Progress in Nonlinear Differential Equations and Their Applications Volume71 Editor HaimBrezis Universite´ PierreetMarieCurie Paris and RutgersUniversity NewBrunswick,N.J. EditorialBoard AntonioAmbrosetti,ScuolaInternationaleSuperiorediStudiAvanzati,Trieste A.Bahri,RutgersUniversity,NewBrunswick FelixBrowder,RutgersUniversity,NewBrunswick LuisCaffarelli,TheUniversityofTexas,Austin LawrenceC.Evans,UniversityofCalifornia,Berkeley MarianoGiaquinta,UniversityofPisa DavidKinderlehrer,Carnegie-MellonUniversity,Pittsburgh SergiuKlainerman,PrincetonUniversity RobertKohn,NewYorkUniversity P.L.Lions,UniversityofParisIX JeanMawhin,Universite´ CatholiquedeLouvain LouisNirenberg,NewYorkUniversity LambertusPeletier,UniversityofLeiden PaulRabinowitz,UniversityofWisconsin,Madison JohnToland,UniversityofBath Satyanad Kichenassamy Fuchsian Reduction Applications to Geometry, Cosmology, and Mathematical Physics Birkha¨user Boston • Basel • Berlin SatyanadKichenassamy Universite´deReimsChampagne-Ardenne Moulin de la Housse, B.P. 1039 F-51687 Reims Cedex 2 France [email protected] MathematicsSubjectClassification(2000):35-02,35A20, 35B65,35J25,35J70,35L45, 35L70,35L80, 35Q05, 35Q51, 35Q75, 53A30, 80A25, 78A60, 83C75, 83F05, 85A15 LibraryofCongressControlNumber:2007932088 ISBN-13:978-0-8176-4352-2 e-ISBN-13:978-0-8176-4637-0 Printedonacid-freepaper. (cid:1)c2007Birkha¨userBoston Allrightsreserved.Thisworkmaynotbetranslatedorcopiedinwholeorinpartwithoutthewrit- tenpermissionofthepublisher(Birkha¨userBoston,c/oSpringerScience+BusinessMediaLLC,233 SpringStreet,NewYork,NY10013,USA),exceptforbriefexcerptsinconnectionwithreviewsor scholarlyanalysis.Useinconnectionwithanyformofinformationstorageandretrieval,electronic adaptation,computersoftware,orbysimilarordissimilarmethodologynowknownorhereafterde- velopedisforbidden. Theuseinthispublicationoftradenames,trademarks,servicemarksandsimilarterms,evenifthey arenotidentifiedassuch,isnottobetakenasanexpressionofopinionastowhetherornottheyare subjecttoproprietaryrights. 9 8 7 6 5 4 3 2 1 www.birkhauser.com (INT/MP) To my parents Preface The nineteenth century saw the systematic study of new “special functions”, suchasthehypergeometric,Legendreandellipticfunctions,thatwererelevant in number theory and geometry, and at the same time useful in applications. Tounderstandthepropertiesofthesefunctions,itbecameimportanttostudy their behavior near their singularities in the complex plane. For linear equa- tions, two cases were distinguished: the Fuchsian case, in which all formal solutions converge,and the non-Fuchsian case. Linear systems of the form du z +A(z)u =0, dz with A holomorphic around the origin, form the prototype of the Fuchsian class. The study of expansions for this class of equations forms the familiar “Fuchs–Frobenius theory,” developed at the end of the nineteenth century by Weierstrass’s school. The classification of singularity types of solutions of nonlinearequationswasincomplete,andthe Painlev´e–Gambierclassification, forsecond-orderscalarequationsofspecialform,leftnohopeoffindinggeneral abstract results. The twentieth century saw, under the pressure of specific problems, the developmentofcorrespondingresultsforpartialdifferentialequations(PDEs): The Euler–Poisson–Darbouxequation λ u + u −Δu=0 tt t t and its elliptic counterpart arise in axisymmetric potential theory and in the methodofsphericalmeans;italsocomesupinspecialreductionsofEinstein’s equations.Inparticular,onerealizedthatequationswithdifferentvaluesofλ couldbe relatedtoeachotherby transformationsu(cid:2)→tmu.Ellipticproblems in corner domains and problems with double characteristics also led to fur- ther generalizations.Thisdevelopmentwasconsideredas fairlymatureinthe 1980s; it was realized that some problems required complicated expansions withlogarithmsandvariablepowers,beyondthescopeofexistingresults,but viii Preface it was assumed that this behavior was nongeneric. Nonlinear problems were practically ignored. The word “Fuchsian” had come to stand for “equations for which all for- mal power series solutions are convergent.” Of course, Fuchsian ODEs have solutions involving logarithms, but by Frobenius’s trick, logarithms could be viewed as limiting cases of powers, and were therefore not thought of as generic. However,in the 1980s difficulties arosewhen it became necessaryto solve Fuchsian problems arising from other parts of mathematics, or other fields. The convergence of the “ambient metric” realizing the embedding of a Rie- mannianmanifoldinaLorentzspacewithahomothetycouldnotbeprovedin evendimensions.When,inthewakeoftheHawking–Penrosesingularitytheo- rems,itbecamenecessarytolookforsingularsolutionsofEinstein’sequations, existing results covered only very special cases, although the field equations appeared similar to the Euler–Poisson–Darbouxequation. Numerical studies of suchspace-times led to spiky behavior:were these spikes artefacts?indica- tions of chaotic behavior? OtherproblemsseemedunrelatedtoFuchsianPDEs.Fortheblowupprob- lemfornonlinearwaveequations,againintheeighties,Ho¨rmander,John,and their coworkerscomputedasymptoticestimatesofthe blowuptime—whichis notaLorentzinvariant.Forelliptic problemsΔu=f(u)withmonotonenon- linearities,solutionswith infinite datadominate allsolutions,andcomeupin several contexts; the boundary behavior of such solutions in bounded C2+α domains is nota consequence ofweightedSchauder estimates.Outside math- ematics, we may mention laser collapse and the weak detonation problem. In astrophysics, stellar models raise similar difficulties; equations are singu- lar at the center, and one would like to have an expansion of solutions near the singularityto startnumericalintegration.Also, the theory ofsolitons has provided,from1982on,aplethoraofformalseriessolutionsforcompletelyin- tegrablePDEs,ofwhichonewouldliketoknowwhethertheyrepresentactual solutions. Do these series have any relevance to nearly integrable problems? The method of Fuchsian reduction, or reduction for short, has provided answers to the above questions. The upshot of reduction is a representation of the solution u of a nonlinear PDE in the typical form u =s+Tmv, wheresisknowninclosedform,issingularforT =0,andmayinvolveafinite number of arbitrary functions. The function v determines the regular part of u.This representationhas the sameadvantagesas anexactsolution,because one can prove that the remainder Tmv is indeed negligible for T small. In particular,it is availablewherenumericalcomputationfails;itenables one to compute whichquantitiesbecome infinite andatwhatrate,andto determine which combinations of the solution and its derivatives remain finite at the singularity. From it, one can also decide the stability of the singularity under Preface ix perturbations, and in particular how the singularity locus may be prescribed or modified. Reduction consists in transforming a PDE F[u] = 0, by changes of vari- ablesandunknowns,into anasymptoticallyscale-invariantPDEorsystemof PDEs Lv =f[v] such that (i) one can introduce appropriate variables (T,x ,...) such that 1 T = 0 is the singularity locus; (ii) L is scale-invariant in the T-direction; (iii) f is “small” as T tends to zero; (iv) bounded solutions v of the reduced equation determine singular u that are singular for T = 0. The right-hand side may involvederivatives of v. After reduction to a first-ordersystem, one is usually led to an equation of the general form (cid:2) (cid:3) d T +A w =f[T,w], dT wheretheright-handsidevanishesforT =0.PDEsofthisformwillbecalled “Fuchsian.” The Fuchsian class is itself invariant under reduction under very general hypotheses on f and A. This justifies the name of the method. Since v is typically obtained from u by subtracting its singularities and dividingbyapowerofT,v willbecalledtherenormalizedunknown.Typically, the reduced Fuchsianequations havenonsmooth coefficients,and logarithmic termsinparticulararetheruleratherthantheexception.Sincethecoefficients and nonlinearities are not required to be analytic, it will even be possible to reduce certain equations with irregular singularities to Fuchsian form. Even thoughLisscale-invariant,smaynothavepower-likebehavior.Also,inmany cases, it is possible to give a geometric interpretation of the terms that make up s. The introduction, Chapter 1, outlines the main steps of the method in algorithmic form. PartI describes a systematic strategy for achieving reduction. A few gen- eralprinciplesthatgovernthe searchforareducedformaregiven.The listof examples of equations amenable to reduction presented in this volume is not meantto be exhaustive.In fact,everynew applicationofreductionso farhas led to a new class of PDEs to which these ideas apply. Part II develops variants of several existence results for hyperbolic and elliptic problems in order to solve the reduced Fuchsian problem, since the transformedproblemisgenerallynotamenabletoclassicalresultsonsingular PDEs. Part III presents applications. It should be accessible after an upper- undergraduate course in analysis, and to nonmathematicians, provided they take for granted the proofs and the theorems from the other parts. Indeed, thediscussionofideasandapplicationshasbeenclearlyseparatedfromstate- ments of theorems and proofs, to enable the volume to be read at various levels. x Preface Part IV collects general-purpose results, on Schauder theory and the dis- tancefunction(Chapter12),andontheNash–Moserinversefunctiontheorem (Chapter 13).Togetherwith the computations workedout in the solutions to the problems, the volume is meant to be self-contained. Most chapters contain a problem section. The solutions worked out at the end of the volume may be taken as further prototypes of application of reduction techniques. A number of forerunners of reduction may be mentioned. 1. TheBriot–BouquetanalysisofsingularitiesofsolutionsofnonlinearODEs offirstorder,continuedbyPainlev´eandhisschoolforequationsofhigher order. It has remained a part of complex analysis. In fact, the catalogue of possible singularities in this limited framework is still not complete in many respects. Most of the equations arising in applications are not covered by this analysis. 2. The regularization of collisions in the N-body problem. This line of thought has gradually waned, perhaps because of the smallness of the radius of convergence of the series in some cases, and again because the relevance to nonanalytic problems was not pursued systematically. 3. AnumberofspecialcasesforsimpleODEshavebeenrediscoveredseveral times; a familiarexample is the constructionofradialsolutions ofnonlin- ear elliptic equations, which leads to Fuchsian ODEs with singularity at r =0. In retrospect, reduction techniques are the natural outgrowth of what is tra- ditionally calledthe “Weierstrassviewpoint” in complex analysis,as opposed totheCauchyandRiemannviewpoints.Thisviewpoint,fromthepresentper- spective,putsexpansionsatthemainfocusofinterest;allrelevantinformation isderivedfromthem.Forthisapproachtoberelevantbeyondcomplexanaly- sis,itwasnecessarytounderstandwhichaspectsoftheWeierstrassviewpoint admit a generalizationto nonanalytic problems with nonlinearities—andthis generalization required a mature theory of nonlinear PDEs which was devel- opedrelativelyrecently.Thedevelopmentofreductiontechniquesintheearly nineties seems to have been stimulated by the convergence of five factors: 1. The emergence of singularities as a legitimate field of study, as opposed to a pathology that merely indicates the failure of global existence or regularity. 2. The existence of a mature theory of elliptic and hyperbolic PDEs, which could be generalized to singular problems. 3. The failure of the searchfor a weak functional setting that would include blowup singularities for the simplest nonlinear wave equations. 4. Therediscoveryofcomplexanalysisstimulatedbytheemergenceofsoliton theory. 5. The availability of a beginning of a theory of Fuchsian PDEs, as opposed to ODEs, albeit developed for very different reasons, as we saw. Preface xi On a more personalnote, a number of mathematicians have,directly or indi- rectly,helpedtheauthorintheemergenceofreductiontechniques:D. Aronson, C. Bardos, L. Boutet de Monvel, P. Garrett, P. D. Lax, W. Littman, L. Nirenberg, P. J. Olver, W. Strauss, D. H. Sattinger, A. Tannenbaum, E. Zeidler. In fact, my indebtedness extends to many other mathematicians whomI havemetorread,including theanonymousreferees.H.Brezis,whose mathematical influence may be felt in severalof my works,deserves a special place. I am also grateful to him for welcoming this volume in this series, and to A. Kostant and A. Paranjpye at Birkha¨user, for their kind help with this project. Paris Satyanad Kichenassamy February 27, 2007