ebook img

Simulation of stochastic Volterra equations driven by space--time L\'evy noise PDF

0.31 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Simulation of stochastic Volterra equations driven by space--time L\'evy noise

Simulation of Stochastic Volterra Equations Driven by Space–Time Le´vy Noise BohanChen,CarstenChong,andClaudiaKlu¨ppelberg 5 1 0 2 n a J 7 AbstractInthispaperweinvestigatetwonumericalschemesforthesimulationof stochastic Volterra equationsdrivenby space–timeLe´vy noise of pure-jumptype. ] R The first one is based on truncating the small jumps of the noise, while the sec- P ond one relies on series representation techniques for infinitely divisible random . variables. Under reasonable assumptions, we prove for both methods Lp- and al- h t most sure convergence of the approximations to the true solution of the Volterra a equation.WegiveexplicitconvergenceratesintermsoftheVolterrakernelandthe m characteristicsofthe noise.A simulationstudyvisualizesthemostimportantpath [ propertiesoftheinvestigatedprocesses. 1 v Key words: Simulation of SPDEs, Simulation of stochastic Volterra equations, 5 Space–time Le´vy noise, Stochastic heat equation, Stochastic partial differential 4 equation 6 1 0 Mathematics Subject Classifications (2010): 60H35, 65C30, 60H20, 60H15, . 60G57,60G51 1 0 5 1 : v i X r a BohanChen·CarstenChong·ClaudiaKlu¨ppelberg TechnischeUniversita¨tMu¨nchen Boltzmannstraße3,85748Garching,Germany BohanChen e-mail:[email protected] CarstenChong e-mail:[email protected] ClaudiaKlu¨ppelberg e-mail:[email protected] 1 2 BohanChen,CarstenChong,andClaudiaKlu¨ppelberg 1 Introduction Theaimofthispaperistoinvestigatedifferentsimulationtechniquesforstochastic Volterraequations(SVEs)oftheform t Y(t,x)=Y (t,x)+ G(t,x;s,y)s (Y(s,y))L (ds,dy), (t,x)∈R ×Rd, (1) 0 + 0 Rd Z Z whereGisadeterministickernelfunction,s aLipschitzcoefficientandL aLe´vy basisonR ×Rd ofpure-jumptypewithnoGaussianpart.Inthepurelytemporal + casewherenospaceisinvolvedandthekernelGissufficientlyregularonthediag- onal{(t;s)∈R ×R : t=s},theexistenceanduniquenessofthesolutionY to(1) + + areestablishedforgeneralsemimartingaleintegratorsin[14].Thespace–timecase (1) is treated in [5] for quite generalLe´vy bases. In particular,G is allowed to be singularonthediagonal,whichtypicallyhappensinthecontextofstochasticpartial differentialequations (SPDEs) where G is the Green’s function of the underlying differentialoperator.MoredetailsontheconnectionbetweenSPDEsandtheSVE (1)arepresentedinSect.2,orcanbefoundin[2,5,20]. SinceinmostcasesthereexistsnoexplicitsolutionformulafortheSVE(1), it is a natural task to develop appropriate simulation algorithms. For SPDEs driven byGaussiannoise,researchonthistopicisratherfaradvanced,seee.g.[7,9,21]. However,forSPDEsdrivenbyjumpnoisessuchasnon-GaussianLe´vybases,the related literature is considerably smaller, see [3] and the work of Hausenblas and coauthors[8, 10, 11]. The case s ≡1 hasbeentreated in [4]. The contributionof ourpapercanbesummarizedasfollows: • We proposeandanalyzetwo approximationschemesfor(1), eachofwhichre- placestheoriginalnoisebya truncatednoisethatonlyhasfinitelymanyatoms oncompactsubsetsofR ×Rd.Forthefirstscheme,wesimplycutoffalljumps + whosesizeissmallerthanaconstant.Forthesecondscheme,weuseseriesrep- resentationtechniquesforthenoiseasin[17]suchthatthejumpstobedropped offare chosenrandomly.Both methodshave alreadybeenappliedsuccessfully tothesimulationofLe´vyprocesses,cf.[1,18]. • Inthe casewhereG originatesfromanSPDE,the crucialdifferenceofournu- merical schemes to the Euler or finite element methods in the referencesmen- tionedbeforeisthatwedonotsimulatesmallspace–timeincrementsofthenoise butsuccessivelythetruejumpsoftheLe´vybasis,whichisaneasiertaskgiven thatoneusuallyonlyknowstheunderlyingLe´vymeasure.Itisimportanttorec- ognizethat this is only possible because the noiseL is of pure-jumptype, and containsneitheraGaussianpartnoradrift.WeshallpointoutinSect.6howto relaxthisassumption. Theremainingarticleisorganizedasfollows:Section2givesthenecessaryback- ground for the SVE (1). In particular, we present sufficient conditions for the ex- istence and uniqueness of solutions, and address the connection between (1) and SPDEs.InSect.3weconstructapproximationstothesolutionY of(1)bytruncat- ing the small jumpsof the Le´vy basis. We provein Thm. 1 their Lp-convergence, SimulationofStochasticVolterraEquationsDrivenbySpace–TimeLe´vyNoise 3 andinsomecasesalsotheiralmostsure(a.s.)convergencetothetargetprocessY. InSect.4weapproximatethedrivingLe´vybasisusingseriesrepresentationmeth- ods. This leads to an algorithmthat producesapproximationsagain convergingin theLp-sense,sometimesalsoalmostsurely,toY,seeThm.2.Inboththeorems,we findexplicitLp-convergenceratesthatonlydependonthekernelGandthecharac- teristicsofL .Section5presentsasimulationstudyforthestochasticheatequation which highlights the typical path behaviour of stochastic Volterra equations. The final Sect. 6 compares the two simulation algorithms developed in this paper and discussessomefurtherdirectionsofthetopic. 2 Preliminaries Westartwithasummaryofnotationsthatwillbeemployedinthispaper. R theset[0,¥ )ofpositiverealnumbers + N thenaturalnumbers{1,2,...} B a stochastic basis (W ,F,F = (Ft)t∈R+,P) satisfying the usual hy- pothesesofcompletenessandright-continuity W¯,W˜ W¯ :=W ×R andW˜ :=W ×R ×Rd whered∈N + + B(Rd) theBorels -fieldonRd B˜ thecollectionofallboundedBorelsetsofR ×Rd b + P thepredictables -fieldonBorthecollectionofallpredictableprocesses W¯ →R P˜ theproductP⊗B(Rd)or thecollectionallP⊗B(Rd)-measurable processesW˜ →R P˜ the collection of sets in P˜ which are a subset of W ×[0,k]×[−k,k]d b forsomek∈N p∗ p∨1 Lp thespaceLp(W ,F,P), p∈(0,¥ ],endowedwiththetopologyinduced bykXkLp :=E[|X|p]1/p∗ L0 thespaceL0(W ,F,P)ofallrandomvariablesonBendowedwiththe topologyofconvergenceinprobability Blpoc the set of all Y ∈P˜ for which kY(t,x)kLp is uniformly bounded on [0,T]×Rd forallT ∈R (p∈(0,¥ ]) + Ac the complement of A within the superset it belongs to (which will be clearfromthecontext) A−B {x−y: x∈A,y∈B} −A {−x: x∈A} Leb theLebesguemeasureonRd (dshouldbeclearfromthecontext) k·k theEuclideannormonRd. C,C(T) twogenericconstantsinR ,onedependentandoneindependentofT, + whose values we do not care of and may therefore change from one placetotheother 4 BohanChen,CarstenChong,andClaudiaKlu¨ppelberg WesupposethatthestochasticbasisBsupportsaLe´vybasis,thatis,amapping L : P˜ →L0 withthefollowingproperties: b • L (0/)=0a.s. • Forallpairwisedisjointsets(Ai)i∈N⊂P˜bwith ¥i=1Ai∈P˜b wehave ¥ ¥ S L A =(cid:229) L (A) inL0. (2) i i i=1 ! i=1 [ • (L (W ×Bi))i∈N is a sequence of independent random variables if (Bi)i∈N are pairwisedisjointsetsinB˜ . b • ForeveryB∈B ,L (W ×B)hasaninfinitelydivisibledistribution. b • L (A)isF-measurablewhenA∈P˜ andA⊆W ×[0,t]×Rd fort∈R . t b + • Foreveryt∈R ,A∈P˜ andW ∈F wehavea.s. + b 0 t L (A∩(W 0×(t,¥ )×Rd))=1W 0L (A∩(W ×(t,¥ )×Rd)). Just as Le´vy processes are semimartingales and thus allow for an Itoˆ integra- tiontheory,Le´vybasesbelongtotheclassofL0-valueds -finiterandommeasures. Therefore,itispossibletodefinethestochasticintegral H(s,y)L (ds,dy) ZR+×Rd forH∈P˜ thatareintegrablewithrespecttoL ,see[6]forthedetails. Similarlyto Le´vyprocesses, thereexisttwo notionsof characteristicsforLe´vy bases:onegoingbackto[15,Prop.2.1]thatisbasedontheLe´vy-Khintchinefor- mulaandisindependentofF,andafiltration-basedonethatisusefulforstochastic analysis[6,Thm.3.2].Forthewholepaper,wewillassumethatbothnotionscoin- cidesuchthatL hasacanonicaldecompositionunderthefiltrationFoftheform L (dt,dx)=B(dt,dx)+L c(dt,dx)+ z1 (m −n )(dt,dx,dz) {|z|≤1} R Z + z1 m (dt,dx,dz), {|z|>1} R Z whereBisas -finitesignedBorelmeasureonR ×Rd,L c aLe´vybasissuchthat + L (W ×B) is normallydistributed with mean 0 and varianceC(B) for all B∈B˜ , b andm aPoissonmeasureonR ×Rd relativetoFwithintensitymeasuren (cf.[12, + Def.II.1.20]).Thereexistsalsoas -finiteBorelmeasurel onR ×Rd suchthat + B(dt,dx)=b(t,x)l (dt,dx), C(dt,dx)=c(t,x)l (dt,dx) and n (dt,dx,dz)=p (t,x,dz)l (dt,dx) (3) withtwofunctionsb: R ×Rd →Randc: R ×Rd →R aswellasatransition + + + kernelp from(R ×Rd,B(R ×Rd))to (R,B(R)) suchthatp (t,x,·)isaLe´vy + + measureforeach(t,x)∈R ×Rd. + SimulationofStochasticVolterraEquationsDrivenbySpace–TimeLe´vyNoise 5 Wehavealreadymentionedintheintroductionthatwewillassume C=0 (4) throughout the paper. For simplicity we will also make two further assumptions: first,thatthereexistb∈RandaLe´vymeasurep suchthatforall(t,x)∈R ×Rd + wehave b(t,x)=b, p (t,x,·)=p and l (dt,dx)=d(t,x); (5) second,that L ∈S ∪V , (6) 0 whereS isthecollectionofallsymmetricLe´vybasesandV istheclassofLe´vy 0 baseswithlocallyfinitevariationandnodrift,definedbythepropertythat |z|1 p (dz)<¥ , and b :=b− z1 p (dz)=0. {|z|≤1} 0 {|z|≤1} R R Z Z Furthermore,ifp hasafinitefirstmoment,thatis, |z|1 p (dz)<¥ , (7) {|z|>1} R Z wedefine B (dt,dx):=b d(t,x), b :=b+ z1 p (dz), 1 1 1 {|z|>1} R Z M(dt,dx):=L (dt,dx)−B (dt,dx)= z(m −n )(dt,dx,dz). 1 R Z Next,letussummarizethemostimportantfactsregardingtheSVE (1).Allde- tailsthatarenotexplainedcanbefoundin[5].First,manySPDEsofevolutiontype drivenbyLe´vynoisecanbewrittenintermsof(1),whereGistheGreen’sfunction of the corresponding differential operator. Most prominently, taking G being the heatkernelinRd,(1)istheso-calledmildformulationofthestochasticheatequa- tion (with constant coefficients and multiplicative noise). Typically for parabolic equations, the heat kernel is very smooth in general but explodeson the diagonal t=sandx=y.Infact,itisonlyp-foldintegrableon[0,T]×Rdforp<1+2/d.In particular,assoonasd≥2,itisnotsquare-integrable,andasaconsequence,noso- lutiontothestochasticheatequationintheform(1)willexistforLe´vynoiseswith non-zeroGaussiancomponent.Thisisanotherreasonforincludingassumption(4) inthispaper. Second, let us address the existence and uniqueness problem for (1). By a so- lution to this equation we mean a predictable process Y ∈ P˜ such that for all (t,x)∈R ×Rd, the stochastic integral on the right-hand side of (1) is well de- + finedandtheequationitselfforeach(t,x)∈[0,T]×Rd holdsa.s.Weidentifytwo solutionsassoonastheyaremodificationsofeachother.Givenanumberp∈(0,2], thefollowingconditionsguaranteeauniquesolutionto(1)inBp by[5,Thm.3.1]: loc 6 BohanChen,CarstenChong,andClaudiaKlu¨ppelberg A1. Y ∈Bp isindependentofL . 0 loc A2. s : R→RisLipschitzcontinuous,thatis,thereexistsC∈R suchthat + |s (x)−s (y)|≤C|x−y|, x,y∈R. (8) A3. G: (R ×Rd)2→RisameasurablefunctionwithG(t,·;s,·)≡0fors>t. + A4. L satisfies(3)–(6)and |z|pp (dz)<¥ . (9) R Z A5. Ifwedefinefor(t,x),(s,y)∈R ×Rd + G˜(t,x;s,y):=|G(t,x;s,y)|1{p>1,L ∈/S}+|G(t,x;s,y)|p, (10) thenwehaveforallT ∈R + T sup G˜(t,x;s,y)d(s,y)<¥ . (11) (t,x)∈[0,T]×RdZ0 ZRd A6. Foralle >0andT ∈R thereexistk∈Nandapartition0=t <...<t =T + 0 k suchthat sup sup ti G˜(t,x;s,y)d(s,y)<e . (12) (t,x)∈[0,T]×Rdi=1,...,kZti−1ZRd ApartfromA1–A6,wewilladdanotherassumptionsinthispaper: A7. Thereexistsasequence(UN)N∈NofcompactsetsincreasingtoRd suchthatfor allT ∈R andcompactsetsK⊆Rd wehave,asN→¥ , + t r1N(T,K):= sup |G(t,x;s,y)|1{p>1,L ∈/S}d(s,y) (t,x)∈[0,T]×K Z0 Z(UN)c t 1/p∗ + |G(t,x;s,y)|pd(s,y) →0. (13) (cid:18)Z0 Z(UN)c (cid:19) ! ConditionsA6andA7areautomaticallysatisfiedif|G(t,x;s,y)|≤g(t−s,x−y) forsomemeasurablefunctiongandA5holdswithGreplacedbyg.ForA6see[5, Rem.3.3(3)];forA7chooseUN :={x∈Rd: kxk≤N}suchthatfor p>1 t sup |G(t,x;s,y)|pd(s,y) (t,x)∈[0,T]×KZ0 Z(UN)c t ≤ sup gp(t−s,x−y)d(s,y) (t,x)∈[0,T]×KZ0 Z(UN)c T T = sup gp(s,y)d(s,y)≤ gp(s,y)d(s,y) x∈KZ0 Zx−(UN)c Z0 ZK−(UN)c →0 as N→¥ SimulationofStochasticVolterraEquationsDrivenbySpace–TimeLe´vyNoise 7 by the fact thatK−(UN)c ↓0. A similar calculationappliesto the case p∈(0,1] andthefirstterminrN(T,K). 1 Example1.WeconcludethissectionwiththestochasticheatequationinRd,whose mildformulationisgivenbytheSVE(1)with exp(−kxk2/(4t)) G(t,x;s,y)=g(t−s,x−y), g(t,x)= 1 (s) (14) (4p t)d/2 [0,t) for(t,x),(s,y)∈R ×Rd.WeassumethatY ands satisfyconditionsA1andA2, + 0 respectively.Furthermore,wesupposethat(3)–(6)arevalid,andthat(9)holdswith some p∈(0,1+2/d). Itis straightforwardto show thatthen A3–A6are satisfied withthesamep.LetusestimatetheraterN(T,K)forT ∈R ,K:={kxk≤R}with 1 + R∈N, andUN :={kxk≤N}. We first consider the case p≤1 orL ∈S. Since K−(UN)c=(UN−R)cforN≥R,thecalculationsafterA7yield(G (·,·)denotesthe upperincompletegammafunctionand p(d):=1+(1−p)d/2) (rN(T,K))p∗ ≤ T gp(t,x)d(t,x)= T ¥ exp(−pr2/(4t))rd−1drdt 1 Z0 Z(UN−R)c Z0 ZN−R (4p t)pd/2 T d p(N−R)2 =C tp(d)−1G , dt 0 2 4t Z (cid:18) (cid:19) d p(N−R)2 p(N−R)2 p(d) =C(T) (p(d))−1G , − 2 4T 4T (cid:18) (cid:19) (cid:18) (cid:19) d p(N−R)2 ×G −p(d), 2 4T (cid:18) (cid:19)! p(N−R)2 ≤C(T)exp − (N−R)d−2, (15) 4T (cid:18) (cid:19) whichtendsto0exponentiallyfastasN→¥ .If p>1andL ∈/S,itfollowsfrom formula(13)thatweneedanextrasummandforrN(T,K),namely(15)with p=1. 1 3 Truncation ofSmall Jumps Inthissectionweapproximateequation(1)bycuttingoffthesmalljumpsofL .To thisend,wefirstdefineforeachN∈N GN(t,x;s,y):=G(t,x;s,y)1 (y), (t,x),(s,y)∈R ×Rd , (16) UN + wherethemeaningofthesetsUN isexplainedinA7.Furthermore,weintroduce 1/p∗ r2N := |z|pp (dz) , r3N := z1{p>1,L ∈/S}p (dz) , (17) (cid:18)Z[−eN,eN] (cid:19) (cid:12)Z[−eN,eN] (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) 8 BohanChen,CarstenChong,andClaudiaKlu¨ppelberg where (e N)N∈N ⊆ (0,1) satisfies e N →0 as N →¥ . Condition A4 implies that rN,rN →0asN→¥ .Next,definingtruncationsoftheLe´vybasisL by 2 3 L N(dt,dx):= zm (dt,dx,dz), (18) Z[−eN,eN]c ourapproximationschemeforthesolutionY to(1)isgivenas: t YN(t,x):=Y (t,x)+ GN(t,x;s,y)s (YN(s,y))L N(ds,dy) (19) 0 0 Rd Z Z for(t,x)∈R ×Rd.Indeed,YN canbesimulatedexactlybecauseforallT ∈R the + + truncationL N only hasa finite (random)numberRN(T) of jumpson [0,T]×UN, say at the space–time locations(t N,x N) with sizes JN. This implies that we have i i i thefollowingalternativerepresentationofYN(t,x)for(t,x)∈[0,T]×Rd: RN(T) YN(t,x)=Y (t,x)+ (cid:229) G(t,x;t N,x N)s (YN(t N,x N))JN1 . (20) 0 i i i i i {tN<t} i i=1 What remains to do is to simulateYN(t N,x N), i=1,...,RN(T), iteratively, from i i whichthevaluesY(t,x)forallother(t,x)∈[0,T]×Rd canbecomputed. Thefollowingalgorithmsummarizesupthesimulationprocedure: Algorithm1.ConsiderafinitegridG thatisasubsetof[0,T]×Rd.Foreachstep N proceedasfollows: 1.DrawaPoissonrandomvariableRN(T)withintensity T RN(T):= 1{z∈[−eN,eN]c}n (dt,dx,dz)=TLeb(UN)p [−e N,e N]c . 0 UN R Z Z Z (cid:0) (cid:1) 2.Fori=1,...,RN(T): a.Drawapair(t N,x N)withuniformdistributionfrom[0,T]×UN. i i b.DrawJN from[−e N,e N]cwithdistributionp /p [−e N,e N]c . i 3.Foreachi=1,...,RN(T)and(t,x)∈G simulateY0((cid:0)tiN,x iN)and(cid:1)Y0(t,x). 4.Foreachi=1,...,RN(T)set i−1 YN(t N,x N):=Y (t N,x N)+(cid:229) G(t N,x N;t N,x N)s (YN(t N,x N))JN . i i 0 i i i i j j j j j j=1 5.Foreach(t,x)∈G defineYN(t,x)via(20). The nexttheoremdeterminesthe convergencebehaviourof the scheme (19)to thetruesolutionY to(1). Theorem1.GrantassumptionsA1–A7underwhich the SVE (1)hasa uniqueso- lutioninBp .ThenYN asdefinedin(19)belongstoBp forallN∈N,andforall loc loc SimulationofStochasticVolterraEquationsDrivenbySpace–TimeLe´vyNoise 9 T ∈R andcompactsetsK⊆Rd thereexists aconstantC(T)∈R independent + + ofN andKsuchthat sup kY(t,x)−YN(t,x)kLp ≤C(T)(r1N(T,K)+r2N+r3N). (21) (t,x)∈[0,T]×K Furthermore,if(cid:229) ¥ (rN(T,K)+rN+rN)p∗ <¥ isfulfilled,thenwealsohavefor n=1 1 2 3 all(t,x)∈[0,T]×KthatYN(t,x)→Y(t,x)a.s.asN→¥ . Proof. Itisobviousthat|GN|≤|G|pointwiseandthatwehaven N≤n forthethird characteristicn N ofL N.Thus,A1–A6arestillsatisfiedwhenGandn arereplaced byGN andn N (ifL ∈S,alsoL N∈S).SoYN asasolutionto(1)withGN andL N instead of G andL belongsto Bp as well. Moreover,forall T ∈R there exists loc + C(T)∈R independentofN∈Nsuchthat + sup kYN(t,x)kLp ≤C(T), N∈N. (22) (t,x)∈[0,T]×Rd Weonlysketchtheproofforthisstatement.Infact,using[5,Lem.6.1(1)]itcanbe shown thatthe left-handside of (22)satisfies an inequality of the same type as in Lem.6.4(3)ofthesamepaper.Inparticular,itisboundedbyaconstantCN(T)that dependsonN onlythrough|GN|andn N,andthatthisconstantisonlyincreasedif we replace |GN| and n N by the larger |G| and n . In this way, we obtain an upper boundC(T)thatdoesnotdependonN. Next,weprovetheconvergenceofYN toY asstatedin(21).Wehave t Y(t,x)−YN(t,x)= [G(t,x;s,y)−GN(t,x;s,y)]s (Y(s,y))L (ds,dy) 0 Rd Z Z t + GN(t,x;s,y)[s (Y(s,y))−s (YN(s,y))]L (ds,dy) 0 Rd Z Z t + GN(t,x;s,y)s (YN(s,y))(L −L N)(ds,dy) 0 Rd Z Z =:IN(t,x)+IN(t,x)+IN(t,x), (t,x)∈R ×Rd. (23) 1 2 3 + If p>1, we have by (8), Ho¨lder’s inequality and the Burkholder-Davis-Gundy- inequality t kI2N(t,x)kLp ≤ GN(t,x;s,y)[s (Y(s,y))−s (YN(s,y))]B1(ds,dy) (cid:13)Z0 ZRd (cid:13)Lp (cid:13) t (cid:13) +(cid:13)(cid:13) GN(t,x;s,y)[s (Y(s,y))−s (YN(s,y))]M(ds,dy(cid:13)(cid:13)) (cid:13)Z0 ZRd (cid:13)Lp (cid:13)(cid:13) t p−1 (cid:13)(cid:13) ≤C(cid:13) |G(t,x;s,y)||B |(ds,dy) (cid:13) 1 (cid:18)Z0 ZRd (cid:19) 1/p t × |G(t,x;s,y)|kY(s,y)−YN(s,y)kp |B |(ds,dy) Lp 1 Z0 ZRd ! 10 BohanChen,CarstenChong,andClaudiaKlu¨ppelberg t 1/p +C |G(t,x;s,y)|pkY(s,y)−YN(s,y)kp d(s,y) . (24) Lp 0 Rd (cid:18)Z Z (cid:19) If p∈(0,1],wehaveL ∈V by(6)and(9),andthusJensen’sinequalitygives 0 t p kI2N(t,x)kLp =E GN(t,x;s,y)[s (Y(s,y))−s (YN(s,y))]zm (ds,dy,dz) 0 Rd (cid:20)(cid:18)Z Z (cid:19) (cid:21) t ≤E |GN(t,x;s,y)[s (Y(s,y))−s (YN(s,y))]z|pn (ds,dy,dz) 0 Rd (cid:20)Z Z (cid:21) t ≤C |G(t,x;s,y)|pkY(s,y)−YN(s,y)kLpd(s,y). (25) 0 Rd Z Z Inserting(24)and(25)backinto(23),wehaveforvN(t,x):=kY(t,x)−YN(t,x)kLp t 1/p vN(t,x)≤C(T) |G(t,x;s,y)|1{p>1,L ∈/S}(vN(s,y))pd(s,y) (cid:18)Z0 ZRd (cid:19) t 1/p∗ + |G(t,x;s,y)|p(vN(s,y))p∗d(s,y) (cid:18)Z0 ZRd (cid:19) ! +kI1N(t,x)+I3N(t,x)kLp , (t,x)∈[0,T]×Rd. ByaGronwall-typeestimate,whichispossiblebecauseofA5(seetheproofof[5, Thm.4.7(3)]foranelaborationofanargumentofthistype),weconclude sup vN(t,x)≤C(T) sup kI1N(t,x)+I3N(t,x)kLp . (t,x)∈[0,T]×K (t,x)∈[0,T]×K whereC(T)doesnotdependonKbecauseof(11).ForIN(t,x)wehavefor p>1 1 t kI1N(t,x)kLp ≤ [G(t,x;s,y)−GN(t,x;s,y)]s (Y(s,y))B1(ds,dy) (cid:13)Z0 ZRd (cid:13)Lp (cid:13) t (cid:13) +(cid:13)(cid:13) [G(t,x;s,y)−GN(t,x;s,y)]s (Y(s,y))M(ds,dy(cid:13)(cid:13)) (cid:13)Z0 ZRd (cid:13)Lp (cid:13) (cid:13) (cid:13) t (cid:13) ≤C(cid:13)1+ sup kY(t,x)kLp |G(t,x;s,y)||(cid:13)B1|(ds,dy) (t,x)∈[0,T]×Rd ! Z0 Z(UN)c t 1/p + |G(t,x;s,y)z|pn (ds,dy,dz) (cid:18)Z0 Z(UN)c (cid:19) ! ≤C(T)rN(T,K), (26) 1 uniformlyin(t,x)∈[0,T]×K.Insimilarfashiononeprovestheestimate(26)for p∈(0,1],perhapswithadifferentC(T).Next,when p>1,(22)implies

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.