LOWER SEMICONTINUITY AND RELAXATION OF LINEAR-GROWTH INTEGRAL FUNCTIONALS UNDER PDE CONSTRAINTS ADOLFOARROYO-RABASA,GUIDODEPHILIPPIS,ANDFILIPRINDLER ABSTRACT. We show general lower semicontinuity and relaxation theorems for linear-growth integral functionals defined on vector measures that satisfy linear PDEsideconstraints(ofarbitraryorder). Theseresultsgeneralizeseveralknown 7 lower semicontinuity and relaxation theorems for BV, BD, and for more general 1 first-orderlinearPDEsideconstrains. Ourproofsarebasedonrecentprogressin 0 theunderstandingofsingularitiesinmeasuresolutionstolinearPDE’sandofthe 2 correspondinggeneralizedconvexityclasses. n KEYWORDS: Lower semicontinuity, functional on measures, A-quasiconvexity, a generalizedYoungmeasure. J 9 DATE:January10,2017. ] P A h. 1. INTRODUCTION t a The theory of linear-growth integral functionals defined on vector-valued mea- m sures satisfying PDEconstraints iscentral tomanyquestions ofthecalculus ofvari- [ ations. In particular, their relaxation and lower semicontinuity properties have at- 1 tracted a lot of attention, see for instance [AD92, FM93, FM99, FLM04, KR10b, v Rin11, BCMS13]. In the present work weunify and extend a large number of these 0 results by proving general lower semicontinuity and relaxation theorems for such 3 2 functionals. Our proofs are based on recent advances in the understanding of the 2 singularities thatmayoccurinmeasuressatisfying (under-determined) linearPDEs. 0 Concretely, let W ⊂ Rd be an open and bounded subset with Ld(¶ W ) = 0 and . 1 consider for a vector Radon measure m ∈ M(W ;RN) on W with values in RN the 0 functional 7 1 dm dm s : F#[m ]:= f x, (x) dx+ f# x, (x) d|m s|(x). (1.1) v ZW (cid:18) dLd (cid:19) ZW (cid:18) d|m s| (cid:19) i X Here, f: W ×RN →RisaBorelintegrand thathaslineargrowthatinfinity, i.e., r a |f(x,A)|≤M(1+|A|) forall(x,A)∈W ×RN, wherebythe(generalized) recessionfunction f(x′,tA′) f#(x,A):=limsup , (x,A)∈W ×RN, t x′→x A′→A t→¥ takesonlyfinitevalues. Furthermore, onthecandidate measures m ∈M(W ;RN)we imposethek’th-order linearPDEsideconstraint Am := (cid:229) Aa ¶ a m =0 inthesenseofdistributions. |a |≤k 1 2 A.ARROYO-RABASA,G.DEPHILIPPIS,ANDF.RINDLER ThecoefficientsAa ∈Rn×N areassumedtobeconstantandwewrite¶ a =¶ a 1...¶ a d 1 d foreverymulti-indexa =(a ,...,a )∈(N∪{0})d with|a |:=|a |+···+|a |≤k. 1 d 1 d Wecallmeasures m ∈M(W ;RN)withAm =0inthesenseofdistributions A-free. WewillalsoassumethatA satisfiesMurat’sconstantrankcondition(see[Mur81, FM99]),thatis,wesuppose thatthereexistsr∈Nsuchthat rank(kerAk(x ))=r forallx ∈Sd−1, (1.2) where Ak(x ):=(2p i)k (cid:229) x a Aa , x a =x a 1···x a d 1 d |a |=k istheprincipalsymbolofA. WealsorecallthenotionofwaveconeassociatedtoA, which plays a fundamental role in the study of A-free fields and first originated in theTartar–Murattheoryofcompensatedcompactness [Tar79,Tar83,Mur78,Mur79, Mur81,DiP85]. Definition1.1. LetA bek’th-orderlinearPDEoperatorasabove,A :=(cid:229) |a |≤kAa ¶ a . Thewaveconeassociated toA istheset L A := kerAk(x )⊂RN. |x[|=1 Note that the wave cone contains those amplitudes along which it is possible to constructhighlyoscillatingA-freefields. MorepreciselyifA ishomogeneous, i.e., A =(cid:229) |a |=kAa ¶ a ,thenP∈L A ifandonlyifthereexistsx 6=0suchthat A(Ph(x·x ))=0 forallh∈Ck(R). Ourfirstmaintheoremconcerns thecasewhen f isAk-quasiconvex initssecond argument, where Ak := (cid:229) Aa ¶ a |a |=k isthe principal part of A. Recall from [FM99]that aBorel function h: RN →Ris calledAk-quasiconvex if h(A)≤ h(A+w(y))dy ZQ forallA∈RN andallQ-periodicw∈C¥ (Q;RN)suchthatAkw=0and wdy=0, Q whereQ:=(−1/2,1/2)d istheunitcubeinRd. R Theorem 1.2 (lower semicontinuity). Let f: W ×RN → [0,¥ ) be a continuous integrand. Assume that f has linear growth at infinity and is Lipschitz in its second argument, uniformlyinx. Assumefurtherthatthereexistsamodulusofcontinuity w suchthat |f(x,A)− f(y,A)|≤w (|x−y|)(1+|A|) forallx,y∈W ,A∈RN. (1.3) andthatthestrongrecession function f(x,tA) f¥ (x,A):= lim existsforall (x,A)∈W ×spanL A. (1.4) t→¥ t Then,thefunctional dm dm s F[m ]:= f x, (x) dx+ f¥ x, (x) d|m s|(x) ZW (cid:18) dLd (cid:19) ZW (cid:18) d|m s| (cid:19) LOWERSEMICONTINUITYANDRELAXATIONOFINTEGRALFUNCTIONALS 3 issequentially weakly*lowersemicontinuous formeasuresinthespace M(W ;RN)∩kerA := m ∈M(W ;RN) : Am =0 (cid:8) (cid:9) ifandonlyif f(x,q)isAk-quasiconvex foreveryx∈W . Note that according to (1.7) below, F[m ] is well defined. Since the strong reces- sionfunction iscomputed onlyatamplitudes thatbelongtospanL A. Remark1.3. TheconclusionofTheorem1.2extendstosequencessuchthatAm → j 0strongly inW−k,q(W ;Rn)forsome1<q<d/(d−1). Notice that f¥ in (1.4) is a limit, and differently from f#, it may fail to exist for A ∈ (spanL A)\L A (for A ∈ L A the existence of f¥ (x,A) follows from the Ak- ¥ quasiconvexity, see Corollary 2.19). If we remove the assumption that f exists for points in the subspace generated by the wave cone L A, we still have the following partiallowersemicontinuity result(cf.[FLM04]). Theorem 1.4 (partial lower semicontinuity). Let f: W ×RN →[0,¥ ) be a con- tinuous integrand. Assume that f has linear growth at infinity and is Lipschitz in its second argument, uniformly in x. Assume further that there exists a modulus of continuity w suchthat |f(x,A)− f(y,A)|≤w (|x−y|)(1+|A|) forallx,y∈W ,A∈RN. (1.5) Then, dm f x, (x) dx≤ liminfF#[m ] : m ⇀∗ m andAm →0inW−k,q , ZW (cid:18) dLd (cid:19) n j→¥ j j j o where dm dm s F#[m ]:= f x, (x) dx+ f# x, (x) d|m s|(x). ZW (cid:18) dLd (cid:19) ZW (cid:18) d|m s| (cid:19) Remark1.5. Asspecial casesofTheorem1.2weget,amongothers, thefollowing well-knownresults: (i) For A = curl, one obtains BV-lower semicontinuity results in the spirit of Ambrosio–Dal Maso [AD92] and Fonseca–Mu¨ller [FM93], also see [KR10b] forthecaseofsignedintegrands. (ii) ForA =curlcurl,where d curlcurlm := (cid:229) ¶ m j+¶ m k−¶ m i−¶ m k (cid:18) ik i ij i jk i ii j(cid:19) i=1 j,k=1,...,d is the second order operator expressing the Saint-Venant compatibility condi- tions (see [FM99, Example 3.10(e)]), we re-prove the lower semicontinuity andrelaxation theorem inthespaceoffunctions ofbounded deformation (BD) from[Rin11]. (iii) Forfirst-orderoperators A,asimilarresultwasprovedin[BCMS13]. (iv) Earlier work in this direction is in [FM99, FLM04], but did not consider the singular part(concentration ofmeasure). If we dispense with the assumption of Ak-quasiconvexity on the integrand, we havethefollowingtworelaxation results: 4 A.ARROYO-RABASA,G.DEPHILIPPIS,ANDF.RINDLER Theorem1.6(relaxation). Let f: W ×RN →[0,¥ )beacontinuous integrandthat isLipschitzinitssecondargument(uniformlyinx),haslineargrowthatinfinity,and is such that there exists a modulus of continuity w as in (1.5). Further we assume that A is a homogeneous partial differential operator and that the strong recession function f¥ (x,A) existsforall (x,A)∈W ×spanL A. Then,forthefunctional G[u]:= f(x,u(x))dx, u∈L1(W ;RN), ZW the(sequentially) weakly*lowersemicontinuous envelope G[m ]:=inf liminfG[u ] : u Ld ⇀∗ m andAu →0inW−k,q j j j n j→¥ o isgivenby dm dm s G[m ]=ZW QA f(cid:18)x,d|m |(x)(cid:19)dx+ZW (QA f)#(cid:18)x,d|m s|(x)(cid:19) d|m s|(x), where QA f(x,q) denotes the A-quasiconvex envelope of f(x, q) with respect to the secondargument(seeDefinition2.16below). If we want to relax in the space M(W ;RN)∩kerA we need to assume that L1(W ;RN)∩kerA is dense in M(W ;RN)∩kerA with respect to a finer topology thanthenaturalweak*topology (inthiscontextalsosee[AR16]). Theorem 1.7. Let f: W ×RN →[0,¥ ) be a continuous integrand that is Lipschitz in its second argument (uniformly in x), has linear growth at infinity, and is such that there exists a modulus of continuity w as in (1.5). Further assume that A is a homogeneous partialdifferential operatorandthatthestrongrecession function f¥ (x,A) existsforall (x,A)∈W ×spanL A. If for all m ∈M(W ;RN) with Am =0 there exists a sequence (u )⊂L1(W ;RN)∩ j kerA suchthat u Ld ⇀∗ m inM(W ;RN) and hu Ldi(W )→hm i(W ), (1.6) j j wherehqiis thearea functional defined in (2.2), then the weakly* lower semicontin- uousenvelope ofthefunctional G[u]:= f(x,u(x))dx, u∈L1(W ;RN)∩kerA, ZW withrespecttoweak*-convergence inthespaceM(W ;RN)∩kerA,isgivenby dm dm s G[m ]=ZW QA f(cid:18)x,d|m |(x)(cid:19)dx+ZW (QA f)#(cid:18)x,d|m s|(x)(cid:19)d|m s|(x). Remark 1.8 (density assumptions). Condition (1.6) is automatically fulfilled in thefollowingcases: (i) For A =curl, the approximation property (for general domains) is proved in the appendix of [KR10a] (also see Lemma B.1 of [Bil03] for Lipschitz domains). The same argument further shows the area-strict approximation property in the BD-case (also see Lemma 2.2 in [BFT00] for a result which coversthestrictconvergence). LOWERSEMICONTINUITYANDRELAXATIONOFINTEGRALFUNCTIONALS 5 (ii) IfW isastrictlystar-shaped domain,i.e.,thereexistsx ∈W suchthat 0 (W −x )⊂t(W −x ) forallt >1, 0 0 then (1.6) holds for every homogeneous operator A. Indeed, for t >1 we can consider the dilation of m defined on t(W −x ) and then mollify it at a 0 sufficiently smallscale. Wereferforinstance to[Mu¨l87]fordetails. As a consequence of Theorem 1.7 and of Remark 1.8 we explicitly state the fol- lowingcorollary,whichextendsthelowersemicontinuityresultof[Rin11]intoafull relaxation result. The only other relaxation result in this direction, albeit for special functions ofbounded deformation, seemstobein[BFT00],other results inthisarea arediscussed in[Rin11]andthereferences therein. Corollary 1.9. Let f: W ×Rd×d → [0,¥ ) be a continuous integrand that is uni- sym formlyLipschitzinitssecondargument,haslineargrowthatinfinity,andissuchthat there exists a modulus of continuity w as in (1.5). Further assume that the strong recession function f¥ (x,A) existsforall (x,A)∈W ×Rd×d. sym Letusconsider thefunctional G[u]:= f(x,Eu(x))dx, ZW defined for u∈LD(W ):={u∈BD(W ) : Esu=0}, where Eu:=(Du+DuT)/2∈ M(W ;Rd×d)isthesymmetrizeddistributional derivative ofu∈BD(W )andwhere sym dEsu Eu=EuLd W + |Esu|, d|Esu| isitsRadon–Nikody´m decomposition withrespecttoLd. Then,thelowersemicontinuousenvelopeofG[u]withrespecttoweak*-convergence inBD(W ),isgivenbythefunctional dEsu G[u]:= SQf(x,Eu(x))dx+ (SQf)# x, (x) d|Esu|(x), ZW ZW (cid:18) d|Esu| (cid:19) where SQf denotes the symmetric-quasiconvex envelope of f with respect to the second argument (i.e., the curlcurl-quasiconvex envelope of f(x, q) in the sense of Definition2.16). OurproofsarefairlyconciseandbasedonnewtoolstostudysingularitiesinPDE- constrained measures. Concretely, we exploit the recent developments on the struc- tureofA-freemeasuresobtainedin[DR16b]. Inparticular, thestudyofthesingular part – up to now the most complicated argument in the proof – now only requires a fairly straightforward (classical) convexity argument. More precisely, the main the- orem of [KK16] establishes that the restriction of f# to the linear space spanned by the wave cone is in fact convex at all points of L A (in the sense that a supporting hyperplane exists). Moreover, by[DR16b], dm s d|m s|(x)∈L A for|m s|-a.e.x∈W . (1.7) Thus, combining these twoassertions, wegain classical convexity for f# at singular points, which can be exploited via the theory of generalized Young measures devel- opedin[DM87,AB97,KR10a]. 6 A.ARROYO-RABASA,G.DEPHILIPPIS,ANDF.RINDLER Remark 1.10 (different notions of recession function). Note that both in Theo- ¥ rem 1.2 and Theorem 1.6 the existence of the strong recession function f is as- sumed, in contrast with the results in [AD92, FM93, BCMS13] where this is not imposed. The need for this assumption comes from the use of Young measure techniques which seem to be better suited to deal with the singular part of the measure, as we alreadydiscussedabove. Intheaforementionedreferencesadirectblowupapproach isinstead performedandthisallowstodealdirectlywiththefunctional in(1.1). The blow up techniques, however, rely strongly on the fact that A is a homogeneous first-order operator. Indeed, it is not hard to check that for all “elementary” A-free measuresoftheform m =P0l , where P0∈L A, l ∈M+(Rd), the scalar measure l is necessarily translation invariant along orthogonal directions tothecharacteristic set X (P ):= x ∈Rd : P ∈kerA(x ) , 0 0 (cid:8) (cid:9) whichturnsouttobeasubspaceofRd wheneverA isafirst-orderoperator. Thesub- spacestructureandtheaforementioned translationinvarianceisthenusedtoperform homogenization-type arguments. Duetothelackoflinearity ofthemap x 7→Ak(x ) fork>1, the structure of elementary A-free measures for general operators is more com- plicated and not yet fully understood (see however [Rin11, DR16a] for the case A =curlcurl).Thisprevents, atthemoment,theuseofa“pure”blow-uptechniques andforces ustopass through thecombination oftheresults of[DR16b,KK16]with theYoungmeasureapproach. Thispaperisorganized asfollows: First,inSection2,weintroduce alltheneces- sary notation and prove a few auxiliary results. Then, in Section 3, we establish the central Jensen-type inequalities, whichimmediately yieldtheproofofTheorems1.2 and1.4inSection4. TheproofsofTheorems1.6and1.7aregiveninSection5. Acknowledgments. A.A.-R.issupportedbyascholarshipfromtheHausdorffCen- terofMathematics and theUniversity ofBonn; the research conducted inthis paper formspartofthefirstauthor’sPh.D.thesisattheUniversityofBonn. G.D.P.issup- ported by the MIUR SIR-grant “Geometric Variational Problems” (RBSI14RVEZ). F.R.acknowledges thesupport fromanEPSRCResearch Fellowshipon“Singulari- tiesinNonlinear PDEs”(EP/L018934/1). 2. NOTATION AND PRELIMINARIES We write M(W ;RN) and M (W ;RN) to denote the space of finite and locally loc finite vector Radon measures on W ⊂RN. Wewritethe Radon–Nikody´m decompo- sitionofm ∈M(W ;RN)as dm m = Ld W +m s, (2.1) dLd where dm ∈L1(W ;RN)and m s∈M(W ;RN)issingular withrespecttoLd. dLd Inorder tokeep asimple presentation, wewill often identify u∈L1(W ;RN)with themeasureuLd ∈M(W ;RN). LOWERSEMICONTINUITYANDRELAXATIONOFINTEGRALFUNCTIONALS 7 2.1. Integrands and Young measures. For f ∈C(W ×RN) define the transforma- tion Aˆ (Sf)(x,Aˆ):=(1−|Aˆ|)f x, , x∈W ,Aˆ ∈BN, (cid:18) 1−|Aˆ|(cid:19) whereBN denotes theopenunitballinRN. Then,Sf ∈C(W ×BN). Weset E(W ;RN):= f ∈C(W ×RN) : Sf extendstoC(W ×BN) . (cid:8) (cid:9) Inparticular, all f ∈E(W ;RN)havelinear growthatinfinity, i.e.,thereexists aposi- tive constant M such that |f(x,A)|≤M(1+|A|)for all x∈W and all A∈RN. With thenorm kfkE(W ;RN):=kSfk¥ , f ∈E(W ;RN), the space E(W ;RN) turns out to be a Banach space. Also, by definition, for each f ∈E(W ;RN)thelimit f(x′,tA′) f¥ (x,A):= lim , x∈W ,A∈RN, x′→x t A′→A t→¥ exists and defines a positively 1-homogeneous function called the strong recession ¥ functionof f. Evenifonedropsthedependenceonx,therecessionfunctionh might not exist for h ∈ C(Rd). Instead, one can always define the generalized recession functions f(x′,tA′) f#(x,A):=limsup , t x′→x A′→A t→¥ f(x′,tA′) f (x,A):=liminf , # x′→x t A′→A t→¥ which again turn out to be positively 1-homogeneous. If f is x-uniformly Lipschitz continuous in the A-variable and there exists a modulus of continuity w : [0,¥ ) → [0,¥ )(increasing, continuous, andw (0)=0)suchthat |f(x,A)− f(y,A)|≤w (|x−y|)(1+|A|), x,y∈W ,A∈RN, thenthedefinitions of f¥ and f# (and f )simplifyto # f(x,tA) ¥ f (x,A):= lim , t→¥ t f(x,tA) f#(x,A):=limsup . t→¥ t AnaturalactionofE(W ;RN)onthespaceM(W ;RN)isgivenby dm dm s m 7→ f x, (x) dx+ f¥ x, (x) d|m s|(x). ZW (cid:18) dLN (cid:19) ZW (cid:18) d|m s| (cid:19) Inparticular,for f(x,A)= 1+|A|2∈E(W ;RN)–forwhich f¥ (A)=|A|,wedefine theareafunctional p dm 2 hm i(W ):= 1+ dx+|m s|(W ), m ∈M(W ;RN). (2.2) ZW r (cid:12)dLN(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) 8 A.ARROYO-RABASA,G.DEPHILIPPIS,ANDF.RINDLER In addition to the well-known weak* convergence of measures, we say that a se- quence(m )converges area-strictly to m inM(W ;RN)if j m ⇀∗ m inM(W ;RN) and hm i(W )→hm i(W ), j j inwhichcasewewrite“m →m area-strictly”. j This notion of convergence turns out to be stronger than the conventional strict convergence ofmeasures,whichmeansthat m ⇀∗ m inM(W ;RN) and |m |(W )→|m |(W ). j j Indeed, the area-strict convergence, as opposed to the usual strict convergence, pro- hibitsone-dimensionaloscillations. Themeaningofarea-strictconvergencebecomes clear when considering the following version of Reshetnyak’s continuity theorem, which entails that the topology generated by area-strict convergence is the coarsest topology underwhichthenatural actionofE(W ;RN)onM(W ;RN)iscontinuous. Theorem 2.1 (Theorem 5 in [KR10b]). For every integrand f ∈ E(W ;RN), the functional dm dm s m 7→ f x, (x) dx+ f¥ x, (x) d|m s|(x), ZW (cid:18) dLN (cid:19) ZW (cid:18) d|m s| (cid:19) isarea-strictly continuous onM(W ;RN). Remark2.2. Noticethatifm ∈M(Rd;RN),thenm e →m area-strictly, wherem e is themollificationofm withafamilyofstandardconvolution kernels, m e :=m ∗r e and r e (x):=e −dr (x/e )forr ∈C¥c(B1)positiveandevenfunctionsatisfying r dx=1. R GeneralizedYoungmeasuresformasetofdualobjectstotheintegrandsinE(W ;RN). Werecall briefly some aspects of this theory, which was introduced by DiPerna and Majdain[DM87]andlaterextended in[AB97,KR10a]. Definition2.3(generalizedYoungmeasure). AgeneralizedYoungmeasure,parametrized byanopensetW ⊂Rd,andwithvaluesinRN isatriple(n x,l n ,n x¥ ),where (i) (n x)x∈W ⊂M(RN)isaparametrized familyofprobability measuresonRN, (ii) l n ∈M+(W )isapositive finiteRadonmeasureonW ,and (iii) (n ¥ ) ⊂ M(SN−1) is a parametrized family of probability measures on x x∈W theunitsphereSN−1. Additionally, werequirethat (iv) themapx7→n isweakly*measurable withrespecttoLd, x (v) themapx7→n x¥ isweakly*measurablewithrespecttol n ,and (vi) x7→h|q|,n i∈L1(W ). x ThesetofallsuchYoungmeasuresisdenotedbyY(W ;RN). Here,weak*measurability meansthatthefunctionsx7→hf(x, q),n i(respectively x x 7→ hf¥ (x, q),n x¥ i) are Lebesgue measurable (respectively l n -measurable) for all Carathe´odory integrands f: W ×RN → R (measurable in their first argument and continuous intheirsecond argument). LOWERSEMICONTINUITYANDRELAXATIONOFINTEGRALFUNCTIONALS 9 Foranintegrand f ∈E(W ;RN)andaYoungmeasuren ∈Y(W ;RN),wedefinethe dualityparingbetween f andn asfollows: f,n := hf(x, q),n xidx+ hf¥ (x,q),n x¥ idl n (x). ZW ZW (cid:10)(cid:10) (cid:11)(cid:11) In many cases it will be sufficient to work with functions f ∈ E(W ;RN) which are Lipschitz continuous. The following density lemma can be found in [KR10a, Lemma3]: Lemma2.4. There exists acountable set of functions {f }={j ⊗h ∈C(W )× m m m C(RN):m∈N}⊂E(W ;RN) such that for two Young measures n ,n ∈Y(W ;RN) 1 2 theimplication hhf ,n ii=hhf ,n ii ∀m∈N =⇒ n =n m 1 m 2 1 2 holds. Moreover,alltheh canbechosentobeLipschitzcontinuous. m SinceY(W ;RN)iscontained inthedualspaceofE(W ;RN)viathedualitypairing hhq,qii,wesaythatasequenceofYoungmeasures(n )⊂Y(W ;RN)convergesweakly* j ton ∈Y(W ;RN),insymbolsn ⇀∗ n ,if j f,n → f,n forall f ∈E(W ;RN). j (cid:10)(cid:10) (cid:11)(cid:11) (cid:10)(cid:10) (cid:11)(cid:11) Fundamental for all Young measure theory is the following compactness result, see[KR10a,Section3.1]foraproof: Lemma2.5(compactness). Let(n )⊂Y(W ;RN)beasequenceofYoungmeasures j satisfying (i) thefunctions x7→h|·|,n iareuniformlybounded inL1(W ), j (ii) supjl n j(W )<¥ . Then,thereexistsasubsequence (notrelabeled)andn ∈Y(W ;RN)suchthatn ⇀∗ n j inY(W ;RN). Young measures generated by means of periodic homogenization can be easily computed, see[BM84]. Lemma2.6(oscillationmeasures). Letw∈L1 (Rd;RN)beaQ-periodicfunction loc andletm∈N. Definethe(Q/m)-periodic functions w (x):=w(mx). m Then, w ⇀w(x):= w(y)dy inL1(W ;RN), m ZQ foreverymeasurableW ⊂Rd. Moreover, thesequence (w ) ⊂L1 (Rd;RN)generates thehomogeneous (local) m loc Young measure n =(d ,0,d )∈Y (Rd;RN) (a Young measure restricted to every w 0 loc compactsubsetofRd),where hh,d i:= h(w(y))dy forallh∈C(Rd)withlineargrowthatinfinity. w ZQ 10 A.ARROYO-RABASA,G.DEPHILIPPIS,ANDF.RINDLER TheRadon–Nikody´mdecomposition(2.1)inducesanaturalembeddingofM(W ;RN) intoY(W ;RN):m 7→d [m ],viatheidentification (d [m ])x :=d ddLmd(x), l d [m ]:=|m s|, (d [m ])¥x :=d dd|mmss|(x). Inthissense,wesaythatthesequenceofmeasures(m )generatestheYoungmeasure j n ifd [m ]⇀∗ n inY(W ;RN),insymbols h m →Y n . j Thebarycenter [n ]∈M(W ;RN)ofaYoungmeasuren ∈Y(W ;RN)isdefinedas [n ]:= id,n =hid,n xiLd W +hid,n x¥ il n . (cid:10)(cid:10) (cid:11)(cid:11) Usingthenotation aboveitisclear thatfor(m )⊂M(W ;RN)itholds that m ⇀∗ [n ] j j asmeasuresonW ifm →Y n . j Remark2.7. Forasequence(m )⊂M(W ;RN)thatarea-strictlyconvergestosome j limitm ∈M(W ;RN),itisrelativelyeasytocharacterizethe(unique)Youngmeasure it generates. Indeed, an immediate consequence of the Separation Lemma 2.4 and Theorem2.1isthat m →m area-strictly inW ⇐⇒ m →Y d [m ]∈Y(W ;RN). j j Insomecasesitwillbenecessarytodeterminethesmallestlinearspacecontaining thesupportofaYoungmeasure. Withthisaiminmind,westatethefollowingversion ofTheorem2.5in[AB97]: Lemma 2.8. Let (u ) be a sequence in L1(W ;RN) generating a Young measure j n ∈Y(W ;RN)andletV beasubspace ofRN suchthatu (x)∈V forLd-a.e. x∈W . j Then, (i) suppn ⊂V forLd-a.e.x∈W , x (ii) suppn x¥ ⊂V∩SN−1 forl n -a.e.x∈W . Finally,wehavethefollowingapproximation lemma,see[AB97,Lemma2.3]for aproof. Lemma2.9. ForeveryuppersemicontinuousBorelintegrand f: W ×RN →Rwith linear growth at infinity, there exists a decreasing sequence (f )⊂E(W ;RN) of the m form f =(cid:229) l(m)j ⊗h (withthechoicesofj ,h depending onm)with m j=1 j j j j inf f = lim f = f, inf f¥ = lim f¥ = f# (pointwise). m∈N m m→¥ m m∈N m m→¥ m Furthermore, the linear growth constants of the h can be chosen to be bounded by j thelineargrowthconstant of f. Byapproximation, wethusget: Corollary 2.10. Let f: W ×RN →Rbeanuppersemicontinuous Borelintegrand. Thenthefunctional n 7→ hf(x, q),n xidx+ hf#(x,q),n x¥ idl n (x) ZW ZW issequentially weakly*uppersemicontinuous onY(W ;RN).