Preface This handbook is the sixth and last volume in the series devoted to stationary partial differentialequations. As theprecedingvolumes, itisacollectionofselfcontained, state of-the-artsurveyswrittenbywell-knownexpertsinthefield. The topics covered by this volume include in particular domain perturbations for boundary value problems, singular solutions of semi-linear elliptic problems, positive solutions to elliptic equations on unbounded domains, symmetry ofsolutions, stationary compressible Navier-Stokes equation, Lotka-Volterrasystems withcross-diffusion, fixed pointtheoryforelliptic boundaryvalueproblems. Ihope thatthese surveyswill be useful forbothbeginnersandexpertsandhelptoawidediffusionoftheserecentanddeepresults inmathematicalscience. I wouldlike to thank all the contributors for theirelegantarticles. I also thankLauren Schultz Yuhasz andMaraVos-SarmientoatElsevierfor the excellentediting workofthis volume. M.Chipot v List of Contributors Daners, D., SchoolofMathematics and Statistics, The University ofSydney, NSW2006, [email protected] (Ch. 1) Davila, J., Departamento de Ingenierfa Matenuitica and CMM, Universidad de Chile, Casilla 170Correo 3, Santiago, [email protected](Ch. 2) v., Kondratiev, Department ofMathematics andMechanics, Moscow State University, Moscow 119899, [email protected] (Ch.3) v., Liskevich, Department ofMathematics, University ofSwansea, Swansea SA2 8Pp, [email protected](Ch. 3) Sobol, Z., Department ofMathematics, University ofSwansea, Swansea SA2 8Pp, UK [email protected](Ch. 3) Pacella, F., Dipartimento diMatematica, Universita di Roma La Sapienza, P.leA. Moro 2, 00185, Roma, [email protected](Ch. 4) Ramaswamy, M., TIFR-CAM, Yelahanka New Town, Bangalore 560065, India [email protected] (Ch. 4) Plotnikov, P.I., Lavryentyev Institute of Hydrodynamics, Siberian Division of Russian Academy of Sciences, Lavryentyev pI: 15, Novosibirsk 630090, Russia [email protected] (Ch. 5) Sokolowski, J., Institut Elie CaI·tan, Laboratoire de Mathematiques, Universite Henri Poincare, Nancy 1, B.P. 239, 54506 Vandoeuvre les Nancy Cedex, France [email protected] (Ch.5) Yamada, Y., Department ofApplied Mathematics, Waseda University, 3-4-1 Ohkubo, Shinjuku-ku, Tokyo 169-8555, Japan [email protected] (Ch.6) Zon, H., Department of Mathematics, University ofAlabama at Birmingham, USA [email protected] (Ch.7). vii CHAPlER 1 Domain Perturbation for Linear and Semi-Linear Boundary Value Problems Daniel Daners SchoolofMathell/£lticsandStatistics. TheUniversityofSydney.NSW2006.Australia E-/Iutil:D.Daners@/I1£1ths.usyd.edu.au URL:www./I1£1ths.usyd.edu.au/u/daners/ Abstract This is a survey on elliptic boundary value problems on varying domains and tools neededforthat. Suchproblemsariseinnumericalanalysis,inshapeoptimisationprob lems and in the investigation ofthe solution structure ofnonlinearelliptic equations. Themethodsare also usefultoobtaincertainresultsfor equationsonnon-smoothdo mainsbyapproximationbysmoothdomains. Domain independentestimates and smoothing properties are an essential tool to deal with domain perturbation problems, especially for non-linear equations. Hence we discuss such estimates extensively, together with some abstract results on linear operators. Asecondmajorpartdealswithspecificdomainperturbationresultsforlinearequa tions with various boundaryconditions. We completelycharacterise convergence for Dirichlet boundary conditions and also give simple sufficient conditions. We then prove boundary homogenisation results for Robin boundary conditions on domains with fast oscillating boundaries, where the boundary condition changes in the limit. We finallymentionsome simpleresults onproblemswithNeumann boundarycondi tions. The final partis concernedaboutnon-linearproblems, using the Leray-Schauder degree to prove the existence of solutions on slightly perturbed domains. We also demonstrate how to use the approximation results to getsolutions to nonlinearequa tionsonunboundeddomains. Keywords: Elliptic boundaryvalue problem, Domainperturbation, Semilinearequa tions,Aprioriestimates,Boundaryhomogenisation HANDBOOKOFDIFFERENTIALEQUATIONS StationaryPartialDifferentialEquations.volume6 EditedbyM.Chipot ©2008ElsevierB.V.Allrightsreserved 2 D.Dallers Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 2. Ellipticboundaryvalueproblemsindivergenceform. . . 5 2.1. Weaksolutionstoellipticboundaryvalueproblems. 5 2.2. Abstractformulationofboundaryvalueproblems. 9 2.3. Formallyadjointproblems . 12 2.4. Globalaprioriestimatesforweaksolutions . . . . . 13 2.4.1. SobolevinequalitiesassociatedwithDirichletproblems . 16 2.4.2. Maz'ya'sinequalityandRobinproblems . 17 2.4.3. SobolevinequalitiesassociatedwithNeumannproblems. 18 2.5. Thepseudo-resolventassociatedwithboundaryvalueproblems 18 3. Semi-linearellipticproblems . 20 3.1. Abstractformulationofsemi-linearproblems 20 3.2. Boundednessofweaksolutions .. 22 4. Abstractresultsonlinearoperators . . . 23 4.1. Convergenceintheoperatornorm . 24 4.2. Aspectralmappingtheorem . . . . 25 4.3. Convergencepropertiesofresolventandspectrum. 26 5. PerturbationsforlinearDirichletproblems 29 5.1. Assumptionsandpreliminaryresults. 29 5.2. Themainconvergenceresult..... 31 5.3. Necessaryconditionsforconvergence 35 5.4. Sufficientconditionsforconvergence 37 5.5. Proofofthemainconvergenceresult. 39 6. VaryingdomainsandRobinboundaryconditions. 42 6.1. Summaryofresults. . . . . . . . . . . . . . 42 6.2. Preliminaryresults . . . . . . . . . . . . . . 45 6.3. Smallmodificationsoftheoriginalboundary 47 6.4. Boundaryhomogenisation: LimitisaDirichletproblem 50 6.5. Boundaryhomogenisation: LimitisaRobinproblem. 53 7. Neumannproblemsonvaryingdomains . 55 7.1. RemarksonNeumannproblems . 55 7.2. ConvergenceresultsforNeumannproblems. 56 8. Approximationbysmoothdataanddomains . . . 58 8.1. Approximationbyoperatorshavingsmoothcoefficients. 58 8.2. Approximationbysmoothdomainsfromtheinterior . . 60 8.3. ApproximationfromtheexteriorforLipschitzdomains. 60 9. Perturbationofsemi-linearproblems . 62 9.1. Basicconvergenceresultsforsemi-linearproblems .. 62 9.2. Existenceofnearbysolutionsforsemi-linearproblems 65 9.3. Applicationstoboundaryvalueproblems 68 9.4. Remarksonlargesolutions..... 70 9.5. Solutionsbydomainapproximation 72 9.6. Problemsonunboundeddomains 73 References. . . . . . . . . . . . . . . . . . 75 DOII/ninperturbationforlinearandsell/i-linearboundaryvalueproblems 3 1. Introduction Thepurposeofthis surveyis tolookatellipticboundaryvalueproblems Anu= f in Qn, Hnu=O oniJQn withallmajortypesofboundaryconditionsonasequenceofopensets Qn inJRN (N ~ 2). We then study conditions under which the solutions converge to a solution of a limit problem Au= f in Q, Hu = 0 on iJQ on some open set Q C JRN. In the simplest case A,An = - ~ is the negative Laplace operator, and Hn•H the Dirichlet, Robin or Neumann boundary operator, but we work with general non-selfadjoint elliptic operators in divergence form. We are interested in verysingularperturbation,notnecessarilyofatype suchthatachangeofvariablescan be appliedtoreducetheproblemontoafixeddomain. Forthetheoryofsmoothperturbations rathercomplementarytoours wereferto [84] andreferences therein. Themainfeatures ofthis expositionare the following: • We presentan Lp-theoryoflinearand semi-linearelliptic boundary value problems withdomainperturbationinview. • We establish domain perturbation results for linear elliptic problems with Dirichlet, RobinandNeumannboundaryconditions, applicable tosemi-linearproblems. • We showhow to use the linearperturbationtheory to deal with semi-linearproblems on bounded and unbounded domains. In particular we show how to get multiple solutions for simple equations, discuss the issue of precise multiplicity and the occurrenceoflargesolutions. • We provide abstract perturbation theorems useful also for perturbations other than domainperturbations. • We provide tools to prove results for linear and nonlinear equations on general domains bymeans ofsmoothingdomainsandoperators (see Section8). Ouraimis tobuildadomainperturbationtheorysuitableforapplications tosemi-linear problems,thatis,problemswheref = f (x. u(x))isafunctionofx aswellasthesolution u(x). Fornonlinearities with growth, polynomial orarbitrary, we needa good theory for the linearproblem in Lp for 1 < P < 00. Good in the contextofdomain perturbations meansthatinallestimatesthereiscontrolondomain dependenceoftheconstantsinvolved. WeestablishsuchatheoryinSection2.1,wherewealsointroducepreciseassumptionson the operators. Startingfrom adefinitionofweaksolutions weprove smoothing properties ofthe corresponding resolvent operators with control on domain dependence. The main results are Theorems 2.4.1 and 2.4.2. In particular we prove that the resolventoperators have smoothing properties independent ofthe domain for Dirichlet and Robin boundary conditions, butnot for Neumann boundary conditions. To be able to work in a common spaceweconsidertheresolventoperatorasamapactingon Lp(JRN), sothatitbecomesa pseudo-resolvent(see Section2.5). 4 D.Dallers The smoothing properties ofthe resolvent operators enable us to reformulate a semi linear boundary value problem as a fixed point problem in Lp(JRN). Which P E (1, 00) wechoosedependsonthegrowthofthenonlinearity. Theruleis, thefasterthegrowth, the larger the choice ofp. We also show that under suitable growth conditions, a solution in Lp is infact in Loo. Again, the focus is ongetting controloverthe domain dependenceof the constantsinvolved. Forapreciseformulation oftheseresults werefertoSection3. Let the resolvent operator corresponding to the linear problems be denoted by RIlCA) and RCA). The key to be able to pass from perturbations ofthe linearto perturbations of thenonlinearproblemis thefollowing propertyoftheresolvents: If!Il ~ ! weakly, thenRIlCA)!1l---+ ROc)!strongly. Henceforalltypes ofboundaryconditionsweprovesuchastatement. IfR(A) iscompact, it turns out that the above property is equivalent to convergence in the operator norm. An issue connected with that is also the convergence ofthe spectrum. We show that the abovepropertyimpliesthe convergenceofeveryfinite partofthe spectrumoftherelevant differential operators. We also show here that itis sufficientto prove convergence ofthe resolvent in Lp(JRN) for some p and some Ato have them for all. These abstractresults are collectedinSection4. ThemostcompleteconvergenceresultsareknownfortheDirichletproblem(Section5). Thelimitproblemis always aDirichletproblemonsomedomain. Weextensivelydiscuss convergence in the operator norm. In particular, we look at necessary and sufficient conditions for pointwise and uniform convergence of the resolvent operators. As a corollary to the characterisation we see that convergence is independent ofthe operator chosen. Wealsogivesimplesufficientconditionsforconvergenceinterms ofpropertiesof the setQIl n Qc. Themainsourcefor theseresults is [58]. ThesituationisrathermorecomplicatedforRobinboundaryconditions,where thetype of boundary condition can change in the limit problem. In Section 6 we present three differentcases. First, we lookatproblems with only a smallperturbation ofiJQ. We can cutholes and add thin pieces outside Q, connectedto Q onlyneara setofcapacity zero. Second, we look at approximating domains with very fast oscillating boundary. In that case the limit problem has Dirichlet boundary conditions. Third, we deal with domains withoscillatingboundary, suchthatthelimitprobleminvolvesRobinboundaryconditions withadifferentweightonthe boundary. The secondandthirdresults are really boundary homogenisationresults. Theseresults arealltakenfrom [51]. The Neumann problem is very badly behaved, and without quite severe restriction on the sequence ofdomains QIl we cannot expect the resolvent to converge in the operator norm. Inparticularthespectrumdoesnotconverge,asalreadynotedin [40,page420]. We only prove a simple convergence resultfitting into the general framework establishedfor the otherboundaryconditions. Afterdealing withlinearequations weconsidersemi-linearequations. Alotofthispart is inspiredbyDancer'spaper [45] andrelatedwork. Theapproachis quite differentsince wetreatlinearequationsfirst, andthenusetheirpropertiestodealwithnonlinearequations. Theideaistousedegreetheorytogetsolutionsonanearbydomain, givenasolutionofthe limitproblem. Thecoreoftheargumentisanabstracttopologicalargumentwhichmaybe usefulalsofor othertypes ofperturbations (see Section9.2). Wealso discuss the issue of DOII/ninperturbationforlinearandsell/i-linearboundaryvalueproblems 5 precisemultiplicityofsolutions andthe phenomenonoflarge solutions. Finally, we show thatthe theoryalsoapplies tounboundedlimitdomains. There are many other motivations to look at domain perturbation problems, so for instance variational inequalities (see [102]), numerical analysis (see [77,107,110,116 119]), potential and scattering theory (see [10,108,113,124]), control and optimisation (see [31,34,82,120]), r-convergence (see [24,42]) and solution structures of nonlinear elliptic equations (see [45,47,52,69]). We mention more references in the discussion on the specific boundary conditions. Someresults go backalong time, see for instance [19] or [40]. The techniques are even older for the Dirichletproblem for harmonic functions withthepioneering work [93]. Finally, therearemanyresults wedonotevenmention, sofor instancefor convergence inthe Loo-normwereferto [8,9,14,23,26]. Furthermore, similarresults can beprovedfor parabolic problems. The key for thatare domain-independentheatkernel estimates. See for instance [7,17,47,52,59,78] and references therein. The above is only a small rather arbitrary selectionofreferences. 2. Ellipticboundaryvalueproblemsindivergenceform The purpose of this section is to give a summary ofresults on elliptic boundary value problems in divergence form with emphasis on estimates with control over the domain dependence. 2.1. Weaksolutions to ellipticboundaryvalueproblems Weconsiderboundaryvalueproblemsoftheform Au = f in Q, (2.1.1) Bu = 0 on iJQ on an open subset ofJRN, not necessarily bounded or connected. Here A is an elliptic operatorindivergenceformandB aboundaryoperatortobe specifiedlaterinthis section. TheoperatorA is oftheform -div(AoV'u +au) +b· V'u +cou (2.1.2) with Ao E Loo(Q. JRN'-;N), a. b E Loo(Q. JRN) and Co E Loo(Q). Moreover, we assume that Ao(x)is positive definite, uniformly with respect to x E Q. More precisely, there exists aconstantao > 0suchthat aol~12 :s ~TAo(x)~ (2.1.3) forall~ E JRN andalmostallx E Q. Wecallao the ellipticityconstant. REMARK 2.1.1. We only defined the operator A on Q, but we can extend it to JRN by setting a = b = 0, Co = 0 and A(x) := aoI on QC. Then the extendedoperator A also 6 D.Dallers satisfies (2.1.3). In particularthe ellipticityproperty (2.1.3) holds. Hence withoutloss of generalitywecauassume thatA isdefinedonJRN. Wefurtherdefine the co-normalderivativeassociatedwith A on aQ by au + -- := (Ao(x)Vu a(x)u) .v, aVA where vis theoutwardpointing unitnormalto aQ. AssumingthataQis thedisjointunion offl, f2 and f3 wedefine the boundaryoperatorE by ulrl on f I (Dirichletb.c.), au Eu := on f2 (Neumann b.c.), (2.1.4) aVA au -- +bou on f3 (Robinb.c.) aVA with bo E Leo(f3) nonnegative. Ifallfunctions involvedare sufficiently smooth, then by theproductrule + + + -vdiv(AoVu au) = (AoVu au) .Vv - div(v(AoVu au)) andtherefore, ifQ admits the divergence theorem, then -lVdiV(Aovu+aU)dX = J{Q(AoVu+au).Vvdx- J{aQ(v(AoVu+au))·vdC5, where C5 is the surfacemeasure on aQ. Hence, ifu is sufficiently smoothand v E ClUJ) with v = 0on fl, then {vAudx= {(AoVu+au).Vv+(a.Vu+cou)vdx+ (bouvdC5. JQ JQ Jr3 The expression on the right-hand side defines a bilinear form. We denote by HI(Q) the usual Sobolev space ofsquare integrable functions having square integrable weak partial derivatives. Moreover, H6(Q) is the closureofthe setoftestfunctions C~(Q) in HI(Q). DEFINITION2.1.2. Foru, v E HI(Q) weset ao(zt,v):= l(Aovu+aU),vv+(b,VU+COU)VdX. Theexpression + { a(u, v) :=ao(zt, v) bouvdC5 J r 3 iscalledthe bilinearformassociatedwith (A,E). DOII/ninperturbationforlinearandsell/i-linearboundaryvalueproblems 7 Ifuisasufficientlysmoothsolutionof(2.1.1), then l a(u. v) = (I. v) := jvdx (2.1.5) forallv E CI(Q)with v = 0onfl. Notethat(2.1.5)doesnotjustmakesenseforclassical solutions of(2.1.1), but for u E HI(Q) as long as the boundary integral is defined. We therefore generalise the notionofsolutionandjustrequire that u is in asuitable subspace V ofHI(Q) and(2.1.5)forall v E V. ASSUMPTION 2.1.3. Werequire that V beaHilbertspace suchthat V is dense in L2(Q), that andthat [3 {u 2 E CI(Q): suppu c Q \ fl. bolul dO' < oo} C V. Iff3 isnonsmoothwereplacethesurfacemeasure0' bythe(N-1)-dimensionalHausdorff measure sothatthe boundaryintegralmakes sense. Wealsorequire that (2.1.6) isanequivalentnormon V. Wenextconsidersome specific specialcases. EXAMPLE 2.1.4. (a) For a homogeneous Dirichlet problem we assume that fl = aQ and let V := H6(Q). For the norm we can choose the usual HI-norm,but on bounded domains we couldjust use the equivalent norm II'YuI12. More generally, on domains Q lying between two hyperplanes of distance D, we can work with the equivalent norm II'Yu112 becauseofFriedrich's inequality (2.1.7) validforall u E H6(Q) (see [1l1, TheoremII.2.D]). (b) For a homogeneous Neumann problem we assume that f2 = aQ and let V := HI(Q) with the usualnorm. (c)Forahomogeneous Robinproblemwe assume thatf3 = aQ. In this expositionwe will always assume that Q is a Lipschitz domain when working with Robin boundary conditions and choose V := HI(Q). On a bounded domain we can work with the equivalentnorm (2.1.8) (see [1l1, TheoremIII.5.C] or [56]). Itis possibletoadmitarbitrarydomains as shownin [11,56]. Wefinally define whatwemeanbyaweaksolutionof(2.1.1).