7 297 314 The Laplaces and Steepest Descents Methods Revisited Chelo Ferreira Departamento de Matem atica Aplicada Universidad de Zaragoza 50013Zaragoza Spain cferreiunizares Jos eLL opez and Pedro Pagola Departamento de Matem atica e Inform atica ID: 76322
Download Pdf The PPT/PDF document "International Mathematical Forum no" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
TheLaplacesandSteepestDescentsMethods wRe(w)Im(w)Re(f(w))0 Steepest descent pathSteepest ascent path Figure1:Typicalplotoftherealpartof)overasimplesaddlepointCrossingthispointwecanndasteepestascentpath(dashedlines)andasteepestdescentpath.Intheremainingofthepaper,wewillcallasymptoticpointstoboth,thecriticalpointsof)andtheendpointsof(ifagivenpointisboth,acriticalpointandanendpoint,wecallitcriticalpoint).Ifthefunctionhasnotone,butseveralasymptoticpoints:,thenthedominantcontributionto)comesfromthemostrelevantasymptoticpoint.Thisisdenedastheasymptoticpointlocatedon(oronifthedeformationisnecessary)atwhich))ismaximal.Thentheasymptoticexpansion)maybederivedapplyingtheLaplacesmethod(when(whena,b)isreal)ortheSteepestDescentsmethod(when)iscomplex)atthatrelevantasymptoticpoint.Itmayhappensthatthisrelevantasymptoticpointisnotunique.Inthiscase,wemustapplyLaplacesmethodorSteepestDescentsmethodatalloftherelevantasymptoticpoints.Itisimpossibletoresumeinfewwordstheimportancethatthesemethodshavehadtoderiveasymptoticapproximationsofintegralsingeneral,andofspecialfunctionsinparticular.Inthisway,itisverydiculttogiveanexhaustivelistofthethousandsofpublicationsdevotedtoconstruct,generalizeorapplytospecicexamplesthesemethods.Considerthatmostoftheknownasymptoticexpansionsofspecialfunctionsderivedfromintegralshavebeenobtainedbyusingoneofthesemethods.Themaintechnicalcomplicationintheapplicationofthesemethodsliesintheidenticationofthesteepestdescentspaths(intheSteepestDescentsmethod)andinthechangeofvariable(inbothmethods).Anothersourceofcomplicationistheexistenceofmorethanonerelevantasymptoticpoint.Recently,someauthorshavesuggestedbymeansofspecicexamplesthatthesemethodsmaybesimplied.Theideaisjusttoexpand)andperhapsalsopartofin(1)attherelevantasymptoticpoint(s)of)without TheLaplacesandSteepestDescentsMethodsexceptat.Iffa,b)isrealandtimesdierentiableon[a,b]andanda,b],thenmaybeacriticalpointofmultiplicityeven:oranendpointif0if0ifthiscase=1).If)iscomplex(analyticbetweenand),thenasaddlepointofofmultiplicityiii)Thefunction)hasaTayorexpansionat )when)iscontinuouson)whenv)Theintegral(2)andtheintegralsvergeabsolutelyanduniformlywithrespectto.Thesameholdsfortheintegralsifthedeformationisrequired.)isrealandsmoothenoughon,and,Laplacesmethodsaysthatwemustperformin(2)thechangeofvariabledenedbyandthenapproximate))at=0).If)and)areanalyticbetween,theSteepestDescentsmethodsaysthatwemustdeformthecontourthesteepestdescentpath(whichrunsthrow)andthenapplyLaplacesmethodtotheresultingintegralover.Inanycase,bothmethodstellusthatthemaincontributionoftheintegrandin(2)tothevalueoftheintegralcomesfromtheneighbourhoodofthepoint.Weusethisideahereinamoresimpleway:weapproximate)atbutwithoutanychangeofvariableneitheradeformationofthecontour.Wejustintroducetheexpansion(3)in(2)andinterchangesumandintegral: TheLaplacesandSteepestDescentsMethods.Then,weneedtoapproximate)atbothofthemsimultaneouslyandwiththesameaccuracy.ThismeansthatweneedthesimultaneousTaylorexpansionof)at.Adetailedexpositionofthismaybefoundin[6].Weneedtoreplaceconditioniii)andiv)insection2by:iii)Thefunction)hasatwo-pointTaylorexpansionat,..., and,for k!(nk)!(1)n+1ng(nk)(w21)knk)(w1) Theremaindertermveries)when)when)iscontinuouson)when)when1andIntroducingtheexpansion(9)in(2)andinterchangingsumandintegralweobtainobtainakk(z)+bkk(z)]+dw,wdwdw. TheLaplacesandSteepestDescentsMethodsThemaincontributionoftheintegrandin(2)tothevalueoftheintegralcomesfromtheneighbourhoodofalloftherelevantasymptoticpoints.Then,weneedtoapproximate)atallofthemsimultaneously.Butasitwillbeclearbelow,thedegreeofaccuracyintheapproximationof)atmustdependonthemultiplicityofaszeroof).Thismeansthatweneedthemulti-pointTaylorexpansionof)atthepointsrepeatedrepeated.Adetailedexpositionofthismaybefoundin[7].Then,conditionsiii)andiv)insection2mustbereplacedby:iii)Thefunction)hasamulti-pointTaylorexpansionat)arepolynomialsofdegree)when.Coecientsmaybecalculatedtakingintoaccountthatthepolynomialintherighthandsideof(19)istheinterpolationpolynomialof)atwithdata,...,N.Orelse[7]: k,j,lk,j,l (+l)!d+l dw+l g(w) Ns=1,sj(wws) w=wj+Nr=1,rj1 (1)!d1 dw+l g(w) )iscontinuouson,...,NIntroducingtheexpansion(19)in(2)andinterchangingsumandintegralweobtain TheLaplacesandSteepestDescentsMethodswherethecontourencirclestheorigininthecounterclockwisedirectionandcontainsnopolesof2).Theshift12inthevariableisintroducedinordertohavereectionsymmetryinthepolynomials.Wewanttoapproximate)forlargeandwrite(28)intheform 2=n! dw,n)=1)=(sinh(2).Theuniquesaddlepointof)isandhasmultiplicity1.Applyingthemethoddescribedinsection2weobtaintheasymptoticexpansion(4)-(6)of2)forlarge=0).Forsimplicityweonlywriteexplicitlythersttwotermsoftheexpansion: 2=nnzn1 2sinh(11+4 2coth1 2zcoth1 2z1 8nz2+O1 ThisisanasymptoticexpansionforlargeforthosevaluesofforwhichtheSteepestDescentsmethodcanbeappliedtotheintegral(29).Thesteepestascentanddescentpathsof)are(seeg.2(a)):Arg((steepestascentpath)+Arg( sin(+Arg(Arg(Arg((steepestdescentpath)Therealpartof)in(29)isnegativeoverastraightpathjoiningtheendsofthesteepestdescentpathattheinnity(dashedlineing.2(a)).Therefore,thecontourmaybedeformedintothesteepestdescentpathwheneverthepoles)arenotinside.Thishappensfor5.2JacobipolynomialsConsiderthefollowingintegralrepresentationoftheJacobipolynomialsfor 21)n (1+ (1+ wherethefunctionx,w)isdenedbyx,wlog(1+)+log(1 TheLaplacesandSteepestDescentsMethods Re(w)Im(w)C z1 2 i Im(w) Re(w)Steepest ascent pathSteepest descent path1-x i1-x-1-x 21-x i2-C (a)(b)Figure2:(a)Steepestdescentpathofthefunctionforpositive.Thecondition2guaranteesthatthepolesofthefunctionsinh(2)lieoutsidethesteepestdescentpathandthecontourmaybedeformedinto.(b)Steepestdescentandascentpathsofthefunction(32).Thebranchpointsofthefunction)=(1(1+lieoutsidethesteepestdescentpathandthecontourmaybedeformedinto.6UniformityaspectsofthemethodIfthefunction)intheintegrand(2)containsarealparameter,saythesaddlepointsofw,a)maydependon.Then,itmayhappenthatwhencrossesacriticalvalue,anasymptoticpointof)changesitscharacterfromnon-relevanttorelevantorvice-versa,orthattworelevantasymptoticpointscoalesce,.Thatis,theasymptoticbehaviouroftheintegralmaychangedrasticallyatcertainvaluesof.Itiswellknownthat,inthiscase,thestandardLaplacesorSteepestDescentsmethodsmustbemodiedtogetasymptoticexpansionsof(2)uniformlyvalidforvaryingaryingchap.7].Thatmodicationsdependonthekindofcoalescenceofasymptoticpoints.Wearespeakingaboutthewellknownuniformasymptoticmethods.Forexample,iftwosaddlepointscoalesce,then(2)maybeapproximatedbyAiryfunctions;ifasaddlepointcoalescewithandendpoint,(2)mayapproximatedbyparaboliccylinderfunctions,Themethoddescribedinsections2-4doesnotneedanymodicationbe-causeitisalreadyauniformmethod.Considerthefollowingtwopossibilities(notmutuallyexcluding)forthecriticalpointsofw,a)whencrossesacertaincriticalvaluea)Twoormorerelevantasymptoticpoints(criticalorendpoints)coalesce.Inthiscasethemethoddescribedinsection4isuniform.Thisisbecausetheexpansionof)issimultaneousatallofthoseasymptoticpoints TheLaplacesandSteepestDescentsMethodsCoecients2,areregularwhen1)if0orif=0and[6].Forexample, =1,=1, =1, =1.Thefunctions),)and)in(23)are(1+ (1+ xb+1 (b x(b,(2)0(x0ext(1+t)bt2dt=ex xb+1 (b x22(b +(Therefore,forlarge,b,c,x xb+1(bc+a(0)1 (b x(b1+a(0)2 (b x2(b2(b x(b+O1 uniformlyforfora0,a1]with2if=0.7ConcludingremarksTheasymptoticexpansionoftheintegral(2)forlargeisdeterminedbythenumber,multiplicityandnatureoftherelevantasymptoticpointsoftheexponent).Then,onlythevalueof)aroundthosepointsisofimportanceandasimultaneousapproximationof)atthosepointsisinorder.Butthisapproximationmusthaveanappropriatecontactorderateachofthosepointsaccordingtotheirmultiplicity.Thisisachievedbythemulti-pointTaylorexpansionof)atthosepointscountingmultiplicitiesultiplicitiesThemethoddescribedinthispaperonlyinvokestheclassicalLaplacesorSteepestDescentsmethodstoshowtheasymptoticbehaviourofthetermsoftheexpansionandoftheremainder.Theasymptoticexpansionisderived TheLaplacesandSteepestDescentsMethodsInthestandardLaplacesorSaddlepointmethods,theasymptoticexpan-sionoftheintegralhasaformsimilarto(4).Thesequence)inthosemethodsisusuallyverysimple:itisproportionalto.Butthecoecientsareusuallyverycomplicated:theyaretheTaylorcoecientsofacom-plicatedfunctionobtainedafterachangeofvariable(includingajacobian).Ontheotherhand,thesequence)inthismethodismorecomplicated(formula(5))butthecoecientsaresimpler:theyaretheTaylorcoe-cientsofthefunction)at(formula(3)).Wecansaythat,withrespecttothestandardmethods,inthismethodthereisamovementofdif-cultyfromthecoecientstothesequence).Thecomplexityofthesequence)willdependofcourseonthepreciseexampleanalyzed(inalltheexamplesconsidered[3],[4],[5]andtheexample6.1thesequenceturnsouttobeverysimple).Iftheintegrationpathinthedenitionof)in(2)containrelevantandnon-relevantasymptoticpointsanduniformityisnotaproblem,wedonotneedtoconsideralloftheasymptoticpointssimultaneouslyinthemethod.Wejustneedtodividetheintegrationpathinseveralpiecescontaininglessasymptoticpoints.Ofcourse,thedominantasymptoticbehaviourcomesfromthepieceofcontainingthedominantasymptoticpoints.Thecontributionoftheremainingpieceswillbeexponentiallysubdominant.Ithasbeenpointedoutinsection6thatthismethodisintrinsicallyuni-form.ClassicaluniformmethodsbasedonstandardLaplacesorSteepestDescentsmethodsarestrongmodicationsofthecorrespondingnon-uniformmethods.Moreover,thereareseveralkindofmodications(andhenceofuniformmethods)dependingonthekindofcoalescenceofasymptoticpoints(saddlepointnearanendpoint,twocoalescingsaddlepoints,).ThisissobecauseeveryuniformversionofthestandardLaplacesorSteepestDescentsmethodsrequiresanspecicchangeofvariable.Butinthemethoddescribedabovewedonotperformanychangeofvariable,justexpand)attheasymptoticpoints,whatevertheirdependenceoftheparameterdescribedinsection6is.Then,themethodisintrinsicallyuniform,andmoreover,itisuniqueindependentlyofthekindofcoalescenceofasymptoticpoints.Anothersourceofasymptoticpoints(relevantforanasymptoticanalysis)arethe(possible)singularitiesof).Webelievethatinthiscasewejustmayintroducethisnewkindofasymptoticpointsinthelistofasymptoticpointsandapplytheaboveideas.Wethinkthatinthiscasethemulti-pointTaylorexpansionof)mustbereplacedbyamulti-pointLaurentorTaylor-Laurentexpansion[7].Thisissubjectoffurtherinvestigation.