• Aucun résultat trouvé

An extremal eigenvalue problem arising in heat conduction Journal de Mathématiques Pures et Appliquées

N/A
N/A
Protected

Academic year: 2022

Partager "An extremal eigenvalue problem arising in heat conduction Journal de Mathématiques Pures et Appliquées"

Copied!
28
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

Journal de Mathématiques Pures et Appliquées

www.elsevier.com/locate/matpur

An extremal eigenvalue problem arising in heat conduction

,✩✩

Grégoire Nadin,Yannick Privat

CNRS,UniversitéPierreetMarieCurie(Univ.Paris 6),UMR7598,LaboratoireJacques-LouisLions, F-75005,Paris,France

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

Articlehistory:

Received20March2015 Availableonline3March2016

MSC:

49J15 49K15 34E05

Keywords:

Extremalproblem Calculusofvariation Sturm–Liouvilleeigenvalue Lebesguedensitytheorem Lateralsurfaceconstraint

This article is devoted to the study of two extremalproblems arising naturally in heat conduction processes. We look for optimal configurations of thermal axisymmetric finsandmodelthisproblem astheissue of(i) minimizing (for the worst shape) or (ii) maximizing (for the best shape) the first eigenvalue of a selfadjointoperatorhavingacompactinverse.Weimposeapointwiselowerbound on the radius of thefin,as well as a lateral surface constraint.Using particular perturbationsandunderasmallnessassumptiononthepointwiselowerbound,one shows that theonlysolution isthecylinder inthefirstcase whereasthere isno solution in thesecond case. Wemoreover construct a maximizing sequence and provide theoptimal valueof theeigenvalue inthiscase. As a byproduct of this result,andto proposearemedyto thenon-existenceinthesecondcase,wealso investigatethewell-posednesscharacterofanotheroptimaldesignproblemsetina classenjoyinggoodcompactnessproperties.

© 2016ElsevierMassonSAS.All rights reserved.

r é s u m é

Cetarticleestdédiéàl’étudedesproblèmesextrémauxsurvenantdefaçonnaturelle danslesprocécédés deconductiondelachaleur.Onrecherchedesconfigurations optimalesd’ailettesthermiquesaxisymétriquesetonmodéliseceproblèmecomme celui(i)deminimiser(pourlapireforme),ou(ii)demaximiser(pourlameilleure forme) la première valeur propre d’un opérateur autoadjoint d’inverse compact.

On imposeuneborne inférieure ponctuelle surle rayon del’ailette, ainsiqu’une contrainte sur la surfacelatérale. En utilisant des perturbations particulières et sousuneconditiondepetitessedelacontraintedeborneinférieure,onmontreque l’uniquesolutionestdonnéeparlecylindredanslepremiercas,tandisqu’iln’ya pasdesolutiondansledeuxièmecas.Onconstruitdeplusunesuitemaximisante et on détermine la valeur propre optimale dans ce cas. Afin de proposer un remèdedansle secondcas,onétudieégalementlecaractère bienposéd’unautre

TheresearchleadingtotheseresultshasreceivedfundingfromtheEuropeanResearchCouncilundertheEuropeanUnion’s Seventh Framework Programme (FP/2007–2013)/ERC Grant Agreementn. 321186 ReaDi “Reaction–DiffusionEquations, PropagationandModelling”heldbyHenriBerestycki.ThisworkwasalsopartiallysupportedbytheFrenchNationalResearch Agency(ANR),withintheprojectNONLOCALANR-14-CE25-0013.

✩✩ ThesecondauthorwaspartiallysupportedbytheANRprojectOPTIFORM,ANR-12-BS01-0007.

* Correspondingauthor.

E-mailaddresses:gregoire.nadin@upmc.fr(G. Nadin),yannick.privat@upmc.fr(Y. Privat).

http://dx.doi.org/10.1016/j.matpur.2016.02.005

0021-7824/© 2016ElsevierMassonSAS.All rights reserved.

(2)

problèmed’optimisationdeformedansuneclassepossédantdebonnespropriétés decompacité.

© 2016ElsevierMassonSAS.All rights reserved.

1. Introduction

The current workis inspiredand motivated by[1], where theauthors consideredtheproblem of maxi- mizing,with respectto thecrosssectionalarea,therateofheat transferthroughabarof givenmass.For the sakeof clarity,we firststate theextremal problem wewill investigate inSection1.1 andwe will thus provide severalexplanationsonthephysicalframeofourstudyinSection1.2.

1.1. Settingof theextremalproblems

Let us introduce the extremal problems we will dealwith. Let a0 > 0.For the reasons evoked inSec- tion1.2,theadmissibleset willconsistofradiia(·) belongingtoW1,∞(0,) suchthat

(H1) Pointwiseconstraint. Thereholdsa(x)≥a0 foreveryx∈[0,];

(H2) Lateralsurfaceconstraint.Thereholds 0

a(x)

1 +a(x)2dx≤S0.

Letus thusintroducetheclassofadmissiblefunctionsdefinedby Sa0,,S0 =

a∈W1,∞(0, ) satisfying (H1) and (H2) , whereS0> a0 isgiven,so thattheclassSa0,,S0 benon-empty.

According toSection1.2, thefunctional weaim atoptimizing isa→λ1(a),where λ1(a) stands forthe firsteigenvalueoftheinverseofacompactoperator(thatwewill denotebyLainthesequel),definedby

λ1(a) = min

ϕ∈H1(0,) ϕ=0

[a, ϕ] (1)

with

[a, ϕ] = α

0a(x)2ϕ(x)2dx+β 0 a(x)

1 +a(x)2ϕ(x)2dx+σa()2ϕ()2

0a(x)2ϕ(x)2dx+δϕ(0)2 , (2) where α,β,δ,σ denotepositivereal numbers.

Note that, according to the Sturm–Liouville eigenfunctions theory, it is standard that λ1(a) is simple (seee.g.[2,3])andthatitsassociatednormalizedeigenfunctiondenotedϕ1,asolvestheordinarydifferential system

−α

a(x)2ϕ1,a(x)

+βa(x)

1 +a(x)2ϕ1,a(x) =λ1(a)a(x)2ϕ1,a(x), x(0, ) γa(0)2ϕ1,a(0) =−λ1(a)ϕ1,a(0)

ϕ1,a() =σαϕ1,a(), (3)

with γ=α/δ.

(3)

This quantityλ1(a) can be viewedas theexponential cooling rate ofdecayof anaxisymmetricbar(or fin)of lengthwith radius a(·).This will beclarified inSection1.2 below.Weare thusled to investigate thetwofollowing extremalproblems:

• Minimization of λ1(a) (worstshapeofafin).

inf1(a), a∈ Sa0,,S0} . (4)

• Maximization of λ1(a) (bestshapeofafin).

sup1(a), a∈ Sa0,,S0} , (5)

Remark1.TheissueofoptimizingeigenvaluesofSturm–Liouvilleoperatorsisalongstory.Forasurveyof suchproblems,onerefersforinstance to[4].Moreover,onealsomentions[5,6]wheretheauthorsdealwith a“lateralsurface”constraintsimilartotheoneconsideredinthisarticle.Nevertheless,totheopinionofthe authors,thetechnics, basedon thestandardchangeof variable forSturm–Liouvilleequation y =x

0 dt a(t)2

withthenotationsofthepaper,cannotbeadaptedinasimplewaytosolvetheproblemsinvestigatedinthis article.Indeed,thischangeofvariablewasusedtointroduceanauxiliaryproblemforwhichoneshowedthat theoptimalvaluecoincidedwiththeoneoftheinitialoptimaldesignproblem.Italsoallowed toconstruct minimizing/maximizing sequences. Unfortunately, such technic does not provide a sharp estimate of the optimalvalue,andwehavetouseanotherapproach.

1.2. Motivations inconvection–conduction theory

Stateoftheartaboutshapeoptimizationinconvection–conductiontheory Amongmanyapplicationsofthe optimaldesignproblemwewillinvestigate,letusmentionthestrongimportanceinthecomputerindustry offindingcoolingfinsinmicroprocessorshavinggoodperformances.

Many engineering works focused on modeling the direct problem in order to assess the efficiency of differentfinshapes. Noticethatthese studiesaremainlynumericalandnomathematicalapproachisused todeterminetheoptimalprofilesoffins(seeforinstance[7–10]).

In amore generalcontext, let us mention several studies dealing essentially with numericalaspects of conduction/convectionproblemsinshapeoptimization.ThemodelusedcombinesaFluidMechanicspartial differentialequationwith aparabolic equationinvolvingatransport term(seee.g.[11,12]).

In [1], the authors investigate the problem of maximizing λ1(a) under a volume constraint, namely

0a=V0, for agivenV0 >0,and underthesimplifiedassumptions β = 0 and σ= + (inother words, ϕ()= 0).BywritingandanalyzingtheEuler–Lagrangeequation,theyprovideanexplicitcharacterization of themaximizing shape with the helpof asymmetrization argument. Theyfind a(x)=C/cosh2(x−), where C is anormalizing constant. The main difference with the present work comes from thefact that thelateralconvectiveheat transfersarenotneglectedanymorehere,leadingto adifferent behavior ofthe maximizingsequences,as itwillbehighlightedinthesequel.Itisinterestingto noticethat,insomesense, theirstudycanbeinterpretedas alimitcaseoftheprobleminvestigatedinthepresentarticle.

In[13],asimilarproblemisinvestigated,wherethelateralconvectiontermβa(x)

1 +a(x)2isreplaced byagivenfunctionβP(x),independentofa.Theauthorsthenminimizethequantity

0 a(x)dxforagiven decayrate λ1. Theyshow that,when β = 0 (equivalently,P 0),theminimizingshape ais thesameas for the“dual problem” studiedin[1]. When β = 0, theyprovide an algorithm enabling to determine the solution.

(4)

Fig. 1.Scheme of the axisymmetric fin.

In[14], theauthors dealtwithasimplifiedone-dimensionalstationary modelofaxisymmetricfintaking intoaccountthelateralheattransfersofthefin.Theyanalyzedtheoptimaldesignproblemandinparticular theexistenceissuesaswell asthedeterminationofmaximizingsequences.

In thiswork, wewill consider amoreaccuratemodelof one-dimensionalthermal barinnon-stationary regime,whereconvectivephenomenafromthesideofthefinareconsidered.Accordingtoourmaintheorems (seeTheorems 2 and3inSection3),weshowinthisarticlethatthistermplaysacrucialrolefordetermining theoptimalshape ofthefin.

Modeling oftheproblem Letusconsider anaxisymmetricfinrepresentedbyadomain Ωaoflength>0 and radiusa(x) atabscissax,asdisplayedinFig. 1, definedinaCartesiancoordinatesystemby

Ωa={(rcosθ, rsinθ, x)|r∈[0, a(x)), θS1, x∈(0, )}, (6) wherea∈W1,∞(0,) issuchthata(x)≥a0 foreveryx∈[0,] witha0 apositiveconstant.

Fig. 1 sums-up the situation and the notations we will use throughout this article. According to the approachandthemodeldescribedin[1,Sections 1,2and6],wemakethetwo followingassumptions:

(i) theconvectivecoefficienth, modelingtheheattransferbetweenthefinsurfaceandthefluidflow,does not depend onthe variable x and θ. This hypothesis allowsto reduce the three-dimensional problem to an axisymmetric one, which justifies thatthe temperature T along the fincan be considered as a functionoft(thetime),randxonly.

(ii) thefincanbeviewedasthermallythinalongther-axis.Asaconsequence,itsradialthermalresistanceis lowenoughincomparisonwiththeconvectiveheattransferhanditisrelevanttoclaimthat∂T /∂r0 almosteverywhereinΩa.ThisiswhywewillimposefromnowonthatthetemperatureT isafunction ofthevariablest(thetime)andxonly.

For instance, if the convective heat transfer coefficient h modeling the heat transfer between the fin surface and the fluid flow, and hr the convective coefficient characterizing the heat transfer over the tip are smallenough,thenthefincanbe viewedas thermallythinalong ther-axis.Therefore, theradialBiot numberBirdeterminingwhetherornotthetemperatureainsideabodywillvarysignificantlywithrespect tothevariablerwillbesmallenough(<0.1 inpractice)toconsidertheone-dimensionalconductionmodel as significant.Wereferto[7]formoredetailsonthesemodelingissues.

Theinletofthefin,aswellasthefluidsurroundingthefinareassumedtobeataconstanttemperatures, denotedrespectivelyTd andT.Consideringprocesseswherethefinaimsatcoolingathermalsystem,i.e.

(5)

wheretheheatflowsfromitsbasistowardsthefluid,wewillassumethat0< T< Td(·) almosteverywhere in(0,).Moreover,wewillconsider abase massM0 attachedattheend pointx= 0.

Let Td L2(0,). The temperature T is then solution of the following parabolic partial differential equation:

a(x)2∂T∂t =ρck ∂x

a(x)2∂T∂x

ρcha(x)

1 +a(x)2(T−T)t >0, x(0, ), cM0∂T∂t(0, t) =ka(0)2∂T∂x(0, t) t >0,

∂T

∂x(, t) =hkr(T(, t)−T) t >0,

T(x,0) =Td(x) x∈(0, ), (7)

wherekdenotesthethermalconductivityofthefin,ρitsdensity,citsspecificheatcapacity.Wewillassume inthe sequelthatthe realnumbers k, M0, ρ,c, hand hr arepositive. Somephysical explanations about thederivationofthetemperaturemodelmaybefoundin[7,15].

Fromnow on,wewillratherusethenotations α= k

ρc, β= h

ρc, γ= k cM0

and σ= hr

ρc, (8)

forthesakeofreadability.Itcanbeprovedusingstandardsemigroupsargumentsandsincea≥a0on[0,], thatthesolutionT ofthepartialdifferentialequation (7)belongstoL2(0,T,H1(0,)).

Asdidtheauthorsof[1],itisconvenienttorepresentthesolutionT intermsofseriesofeigenfunctions.

Forthatpurpose,letusintroducetheoperator

La: C0([0, ]) −→ H1(0, ), f −→ ϕa,

(9)

whereϕadenotes theuniquesolutionoftheo.d.e.

α

a(x)2ϕa(x)

−βa(x)

1 +a(x)2ϕa(x) =a(x)2f(x), x(0, ), γa(0)2ϕa(0) =f(0),

ϕa() =σαϕa(). (10)

According to Lax–Milgram’s theorem and since a W1,(0,) and a a0 on [0,], this system has a unique solutionthat belongsto H1(0,). Let us introducethe inner-product ·,·a in thespace C0([0,]) definedby

f, ga= 0

a(x)2f(x)g(x)dx+α

γf(0)g(0),

forevery(f,g)∈(C0([0,]))2.ThecompletionofC0([0,]) forthetopologyinheritedfromtheinner-product

·,·aisaHilbertspace,andthedefinitionof·,·a extendsclearlytoelements ofthatspace.Wedenoteit byCa.Wealsodefinethenorm · a inducedbytheinnerproduct ·,·a.

Withaslightabuseofnotation,letusstilldenotebyLathisextension.Onehasthusthefollowingresult, whoseproof ispostponedto Appendix A.

(6)

Lemma 1.The operatorLais selfadjointandcompactinCa.

As a consequence of Lemma 1, the operator La is diagonalizable in Ca and there exist a sequence of positive real numbers (λn)nIN diverging to + and a sequence (ϕn)nIN of elements of Ca such that Laϕn= λ1

nϕn foreveryn∈IN, whichrewrites

−α

a(x)2ϕn(x)

+βa(x)

1 +a(x)2ϕn(x) =λn(a)a(x)2ϕn(x), x(0, ), γa(0)2ϕn(0) =−λn(a)ϕn(0),

ϕn() =σαϕn(). (11)

Moreover, accordingto theso-calledmin–max principlebyCourant–Fisher(seee.g.[2]),thereholds λn(a) = max

EH1(0,) s.t. dimE=n

ϕminE ϕ=0

[a, ϕ],

where [a,ϕ] is definedby(2).

These considerationsallowus todecomposethesolutionT of(7)as T(t, x)−T=

+ n=1

Td−T, ϕnae−λntϕn(x), (12)

Toward anextremal problem Accordingto(12),onehasthefollowingasymptoticforthesolutionof(7) T(t,·)−T

t+Td(·)−T, ϕ1aeλ1(a)tϕ1,

providedthatTd(·)−T benon-orthogonaltoϕ1fortheinner-product·,·a,whichisnonrestrictiveand will beassumedfrom now.

Sincewearelookingfortheshapeofafinoptimizingitscoolingproperties,itisthennaturaltoconsider:

• the problem offinding the bestshapeof a thermalfin, bymaximizing the firsteigenvalueλ1(a) with respecttothefunctiona,sothatthetemperatureofthematerialtocoolwill becomeclosetothefluid temperatureT asquickaspossible.

• the problem of finding the worst shape of a thermal fin, by minimizing λ1(a) with respect to the functiona.

Finally,letusbriefly commentonthechoiceof theadmissiblesetof radiia.Wewillimpose:

(i) a regularity assumption, namely a W1,(0,), to guarantee that the surface element be defined almost everywhere.

(ii) a pointwise lower bound assumption that prevent the fin to collapse: there exists a0 > 0 such that a(x) a0 for every x [0,]. Moreover, to consider a class of shapes as large as possible, we will choosea0 suitablysmall(theprecisesense oftheword“small” willbe madeexplicitinthestatement ofthemain theoremsofthispaper).

(iii) a global lateralsurfaceassumption,to modelalimitation onthe manufacturingcost. Moreprecisely, weassumeanupperboundonthelateralsurfaceofΩa,thatisgivenby

lateral surface of Ωa= 2π

0

a(x)

1 +a(x)2dx.

Inthenextsection,wesum-up thepreviousconsiderationsandstatetheextremalproblemswewill solve.

(7)

2. SolvingofProblem(4) (lookingforthe worstshape)

2.1. Main results

Thissectionisdevotedto theinvestigationof Problem(4).

As highlighted in [14, Lemma 3.1], the class Sa0,,S0 does not share nice compactness properties. In particular, it is not closed nor bounded in W1,∞(0,) (endowedwith the strong topology),whereas it is boundedinL(0,).Thisdrives ustointroduceanewoptimaldesignprobleminasubclassofW1,(0,) enjoyinggoodtopological properties.

Tothisaim, letM > a0 andletusdefine thetruncated class SaM0,,S0=

a∈W1,(0, ) satisfying (H1), (H2) anda

1 +a2≤M a.e. in (0, )

. (13) SinceSa0,,S0 isaboundedsetofL(0,),itiseasytoseethatSaM0,,S0 isboundedandclosedinW1,∞(0,).

Inparticular,itinheritsusefulcompactnesspropertiesinaweaksensethatwillbemadepreciseinthesequel.

Inthefollowingtheorem,weinvestigatetheminimizationofa→λ1(a) overtheclassSaM0,,S0.

Theorem1.Letα,β,δ,a0, andS0 be positivereal numberssuchthat S0> a0and σ≥0.The constant functiona(·)=a0 minimizesthefunctional λ1 over theclassSaM0,,S0.

Asaconsequence,we inferthefollowingresultfortheoriginalproblem (4).

Corollary1.Foreverypositiverealnumbersα,β,δ,everypositivenumbersa0,andS0suchthatS0> a0, andσ≥0,theconstant functiona(·)=a0 istheunique solutionof theextremalproblem (4).

This result is quite natural from a physical point of view: in order to cool the material as slowly as possible along a fin with prescribed lateral surface, one needs to use a very long fin, so that the spatial temperaturedecays,and thusthelateralheattransferisverysmooth.

The approachused to provethe results aboverests upon the use of aparticular perturbation that we willintroduceinSection2.2.TheproofsofTheorem 1and Corollary 1arethen gatheredinSection2.3.

Remark2.Theoptimal valueofthe functionλ1(·) can be explicitlycomputed. Forexample, assumethat a0<

βδ σ

1/3

.Recallthat,formodelingreasons,suchasmallnessassumptionona0isofparticularinterest intheframeworkofourstudy,as underlinedinSection1.2.Hence,accordingto(1) andconsideringϕ≡1 asatestfunction,weclaimthat

a0λ1(a0)−β= min

ϕH1(0,) ϕ=0

αa30

0ϕ(x)2dx+σa30ϕ()2−βδϕ(0)2 a20

0ϕ(x)2dx+δϕ(0)2 ≤σa30−βδ a20+δ <0.

Astraightforwardcomputationleadstothefollowing expressionoftheassociatedeigenfunction ϕ1,a0(x) =A

cosh(ω1(a0)x) λ1(a0)

γa20ω1(a0)sinh(ω1(a0)x)

withω1(a0)2= β−λ1(a0)a0

αa0 ,whereAdenotesthenormalizationconstantforthenorm·a0.Theboundary conditionatx=yieldsthat,λ=λ1(a0) isthefirstpositiverootofthetranscendentalequation

(8)

Fig. 2.Construction ofλ1(a0).

λ

γa20ω =αωsinh(ω) +σcosh(ω)

αωcosh(ω) +σsinh(ω) with ω2= β−λ1(a0)a0 αa0

.

Theconstructionofλ1(a0) isillustratedonFig. 2.Noticethatanapproximatedvalueoftheeigenvaluecan be easily obtained numerically,by solvingfor instance thetranscendental equation abovewith a Newton method.

2.2. Akey technicallemma

Thissectionisdevotedtothedescriptionofparticularperturbationsthatwewillusetosolveatthesame time theproblems ofminimizingand maximizingλ1over theclass SaM0,,S0.

Let ε>0, S0≥a0, a∈ SaM0,,S0 suchthata(·)≡a0, andset b =a√

1 +a2. Letus introducethe two familiesofperturbationswewilluseinthesequel.Theirconstructionisbasedonthestraightforwardclaim holding,uptoanull(Lebesgue)measureset:

(0, ) ={b=M} ∪ {a=a0} ∪ {a0< b < M} since {b=a0}={a=a0}.

Perturbation of type I (worst shape) Assumethatthefunctiona isnotidenticallyequalto a0. It follows thattheset {a0+c< b≤M}is ofpositivemeasure forsomec>0.We willthen considerthe particular perturbation ofboftheform

bε=b−cχVx

0(ε),

wherex0denotesaLebesguepointoftheset{a0+c< b≤M}andVx0(ε)={a0< b≤M}∩(x0−ε,x0+ε), and

0bε(x)dx≤S0.

Perturbation oftypeII(bestshape) AssumethatM islarge,morepreciselythatM > S0.Therefore,one hasnecessarilyb≡M inthesensethatthemeasureoftheset {b=M}isstrictly lowerthan.

Assume that the set {a0 < b < M} has a positive measure. We could then take c > 0 such that {a0+c≤b≤M−c}hasapositivemeasure andconsidertheparticularperturbationsofb oftheform

bε=b+c

χVx0(ε)−rεχVy0(ε)

,

where c>0,x0isaLesbeguepointof{a0≤b≤M−c},y0=x0 isaLesbeguepointof{a0+c< b≤M}, themeasurable setsVx0(ε) andVy0(ε) aredefinedby

(9)

Vx0(ε) ={a0≤b < M−c} ∩(x0−ε, x0+ε), Vy0(ε) ={a0+c < b≤M} ∩(y0−ε, y0+ε), with ε is small enough so thatthese two sets do not intersect, and rε := |V|Vx0(ε)|

y0(ε)|. The Lebesgue density theorem yieldsthatlimε0rε= 1.Obviously,sincea(·)≡a0,there holdsa0≤bε≤M almosteverywhere in(0,).

Using a choice of function bε as above, we will now construct a perturbation aε of a within the class SaM0,,S0.Thisisthecontentofthefollowing lemma:

Lemma2.Letbε∈L(0,)beaperturbationofbeither oftypeI,orII.Then,thereexistsafamily(aε)ε>0 suchthat

aε∈ SaM0,,S0 forevery ε>0,

aε

1 +aε2=bε almosteverywherein (0,)andforevery ε>0,

onehasthereminderestimate:aε−aL(0,)≤Cε2,wheretheconstantC onlydependsonM,c and theconstantsα,β,δ,a0,,S0 andσ.

The statementof this lemma is close to [14, Lemma 3.4]. Nevertheless, anotable difference lies in the factthatwehavetodealwithperturbationsofbthatarethesumsofcharacteristicfunctionsofmeasurable sets, insteadofopen sets.

Letusprovideaqualitativeinterpretationofthislemma.Theperturbationaεcanbeseenasaninfinites- imal perturbation(in L)oftheoriginal element a. Theoscillationscreatedonthegraphof aεaremade sothatthelateralsurfaceelementbεisanapproximationoftheDiracmeasureatthefirstorder.Themain difficultyconsistsinbuilding thefunctionaε insuchaway thatitisanadmissibleelementofSaM0,,S0. Proof. The constructions of the function aε satisfying the aforementioned assertions when bε is eitherof type I,or II are quite close. Nevertheless, since the case of aperturbation of typeII requires a little bit moretechnicity,wefocusonitinthisproof.Thecontentoftheproofcanthenbeeasilyadapted(andeven simplified)to deducetheconstructionofaεforperturbationsoftype I.

Firststep:casewherebissmooth Letusassumethatb∈C([0,]) andletusnowdescribetheconstruc- tionofaε.WewillconsiderhereageneralconstantM > a0+ 2csuchthat{a0≤b< M−c}hasapositive Lebesguemeasure.

Letusfirstset

aε(x) =a(x) if x /∈ Vx0(ε)∪ Vy0(ε).

Withoutlossofgenerality,wewillfocuswithinthisproofonthecharacterizationoftheperturbationaε−a onVx0(ε),thedefinitionofaε−aonVy0(ε) beingsimilar (inabsolutevalue).

Hence,letus defineaε onVx0(ε). Sinceb iscontinuous, thisset isafinite unionofopen intervalsIi, in otherwords

Vx0(ε) ={x∈(x0−ε, x0+ε) s.t.b(x)< M−c}=Ni=1Ii.

ThedifficultyliesincontrollingtheLdistance betweenaandaε. Thealgorithmicproceduretodefine aεwrites asfollows:

i oneconsidersaregularsubdivisionofIi intok intervalsoflengthη,withη <min{c2/2M,ε2}. ii oneverysubinterval(¯x,x¯+η) ofthissubdivision,onecreatesoneoscillationbysetting

(10)

aε=aη,2 on (¯x, ξ) and aε=aη,1 on (ξ,x¯+η), wherethefunctionsaη,1andaη,2satisfyinparticularaη,i

1 +aη,i2 =bεfori∈ {1,2},aη,1isdecreasing andaη,2 isincreasing.

Moreprecisely,wedefinethefunctionaη,2 asasolutionof theCauchyproblem aη,2(x) =

bε(x)2−aη,2(x)2

aη,2(x) x∈x,x¯+η) aη,2x) =a(¯x),

andthefunctionaη,1as asolutionoftheCauchyproblem

aη,1(x) =bε(x)aη,12−a(x)η,1(x)2 x∈x,x¯+η), aη,1x+η) =a(¯x+η),

whereξ∈x,x¯+η) ischosensothataη,1(ξ)=aη,2(ξ).

Itremainstoverifythatsuchaconstructionispossible.First,|(a2η,2)|= 2aη,2|aη,2|≤2M sincebε≤M, whichyields

|a2η,2(x)−a2(x)| ≤2M|x−x| ≤2M η

for all x∈ [x,x+η) such that aη,2(x) is defined. Moreover, as bε = b+c on (¯x,x¯+η), this function is smoothand onehas

b2ε(x)−a2(x) =

c+a(x)

1 +a(x)22

−a2(x)≥c2.

Hence, b2ε(x)−a2η,2(x)≥c22M η > 0 due to ourchoice forη. Applying the Cauchy–Lipschitz theorem yields that aη,2 is uniquelydefined at least on[x,x+η). The samearguments yield thataη,1 is uniquely definedon(x−η,x].

We now need to check that the graphsof aη,1 and aη,2 intersect at apoint whose abscissa belongs to [¯x,x¯+η],whichcomesto

aη,2x+η)≥a(¯x+η) and aη,1x)≥a(¯x).

accordingtotheintermediatevaluetheorem.

Letus showthataη,i≥aeverywherein(¯x,x¯+η] fori∈ {1,2}. Becauseofthesymmetricaldefinitions of aη,1 and aη,2, it suffices to prove this factfor i = 2. Assume by contradiction and by continuity of a and aη,2 theexistenceofxη x,x¯+η] andδη >0 suchthata(x)> aη,2(x) on(xη,xη+δη). Then,since bε=b+c onthis interval,weclaimthat

aη,2(x) =

bε(x)2 aη,2(x)2 1>

b(x)2

a(x)2 1 =a(x), foralmost everyx∈(xη,xη+δη).Integratingbothsidesofthisinequalityleadsto

xηx

xη

aη,2(x)dx >

xηx

xη

a(x)dx,

(11)

Fig. 3.Left: Zoom on one oscillation. Right: the perturbationaε.

foreveryδx(0,δη),whichrewritesaη,2(xηx)> a(xηx) sinceaη,2(xη)=a(xη).Thisisincontradiction withtheassumptionaboveandprovesthatforeveryx∈x,x+¯ η],thereholdsaη,2(x)≥a(x).Thisjustifies thedefinitionabove.TheconstructionofaεisillustratedonFig. 3.

Moreover,thelateralsurfaceconstraintremainssatisfiedbyaεsince

0

bε(x)dx= 0

b(x)dx+c(|Vx0(ε)| −rε|Vy0(ε)|)≤S0,

andobviously bε≤b≤M on(0,).Thus,there holdsaε∈ SaM0,,S0.

Atthisstep,thefunctionaεdefinedaspreviouslysatisfiesthetwofirstassertionsofthelemma.Itremains nowto estimatetheL-norm ofaε−a. Since

1

2(aε(x)2−aεx)2) = x

¯ x

aε(s)aε(s)ds≤ x

¯ x

bε(x)dx

and 1

2(a(x)2−a(¯x)2) = x

¯ x

a(s)a(s)ds≤ aL(0,)aL(0,)

foreveryx∈x,x¯+η),there holds

a(x)2≤aε(x)22

¯ x+η

¯ x

bε(x)dx+a(¯x)2

2η(b+cL(0,)+ 2aL(0,)aL(0,)) +a(x)2,

(12)

foreveryx∈x,x¯+η]. Moreover,sincea∈W1,∞(0,) andaccordingto(H1),onehas 0≤aε(x)−a(x)≤ (b+cL(0,)+aL(0,)aL(0,))

a0 ε2

for everyx∈x,x¯+η].As aL(0,)≤M/a0 and, accordingto Lemma 2in[14], aL(0,) is bounded by aconstantwhichonlydepends onS0 and ,it followsthataε−aLx,¯x+η)≤Cε2, foraconstantC onlydepending onM,candtheconstants ofourproblem.

Theexpectedconclusionthen followsinthecasewhereb issmooth.

The general case Byusingstandarddensitytheorems, thereexistsasequence(an)n∈IN inC([0,]) con- vergingto aintheSobolevspace W1,1(0,). Letusintroducebn =an

1 +an2.Weclaimthat,using the previous assumptionsonaand b, itisnotrestrictive toassume (forexamplebyconsideringconvolutions) thatan ≥a0in[0,] andthat|an|≤M/a0.Hence,(an)n∈INand(bn)nareboundedinL(0,) respectively bytwopositiveconstantsC >˜ 0 accordingto[14,Lemma3.1]andM˜.Therefore,stilldenotingwithaslight abuseofnotationby(an)nINanyextractedsubsequenceof(an)nIN,theArzelà–Ascolitheoremyieldsthat (an)nINconvergesto ainL(0,).

First,(bn)nINconverges tobinL1(0,).Indeed,thereholds

0

|bn(x)−b(x)|dx= 0

an(x)

1 +an2(x)−a(x)

1 +a(x)2dx

0

(an(x)2−a(x)2)(1 +an(x)2) bn(x) +b(x)

dx

+

0

a(x)2(an(x)2−a(x)2) bn(x) +b(x)

dx

(1 +M/a0)(C+aL(0,))

2a0 an−aL(0,)

+a2L(0,)(M/a0+aL(0,))

2a0 an−aL1(0,),

sincebn≥a0andb≥a0a.e.in(0,),andtheright-handsideconvergesto0 asn→+.Replacing(bn)n∈IN byawell-chosenextractedsubsequence,wecanthusassumethat(bn)n∈INconvergestobalmosteverywhere in (0,) and, thus, only consider Lebesgue points x0,y0 such thatbn(x0) →b(x0) and bn(y0) →b(y0) as n→+. Hence,themeasure ofthesets

Vxn0(ε) ={x∈(x0−ε, x0+ε), s.t.bn(x)< M −c} and Vyn0(ε) ={x∈(y0−ε, y0+ε), s.t.bn(x)> a0+c} are positivewhenevernislargeenough.

Let us now apply theconstruction of thefirst step to theelements an. It follows thatfor everyε>0, there exists(an,ε)n∈INsuchthat

bn,ε:=an,ε

1 +an,ε2 =bn+c

χVxn

0(ε)−|Vxn0(ε)|

|Vyn0(ε)Vyn

0(ε)

(14)

(13)

and

0≤an,ε(x)−an(x) bn+cL(0,)+anL(0,)anL(0,)

a0/2 ε2 M˜ +c+MC/a˜ 0

a0/2 ε2 (15) for everyx [0,] and n IN.In particular, for a given n∈ IN,the family (an,ε)ε>0 converges to an as ε0.

Moreover,byconstruction,(an,ε)ε>0 isauniformlyLipschitzfunctionsfamilywithrespectto n(andε) andaccordingtotheArzelà–Ascolitheorem,itconvergesuptoasubsequencetoafunctionaε∈W1,(0,).

Lettingntendto+in(15)yieldsthat

0≤aε(x)−a(x)≤M˜ +c+MC/a˜ 0

a0/2 ε2 foreveryx∈[0,].Inparticular, itfollowsthataε≥a≥a0in[0,].

Next,since(bn)nIN convergestob inL1(0,),onegets

n→+∞lim bn,ε=b+c

χVx0(ε)−rεχVy0(ε)

=bε in L1(0, ), bypassingtothelimitin(14)

Moreover,usingthesamedecompositionasabove,weclaimthat

(an,ε)2(aε)2= (bn−b)(bn,ε+bε)(a2n,ε−a2ε)(1 +an,ε2 ) a2ε

for almost every x (0,)∩ {an,ε = aε}. Using the C0-boundedness of the families (an,ε)n∈IN,ε>0, (bn,ε)n∈IN,ε>0, thestrong C0 convergence of(an,ε)n∈IN to aε andthe L1-convergenceof (bn)n∈IN to b, one gets that(an,ε)2 converges to (aε)2 inL1(0,). Therefore, using the samereasonings as above,it follows:

that(bn,ε)nINconvergesto bε:=aε

1 +aε2inL1(0,).

Theproofofthelemmaisthen complete. 2 2.3. Proofs ofTheorem 1 andCorollary 1

Proof of Theorem1 Wefirstinvestigatethe existenceofaminimizer withintheclassSaM0,,S0.Inviewof that,wewillneedthefollowingstraightforwardlemma.

Lemma3.Let(un)n∈IN be asequenceof L2(0,)converging tosomefunctionuweaklyin L2(0,).Assume moreover that(un)n∈IN isboundedinL(0,).Then,

1 +u2n belongstoL2(0,)forevery n∈IN andthe sequence

1 +u2n

nINconvergesweaklyinL2(0,)toafunctionU satisfying

1 +u2≤U a.e.in(0,).

Proof. Even if this result is straightforward, we nevertheless provide elements of proof for the sake of completeness. Since

1 +u2n 1+|un| a.e.in (0,), thefirst claim follows. By assumption,there exists u >0 suchthat|un|≤ua.e.in(0,),foreveryn∈IN.Noticemoreoverthatthefunctionalv→√

1 +v2is convex andcontinuous forthe strongL2-topologyontheset Uu={v ∈L2(0,)| v ≤u}.Indeed, the convexityisobvious andthecontinuity isobtainedbyconsideringasequenceofUu denoted(vn)nIN that convergesstronglyinL2 toafunctionv,andbywriting

0

1 +vn2

1 +v22

dx= 0

(vn−v)2(vn+v)2 1 +vn2+

1 +v2

2dx≤u2vn−v2L2.

Références

Documents relatifs

Optical measurement of the hyperfine splitting of the 1D2 metastable state of Pr3+ in LaF3 by enhanced and saturated

Key words: Convective heat transfer coefficient, correlation, phase change material, heat storage system, transient forced convection, numerical

- Déterminer la motivation des parents et les éléments déclencheurs qui favoriseraient une participation parentale dans un programme de prise en charge du

In the present study, the spanwise-averaged absolute streamwise vorticity flux is computed only with the longitudinal vortex component, and is used for the discussion on the

Through this work the following contributions were made: (1) the fifty freedom and constraint space types were found that may be used to synthesize both parallel

metagenomes from planted soil showed that ryegrass rhizosphere selected for an active 444. population with specific functions compared to bare

The question is bound to arise with new acuteness: are so many arts and economics graduates really necessary, when Japan, like Western industrial countries, is

Figure 5 displays the profiles of the two-point correlations ρ vv (y) and the length-scale- based resolution indicators φ y (l) in the wall-normal direction computed for the