• Aucun résultat trouvé

Uniqueness in a class of Hamilton–Jacobi equations with constraints

N/A
N/A
Protected

Academic year: 2022

Partager "Uniqueness in a class of Hamilton–Jacobi equations with constraints"

Copied!
6
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

C. R. Acad. Sci. Paris, Ser. I

www.sciencedirect.com

Mathematical analysis/Partial differential equations

Uniqueness in a class of Hamilton–Jacobi equations with constraints

Unicité pour une classe d’équations de Hamilton–Jacobi avec contraintes Sepideh Mirrahimi, Jean-Michel Roquejoffre

Institutdemathématiques(UMRCNRS5219),UniversitéPaul-Sabatier,118,routedeNarbonne,31062Toulousecedex,France

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received13February2015 Accepted18March2015 Availableonlinexxxx PresentedbyHaïmBrézis

Inthisnote,wediscussaclassoftime-dependentHamilton–Jacobiequations depending onafunctionoftime,thisfunctionbeingchoseninordertokeepthemaximum ofthe solution ofthe constant value 0. The main result of the note is that the full problem hasauniqueclassicalsolution.Themotivationisaselection–mutationmodelthat,inthe limitofsmalldiffusion,exhibitsconcentrationonthezero-levelsetofthesolutiontothe Hamilton–Jacobiequation.Theuniquenessresultthatweproveimpliesstrongconvergence anderrorestimatesfortheselection–mutationmodel.

©2015Académiedessciences.PublishedbyElsevierMassonSAS.All rights reserved.

r é s um é

Danscettenote,ondiscuteuneclassed’équationsdeHamilton–Jacobidépendantdutemps et une fonction inconnue du temps choisie pour que le maximum de la solution de l’équation deHamilton–Jacobi prennetout letempslavaleur0.Le résultatprincipalde cettenoteestqueleproblèmecompletadmetuneuniquesolutionclassique.Lamotivation est unmodèledesélection–mutationqui, danslalimited’unediffusiviténulle,présente une concentration sur la ligne de niveau 0 de la solution de l’équation de Hamilton–

Jacobi. Le résultatd’unicité que nousdémontrons implique uneconvergence forte, avec estimationsd’erreurpourlemodèledesélection–mutation.

©2015Académiedessciences.PublishedbyElsevierMassonSAS.All rights reserved.

Versionfrançaiseabrégée

Onprésentedanscettenoteunrésultatd’unicitépourleproblèmedeHamilton–Jacobisuivant,d’inconnues

(

u

(

t

,

x

),

I

(

t

))

:

tu

= |∇

u

|

2

+

R

(

x

,

I

), (

t

>

0

,

x

∈ R

d

),

max

x u

(

t

,

x

) =

0

,

I

(

0

) =

I0

>

0

,

u

(

0

,

x

) =

u0

(

x

).

(1)

E-mailaddresses:[email protected](S. Mirrahimi),[email protected](J.-M. Roquejoffre).

http://dx.doi.org/10.1016/j.crma.2015.03.005

1631-073X/©2015Académiedessciences.PublishedbyElsevierMassonSAS.All rights reserved.

(2)

La donnée R

(

x

,

I

)

vérifiedeshypothèsesde stricte concavitépar rapportà xetde monotonieparrapport à I,explicitées plus bas.La donnéeinitiale u0 vérifieaussi deshypothèsesspéciales de concavité.Ainsi, lafonctionI

(

t

)

doitêtrechoisie pourquelasolutionu

(

t

,

x

)

del’équationdeHamilton–Jacobiait,àchaqueinstant,unmaximumégalà0.

Nousavonsalorsle

Theorème0.1.OnchoisitRC2,etonsupposel’existencedeIM

>

0telquemax

x∈RdR

(

x

,

IM

)

=0=R

(

0

,

IM

)

.Deplus,R estsupposée strictementconcaveet,pour|x|grand,compriseentredeuxparaboles.Ladonnéeinitialeu0estégalementàdécroissancequadratique etstrictementconcave.

Le problème(1)auneuniquesolution

(

u

,

I

)

,où u estunesolution classiquedel’équationdeHamilton–Jacobi.Deplus,uLloc

R+;Wloc3,∞

(

Rd

)

Wloc1,∞

R+;Lloc

(

Rd

)

×W1,∞

(

R

)

.

L’existencepour(1)aétédémontréeenplusieursendroits,voirparexemple[8]ou[1].L’unicitéestdoncnotrerésultat principal ;c’étaitunproblèmeouvert.L’unicitéétaiteneffetconnueseulementpouruncastrèsparticulier(voir[8]).

Lemodèle(1)intervientdanslalimite

ε

0 dessolutionsde

tnε

ε

nε

=

nε

ε

R

x

,

Iε

(

t

)

(

t

>

0

,

x

∈ R

d

),

Iε

(

t

) =

Rd

ψ (

x

)

nε

(

t

,

x

)

dx

,

(2)

(

t

,

x

)

estladensitéd’unepopulationcaractériséeparun traitbiologiquex d-dimensionnel. Lacompétitionpourune ressource unique est représentée par

(

t

)

,

ψ >

0, régulière donnée. Le terme R

(

x

,

I

)

est le taux de reproduction. Les hypoth`ses de concavité sontdes hypothèses techniques,mais pertinentes au plan biologique.Par une transformationde Hopf–Cole=exp

(

/ ε )

,onseramèneàl’équationsur suivante :

tuε

= ε

uε

+ |∇

uε

|

2

+

R

(

x

,

Iε

)

(3)

qui, danslalimite

ε

0,donnel’équationsur u.Ons’attendalorsà cequenε seconcentreaux pointsoù u estproche de 0.Et,danscettelimite, apparaîtcommeunesortedemultiplicateurdeLagrange.

La convergence de (3) vers (1) à une sous-suite près est connue depuis[8]. Le Théorème 0.1 donne la convergence de toute la famille, ainsique desestimations d’erreur. Soit

(

t

)

lepoint où

(

t

, .)

atteint son maximum. On suppose l’existencedeI0 telque0

<

I0Iε

(

0

)

:=

Rd

ψ(

x

)

n0ε

(

x

)

dx

<

IM,etonsuppose :

n0ε

=

eu0ε/ε

=

r

ε

d/2e

u0

,

avec u0

C2

(R

d

)

and max

x∈Rdu0

(

x

) =

0

.

Lerésultatestalorsle

Theorème0.2.Soitnεlasolutionde(2)etuεdéfiniepar(3).Nousavonslesdéveloppementsasymptotiquessuivants :

Iε

=

I

+ ε

I1

+

o

(

1

),

xε

=

x

+ ε

x1

+

o

(

1

),

uε

=

u

+ ε

log

(

r

ε

d2

) + ε

u1

+

o

(

1

).

Lestermes I1,x1 etu1vontêtreprésentésdans[7].Cerésultatimpliquelecorollaire :

Corollaire0.3.Nousavonsl’approximationsuivantepournε:

nε

(

t

,

x

) =

r

ε

d2

exp

(

u1

+

u

ε ) +

o

(

1

)

.

Enparticulier,lorsque

ε

0,toutelasuite

(

)

εconverge : nε

(

t

,

x

) −→ ¯ ρ (

t

) δ

x

− ¯

x

(

t

)

au sens des mesures

,

avec

ρ (

t

)

= I

(

t

) ψ(

x

(

t

))

.

End’autrestermes,lapopulationseconcentre suruntrait dominant,quiévolue avecletemps.Onnote quelaconver- gencedeàunesous-suiteprésétaitdéjàétabliedans[5].

Touscesrésultatsserontdétaillésdans[6]et[7].

(3)

1. Introduction

Thepurposeofthisnoteistodiscussuniquenessinthefollowingproblem,withunknowns

(

I

(

t

),

u

(

t

,

x

))

:

tu

= |∇

u

|

2

+

R

(

x

,

I

) (

t

>

0

,

x

∈ R

d

),

max

x u

(

t

,

x

) =

0

,

I

(

0

) =

I0

>

0

,

u

(

0

,

x

) =

u0

(

x

),

(4)

whereI0

>

0 andu0isaconcave,quadraticfunction:

L0

L1

|

x

|

2

u0

(

x

)

L0

L1

|

x

|

2

,

2L1

D2u0

≤ −

2L1

,

D3u0

L

(R

d

).

Theconstraintonthemaximumofu

(

t

, .)

makestheproblemnonstandard.Ourmainresultis

Theorem1.1.ChooseRC2,andsupposethatthereisIM

>

0suchthatmax

x∈RdR

(

x

,

IM

)

=0=R

(

0

,

IM

)

.Alsoassumethefollowing concavityandregularitypropertiesforR:

K1

|

x

|

2

R

(

x

,

I

)

K0

K1

|

x

|

2

,

for 0

I

IM

,

2K1

D2R

(

x

,

I

) ≤ −

2K1

<

0and D3R

( · ,

I

)

L

( R

d

)

for0

I

IM

,

K2

IR

≤ −

K2

,

|

I x2

iR

(

x

,

I

) | + |

3RI xixj

(

x

,

I

) | ≤

K3

,

for0

I

IM

,

and i

,

j

=

1

,

2

, · · · ,

d

.

Problem (4)has a unique solution

(

u

,

I

)

, where u solves the Hamilton–Jacobi equation in the classical sense. Moreover uLloc

R+;W3loc,

(

Rd

)

Wloc1,

R+;Lloc

(

Rd

)

×W1,∞

(

R

)

.

Existencefor(4)hasbeenprovedinvariouscontexts(see[8,1,5]).Thus,ourcontributionisuniqueness,whichhasupto nowbeenanopenproblem.Theuniquenesshasindeedbeenknownonlyforaveryparticularcase(see[8]).

Therestofthenote isorganizedasfollows.InSection 2,weexplain themotivationand,inparticular,themeaning of the various assumptions.In Section 3, we revisit existence for(4),which will entail an unconventionalODE formulation foruniqueness.Section4,whichisthemainpartofthenote,providesafairlycompletesketchoftheuniquenessproof.In Section5,wegiveanapplication.

2. Backgroundandmotivation

Model(4)arisesinthelimit

ε

0 ofthesolutionstotheproblem

tnε

ε

nε

=

nε

ε

R

x

,

Iε

(

t

)

(

t

>

0

,

x

∈ R

d

),

Iε

(

t

) =

Rd

ψ (

x

)

nε

(

t

,

x

)

dx

,

(5)

where

(

t

,

x

)

isthedensityofapopulation characterizedbya d-dimensionalbiologicaltrait x.Thepopulationcompetes for a single resource, thisis represented by

(

t

)

, where

ψ

is a givenpositive smooth function. The term R

(

x

,

I

)

isthe reproduction rate; it is, ascan be expected, very negative for large x anddecreases asthe competition increases. Such modelscanbederived fromindividualbasedstochasticprocessesinthelimitoflargepopulations(see[2]).Theconcavity assumptionon Risatechnicalone,althoughbiologicallyrelevant.TheHopf–Coletransformation=exp

(

/ ε )

yieldsthe equation

tuε

= ε

uε

+ |∇

uε

|

2

+

R

(

x

,

Iε

)

(6)

which,in thelimit

ε

0, yields theequation for u. Now, beinguniformlypositive andboundedin

ε

,the Hopf–Cole transformationleadstotheconstraintonu.Moreover,oneexpectsthat concentratesatthepointswhereuiscloseto0 andthefunction appears,inthelimit,asasortofLagrangemultiplier.

Thisapproach,basedon theHopf–Coletransformation,to study(5) hasbeenintroducedin[4]andthendeveloped in differentcontexts (see forinstance[8,1,3,5]).Long-timeasymptotics ofsuch modelshasalsobeen studiedin[9]andthe referencestherein.

3. Existence

Existenceofasolutionto(4)isobtainedbyletting

ε

0 in(6).Themainstepisthefollowingtheorem.

(4)

Theorem3.1(Uniformestimatesforuε,[5]).Thereexists Im

>

0suchthat0

<

Im

(

t

)

IM+C

ε

2.Moreover,wehavethe followingestimatesonuε

L0

L1

|

x

|

2

ε

2dL1t

uε

(

t

,

x

)

L0

L1

|

x

|

2

+

K0

+

2d

ε

L1

t

,

L1

2t K1

D2uε

(

t

,

x

) ≤ −

2L1

,

D3uε

(

t

, ·)

L

C

(

T

),

for t

∈ [

0

,

T

].

(7)

Theboundsfor canbeobtainedforanyuniformlyboundedfunction,notonlyforthatof(5).Thisremarkwillbe animportantingredientoftheuniquenessproof.

4. Uniqueness

Foragivencontinuousfunction I

(

t

)

suchthat0

<

I

(

t

) <

IM,onemayconstructasolutionto

tu= |∇u|2+R

(

x

,

I

)

with initial datumu0.JustasinTheorem 3.1,thissolutionsatisfiesestimates(7).Andso,foru

(

t

, .)

beingstrictlyconcaveand quadratically decreasing,there existsa unique function x¯

(

t

)

such that u

(

t

,

x¯

(

t

))

=max

x∈Rdu

(

t

,

x

)

.Assume that I

(

t

)

ischosen such thatu

(

t

,

x¯

(

t

))

=0.Then,fromtheequationonu,wededucethat R

x

(

t

),

I

(

t

))

=0.Noticealsothat,because

IR

<

0, we have R

(

¯x

(

t

),

0

) >

0.Finally,differentiating ∇u

(

t

,

x¯

(

t

))

=0 andpluggingintheequationforu,weobtainan ODEfor¯x:

˙¯

x

(

t

)

=

D2u

t

,

¯x

(

t

)

1

xR

¯ x

(

t

),

¯I

(

t

)

.

Theideaisthustochangetheconstrainedproblem(4)bythefollowingslightlynonstandarddifferentialsystem:

⎧ ⎨

R

x

(

t

),

I

(

t

)

=

0

,

fort

∈ [

0

,

T

] ,

˙

x

(

t

) =

D2u

t

,

x

¯ (

t

)

1

xR

¯

x

(

t

), ¯

I

(

t

)

,

fort

∈ [

0

,

T

],

tu

= |∇

u

|

2

+

R

(

x

,

I

),

in

[

0

,

T

] × R

d

,

(8)

withinitialconditions

I

(

0

) =

I0

,

u

(

0

, ·) =

u0

(·),

x

(

0

) =

x0

,

such thatR

(

x0

,

I0

) =

0

.

(9) Note that (8)isreally adifferential systembecausethe assumptions on R implythat I

(

t

)

canimplicitly beexpressed in termsofx¯

(

t

)

.Anditisslightlynonstandard becausex¯ solves anODEwhosenonlinearitydependsonu.Finally,notethat, assoonasusatisfiestheconcavityandregularityestimates(7),system(8)isequivalenttotheconstrainedproblem(4).

Thissuggeststouseasimplefixed-point argumenttoproveuniquenessof(8)(andso,of(4)).Whichinturnsuggests to set up thefollowing scheme: starting fromx

(

t

)

C

(

[0

,

T]

,

Rd

)

,such that x

(

0

)

=x0, where R

(

x0

,

0

) >

0. Let I

(

t

)

solve R

(

x

(

t

),

I

(

t

))

=0 on[0

,

T]withR

(

x0

,

I0

)

=0.Letv

(

t

, .)

betheuniquesolutionto

tv

= |∇

v

|

2

+

R

(

x

,

I

),

v

(

0

,

x

) =

u0

(

x

).

(10)

Let y

(

t

)

solve ˙y

(

t

)

=

D2v

t

,

x

(

t

)

1

xR

x

(

t

),

I

(

t

)

on[0

,

T] withinitial datum y

(

0

)

=x0.Setting y:=

(

x

)

,wenotice that uniqueness is proved assoon as we have proved that

has a unique fixed point. One additional feature about a solution

I

,

u

,

¯x

)

to(8):

Lemma4.1.Thefunction¯I

(

t

)

isincreasing.

Weclaimthatourproblemreducestoprovingthefollowingtheorem.

Theorem4.2.ThereexistsC

>

0universaland

δ >

0,whichissmallas R

(

x0

,

0

)

tendsto0,suchthat

isacontractionfrom C

(

[0

, δ

]

,

BC

(

x0

))

toitself;hereBr

(

a

)

denotestheballofcentrea∈Rdandradiusr

>

0.

Noteindeedthat,byLemma 4.1,wehave,because

IR

<

0:

R

x

(δ),

0

) =

R

x

(δ),

0

)

R

x

(δ), ¯

I

(δ))

c

¯

I

(δ)

c I0

,

forsomeuniversalc

>

0.HenceTheorem 4.2canbeiteratedtoyieldglobalexistenceanduniqueness.

Letusgive anoverviewoftheproofofTheorem 4.2.ForIC

(

[0

, δ

]; [0

,

IM]

)

,let V

(

I

)

bethe(unique)solutionto(10).

Themainstepisthefollowing.

Lemma4.3.LetI1

,

I2C

(

[0

, δ

]; [0

,

IM]

)

.Then

V

(

I1

)

V

(

I2

)

W2,∞([0]×Rd)

C

I1

I2

L([0,δ])

δ.

(5)

Thislemma,onceproved,opensthewaytoTheorem 4.2.Indeedtheequation R

(

x

,

I

)

=0 yieldsasmoothmappingxI, and IV is a Lipschitz mappingthanks to Lemma 4.3. Moreover, theequation y˙

(

t

)

=

D2v

t

,

x

(

t

)

1

xR

x

(

t

),

I

(

t

)

yieldsaLipschitzmappingvybytheestimatesforv givenbyTheorem 3.1.

Lemma 4.3ismoreinvolved.IfI1andI2areasintheassumptionsofthelemma,thefunctionr=V

(

I1

)

V

(

I2

)

solves

tr

= (∇

v1

+ ∇

v2

) · ∇

r

+

R

(

x

,

I2

)

R

(

x

,

I1

),

in

[

0

, δ] × R

d

,

r

(

0

,

x

) =

0

,

for allx

∈ R

d (11)

with vi=V

(

Ii

)

. Note that the above equation has a unique classical solution that can be computed by the method of characteristics.Thecharacteristicssolve

˙

γ (

t

) = −∇

v1

(

t

, γ ) − ∇

v2

(

t

, γ ),

(12)

and,duetotheestimatesofTheorem 3.1,theyexistglobally.So,onemaysuccessivelyexpressr givenbyintegrationalong characteristics,andestimateitsderivativesrecursively.

5. Application

Convergence of (6)to (4)had alreadybeen proved in[8,1],along subsequences. The uniqueness partof Theorem 1.1 yields theconvergence ofthefull familyofsolutions to (6),insteadofconvergence alonga subsequence. Moreoverit allowsanexpansionof, and (themaximumpointofateachtime)intermsof

ε

.Herearetheresults.

AssumethatthereisI0suchthat0

<

I0Iε

(

0

)

:=

Rd

ψ(

x

)

n0ε

(

x

)

dx

<

IM,andthat

n0ε

=

eu0ε/ε

=

r

ε

d/2e

u0

,

with u0

C2

(R

d

)

and max

x∈Rdu0

(

x

) =

0

.

Theresultisthefollowingtheorem.

Theorem5.1.Letnεbethesolutionto(5)anduεbedefinedby(6).Wehavethefollowingasymptoticexpansions

Iε

=

I

+ ε

I1

+

o

(

1

),

xε

=

x

+ ε

x1

+

o

(

1

),

uε

=

u

+ ε

log

(

r

ε

d2

) + ε

u1

+

o

(

1

).

ThetermsI1,x1 andu1 willbeprovidedin[7].Thisyieldsthefollowingcorollary.

Corollary5.2.Wehavethefollowingapproximationfornε:

nε

(

t

,

x

) =

r

ε

d2

exp

u1

(

t

,

x

) +

u

(

t

,

x

) ε

+

o

(

1

)

.

Inparticular,as

ε

0,thewholesequence

(

)

εconverges:

nε

(

t

,

x

) −→ ¯ ρ (

t

) δ

x

− ¯

x

(

t

)

,

weakly in the sense of measures

,

with

ρ (

t

)

= I

(

t

) ψ(

x

(

t

))

.

Inotherwords,thepopulationdensityconcentratesonadominanttraitthat evolvesovertime.We notethatthecon- vergenceof alongsubsequenceswasalreadyestablishedin[5].

Theaboveresultswillbedetailedin[6]and[7].

Acknowledgements

S.MirrahimiwaspartiallyfundedbytheFrenchANRprojectsKIBORDANR-13-BS01-0004andMODEVOLANR-13-JS01- 0009. J.-M. Roquejoffre has received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP/2007–2013)/ERC Grant Agreementn. 321186 – ReaDi –“Reaction-Diffusion Equations, Propa- gationandModelling”held byHenri Berestycki.He wasalso partiallysupported bythe FrenchNationalResearchAgency (ANR),withintheprojectNONLOCALANR-14-CE25-0013.BothauthorsaregratefultotheLabexCIMIfororganizingaPDE- probabilityquarterinToulouse,whichprovidedquiteastimulatingenvironmenttothiscollaboration.

(6)

References

[1]G.Barles,S.Mirrahimi,B.Perthame,ConcentrationinLotka–Volterraparabolicorintegralequations:ageneralconvergenceresult,MethodsAppl.Anal.

16 (3)(2009)321–340.

[2]N.Champagnat,R.Ferrière,S.Méléard,Individual-BasedProbabilisticModelsofAdaptiveEvolutionandVariousScalingApproximations,Progressin Probability,vol. 59,Birkhäuser,Basel,Switzerland,2008.

[3]N.Champagnat,P.-E.Jabin,Theevolutionarylimitformodelsofpopulationsinteractingcompetitivelyviaseveralresources,J.Differ.Equ.261(2011) 179–195.

[4]O.Diekmann,P.-E.Jabin,S.Mischler,B.Perthame,Thedynamicsofadaptation:anilluminatingexampleandaHamilton–Jacobiapproach,Theor.Popul.

Biol.67 (4)(2005)257–271.

[5]A.Lorz,S.Mirrahimi,B.Perthame,Diracmassdynamicsinamultidimensionalnonlocalparabolicequation,Commun.PartialDiffer.Equ.36(2011) 1071–1098.

[6] S.Mirrahimi,J.-M.Roquejoffre,UniquenessinaHamilton–Jacobiequationwithconstraints,inpreparation.

[7] S.Mirrahimi,J.-M.Roquejoffre,Approximationofsolutionsofselection–mutationmodelsanderrorestimates,inpreparation.

[8]B.Perthame,G.Barles,DiracconcentrationsinLotka–VolterraparabolicPDEs,IndianaUniv.Math.J.57(2008)3275–3301.

[9]G.Raoul,Étudequalitativeetnumériqued’équationsauxdérivéespartiellesissuesdessciencesdelanature,PhDthesis,ENSCachan,France,2009.

Références

Documents relatifs

Let u be the unique uniformly continuous viscosity solution of (1.4). Minimal time function.. To prove the claim of the theorem, we will use arguments from control theory. Let x be

MURAT, Uniqueness and the maximum principle for quasilinear elliptic equations with quadratic growth conditions, Arch.. Ra-

It is explained in [15] that the proof of the comparison principle between sub- and super-solutions of the continuous Hamilton-Jacobi equation can be adapted in order to derive

Dans cette note, on discute une classe d’´ equations de Hamilton-Jacobi d´ ependant du temps, et d’une fonction inconnue du temps choisie pour que le maximum de la solution de

Key-Words: Hamilton-Jacobi equation with constraint, uniqueness, constructive existence result, selection-mutation models..

Mimicking the previous proof of uniqueness (given in Sec. 2.2) at the discrete level it is possible to obtain error esti- mates, that is an estimate between a strong solution (we

degenerate parabolic equations, Hamilton-Jacobi-Bellman equations, viscosity solutions, un- bounded solutions, maximum principle, backward stochastic differential equations,

[r]