Contents lists available atScienceDirect
Journal de Mathématiques Pures et Appliquées
www.elsevier.com/locate/matpur
Optimal location of resources maximizing the total population size in logistic models
✩,✩✩Idriss Mazaria, Grégoire Nadinb, Yannick Privatc,∗
a SorbonneUniversités,UPMCUnivParis06,UMR7598,LaboratoireJacques-LouisLions,F-75005, Paris,France
bCNRS,SorbonneUniversités,UPMCUnivParis06,UMR7598,LaboratoireJacques-LouisLions, F-75005,Paris,France
cIRMA,UniversitédeStrasbourg,CNRSUMR7501,7rueRenéDescartes,67084Strasbourg,France
a r t i cl e i n f o a b s t r a c t
Articlehistory:
Received16May2018 Availableonlinexxxx
MSC:
49K20 35Q92 49J30 34B15
Keywords:
Diffusivelogisticequation Rearrangementinequalities Symmetrization
Optimalcontrol Shapeoptimization Optimalityconditions
Inthis article, weconsider a species whose populationdensity solvesthe steady diffusivelogisticequationinaheterogeneousenvironmentmodeledwiththehelpof aspatiallynonconstantcoefficientstandingforaresourcesdistribution.Weaddress the issue of maximizing the total population size with respect to the resources distribution,consideringsomeuniformpointwiseboundsaswellasprescribingthe totalamountofresources.Byassumingthediffusionrateofthespecieslargeenough, weprovethatanyoptimalconfigurationisbang-bang(inotherwordsanextreme point oftheadmissible set)meaning thatthisproblem can berecastas a shape optimizationproblem,theunknowndomainstandingfortheresourceslocation.In theone-dimensionalcase, thisproblem isdeeplyanalyzed,andforlargediffusion rates,all optimalconfigurationsareexhibited.Thisstudyiscompletedby several numericalsimulationsintheonedimensionalcase.
©2019ElsevierMassonSAS.Allrightsreserved.
r é s u m é
Dans cet article, nous considérons une espèce dont la densité de population résoutl’équationlogistiquediffusivestationairedansunenvironnementhétérogène modélisé à l’aide d’un coefficient non constant en espace représentant une distributionderessources.Nousabordonslaquestiondelamaximisationdelataille totaledelapopulationparrapportàlarépartitiondesressources,enconsidérant des contraintes ponctuelles uniformes et en prescrivant la quantité totale des ressources. En supposant que le taux de diffusion de l’espèce est suffisamment grand, nous prouvons que toute configuration optimale est bang-bang (c’est-à- dire unpointextrêmaldel’ensemble admissible),ce quisignifie que ceproblème peut être transformé en problème d’optimisation de forme, le domaine inconnu représentantl’emplacementdesressources.Danslecasunidimensionnel,ceproblème estanalyséprécisément,ainsiquepourdesvitessesdediffusionélevées,toutesles
✩ TheauthorswerepartiallysupportedbytheProject“Analysisandsimulationofoptimalshapes–applicationtolifesciences”
oftheParisCityHall.
✩✩ I. Mazari andY. Privat were partially supportedby the ANR Project SHAPO (ANR-18-CE40-0013– SHAPO – SHAPE OPTIMIZATION(2018)).
* Correspondingauthor.
E-mailaddresses:idriss.mazari@sorbonne-universite.fr(I. Mazari),gregoire.nadin@sorbonne-universite.fr(G. Nadin), yannick.privat@unistra.fr(Y. Privat).
https://doi.org/10.1016/j.matpur.2019.10.008
0021-7824/©2019ElsevierMassonSAS.Allrightsreserved.
configurations optimales sont exhibées. Cette étude est complétée par plusieurs simulationsnumériquesdanslecasunidimensionnel.
©2019ElsevierMassonSAS.Allrightsreserved.
1. Introduction
1.1. Motivationsandstateof theart
Inthisarticle,weinvestigateanoptimalcontrolproblemarisinginpopulationdynamics.Letusconsider the populationdensityθm,μ ofagivenspeciesevolvinginabounded andconnected domainΩ in Rn with n ∈ N∗, having aC2 boundary.In whatfollows, wewill assumethat θm,μ is thepositive solutionof the steady logistic-diffusiveequation (denoted(LDE) inthesequel) whichwrites
μΔθm,μ(x) + (m(x)−θm,μ(x))θm,μ(x) = 0 x∈Ω,
∂θm,μ
∂ν = 0 x∈∂Ω, (LDE)
where m ∈L∞(Ω) stands for the resources distribution and μ >0 stands for thedispersal abilityof the species,alsocalleddiffusionrate.Fromabiologicalpointofview,therealnumberm(x) isthelocalintrinsic growthrateofspeciesatlocationxofthehabitatΩ andcanbeseenasameasureoftheresourcesavailable at x.
As will be explained below,we will only consider non-negative resourcedistributions m, i.e. such that m ∈L∞+(Ω) ={m∈L∞(Ω), m≥0 a.e}.Inviewof investigatingthe influenceofspatial heterogeneityon themodel,weconsider theoptimalcontrolproblemofmaximizing thefunctional
Fμ:L∞+(Ω)m→
Ω
θm,μ,
wherethenotationffl
denotestheaverageoperator,inotherwordsffl
Ωf = |Ω1|´
Ωf.ThefunctionalF stands for thetotal populationsize,inorder to furtherourunderstandingof spatial heterogeneityonpopulation dynamics.
In theframeworkofpopulation dynamics,thedensityθm,μ solvingEquation(LDE) canbe interpreted as asteadystateassociatedto thefollowingevolutionequation
⎧⎪
⎨
⎪⎩
∂u
∂t(t, x) =μΔu(t, x) +u(t, x)(m(x)−u(t, x)) t >0, x∈Ω
∂u
∂ν(t, x) = 0 t >0, x∈∂Ω
u(0, x) =u0(x)≥0, u0= 0 x∈Ω
(LDEE)
modelingthespatiotemporalbehaviorofapopulationdensityuinadomainΩ withthespatiallyheteroge- neousresourcetermm.
ThepioneeringworksbyFisher[1],Kolmogorov-Petrovski-Piskounov [2] and Skellam[3] onthelogistic diffusive equation weremainly concerned withthe spatiallyhomogeneous case.Thereafter, many authors investigated the influence of spatial heterogeneity on population dynamics and species survival. In [4], propagation properties ina patch model environment are studied. In [5], aspectral condition for species survivalinheterogeneousenvironmentshasbeenderived,while[6] dealswiththeinfluenceoffragmentation and concentrationof resources onpopulation dynamics.These workswerefollowedby [7] dedicatedto an optimaldesignproblem,thatwill becommentedinthesequel.
Investigating existence anduniquenessproperties ofsolutionsfor thetwo previousequations as well as theirregularity propertiesboils downtothestudyofspectral propertiesforthelinearizedoperator
L:D(L)f →μΔf+mf,
wherethedomainofLisD(L)={f ∈L2(Ω)|Δf ∈L2(Ω)}andofitsfirsteigenvalueλ1(m,μ),character- izedbytheCourant-Fischerformula
λ1(m, μ) := sup
f∈W1,2(Ω),´
Ωf2=1
⎧⎨
⎩−μ ˆ
Ω
|∇f|2+ ˆ
Ω
mf2
⎫⎬
⎭. (1) Indeed,thepositivenessofλ(m,μ) isasufficientconditionensuringthewell-posednessofequations(LDEE) and (LDE) ([7]). Then, Equation(LDE) has auniquepositivesolution θm,μ ∈W1,2(Ω).Furthermore,for anyp≥1,θm,μ belongstoW2,p(Ω),and thereholds
0<inf
Ω
θm,μ≤θm,μ≤ mL∞(Ω). (2)
Moreover, thesteady stateθm,μ is globallyasymptotically stable:for anyu0 ∈W1,2(Ω) such thatu0 ≥0 a.e.inΩ andu0= 0,onehas
u(t,·)−θm,μL∞(Ω)t→+∞−→ 0,
where udenotes the unique solutionof (LDEE) with initial state u0 (belonging to L2(0,T;W1,2(Ω)) for everyT >0).
The importance of λ1(m,μ) forstability issues related to population dynamics models was first noted in simple cases by Ludwig, Aronson and Weinberger [8]. Let us mention [9] where the case of diffusive Lotka-Volterraequationsisinvestigated.
To guarantee that λ1(m,μ)> 0, it is enough to work with distributions of resources m satisfying the assumption
m∈L∞+(Ω) where L∞+(Ω) =
⎧⎨
⎩m∈L∞(Ω), ˆ
Ω
m >0
⎫⎬
⎭. (H1) Notethattheissueofmaximizingthisprincipaleigenvaluewasaddressedforinstancein[10–14].
Inthe survey article[15], Lou suggeststhefollowing problem: theparameter μ>0 beingfixed, which weightmmaximizesthetotalpopulation sizeamong alluniformly boundedelementsof L∞(Ω)?
Inthis article, weaim at providingpartial answersto this issue,and moregenerallynew resultsabout theinfluenceof thespatial heterogeneitym(·) onthetotalpopulationsize.
Forthatpurpose,letusintroducethetotalpopulation sizefunctional,definedforagivenμ>0 by Fμ:L∞+(Ω)m−→
Ω
θm,μ, (3)
whereθm,μ denotesthesolutionofequation(LDE).
Letus mentionseveral previousworksdealingwiththemaximizationof thetotalpopulation size func- tional.Itisshownin[16] that,amongallweightsmsuchthatffl
Ωm=m0,thereholds Fμ(m)≥ Fμ(m0)= m0; this inequality is strict whenever m is nonconstant. Moreover, it is also shown that the problem of maximizingFμ overL∞+(Ω) has nosolution.
Remark 1.The fact that m ≡ m0 is a minimum for Fμ among the resources distributions m satisfying ffl
Ωm=m0relies onthefollowingobservation:multiplying(LDE) by θ1
m,μ and integratingbypartsyields μ
Ω
|∇θm,μ|2 θm,μ2 +
Ω
(m−θm,μ) = 0 (4)
and therefore, Fμ(m)=m0+μ´
Ω
|∇θm,μ|2
θm,μ2 ≥m0 =Fμ(m0) forall m∈ L∞+(Ω) such thatffl
Ωm =m0. It follows thattheconstantfunctionequaltom0isaglobalminimizerofFμ over{m∈L∞+(Ω),ffl
Ωm=m0}. Intherecentarticle[17],itisshownthat,whenΩ= (0,),onehas
∀μ >0, ∀m∈L∞+(Ω)|m≥0 a.e.,
Ω
θm,μ ≤3
Ω
m.
This inequalityissharp, althoughtheright-handsideisneverreached,andtheauthorsexhibit asequence (mk,μk)k∈N suchthatffl
Ωθmk,μk/ffl
Ωmk→3 ask→+∞,butforsuchasequencethereholdsmkL∞(Ω)→ +∞andμk →0 ask→+∞.
In[15],itisprovedthat,withoutL1 orL∞boundsontheweightfunctionm,themaximizationproblem is ill-posed. It isthus naturalto introducetwo parametersκ,m0 >0,and to restrict ourinvestigation to theclass
Mm0,κ(Ω) :=
⎧⎨
⎩m∈L∞(Ω),0≤m≤κa.e ,
Ω
m=m0
⎫⎬
⎭. (5) It isnotablethatin[18], themoregeneralfunctionalJB definedby
JB(m) = ˆ
Ω
(θm,μ−Bm2) forB ≥0
is introduced.Inthe caseB= 0, theauthorsapply theso-calledPontryagin principle,showtheGâteaux- differentiabilityofJB andcarryoutnumericalsimulationsbackinguptheconjecturethatmaximizersofJ0 over Mm0,κ(Ω) areofbang-bangtype.
However, proving this bang-bang property is a challenge. The analysis of optimal conditions is quite complex, because the sensitivity of the functional “total population size” with respect to m(·) isdirectly related to thesolution of anadjoint state, solvinga linearizedversion of (LDE).Deriving and exploiting the propertiesofoptimal configurationstherefore requires athoroughunderstandingof theθm,μ behavior as wellastheassociatedstate.Todothis,weareintroducinganewasymptoticmethodto exploitoptimal conditions.
Wewill investigatetwo propertiesofthemaximizersofthetotalpopulationsize functionFμ.
1. Pointwiseconstraints.Themainissuethatwillbeaddressedinwhatfollowsisthebang-bangcharacter ofoptimalweightsm∗(·),inotherwords,whetherm∗ isequalto0orκalmosteverywhere.Notingthat Mm0,κ(Ω) isaconvexset andthatbang-bang functionsare theextreme pointsofthisconvexset,this questionrewrites:
Arethemaximizersm∗ extremepointsof thesetMm0,κ(Ω)?
Fig. 1.Ω = (0,1)2. The left distribution is “concentrated” (connected) whereas the right one is fragmented (disconnected).
Inourmainresult(Theorem1)weprovideapositiveanswerforlargediffusivities.Itisnotablethatour proof restsuponawell-adaptedexpansionofthesolutionθm,μ of(LDE) withrespectto thediffusivity μ.
This approachcould be consideredunusual, since such resultsare usually obtainedby an analysis of the propertiesof theadjointstate (orswitching function). However,sincethe switchingfunction very implicitly dependsonthedesignvariablem(·),wedidnotobtainthisresultinthis way.
2. Concentration-fragmentation. Itis wellknownthatresourceconcentration(whichmeansthatthedis- tribution of resourcesm decreasesinalldirections,see Definition2foraspecific statement)promotes the survival of species[7]. On the contrary, we will saythat aresource distributionm =κχE, where E isasubsetofΩ,isfragmented whentheE set isdisconnected.IntheFig.1, Ω isasquare,and the intuitivenotionofconcentration-fragmentationofresourcedistributionisillustrated.
A naturalissuerelatedto qualitativepropertiesofmaximizersisthus Aremaximizersm∗ concentrated? Fragmentated?
InTheorem2,weconsiderthecaseofaorthotopeshapehabitat,andweshowthatconcentrationoccurs forlargediffusivities:ifmμmaximizesFμoverMm0,κ(Ω),thesequence{mμ}μ>0stronglyconvergesin L1(Ω) toaconcentrateddistributionas μ→ ∞.
In the one-dimensional case, we also prove thatif the diffusivity is large enough, there are only two maximizers, thatareplottedin Fig.2(seeTheorem 3).
Finally, inthe one-dimensionalcase, weobtain asurprising result:fragmentation may bebetter than concentrationforsmalldiffusivities(seeTheorem 4andFig.3).
This issurprising becauseinmanyproblemsofoptimizingthelogistic-diffusiveequation,itisexpected thatthebestdispositionofresourceswillbe concentrated.
1.2. Main results
Inthewholearticle, thenotation χI will be used todenote the characteristicfunction ofameasurable subsetI ofRn,inotherwords, thefunctionequalto1inI and0elsewhere.
For the reasons mentioned inSection 1.1, it is biologically relevant to consider theclass of admissible weights Mm0,κ(Ω) defined by (5), where κ > 0 and m0 > 0 denote two positive parameters such that m0< κ (sothatthis setisnontrivial).
Wewillhenceforthconsiderthefollowing optimaldesignproblem.
Fig. 2.Ω = (0,1). Plot of the only two maximizers ofFμoverMm0,κ(Ω).
Fig. 3.Ω = (0,1). A double crenel (on the left) is better than a single one (on the right).
Optimaldesign problem. Fix n ∈N∗, μ >0, κ>0, m0 ∈ (0,κ) and let Ω be a bounded connected domain ofRn having aC2 boundary. Weconsidertheoptimization problem
sup
m∈Mκ,m0(Ω)Fμ(m). (Pμn)
As will be highlighted in thesequel, the existence of amaximizer follows from adirect argument. We will thusinvestigatethequalitativepropertiesofmaximizersdescribed intheprevioussection(bang-bang character,concentration/fragmentationphenomena).
Forthesakeofreadability,almostalltheproofsarepostponedtoSection2.
Let us stress that the bang-bang character of maximizer is of practical interest in view of spreading resources inanoptimalway.Indeed,inthecasewhereamaximizerm∗ writesm∗=κχE,thetotalsize of population ismaximizedbylocatingalltheresourcesonE.
1.2.1. Firstproperty
Westart withapreliminaryresultrelated tothe saturationofpointwise constraintsforProblem (Pμn), valid for all diffusivities μ. It is obtained by exploitingthe first order optimality conditions for Problem (Pμn),writtenintermsofanadjointstate.
Proposition1.Letn∈N∗,μ>0,κ>0,m0∈(0,κ).Letm∗ beasolutionof Problem (Pμn).Then, theset {m=κ}∪ {m= 0}hasapositive measure
1.2.2. Thebang-bangproperty holdsforlarge diffusivities
For large values of μ, we will prove that the variational problem can be recast in terms of a shape optimizationproblem,as underlinedinthenextresults.
Theorem 1.Letn ∈N∗,κ>0,m0 ∈ (0,κ). Thereexists a positivenumber μ∗ =μ∗(Ω,κ,m0) such that, forevery μ≥μ∗,thefunctional Fμ isstrictly convex.Asaconsequence, forμ≥μ∗,any maximizerof Fμ
overMm0,κ(Ω) (orsimilarlyany solution ofProblem (Pμn))ismoreover of bang-bangtype.1
WeemphasizethattheproofofTheorem1is quiteoriginal, sinceitdoesnotrestupontheexploitation ofadjointstateproperties,butupontheuseofapowerseriesexpansionsinthediffusivityμofthesolution θm,μof(LDE),aswellastheirderivativewithrespecttothedesignvariablem.Inparticular,thisexpansion isusedtoprovethat,ifμislargeenough,thenthefunctionFμ isstrictlyconvex.Sincetheextremepoints ofMm0,κ(Ω) arebang-bangresourcesfunctions,theconclusionreadilyfollows.
Thistheorem justifiesthatProblem(Pμn) canbe recast asashapeoptimizationproblem.Indeed,every maximizerm∗ isoftheform m∗=κχE whereE isameasurablesubsetsuchthat|E|=m0|Ω|/κ.
Remark2.Wecanrewritethis resultintermsofshapeoptimization,byconsideringasmain unknownthe subset E of Ω where resources are located: indeed, under the assumptions of Theorem 1, there exists a positivenumberμ∗=μ∗(Ω,κ,m0) suchthat,foreveryμ≥μ∗, theshape optimizationproblem
sup
E⊂Ω,|E|=m0|Ω|/κFμ(κχE), (6) where thesupremum istaken overallmeasurable subset E ⊂Ω such that|E|=m0|Ω|/κ,has asolution.
We underline the fact that, for such a shape functional, which is “non-energetic” in the sense that the solutionof thePDEinvolved cannotbe seenas aminimizerof thesamefunctional,proving theexistence ofmaximizersisusuallyintricate.
1.2.3. Concentrationoccurs forlargediffusivities
Inthissection,westatetworesultssuggestingconcentrationpropertiesforthesolutionsofProblem(Pμn) mayholdforlargediffusivities.
Forthatpurpose,letusintroducethefunctionspace
X :=W1,2(Ω)∩
⎧⎨
⎩u∈W1,2(Ω),
Ω
u= 0
⎫⎬
⎭ (7)
andtheenergyfunctional
1 Inotherwords,itis,anelementofMm0,κ(Ω) equala.e.to0orκinΩ.
Em:X u→ 1 2
Ω
|∇u|2−m0 Ω
mu. (8)
Theorem 2.[Γ-convergenceproperty]
1. LetΩbe adomain witha C2 boundary.Forany μ>0,letmμ bea solutionof Problem (Pμn).Any L1 closure pointof {mμ}μ>0 asμ→ ∞ isasolution oftheoptimaldesignproblem
m∈Mminm0,κ(Ω)min
u∈XEm(u). (9)
2. In the case of atwo dimensional orthotope Ω= (0;a1)×(0;a2), any solution of theasymptotic opti- mization problem(9)decreases in everydirection.
As willbeclearintheproof,thistheorem isaΓ-convergence property.
Intheone-dimensionalcase,onecanrefinethisresultbyshowingthat,forμlargeenough,themaximizer is astepfunction.
Theorem 3.Letusassume thatn= 1 andΩ= (0,1).Letκ>0,m0∈(0,κ).Thereexistsμ >ˆ 0suchthat, forany μ≥μ,ˆ anysolution m ofProblem (Pμn)isequala.e.toeither m˜ orm(1˜ − ·),where m˜ =κχ(1−,1)
and =m0/κ.
1.2.4. Fragmentationmay occurforsmall diffusivities
Let us concludeby underlining thatthe statementof Theorem 3cannot be truefor allμ >0.Indeed, weprovideanexampleinSection3.1whereadouble-crenelgrowthrategivesalargertotalpopulationsize than thesimplecrenelm˜ ofTheorem3.
Theorem 4.The function m˜ =κχ(1−,1) (and m(1˜ − ·)=κχ(0,)) does not solve Problem (Pμn) for small valuesof μ.Moreprecisely,if weextendm˜ outside of(0,1) by periodicity,there existsμ>0suchthat
Fμ
m(2˜ ·)
>Fμ( ˜m).
This result is quite unusual.For the optimization of the first eigenvalue λ1(m,μ) defined by (1) with respect tomontheinterval(0;1)
sup
m∈M((0;1))λ1(m, μ)
we know(see[7])thattheonlysolutionsarem˜ andm(1˜ − ·),foranyμ.
It isnotablethatthefollowing resultisabyproductofTheorem 4above.
Corollary 1.Thereexistsμ>0suchthat theproblems sup
m∈M((0;1))λ1(m, μ) and
sup
m∈M((0;1))Fμ(m) do not havethesamemaximizers.
Forfurthercommentsontherelationshipbetweenthemaineigenvalueandthetotalsizeofthepopulation, wereferto[19],whereanasymptoticanalysisofthemaineigenvalue(relativetoμasμ→+∞)isperformed, andthereferencestherein.
Weconcludethissectionbymentioningtherecentwork[20],thatwasreportedtouswhenwewrotethis article.Theyshowthat,ifweassumetheoptimaldistributionofregularresources(moreprecisely,Riemann integrable),thenitisnecessarilyofbang-bangtype.Theirproofisbasedonaperturbationargumentvalid forallμ>0.However,provingsuchregularityisgenerallyquitedifficult.Ourproof,althoughitisnotvalid forallμ,isnotbasedonsucharegularityassumption,butthesetwocombinedresultsseemtosuggestthat allmaximizersofthisproblemare ofbang-bangtype.
1.3. Tools andnotations
Inthissection,wegathersomeusefultoolswewilluseto provethemainresults.
Rearrangementsoffunctionsandprincipaleigenvalue Letusfirstrecallseveralmonotonicityandregularity relatedtotheprincipaleigenvalueoftheoperatorL.
Proposition2.[9] Let m∈L∞+(Ω) andμ>0.
(i) Themapping R∗+μ→λ1(m,μ)iscontinuous andnon-increasing.
(ii) Ifm≤m1,then λ1(m,μ)≤λ1(m1,μ),andtheequality istrueif,and onlyif m=m1 a.e.in Ω.
In the proof of Theorem 3, we will use rearrangement inequalities at length. Let us briefly recall the notionofdecreasingrearrangement.
Definition1.Foragivenfunctionb∈L2(0,1),onedefinesitsmonotonedecreasing(resp.monotoneincreas- ing)rearrangementbdr (resp.bbr)on(0,1) bybdr(x)= sup{c∈R|x∈Ω∗c},whereΩ∗c = (1− |Ωc|,1) with Ωc={b> c}(resp.bbr(·)=bdr(1− ·)).
The functions bdr and bbr enjoy nice properties. In particular, the Pólya-Szegö and Hardy-Littlewood inequalitiesallowtocompareintegralquantitiesdependingonb,bdr,bbr andtheirderivative.
Theorem([21,22]). Letubeanon-negative andmeasurable function.
(i) Ifψ isany measurablefunctionfrom R+ toR,then ˆ1
0
ψ(u) = ˆ1
0
ψ(udr) = ˆ1
0
ψ(ubr) (equimeasurability);
(ii) Ifubelongs toW1,p(0,1)with 1≤p,then ˆ1
0
(u)p≥ ˆ1
0
(ubr)p= ˆ1
0
(udr)p (Pólya inequality);
(iii) If u,v belong toL2(0,1),then ˆ1
0
uv≤ ˆ1
0
ubrvbr= ˆ1
0
udrvdr (Hardy-Littlewood inequality);
The equality case inthe Pólya-Szegö inequality is the object of the Brothers-Ziemertheorem (see e.g.
[23]).
Symmetric decreasingfunctions Inhigherdimensions,inorder tohighlightconcentrationphenomena,we will useanother notion of symmetry, namely monotone symmetric rearrangements thatare extensions of monotone rearrangementsinonedimension. Here,Ω denotesthen-dimensionalorthotopen
i=1(ai,bi).
Definition 2.Foragiven functionb∈L1(Ω),onedefines itssymmetricdecreasingrearrangementbsd onΩ as follows: first fix then−1 variablesx2,. . . ,xn. Define b1,sd as the monotone decreasingrearrangement of x→b(x,x2,. . . ,xn).Then fixx1,x3,. . . ,xn anddefine b2,sd as themonotonedecreasingrearrangement of x→ b1,sd(x1,x,. . . ,xn). Performsuch monotone decreasingrearrangements successively. Theresulting functionisthesymmetricdecreasingrearrangementofb.
We define thesymmetric increasing rearrangementin adirection i asimilar fashion and write it bi,id. Note that,inhigherdimensions,thedefinitionofdecreasingrearrangementstrongly dependsontheorder inwhichthevariablesaretaken.
Similarly to the one-dimensional case, the Pólya-Szegö and Hardy-Littlewood inequalities allow us to compare integralquantities.
Theorem ([7,24]). Letubeanon-negative andmeasurablefunction definedon aboxΩ=n
i=1(0;ai).
(i) If ψ isanymeasurable functionfromR+ toR,then ˆ
Ω
ψ(u) = ˆ
Ω
ψ(usd) = ˆ
Ω
ψ(usd) (equimeasurability);
(ii) If u belongs to W1,p(Ω) with 1 ≤ p, then, for every i ∈ NN, [ai;bi] = ω1,i∪ω2,i∪ω3,i, where the map (x1,. . . ,xi−1,xi+1,. . . ,xN)→u(x1,. . . ,xn)is decreasing if xi ∈ω1,i,increasing if xi ∈ω2,i and constant ifxi∈ω3,i.
ˆ
Ω
|∇u|p≥ ˆ
Ω
|∇usd|p (Pólya inequality);
Furthermore, if, for any i ∈ Nn, ´
Ω|∇u|p = ´
Ω|∇ui,sd|p then there exist three measurable subsets ωi,1,ωi,2 andωi,3 of(0;ai)such that
1. (0;ai)=ωi,1∪ωi,2ωi,3, 2. u=ui,bd oni−1
k=1(0;ak)×ωi,1×n
k=i+1(0;ak), 3. u=ui,id oni−1
k=1(0;ak)×ωi,2×n
k=i+1(0;ak), 4. u=ui,bd=ui,id on i−1
k=1(0;ak)×ωi,3×n
k=i+1(0;ak).
(iii) If u,v belongtoL2(Ω),then ˆ
Ω
uv≤ˆ
Ω
usdvsd (Hardy-Littlewood inequality);
Poincaréconstantsandellipticregularityresults Wewilldenotebyc(p) theoptimalpositiveconstantsuch thatforeveryp∈[1,+∞),f ∈Lp(Ω) andu∈W1,p(Ω) satisfying
Δu=f inD(Ω),
thereholds
uW2,p(Ω)≤c(p)
fLp(Ω)+uLp(Ω)
.
The optimal constant in the Poincaré-Wirtinger inequality will be denoted by CP W(p) (Ω). This inequality reads: foreveryu∈W1,p(Ω),
u−
Ω
u Lp(Ω)
≤CP W(p) (Ω)∇uLp(Ω). (10)
Wewillalsousethefollowingregularity results:
Theorem. ([25, Theorem 9.1]) Let Ω be a C2 domain. There exists a constant CΩ > 0such that, if f ∈ L∞(Ω) andu∈W1,2(Ω) solve
−Δu=f in Ω,
∂u
∂ν = 0 on ∂Ω, (11)
then
∇uL1(Ω)≤CΩfL1(Ω). (12) Theorem. ([26, Theorem 1.1]) Let Ω be a C2 domain. There exists a constant CΩ > 0such that, if f ∈ L∞(Ω) andu∈W1,2(Ω) solve
−Δu=f in Ω,
∂u
∂ν = 0 on ∂Ω, (13)
then
∇uL∞(Ω)≤CΩfL∞(Ω). (14) Remark3.Thisresultisinfactacorollary[26,Theorem1.1].InthisarticleitisprovedthattheL∞norm ofthegradientoff isboundedbytheLorentznormLn,1(Ω) off,whichisautomaticallycontrolledbythe L∞(Ω) normoff.
Note that Stampacchia’s original result deals with Dirichlet boundary conditions. However, the same dualityargumentsprovidetheresultforNeumannboundaryconditions.
Theorem. ([27]) Let r ∈ (1;+∞). There exists Cr > 0 such that, if f ∈Lr(Ω) and if u∈ W1,r(Ω) be a solution of
−Δu= div(f) inΩ,
∂u
∂ν = 0 on∂Ω, (15)
thenthere holds
∇uLr(Ω)≤CrfLr(Ω). (16)
2. Proofsofthe mainresults
2.1. Firstorderoptimalityconditions forProblem (Pμn)
To provethemain results,wefirstneedtostatethefirstorder optimalityconditionsforProblem(Pμn).
Forthatpurpose,letusintroducethetangentconetoMm0,κ(Ω) atanypointofthis set.
Definition 3.([28, chapter 7]) For every m ∈ Mm0,κ(Ω), the tangent cone to the set Mm0,κ(Ω) at m, denoted by Tm,Mm0,κ(Ω) is the set of functions h ∈ L∞(Ω) such that, for any sequence of positive real numbersεn decreasingto 0,there existsasequenceoffunctionshn∈L∞(Ω) converging tohasn→+∞, and m+εnhn∈ Mm0,κ(Ω) foreveryn∈N.
We will show that, for any m ∈ Mm0,κ(Ω) and any admissible perturbation h ∈ Tm,Mm0,κ(Ω), the functionalFμ istwiceGâteaux-differentiableatmindirectionh.Todothat,wewillshowthatthesolution mapping
S:m∈ Mm0,κ(Ω)→θm,μ∈L2(Ω),
where θm,μ denotes thesolutionof(LDE),istwiceGâteaux-differentiable.Inthisview,weprovide several L2(Ω) estimates ofthesolutionθm,μ.
Lemma 1.([18])The mapppingS is twiceGâteaux-differentiable.
For the sakeofsimplicity, we will denote by θ˙m,μ =dS(m)[h] theGâteaux-differentialof θm,μ at min directionhandbyθ¨m,μ=d2S(m)[h,h] itssecondorder derivativeat mindirectionh.
Elementary computationsshowthatθ˙m,μ solvesthePDE
μΔ ˙θm,μ+ (m−2θm,μ) ˙θm,μ =−hθm,μ in Ω,
∂θ˙m,μ
∂ν = 0 on∂Ω, (17)
whereasθ¨m,μ solvesthePDE
μΔ¨θm,μ+ ¨θm,μ(m−2θm,μ) =−2
hθ˙m,μ−θ˙2m,μ
in Ω,
∂θ¨m,μ
∂ν = 0 on∂Ω. (18)
It followsthat,forallμ>0,theapplicationFμ isGâteaux-differentiablewithrespectto mindirection handitsGâteauxderivativewrites
dFμ(m)[h] = ˆ
Ω
θ˙m,μ.
SincetheexpressionofdFμ(m)[h] aboveisnotworkable,weneedtointroducetheadjointstatepm,μtothe equation satisfiedbyθ˙m,μ,i.ethesolutionoftheequation
μΔpm,μ+pm,μ(m−2θm,μ) = 1 in Ω,
∂pm,μ
∂ν = 0 on∂Ω. (19)
Note that pm,μ belongs to W1,2(Ω) and is unique, accordingto theFredholm alternative. Infact, we can prove thefollowing regularity resultson pm,μ: pm,μ ∈ L∞(Ω),pm,μL∞(Ω) ≤M,where M isuniform in
m ∈ Mm0,κ(Ω) and, for any p ∈ [1;+∞), pm,μ ∈ W2,p(Ω), so that Sobolev embeddings guarantee that pm,μ∈C1,α(Ω).
Now, multiplying the main equation of (19) by θ˙m,μ and integrating two times by parts leads to the expression
dFμ(m)[h] =− ˆ
Ω
hθm,μpm,μ.
Nowconsideramaximizerm.ForeveryperturbationhintheconeTm,Mm0,κ(Ω),thereholdsdFμ(m)[h]≥0.
Theanalysis ofsuch optimalitycondition isstandardinoptimalcontroltheory(see forexample[29]) and leadstothefollowingresult.
Proposition3.Letusdefine ϕm,μ =θm,μpm,μ,whereθm,μ andpm,μ solverespectivelyequations(LDE)and (19).Thereexistsc∈R suchthat
{ϕm,μ < c}={m=κ}, {ϕm,μ=c}={0< m < κ}, {ϕm,μ > c}={m= 0}. 2.2. Proof ofProposition 1
Aneasybuttediouscomputationshowsthatthefunctionϕm,μ introducedinProposition3isC1,α(Ω)∩ W1,2(Ω) function,asaproductoftwo C1,α functionsandsatisfies(in aW1,2 weaksense)
μΔϕm,μ−2μ
∇ϕm,μ,∇θθm,μ
m,μ
+ϕm,μ
2μ|∇θθm,μ|2
m,μ2 + 2m−3θm,μ
=θm,μ in Ω,
∂ϕm,μ
∂ν = 0 on∂Ω, (20)
where ·,· stands for the usual Euclidean inner product. To provethat |{m = 0}|+|{m = κ}|> 0,we argueby contradiction,byassuming that|{m=κ}|=|{m= 0}|= 0. Therefore,ϕm,μ =c a.e.inΩ and, accordingto(20),thereholds
c
2μ|∇θm,μ|2
θm,μ2 + 2m−3θm,μ
=θm,μ Integratingthisidentityand usingthatθm,μ >0 inΩ andc= 0, weget
2c
⎛
⎝μˆ
Ω
|∇θm,μ|2 θm,μ2 +
ˆ
Ω
(m−θm,μ)
⎞
⎠= (c+ 1) ˆ
Ω
θm,μ.
Equation(4) yields thattheleft-handsideequals 0,sothatonehasc=−1.Comingbacktotheequation satisfiedbyϕm,μ leadsto
m=θm,μ−μ|∇θm,μ|2 θm,μ2 .
Thelogisticdiffusiveequation(LDE) isthentransformedinto μθm,μΔθm,μ−μ|∇θm,μ|2= 0.
Integrating this equation by parts yields ´
Ω|∇θm,μ|2 = 0.Thus, θm,μ is constant,and so is m. In other words,m=m0,which, accordingto(4) (seeRemark1)isimpossible.Theexpectedresultfollows.
2.3. Proofof Theorem1
TheproofofTheorem1isbasedonacarefulasymptoticanalysiswithrespecttothediffusivityvariable μ.
Letus firstexplaintheoutlinesoftheproof.
Let us fix m ∈ Mm0,κ(Ω) and h ∈ L∞(Ω). In the sequel, the dot or double dot notation f˙ or f¨will respectivelydenote firstandsecondorder Gâteaux-differentialoff atm indirectionh.
According to Lemma 1, Fμ is twice Gâteaux-differentiable and its second order Gâteaux-derivative is given by
d2Fμ(m)[h, h] = ˆ
Ω
θ¨m,μ,
where θ¨m,μ isthesecondGâteauxderivativeofS,definedastheuniquesolutionof(18).
Letm1 andm2be twoelementsofMm0,κ(Ω) anddefine φμ: [0; 1]t→ Fμ
tm2+ (1−t)m1
−tFμ(m2)−(1−t)Fμ(m1).
Onehas
d2φμ dt2 (t) =
ˆ
Ω
θ¨(1−t)m1+tm2,μ, and φμ(0) =φμ(1) = 0,
where θ¨(1−t)m1+tm2,μ mustbe interpretedas abilinear formfrom L∞(Ω) toW1,2(Ω),evaluatedtwo times at thesamedirectionm2−m1.Hence,togetthestrictconvexityofFμ,itsufficestoshowthat,whenever μ islargeenough,
ˆ
Ω
θ¨tm2+(1−t)m1,μ>0
as soon as m1 =m2 (in L∞(Ω)) and t ∈ (0,1),or equivalently thatd2Fμ(m)[h,h] > 0 as soon as m ∈ Mm0,κ(Ω) andh∈L∞(Ω).Notethatsinceh=m2−m1,itispossibletoassumewithoutlossofgenerality thathL∞(Ω)≤2κ.
Theproofisbasedonanasymptoticexpansionofθm,μintoamaintermandareminderone,withrespect to thediffusivityμ.Itiswell-known(seee.g.[16,Lemma2.2])thatonehas
θm,μ W
1,2(Ω)
−−−−−−→
μ→∞ m0. (21)
However,sinceweareworkingwithresourcesdistributionslivinginMm0,κ(Ω),suchaconvergenceproperty doesnotallowustoexploititforderivingoptimalitypropertiesforProblem(Pμn).
Forthisreason,inwhatfollows,wefindafirstorderterminthisasymptoticexpansion.Togetaninsight into theproof’smain idea,letus firstproceedinaformalway, bylookingforafunctionη1,msuchthat
θm,μ ≈m0+η1,m μ
as μ→ ∞.Pluggingthisformal expansionin(LDE) andidentifyingat order 1μ yieldsthatη1,m satisfies
Δη1,m+m0(m−m0) = 0 in Ω,
∂η1,m
∂ν = 0 on∂Ω.
To make this equation well-posed, it is convenient to introduce the function ˆη1,m defined as the unique solutiontothesystem
⎧⎪
⎨
⎪⎩
Δˆη1,m+m0(m−m0) = 0 in Ω,
∂ηˆ1,m
∂ν = 0 on∂Ω,
ffl
Ωηˆ1,m= 0, andto determineaconstantβ1,m suchthat
η1,m= ˆη1,m+β1,m.
Inviewofidentifyingtheconstantβ1,m,weintegrateequation(LDE) toget ˆ
Ω
θm,μ(m−θm,μ) = 0
whichyields,at theorder 1μ,
β1,m= 1 m0
Ω
ˆ
η1,m(m−m0) = 1 m20
Ω
|∇ηˆ1,m|2.
Therefore,onehasformally
Ω
θm,μ ≈m0+1 μ
Ω
|∇ηˆ1,m|2.
As will be proved in Step 1 (paragraph 2.3), the mapping Mm0,κ(Ω) m → β1,m is convex so that, at the order μ1, the mapping Mm0,κ(Ω) m → ffl
Ωθm,μ is convex. We will prove the validity of all the claimsabove,bytakingintoaccountremaindertermsintheasymptoticexpansionabove,toprovethatthe mappingFμ:m→ffl
Ωθm,μ isitselfconvexwheneverμislargeenough.
Remark4.Onecouldalso noticethatthe quantityβ1,m aroseinthe recentpaper[30],where theauthors determinethelargetimebehaviorofadiffusiveLotka-Volterracompetitivesystembetweentwopopulations with growth rates m1 and m2. If β1,m1 > β1,m2, then when μ is large enough, the solution converges as t → +∞ to the steady state solution of ascalar equation associated with the growth rate m1. In other words, thespecieswithgrowth ratem1 chasestheother one.Inthepresentarticle, asabyproduct ofour results,wemaximize thefunctionm→β1,m. Thisremark impliesthatthisintermediateresultmight find otherapplications ofitsown.
Letusnowformalizerigorouslythereasoningoutlined above,byconsideringanexpansionoftheform θm,μ=m0+η1,m
μ +Rm,μ
μ2 . Hence,onehasforallm∈ Mm0,κ(Ω),
d2Fμ(m)[h, h] = 1 μ
ˆ
Ω
¨
η1,m+ 1 μ2
ˆ
Ω
R¨m,μ
Wewill showthatthere holds
d2Fμ(m)[h, h]≥C(h) μ
1−Λ
μ
(22) forallμ>0,whereC(h) andΛ denotesomepositiveconstants.
The strict convexity ofFμ will then follow.Concerning the bang-bang characterof maximizers, notice thattheadmissiblesetMm0,κ(Ω) isconvex,andthatitsextremepointsareexactlythebang-bangfunctions of Mm0,κ(Ω).Oncethe strictconvexityof Fμ showed,wethen easilyinferthatFμ reaches itsmaxima at extreme points, in other words that any maximizer is bang-bang.Indeed, assuming by contradiction the existence of amaximizerwritingtm1+ (1−t)m2 with t∈(0,1),m1 andm2, two elements of Mm0,κ(Ω) suchthatm1=m2 onapositiveLebesguemeasure set,onehas
Fμ(tm1+ (1−t)m2)< tFμ(m1) + (1−t)Fμ(m2)<max{Fμ(m1),Fμ(m2)}, byconvexityofFμ,whencethecontradiction.
Therestoftheproofisdevotedtotheproofoftheinequality(22).Itisdividedintothefollowingsteps:
Step 1. Uniformestimateof ´
Ωη¨1,m withrespectto μ.
Step 2. DefinitionandexpansionoftheremaindertermRm,μ. Step 3. Uniformestimateof Rm,μ withrespecttoμ.
Step 1:minoration of´
Ωη¨1,m Onecomputessuccessively β˙1,m= 1
m0 Ω
η˙ˆ1,mm+ ˆη1,mh
, β¨1,m= 1 m0
Ω
2 ˙ˆη1,mh+ ¨η1,mh
(23)
where η˙ˆ1,m solvestheequation
Δ ˙ˆη1,m+m0h= 0 in Ω
∂η˙ˆ1,m
∂ν = 0, on∂Ω with
ˆ
Ω
˙ˆ
η1,m= 0. (24)
Notice moreoverthatη¨1,m = 0, sinceη˙ˆ1,m islinear withrespect to h.Moreover, multiplyingthe equation abovebyη˙ˆ1,m andintegrating bypartsyields
β¨1,m = 2 m20
Ω
|∇η˙ˆ1,m|2>0 (25)
wheneverh= 0, accordingto (23).Finally,weobtain ˆ
Ω
¨
η1,m =|Ω|β¨1,m+ ˆ
Ω
¨
η1,m=|Ω|β¨1,m= 2 m20
ˆ
Ω
|∇η˙ˆ1,m|2.
It isthennotablethat´
Ωη¨1,m≥0.