. W e consider th e balls QV2hr an< ^ Qthr • Sinc e ft < 1 , the set Q\ hr \ D contain s a bal l o f radius hr (Figure 10). By Lemm a 4.1 , for s = e — 2 we obtain sup u(x)

> I 1 + rj ,^*Z

|

M(r

{2hrky

x£DnQy8hrh

k).

Setting C

V

'{Q\hrk) _ (2hrky

=

C a{Ql) 77-

= 7,

we find tha t 7 depend s onl y o n e, and since M(r f e + i) ^ su

p i6(x)

,

zeDnQ!"rfc

the inequalit y M(rfc+1)^(l +

7

)M(r f c )

holds. Henc e (27) M(r

k)^a(l

+

j) k.

From (26 ) and (27) it follow s tha t lnr fc - l n r 0 ln(l + 8ft) *

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

k/4

§6. T H E B E H A V I O R O F S O L U T I O N S A T I N F I N I T Y

33

For sufficientl y larg e / c we hav e M{rk) > a e l n ( 1 + 7 )

N W ^ ar*

l n (1 + 8 h )

Consequently, fo r sufficientl y larg e r k w e obtai n

where c i s a constan t dependin g o n e , whic h yield s (25 ) i n vie w o f th e maximu m principle. • THEOREM

6.3 . Denote by K h the cone 2 n-l

i=l

Let h > 1 . Le £ D be an unbounded domain and let K^ belong to the complement of D. Let L be a uniformly elliptic operator in D and let a subelliptic function u(x) be continuous in the closure of D and be nonpositive on the boundary of D. Then one of the following assertions holds: (a) u(x) < 0 everywhere in D; (b) lim M(r)/r ch ' > 0, r—• o o

where M{r) — maxii(x) |x|=r

and c is a constant depending on e and s = e — 2 . P R O O F . W e assum e tha t assertio n (a ) i s false . Se t H — l/h. Ther e i s a poin t x° e D suc h tha t u(x°) = a > 0 . W e se t |rrr° | = r 0 an d r k = r 0 • 4 fc, k = 1 , 2 , . . ., and conside r th e ball s Q® k and Q® k i . Th e differenc e Q® k \ D contain s a bal l o f radius \r kH (assum e tha t it s cente r i s place d a t th e poin t x'). Settin g s — e — 2 and applyin g Lemm a 4. 1 an d th e maximu m principle , w e obtai n

/C

s(Qi

) \

= (l + ^-C s{Cfx)\ M{r

k.x)

=

( 1 + 7 tf 8 )M(r f c _ 1 ),

fc = l , 2 , . . . , where 7 > 0 is a constan t dependin g o n e. Henc e (28) M(r

k)>a{l

+

iH

8 k

).

Since lnr f c + i = lnr 0 + (f c + l ) l n 4 , we can expres s k a s follows : lnr f c + i - l n r 0 - I n 4 k ~I n4 2

The con e Kh differ s fro m E n \ B^ b y onl y th e propert y tha t i t consist s o f a singl e cavity .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

34

1. ELLIPTI C EQUATION S I N NONDIVERGENC E FOR M

Substituting thi s expressio n i n (28) , we obtai n M(rfc)>a(l + 7i7s)(lnrfc+1-lnro-ln4)/ln4 and, fo r sufficientl y larg e k, M(rk) >

a(l +

7

t f ' ) ( > n r f c +1 ) / 2 1 n 4 =

^ ( 1 + ^ ) 1 / 2I n4^ ^

The constan t c is taken s o as to satisf y th e inequalit y ln( l + ^/Hs) > 2cln4H s (th e choice i s possibl e becaus e H < 1 ) . Applyin g th e maximu m principle , w e arriv e a t the require d inequality . • 6.1 . I n Theore m 6.3 , we derive d th e inequalit y M(r) > ar ch * (fo r sufficiently larg e r) . Prov e tha t M(r) > ar Clh fo r s > 1 and sufficientl y larg e r . EXERCISE

In Theore m 6. 3 w e assume d tha t h > 1 . Th e choic e o f th e constan t 1 is no t essential. W e ca n tak e an y othe r constant . Th e poin t i s that i n th e estimat e

r—>oo ^

the sam e constan t c ca n b e take n onl y fo r thos e h tha t ar e greate r tha n som e hoThe constan t c decreases a s ho decreases. The cas e wher e K^ become s th e half-spac e (h = 0 ) i s o f specia l interest . I n this case , w e obtain a n exac t estimate ; namely , th e increas e i s not slowe r tha n tha t of a linea r function. 3 Considerin g th e Phragmen-Lindelo f theore m fo r a harmoni c function i n a half-plane , w e se e tha t i f th e harmoni c functio n i s define d i n th e upper half-plane , i s nonpositiv e o n th e x-axis , an d i s positiv e a t som e point , the n it necessaril y increase s a t infinit y no t mor e slowl y tha n a linea r function . W e wil l see tha t thi s fac t i s immanen t fo r subellipti c function s fo r operator s (1 /) . THEOREM 6.4 . Let an unbounded domain D lie in the half-space x n > 0 . Let L be a uniformly elliptic operator in D and let a subelliptic function u(x) be continuous in the closure of D and nonpositive on the boundary of D. Then one of the following assertions holds: (a) u(x) ^ 0 everywhere in D; (b)

(29) li

m — ^> 0, i—>oo T

where M(r) — max| r r | = r u{x). P R O O F . W e assum e tha t assertio n (a ) i s false , i.e. , ther e exist s a poin t x° = ( x ? , . . . , x°) £ D suc h tha t u(x°) = a > 0. W e se t

v(x)=u{x)-—. Denote b y D' th e se t o f point s x G D a t whic h v{x) > 0 and b y D" th e componen t of D' containin g th e poin t x°. However, i n thi s estimate , th e lowe r limi t wil l b e replace d wit h th e uppe r limit .

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

§6. T H E B E H A V I O R O F S O L U T I O N S A T I N F I N I T Y

35

By th e maximu m principle , D" i s an unbounde d domain . Le t c be the constan t from Theore m 6. 2 correspondin g t o th e ellipticit y constan t o f th e operato r L. W e choose hj 0 < h < 1 , so a s t o satisf y th e conditio n

Take Bh tha t ha s th e sam e meanin g a s i n Theore m 6.2 . Tw o case s ar e possible . Case 1 . Ther e i s R suc h tha t

D"\Q°RcBh. Case 2 . Ther e i s a sequenc e {x m} suc h tha t x m G D" \ Bh an d \x rn\— » o c a s m— » oo. In Cas e 1 , we appl y Theore m 6.2 . In Cas e 2 , we hav e M (

"

m ) =

2^

+ V ( x m ) >

2^|

> 6 | a ; m |

'

where b is a constant . Thus, (29 ) hold s i n bot h cases . • T E S T QUESTION .

Wh y canno t li m b e replace d b y li m i n ou r proof ?

EXERCISE 6.2 . Sho w tha t fo r a harmoni c functio n define d o n th e plan e th e exact valu e o f c is TT/2 i n Theore m 6. 1 an d ir in Theore m 6.2 .

Unsolved problems . PROBLEM 1 . Le t Kh an d D hav e th e sam e meanin g a s i n Theore m 6.3 . I s i t true tha t fo r an y e > 0 there i s a smal l numbe r h > 0 such tha t

— Mir) lim —^ >

0

for a functio n u(x) tha t i s subharmoni c i n D, i s continuou s i n D, vanishe s o n dD, and i s positive i n D? 2 . Le t K' h b e th e con e YA=I X 1 < ^ ^ L x n > 0 . Le t D b e a n unbounded domai n i n K' h, le t L b e a uniforml y ellipti c operato r i n D, an d le t u(x) be subellipti c an d continuou s i n D , positiv e i n D an d vanis h o n dD. (a) I s i t tru e tha t T ^ M{r) lim = oc : PROBLEM

r—>oo T

(b) I s i t tru e tha t — M(r) lim — r ^ > 0 r—>oo T

e

for som e positiv e el The answe r i s unknow n eve n i n th e plan e case . PROBLEM 3 . Le t a functio n u(x,y) satisf y th e ellipti c equatio n (*) a(x,

y)u xx + 26(rr , y)uxy + c(x , y)uyy = 0

on the half-plane y > 0 and be continuous up to the x-axis. Suppos e that u(x,y) > 0 for y > 0 and u = 0 for y = 0 . Is i t tru e tha t ^(x , 2/) is a linea r function ?

Licensed to AMS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/publications/ebooks/terms

36

1. ELLIPTI C EQUATION S I N NONDIVERGENC E FOR M

FIGURE1

1

PROBLEM 4 . Equatio n (* ) ca n b e considere d o n th e entir e plane . Th e solutio n u(x,y) i s define d o n th e entir e plan e an d

M(r) lim