Contents lists available atScienceDirect
Journal
of
Mathematical
Analysis
and
Applications
www.elsevier.com/locate/jmaa
Sharp
estimates
for
the
anisotropic
torsional
rigidity
and the principal
frequency
Giuseppe Buttazzoa,∗, Serena GuarinoLo Biancob, Michele Marinic
a
DipartimentodiMatematica,UniversitàdiPisa,LargoB.Pontecorvo5,56126Pisa,Italy
b
DipartimentodiMatematicaeApplicazioni“RenatoCaccioppoli”,UniversitàdegliStudidiNapoli FedericoII,ViaCintia,MonteS.Angelo,80126Napoli,Italy
cScuolaInternazionaleSuperiorediStudiAvanzati,viaBonomea265,34136Trieste,Italy
a r t i cl e i n f o a b s t r a c t
Articlehistory:
Received25December2016 Availableonline23March2017 Submittedby G. Moscariello Keywords: Torsionalrigidity Shapeoptimization Principaleigenvalue Convexdomains
Inthispaperwegeneralizesomeclassicalestimatesinvolvingthetorsionalrigidity andtheprincipalfrequencyofaconvexdomaintoaclassoffunctionalsrelatedto someanisotropicnonlinear operators.
©2017ElsevierInc.Allrightsreserved.
1. Introduction
Let hK be the norm associated to aconvex body K (see Section 3for moredetails); given a domain
Ω⊂ RN withfinitemeasure,wedefinetheK-principalfrequency,λK
1 ,andtheK-torsionalrigidity,TK,as
λK1 (Ω) = min u∈W1,2 0 (Ω)\{0} Ωh2K(∇u) dx Ωu2dx , (1.1) and TK(Ω) = max u∈W1,2 0 (Ω)\{0} Ωu dx 2 Ωh2K(∇u) dx . (1.2) * Correspondingauthor.
E-mailaddresses:buttazzo@dm.unipi.it(G. Buttazzo),serena.guarinolobianco@unina.it(S. GuarinoLoBianco),
mmarini@sissa.it(M. Marini).
http://dx.doi.org/10.1016/j.jmaa.2017.03.055
It is convenient to introducethe functionHK = hK2 /2; when HK is sufficiently smooth,we canwrite the
Euler–Lagrangeequations fortheminimizersoftheproblems(1.1)and(1.2)togetaPDEinterpretationof theabovequantities.Indeed,theK-principalfrequencyisrelated totheeigenvalueproblem
−ΔKu = λK1u in Ω, u = 0 on ∂Ω, (1.3)
while theK-torsionalrigidityistheL1norm ofthesolutionu of:
−ΔKu = 1 in Ω, u = 0 on ∂Ω. (1.4)
Here ΔK denotestheFinsler–Laplace operator givenby
ΔKu = div(DHK(∇u)). (1.5)
IntheEuclideancase,occurringwhenK istheunitaryballB,(andhK(x)=|x|)theoperatorgivenin(1.5)
coincides withtheLaplacianand λ1 andT aretheusualfirstDirichleteigenvalueandtorsionalrigidity.
As in the linear case, the quantities defined in (1.1) and (1.2) are monotone, in opposite sense, with respect tothesetinclusion,i.e. if Ω1⊂ Ω2 then
λK1(Ω1)≥ λK1 (Ω2) and TK(Ω1)≤ TK(Ω2). (1.6)
Moreover, sinceHK isahomogeneousfunctionofdegree2,thefollowingscalingshold true:
λK1(tΩ) = t−2λK1 (Ω) and TK(tΩ) = tN +2TK(Ω), t > 0. (1.7)
Shapeoptimizationproblems involvingλ1 andT ,or evenmoregeneralspectral functionalsof theform
F(Ω)= Φ(λ1(Ω),T (Ω)),are widelystudiedintheliterature(seefor instance[2–6,11,12])and, asitis well
known,itispossibletogetbothlowerandupperboundsfortheprincipalfrequencyandthetorsionalrigidity intermsofquantitiesassociated tothegeometry ofthedomainΩ,suchas,forinstancetheperimeterand the volume (justthinkto theFaber–Krahn inequality and the Saint-Venant theorem, see for instance the recentbook[11]).
Asitshouldnotbeunexpected,ifweimposesomefurtherconstraintsintheclassofadmissibledomains, we can get strongerestimates. The class of convex domains, for instance, hasbeen consideredby several authors: on one hand the a priori assumption of the convexity of the domain naturally arises in many situations;ontheother,theclassofconvexsetshasstrongcompactnesspropertieswhichensuretheexistence of extremaldomainsforagreat numberofgeometricinequalities.
In this paper we are interested in estimates of the principal frequency and the torsional rigidity of a convex domainintermsoftheinradius,RΩ,i.e. theradius ofthebiggestball containedinΩ.
Animmediateconsequenceof(1.6)and(1.7)isthat,fortheEuclidean case
λ1(Ω)≤ λ1(BRΩ) = λ1(B1)R
−2
Ω . (1.8)
A classicalresultbyJ.Hersch(see[13])shows thatforanyconvex domainΩ⊂ R2 itholds
π2
4R
−2
Ω ≤ λ1(Ω), (1.9)
andtheinequalityissharp:ifweallowunboundeddomains,equalitycaseoccurswhenΩ isastrip,otherwise itisreachedonlyasymptotically,byasequenceofrectangleswithsidesa b.Hersch’stechniquehasbeen extended to convex domains of RN by M.H. Protter in [16] who proved the validity of (1.9) in every
Concerning the torsional rigidity, in [15] Polya and Szego proved that, for any domain Ω ⊂ R2, the
followinginequalityholdstrue
T (Ω) |Ω| ≥ 1 8R 2 Ω, (1.10)
where|Ω|, denotestheLebesguemeasure ofΩ.Again,this inequalityissharp,becominganequalityifΩ is aball.
Anupperboundfor thetorsionalrigiditywasobtainedbyE. Makaiin[14],whoprovedthat,forevery convexdomain Ω⊂ R2,itholdstruethat
T (Ω) |Ω| ≤ 1 3R 2 Ω. (1.11)
Inequality(1.11)isalsosharp,andthebestconstantisachieved,again,ifweconsiderasequenceofrectangles withsidesa b.
Theaimofthispaperistoextendtotheanisotropiccaseandtoageneraldimensiontheestimates(1.8),
(1.9),(1.10)and (1.11).Byvirtueof(1.6),and (1.7)itseemsreasonable tofindestimatesoftheform
c(N, K)(RKΩ)−2≤ λK1 (Ω)≤ C(N, K)(RΩK)−2 and c(N, K)(RKΩ)2≤ T K(Ω) |Ω| ≤ C(N, K)(RKΩ)2, where RK
Ω istheanisotropicinradius,i.e. thelargestnumbert suchthatx+ tK ⊆ Ω,forsomex∈ Ω. We
show that, surprisingly, for the best constants in the aboveformulas, there is no dependence on K, this meansthattheboundswhichareinforceintheEuclideancase,holdtrueforeverychoiceofthenonlinear operatorΔK.
Inthefollowingsection,weillustratemorepreciselyalltheresultsthatweprovethroughoutthispaper; herewe limitourselvesto stress that,besides theirown interest, theproofs ofsuch resultsmayprovide a moregeometrical insightfor the specialcaseK = B as well,and mayhelpto shed lighton whatare the mostrelevantassumptions thatwe needto impose onanonlinear operator, inorderto expectthose kind ofestimates.
2. Mainresults
In Section4, we extend to theanisotropic caseinequalities (1.10) and (1.11). More precisely, in Theo-rem 4.3,weshowthat,foranyconvexdomainΩ⊂ RN,andany2-homogeneousC2
+-regular1function,HK,
thefollowingapriori boundsontheK-torsionalrigidityholdstrue 1 N (N + 2)(R K Ω)2≤ TK(Ω) |Ω| ≤ 1 3(R K Ω)2. (2.1)
As for the corresponding linear case, equality can be achieved in the first inequality of (2.1) when Ω coincideswithK up totranslationsanddilations,while(asymptotic)equalityinthesecond inequalitycan
be obtainedbyconsideringasequenceof rectangleswithsides a b,see Proposition 4.4(notice that1/3 is thebest constantineverydimension).
As far as we know, inthe case K = B, even if the proof is significantly easier, there is not a specific reference forestimatesoftheform (2.1)ingeneraldimensionavailableintheliterature.
The proofs of both Theorems 4.2 and 4.3 are based on the choice of a suitable one dimensional test function.Since wedefinedthetorsionalasasupremum,itis easytogetanestimate frombelowusingthe variational formulation, while to get abound from above it is important to use the PDE interpretation explained in(1.4).
InSection5,weextendtheresultbyHerschandProtterbyprovingtheformula
π2
4 (R
K
Ω)−2≤ λK1 (Ω)≤ λ1(B)(RKΩ)−2. (2.2)
As inthe linearcase thefirst inequalityholds trueas an equality ifwe consider astrip (or asequence of “thin”rectangles, ifwedonotallowunboundeddomains).Thecaseofequalityinthesecond inequality occurs when wechooseΩ= x+ rK,forsomex∈ RN and r > 0;as inthelinearcase, thelatterestimate followsbyvirtueofthemonotonicityandthescalinglaws.WeproveinProposition 5.1thatλK
1 (K)= λ1(B).
In Section 6, we remove any regularity assumption on the convex body K. This result easily follows by approximating K with smooth convex bodies Kn and then by passing to the limit the corresponding
inequalitiesforthesequenceKn.
We concludethissection byremarkingthat,inarecentpaper [9]authors giveasharpestimate forthe anisotropic principalfrequencyand torsionalrigidity2 intermsofthe anisotropicperimeterand Lebesgue
measure.We pointoutthatthinrectanglesare limitsets fortheanisotropictorsionalrigidityas well;but curiously, whilesuch asequencemaximizes theratio betweentorsionalrigidityand volumeamong allthe setsofgivenanisotropicinradius,itminimizesthesameratioamongallsetsofgivenanisotropicperimeter. 3. Preliminaries
Inthissection,werecallsomebasicnotionsconcerningconvexbodiesandanisotropicdifferential opera-tors.
Associated withaconvexbodyK⊂ RN,there isthesupportfunction,hK :RN → R definedas
hK(x) = max
x· y : y ∈ K.
Thesupportfunctionisaconvexandpositive1-homogeneousfunction,whenK issymmetricwith respect to theorigin,hK isactuallyanorm, whoseunitball,K∗={x : hK(x)≤ 1},iscalledpolar body ofK.
We denotebydK theanisotropicdistance inducedbyK. Inparticular, givenaset A and apoint x we
denote distK(x, A) = inf hK∗(a− x) : a ∈ A . (3.1)
It isusefultodefine theanisotropic inradius intermsofsuchadistanceas
RKΩ = supdistK(x, ∂Ω) : x∈ Ω
.
We now recall someduality relations betweenK and K∗ that weuse inthe following sections and we referto [18, Chapter 1]foramorecomprehensiveaccount onthis subject.
2
Theyconsideraslightlymoregeneralclassofoperatorssincetheyallow,inournotations,functionsHKwhicharehomogeneous
K∗∗ = K, namely ∗ is an involution in the set of convex bodies; this means thatin all the following relations the role played by K and K∗ can be interchanged. In particular hK∗ is a norm as well, K =
x : hK∗(x)≤ 1
is thecorresponding unitballand (RN,hK∗),as aBanachspace, isthedual spaceof,
(RN,h K).
HK∗ = h2
K∗/2 is theLegendre–Fenchel transformation of HK = h2K/2; this entails that, whenboth the
functionsabovearedifferentiable,thegradientmappingsDHK andDHK∗ areonetheinverseoftheother,
namely
DHK(DHK∗(x)) = x for every x∈ RN. (3.2)
A completedescription ofthe differentialtheory forconvex bodiesand supportfunctionscanbe found in[18],herewelimitourselvestorecallthattheobstructiontotheregularityofhK isthepossiblepresence
of “flat” parts in the boundary of K; inparticular, if K is strictly convex, then K∗ (and thence hK) is
differentiableand
DhK(x) = y, (3.3)
where y is the only point in ∂K such that theouter normal unitto theboundary of K at y is x/|x|, in otherwords,DhK is the0-homogeneous extensionoftheinversefunctionof theGaussmap.Conversely,if
K hasdifferentiableboundary,thenhK∗ isdifferentiableandK∗ isstrictlyconvex.Moreover, forx∈ ∂K
DhK∗(x)/|DhK∗(x)| = νK(x), (3.4)
whereνK(x) is theouterunitnormalto ∂K atx.
Astraightforwardconsequenceof(3.3)isthefollowing importantformula:
hK∗(DhK(x)) = 1 for every x= 0. (3.5)
AsimilarcriterionholdstrueforhigherorderdifferentiabilityofHK:indeed,whenK isC+2-regular,i.e.
theboundaryofK canbewrittenasthegraphofatwicedifferentiablefunctionwithpositiveHessian,then both HK and HK∗ areC2-regular,and theirHessianmatrices arepositive.Byvirtueof theconsiderations
abovethedifferentialoperatorΔK,actingonC2functionsas
ΔKu(x) = div(DHK(∇u(x))),
is uniformly elliptic, and classicalexistence and regularity results for the solutionsof problems (1.3)and
(1.4)canbe applied. 4. K-torsionalrigidity
Throughout thissectionwe assumethatK isaC+2-regular symmetricconvexbody,so thatitssupport function,hK,isasmoothnormonRN.Wearegoingtoshowthatballs(withrespecttothemetricinduced
byhK)areextremalbodiesfor(2.1);inthefollowinglemmaweexplicitlycomputetheK-torsionalrigidity
forthosesets. Asweshallsee,this quantitydoesnotdependonthechoiceofthenorm. Lemma4.1. LetTK beasin (1.2),then
TK(rK)
|rK| = r2
Proof. We startbyconstructingthesolutionoftheproblem(1.4).From (3.2)we deduce ΔKHK∗(x) = div [DHK(DHK∗(x))] = div x = N.
Then thefunctionu= (r2− h2
K∗)/2N isasolutionof −ΔKu = 1 in rK, u = 0 on ∂(rK). Wehave rK h2K∗dx = r 0 {hK∗=t} h2 K∗ |DhK∗| dHN−1dt = r 0 t2 ∂(tK) 1 |DhK∗| dHN−1dt = r 0 tN +1 ∂K 1 |DhK∗| dHN−1dt = r N +2 N + 2 ∂K 1 |DhK∗| dHN−1 ,
where weusedtheco-areaandthechangeofvariableformulas.Letusevaluatethelastintegral.Thanksto
(3.4)and (3.5),and sincethesupportfunctionis1-homogeneous,weget ∂K 1 |DhK∗| dHN−1= ∂K hk Dh K∗ |DhK∗| dHN−1 = ∂K hK(νK(x)) dHN−1 = ∂K x· νK(x) dHN−1 = K div x dx = N|K|. Therefore weobtain TK(rK) = rK r2− hK∗ 2N dx = 1 2N r2|rK| − r N +2 N + 2N|K| = r 2|rK| N (N + 2) as required. 2
In the following proposition we prove a lower bound for the K-torsional rigidity for strictly convex domains. Insuchacase,thefunction
u(x) =1−
hΩ∗(x)
2
2N (4.1)
used in the above lemma to compute the K-torsional rigidity of balls is no longer the solution of the anisotropic torsion problem,butit still vanisheson theboundaryof the domain,and itcanbe used as a test function.Noticethat,sinceTK maximizesthequotient
⎛ ⎝ Ω u dx ⎞ ⎠ 2⎛ ⎝ Ω h2K(∇u) dx ⎞ ⎠ −1 ,
theneverytestfunctionprovidesalowerboundforthetorsion.Ourchoiceofthetestfunctionu ismotivated bythefactthatitprovides theoptimallower bound,as statedinTheorem 4.2below.Theassumption on thestrictconvexityofΩ isonlytechnicalandisremovedin
Theorem4.2. LetΩ⊂ RN be aconvexboundeddomain,then
TK(Ω) |Ω| ≥ 1 N (N + 2)(R K Ω)2. (4.2)
Moreover(4.2)issharp, andequalityoccurs onlyif Ω= x+ rK,forsome x∈ RN andr≥ 0.
Proof. We firstproveinequality(4.2)undertheassumption thatΩ isstrictly convex. Upto atranslation ofthedomain,wecanalwaysassumethatRKΩK⊆ Ω,sothathΩ≥ RKΩhK.Letu bethefunctionin(4.1);
u isanadmissiblefunctionfortheproblem(1.2),hence
TK(Ω) = max u∈W1,2 0 (Ω)\{0} Ωu dx 2 Ωh2K(∇u) dx ≥ Ωu dx 2 Ωh2K(∇u) dx .
SinceΩ isastrictlyconvexdomain,thefunctionhΩ∗isdifferentiableand,byrepeatingsimilarcomputations
asthose inLemma 4.1,wehave Ω h2Ω∗dx = 1 0 {h2 Ω∗=t} t |Dh2 Ω∗| dHN−1dt = 1 0 t 2 {h2 Ω∗=t} hΩ(DhΩ∗) hΩ∗|DhΩ∗| dHN−1dt = 1 0 tN/2 2 ∂Ω hΩ(νΩ(x)) dHN−1dt = 1 N + 2 ∂Ω x· νΩ(x) dHN−1 = 1 N + 2 Ω div x dx = N|Ω| N + 2. Hence ⎛ ⎝ Ω u dx ⎞ ⎠ 2 = |Ω| 2 N2(N + 2)2 . (4.3)
Wenowcomputethedenominator.Again,theco-areaandthechangeofvariableformulasgive Ω h2K(∇u) dx = 1 4N2 Ω h2K(Dh2Ω∗) dx
= 1 4N2 1 0 {h2 Ω∗=t} h2K(Dh2Ω∗) |Dh2 Ω∗| dHN−1dt (4.4) = 1 4N2 1 0 t(N−1)/2 ∂Ω h2 K(Dh2Ω∗) |Dh2 Ω∗| dHN−1dt.
Byusingthehomogeneityofthesupportfunctionsandbyrecallingequations(3.5)and(3.4),wehave
h2 K(Dh2Ω∗(x)) |Dh2 Ω∗(x)| = |Dh 2 Ω∗(x)| hΩ(DhΩ∗(x))· h2 K(Dh2Ω∗(x)) |Dh2 Ω∗(x)|2 = 2hΩ∗(x) hΩ(νΩ(x))· h 2 K(νΩ(x)) = 2t1/2h2K(νΩ(x)) hΩ(νΩ(x)) . (4.5)
Plugging(4.5)into(4.4)weget Ω h2K(∇u) dx = 1 4N2 1 0 tN/2 ∂Ω 2hΩ(νΩ(x)) h2K(νΩ(x)) h2 Ω(νΩ(x)) dHN−1(x) dt ≤ 1 4N2 1 0 tN/2 ∂Ω 2hΩ(νΩ(x)) (RK Ω)2 dHN−1(x) dt = 1 N (N + 2) |Ω| (RK Ω)2 , (4.6)
where weusedthefactthathΩ≥ RΩKhK.Combining (4.3)and(4.6)wefind:
TK(Ω) |Ω| ≥ 1 N (N + 2)(R K Ω)2, as required.
We are now left to show the validity of (4.2) without the assumption on the strict convexity of the domain Ω.
We recallthatstrictly convexbodies areadensesubset oftheset of convexbodieswith respect tothe topologyinduced bytheHausdorffdistance.Inparticular (seefor instance[17])thereexists asequenceof convex bodies Ωn ⊂ Ω such that hΩnS
N−1 converges to h
ΩSN−1 inthe C0-norm. Such a convergence
ensuresthatasn→ ∞
|Ωn| → |Ω| and RKΩn→ R
K
Ω. (4.7)
From (1.6),itfollowsthat
T (Ω)≥ T (Ωn),
and byapplying(4.2)toeachΩn equation(4.2),wefind
TK(Ω)≥ |Ωn|
N (N + 2)(R
K Ωn)
2,
which, combinedwith(4.7),givesthedesiredresult.
Notice that,byvirtueof Lemma 4.1, inequality(4.2)issharp; moreover,ifit holdstrueas an identity, then (4.6)as wellmustbe anequality,inparticular
∂Ω hΩ(νΩ(x)) h2 K(νΩ(x)) h2 Ω(νΩ(x)) dHN−1(x) = ∂Ω hΩ(νΩ(x)) (RK Ω)2 dHN−1(x).
Since the integrand function inthe left-hand side is pointwise lower than the onein the right-handside, and since they are continuous functions, they must coincide everywhere. Namely hΩ = RKΩhK, that is
Ω= RK ΩK. 2
Inthefollowing theorem weextendMakai’sresult(1.11) toageneraldimensionand to theanisotropic case. The proof given in[14] is mainlybased on a cleveruse of the Schwarz inequality, on an inequality involvingrealfunctionsandtheirderivatives,andontheinequality
|v|2≥ v2
i. (4.8)
Thesecondingredientisappliedtoafunctionwhichisaparametrizationoftheboundaryofthedomain (werecallthatinMakai’ssettingtheboundaryisasetof“dimensionone”),while(4.8)isatrivialinequality fortheEuclideannorm,butisfalseforagenericnorm.Noticethatthegeometricfeatureofthespherethat ensures the validity of (4.8) is the fact that each radius connecting the center to a boundary point is perpendiculartothetangentspaceatthatpoint.
Inourproofwehavetochangestrategy:asinMakais’sproofwedivideΩ intosuitablesubdomains;then weconstruct, foreachsubdomain,afamilyofonedimensionalobstaclefunctions,andfinally weusesome lineartransformationstogetananalogueof(4.8).
Theorem4.3. LetΩ⊂ RN be aboundedconvexbody,then
TK(Ω) |Ω| ≤ 1 3 RKΩ 2. (4.9)
Proof. Weprovetheclaim whenΩ is apolytope, andthen thevalidityof(4.9) followsbyapproximation. Letusdenote byF1,. . . ,Fn thefacetsofΩ,anddivideΩ intosubdomainsΩj definedas
Ωj=
x∈ Ω : dK(x, Fj)≤ dK(x, Fl) for every l= j
,
wheredK(x,Fj) isthedistancefunctiondefinedin(3.1).Foreveryj = 1,. . . ,n and ε> 0 wedefine
uεj(x) =−dK(x, Fj)
2
2 (1 + ε) + R
K
ΩdK(x, Fj)(1 + 2ε).
Noticethat,ifx∈ Ωi∩ Ωj,thenuεi(x)= uεj(x).
Wedenotebyνj theouterunitnormalto Fj.SincedK(x,Fj)= dist(x,Fj)h−1K (νj),thefunctionsuεj are
smoothintheinteriorofΩj;moreover,since∇dist(x,Fj)=−νj,wehave
∇uε j(x) = dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε) νj. Thus ΔK(uεj) = div DHK(νj) dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε) = 1 + ε h2 K(νj) DHK(νj)· (−νj) =−1 − ε.
Letusconsiderthefunctionuε: Ω→ R definedbyuε Ωj= u
ε
jΩj.Wenowprovethat,ifv isthesolutionof
theanisotropictorsionproblem onΩ,thenv(x)≤ uε(x) inΩ.Suppose,bycontradiction,thatthere exists a pointx∈ Ω such thatuε(x)< v(x); sincewe know thatuε= v on ∂Ω, thenthe functionuε− v hasa local minimum,sayx0insideΩ.
We show that x0 cannot be neither an interior point nora boundarypoint of each Ωj. If x0 ∈ int Ωj,
then ∇uε(x
0)=∇v(x0),so
D2HK(∇uε(x0)) = D2HK(∇v(x0)). (4.10)
Moreover,since∇uε(x
0)= 0 inΩj,thenalso∇v(x0)= 0.Thusv solves,inaneighborhoodofx0anelliptic
equation withcoefficientsinC0,α,and thenceisC2,α-regular(seeforinstance[10, Chapter 6]).Then
D2(uε− v)(x0)≥ 0. (4.11)
Inparticular,using(4.10)and(4.11),andsincethetraceoftheproductoftwopositive-definitematricesis positive,wehavethat
−ε = ΔKuε(x0)− ΔKv(x0) = trD2HK(∇uε(x0))D2uε(x0) − trD2HK(∇v(x0))D2v(x0) = trD2HK(∇uε(x0)) D2uε(x0)− D2v(x0) = trD2HK(∇uε(x0))D2(uε− v)(x0) ≥ 0, whichisacontradiction.
Wearelefttoshowthecontradictioninthecasewhenx0∈ Ωi∩Ωj,i= j.Letw(x)= v(x)−v(x0)+uε(x0);
then w(x0) = uε(x0) and, thanks to our assumption on x0, there exists a positive number r, such that
uε(x)≥ w(x),foreveryx∈ (x0+ rK)⊂ Ω.
Since, uε
j isaconcave functionofthedistance,wehavethat
uεj(x)≤ uεj(x0) + νj· (x − x0) dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε) (4.12)
for everyx∈ (x0+ rK).Withoutloss ofgeneralitywecanchooser < RKΩε/(1+ ε);thischoiceofr allows
us toconcludethat
dK(x, Fj)≤ r + dK(x0, Fj),
and thusthefactor
dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε),
appearingintheright-handsideof (4.12),nevervanishes,foreveryx∈ (x0+ rK).Indeed
dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε)≤r(1 + ε) hK(νj) +dK(x0, Fj)(1 + ε) hK(νj) − RK Ω hK(νj) (1 + 2ε) <dK(x0, Fj)(1 + ε) hK(νj) − RK Ω hK(νj) (1 + ε)≤ 0.
Moreover, since r < RK
Ωε/(1+ ε),for any other l such thatx0 ∈ Ωl, inside x0+ rK, uεl is anincreasing
functionoftheanisotropicdistancedK,then itfollowsfromthedefinitionofthesetsΩl that
uεl(x)≤ uεj(x) for every x∈ (x0+ rK)∩ Ωl,
thatentailsthat
uε(x)≤ uεj(x) for every x∈ (x0+ rK). (4.13)
Plugging(4.13)into(4.12) andrecallingthatuε(x
0)= uεj(x0),weobtain uε(x)≤ uε(x0) + νj· (x − x0) dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε) (4.14) foreveryx∈ (x0+ rK).
Arguinganalogouslyforthefunctionuε
i wefind uε(x)≤ uε(x0) + νi· (x − x0) dist(x, Fi) h2 K(νi) (1 + ε)− R K Ω hK(νi) (1 + 2ε) (4.15) foreveryx∈ (x0+ rK).Byimposingin(4.14)and(4.15) theconditionsv≤ uε andv = uεatx0,wefind
v(x)− v(x0)≤ νj· (x − x0) dist(x, Fj) h2 K(νj) (1 + ε)− R K Ω hK(νj) (1 + 2ε) and v(x)− v(x0)≤ νi· (x − x0) dist(x, Fi) h2 K(νi) (1 + ε)− R K Ω hK(νi) (1 + 2ε) .
Since v isdifferentiable at x0, bymultiplying both sides by the factor|x− x0|−1 and taking thelimit as
x→ x0weobtainthat∇u(x0) mustbeproportional bothtoνi andνj, thatisacontradiction.
Wecannow usethefunctionuj as abarrierfunction toget anestimateof thetorsionrigidityofΩ. In
order to compute Ω
ju
ε
j(x)dx,it isimportant to make surethateach subdomain Ωj can be writtenas a
graphoverthefacetFj.ThisisalwaysthecasewhenK istheEuclideanball,sinceitsradiusisorthogonal
tothetangentspace,whilethisis nottrueforageneralconvexbodyK.
Togetthesamecondition,we consideralineartransformation L (seeFig. 1),suchthatLνj⊥ = Id and
LDhK(νj)= hK(νj)νj.Noticethat,sinceDhK(νj)· νj = hK(νj)> 0,L iswell definedand
det L = 1. (4.16) MoreoverhK(νj)= hLK(νj),indeed hK(νj) = sup x∈Kx· νj = supy∈LKL −1y· ν j = sup y∈LK L−1y· νj+ (y· νj)L−1νj· νj , (4.17)
wherey = y+(y·νj)νj,andy·νj= 0.SinceLνj⊥= Id,thenL−1y·νj,andsinceDhK(νj)= hK(νj)L−1νj,
thenL−1νj· νj = 1 sothat,byrecalling(4.17)
hK(νj) = sup
Fig. 1. The linear transformation L.
WeshowthatourchoiceofthelinearapplicationL allowsusto describethesubdomainLΩj as
LΩj=
y + rνj : y∈ LFj, r∈ [0, ρj(y)]
.
ForthisweneedonlytoprovethatLΩjiscontainedinthecylinderCj = LFj× Rνj,sinceLΩj isaconvex
set.Westartnoticingthat,sincedistK(x,Fj)= distLK(Lx,LFj),
LΩj=
x∈ LΩ : dLK(x, LFj)≤ dLK(x, LFl), for every l= j
.
Suppose,bycontradiction,thatthereexistsapointx∈ LΩj\ Cj.Letr bethelargest numbersuchthat
x + rLK ⊆ LΩ.Itfollowsfrom thedefinitionofLΩj,thatthere existsapointy∈ x + rLK ∩ LFj.
Sincex + rLK ⊆ LΩ,thenormalconeofLΩ iscontainedinthenormalconeofx + rLK atthepointy,
and thenνj istheouter normalunitto x + rLK aty.Thereforex = y + rνj ∈ Cj.
Now wecompute Ωj uεj(x) dx = LΩj uεj(L−1y) dy = LΩj −dist2(x, Fj) 2h2 K(νj) (1 + ε) + R K Ωdist(x, Fj) hK(νj) (1 + 2ε) dx.
We used (4.16) for the first equality, we use thefact that distK(x,Fj) = distLK(Lx,LFj) and hK(νj)=
hLK(νj) forthesecond one.Finally
Ωj uεj(x) dx = LΩj −dist2(x, Fj) 2h2 K(νj) +R K Ωdist(x, Fj) hK(νj) dx + O(ε) = LFj ρj(y) 0 −dist2(y + rνj, Fj) 2h2 K(νj) +R K Ωdist(y + rνj, Fj) hK(νj) dr dy + O(ε) = LFj ρj(y) 0 − r2 2h2 K(νj) + R K Ωr hK(νj) dr dy + O(ε) = LFi − ρ3j(y) 6h2 K(νj) +R K Ωρ2j(y) 2hK(νj) dy + O(ε).
Sinceρj(y)≤ RKΩhK(νj),foreveryy∈ LFj,then − ρ2j 6h2 K(νj) + RKΩ ρj(y) 2hK(νj) ≤ 1 3 RKΩ 2. (4.18) Using(4.18) wefind Ωj uεj(x) dx≤ LFj ρj(y) RK Ω 2 3 dy + O(ε) =1 3 RKΩ 2|LΩj| + O(ε). Since|LΩj|=|Ωj|, weget TK(Ω)≤ Ω uεdx = j Ωj uεjdx ≤ 1 3 RKΩ 2 j |Ωj| + O(ε) = 1 3 RKΩ 2|Ω| + O(ε) asrequired. 2
Remark.The mostimportant inequality playing arole inthe proof of thetheorem above is (4.18). This inequalitymustbestrictforsomeboundarypoint,unfortunatelythisisnotenoughtoconcludethatequality in(4.9) isnever attained, becauseour proof isbased onan approximationprocedure. Inorder to get the strictsignin(4.9)wemusttotakeintoaccount andestimatestheremindertermsin(4.18).
− ρ 2 j 6h2 K(νj) + RKΩ ρj(y) 2hK(νj) =1 3 RKΩ 2−1 6 RKΩ − ρj hK(νj) 2 + ρjR K Ω 6hK(νj)− (RK Ω)2 6 ≤1 3 RΩK 2+ ρjR K Ω 6hK(νj)− (RKΩ)2 6 .
Arguingasintheproofabovewe canconcludethat Ωj v≤1 3 RΩK 2|Ω| + 1 6 Fj ρ2 jRKΩ hK(νj)− ρj (RKΩ)2 dy = 1 3 RKΩ 2|Ωj| + 1 6 Ωj 2RKΩdistK(x, Fj)− (RΩK)2 dx. Summingoverj wefind TK(Ω)≤ 1 3 RKΩ 2|Ωj| + RK Ω 6 Ω dK(x, ∂Ω)− RKΩ dx. (4.19)
Taking into account Equation(4.19) in order to show that theminimum in (4.9)is never attained we havejustto showthat,foreveryconvexbodyΩ
2
Ω
Fig. 2. The set Ωε= Cε∪ Dε.
LetusdenotebyΩt={x∈ Ω : dK(x,∂Ω)≥ t}.Since|∇dK(x)|= hK(νt(x))−1,whereνt(x) istheouter
normalto ∂Ωt atx,thankstheco-areaformulawecanwrite
Ω dK(x, ∂Ω) = RKΩ 0 ∂Ωt dk(x)hK(νt(x)) dHN−1(x)dt = RKΩ 0 tPK(Ωt) dt,
where PK denotestheanisotropicperimeter.
Ontheotherhand
RKΩ 0 dK(x, ∂Ω) dx = RKΩ 0 |Ωt| dt.
AsitiswellknowntPK(Ωt)+|Ωt|<|Ωt+ tK|;thisfactcanbeseenasaconsequenceofthecharacterization
of mixedvolumes(see,forinstance[18, Theorem 5.1.7]).SinceΩt+ tK⊆ Ω,wehave
2 Ω dK(x, ∂Ω) < RK Ω 0 |Ωt+ tK| dt ≤ RK Ω 0 |Ω| dt = RK Ω|Ω|.
Nonetheless, thefollowingexampletellsusthat(4.9)issharp. Proposition 4.4. LetΩε therectangle [−ε,ε]× [−a
2,a2]× . . . × [−aN,aN].Then lim ε→0+ TK(Ωε) RK Ωε 2 |Ωε| = 1 3.
Proof. LetΩε= Cε∪ Dε,whereCε= [−ε,ε]× [−a2+ ε,a2− ε]× . . . × [−aN+ ε,aN− ε],andDε= Ωε\ Cε,
as inFig. 2.Settingx= (x1,z) withz ∈ RN−1 anda= (a2,. . . ,aN),we considerthe functionuε defined
by ⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩ uε(x 1, z) = ε2− x2 1 2h2 K(e1) in Cε uε(x 1, z) = min |a − z|, | − a − z| ε2− x21 2εh2 K(e1) in Dε.
WecanestimatetheK-torsionalrigidity
TK(Ωε)≥ Ωεuε 2 Ωεh2K(∇uε) = Cεuε+ Dεuε 2 Cεh2K(∇uε) + Dεh2K(∇uε) .
Wenowcompute Cε uεdx = Cε ε2− x2 1 2h2 K(e1) dx = |C ε|ε2 3h2 K(e1) and Cε h2K(∇uε) dx = Cε x2 1 h2 K(e1) dx = |C ε|ε2 3h2 K(e1) .
WenoticethatbothDεuεdx and
DεHK(∇uε)dx arenegligible,sincetheygoto zeroasεN +2;moreover
|Dε| is negligibleas well,sinceitgoestozeroas εN.Byrecallingthat
RKΩε 2 = ε 2 h2 K(e1) , wehavethat lim ε→0 TK(Ωε) RK Ωε 2 |Ωε| ≥ limε→0 Cεuεdx 2 RK Ωε 2 |Ωε| Cεh2K(∇uε) dx = 1 3 which,togetherwith Theorem 4.3,concludestheproof. 2
5. Eigenvalue
Inthis section we extendto the anisotropiccase inequalities(1.8) and(1.9). As inthe linearcase, the upper bound for the principal frequency relies on the monotonicity and scaling laws. As we mentioned, allthe bestconstantinthose inequalitiesdo notdependon K;inthe followingproposition weshow that
λK
1(K)= j02,andthisgivesthesecondinequalityin(2.2).
AsisSection4,wealwaysassumethatK isaC2
+ regularsymmetricconvexbody.
Proposition5.1. LetΩ⊂ RN be aconvexbody,then
λK1(Ω)≤ λ1(B)(RΩK)−2.
Proof. From(1.6)and(1.7)wehavethat
λK1(Ω)≤ λK1 (K)(RKΩ)−2.
NowhavetoshowthatλK
1(K)λ1(B).Sincethere existsapositivefunctiong solving
−ΔKg = λK1g in K, g = 0 on ∂K,
wecanassumethatg isradial,namely g(x)= fH∗(x) (see[1]).Adirectcomputationshowsthat
f(t) + N f(t) =−λK1 f (t) and f (1) = 0. Indeed −λK 1 g = ΔKg = div DHK(∇g) = divDHK f(HK∗)DHK∗ = divf(HK∗)x = f(HK∗)DHK∗(x)· x + Nf(HK∗).
Settingg(x)= f (|x|2/2) iteasyto checkthat
Δg =−λK1 g in B, g = 0 on ∂B.
Since g > 0,theng isthefirstDirichleteigenfunctionoftheballB andthenceλK1 (K)= λ1(B). 2
Now we provethree technical lemmas thatare usefulinthe proof ofthe lower bound forthe principal frequency. Lemmas 5.2 and 5.3 are generalizations of a lemma by D. Gale and referred to as a private communicationbyM.H.Protterin[16].
Lemma5.2.LetΩ beaconvexdomain,letK⊆ Ω beaconvexbodywithdifferentiableboundary,andsuppose that foreveryτ > 1 andx∈ RN,x+ τ K Ω.Thenthereexisti∈ N andx1,. . . ,xi ∈ ∂Ω∩ ∂K suchthat,
denoted by νj theouterunit normalto∂K atxj,forevery ν∈ SN−1,there exists,j,such thatν· νj≥ 0.
Proof. Let ν ∈ SN−1; for ε > 0, we consider the set Kε = K + εν. Notice that Kε cannot be a proper
subset of Ω,since, otherwise,also asmalldilation ofKε shouldbe,against ourassumptions. Then, there
exists με∈ SN−1 suchthat
hΩ(με)≤ hKε(με) = hK(με) + εμε· ν.
Notice thatμε isthe outerunitnormal to∂Kε at somepointxε∈ Kε\ Ω.Since K ⊆ Ω,thenhK ≤ hΩ,
and thenμε· ν ≥ 0.
Uptoextractasubsequencewecanalwaysassumethat,asε goesto0,μεconvergestosomeunitvector
μ and xε convergestosomepointx∈ ∂K ∩ ∂Ω withthepropertythatμ· ν ≥ 0.
We haveproved thatCx=
ν ∈ SN−1 : ν· νx≥ 0
, x∈ ∂K ∩ ∂Ω isacoveringof theunitsphere;we are lefttoshowthatwecanextract fromCxafinitecoveringofthesphere.
WeproceedbyinductiononthedimensionN .IfN = 1 thereisnothingtoprove,sincethesphereS0isa finiteset.LetnowN > 1 andletOx=
ν ∈ SN−1 : ν·νx> 0
.IfSN−1 ⊆ ∪xOx,then,bycompactness,we
canextractafinitecoveringofthesphereSN−1.Otherwise,thereexistsaunitvectorν,suchthatν·νx≤ 0,
foreveryx∈ ∂K ∩ ∂Ω.
Let A=x∈ ∂K ∩ ∂Ω : νx· ν = 0
; weclaimthat{Cx}x∈A is acoveringofSN−1∩ ν⊥ =SN−2.Let
η ∈ SN−1, such that η· ν = 0 and let, for ε > 0, ηε = sin(ε)η + cos(ε)ν. Let xε ∈ ∂K ∩ ∂Ω,such that
νxε· ηε≥ 0.Sinceνxε· ν ≤ 0,wemusthaveνxε· η ≥ 0.
Upto extracting asubsequence xε convergesto somepointx. Clearlyν
x· ν ≥ 0 and, since∂K∩ ∂Ω is
closed, alsoνx· ν ≤ 0 andthusx∈ A.Thefactthatνx· η ≥ 0 proves ourclaim.
WecanfinallyapplyourinductionhypothesistofindafinitesubsetA⊆ A suchthatSN−2⊂ ∩x∈ACx,
and itisstraightforwardto checkthatalsoSN−2⊂
x∈ACx. 2
Lemma 5.3. LetΩ and K as above, thenthere exists apolyhedral convex domain T such that Ω⊆ T and,
forevery τ > 1 andx∈ RN,x+ τ K Ω.
Proof. Let x1,. . . ,xi and ν1,. . . ,νi be as inLemma 5.2. Weset T ={x : (x− xi)· νi ≤ 0}. SinceT is
theintersectionofsupportinghalf-spacesofΩ,thenΩ⊆ T .
Supposenow, thatthere existsx,τ > 1 suchthatx + τ K⊆ T .Letj be suchthatνj· x ≥ 0.Thepoint
xj = τ xj+ x∈ x + τK verifies
(xj− xj)· νj≥ (τ − 1)xj· νj> 0,
Lemma5.4. Forevery invertible linearmappingL:RN→ RN wehave
λK1 (Ω) = λLK1 (LΩ).
Proof. Bythedefinitionofsupportfunction,wehave
hL−1K(x) = sup
y∈L−1Ky, x = supz∈KL
−1z, x = sup z∈Kz, L
−tx = h
K(L−tx),
whereL−tdenotes theinverseofthetransposeofL.
Denotingbyv(x)= u(L−1x),againbyasimplecomputationwehave
∇v(x) = L−t∇u(L−1x)
and
hLK(∇v(x)) = hK(Lt∇v(x)) = hK(∇u(L−1x)).
Then,using thecomputationaboveandthechangeofvariable formula,wehave LΩh2LK(∇v) dx LΩv2dx = LΩ h2K(∇u ◦ L−1) dx LΩ(u◦ L−1)2dx = Ωh2K(∇u) dx Ωu2dx ,
provingthelemma. 2
Wearenow readyto provethevalidityof(2.2).Theproof iscarriedoutinthesamespiritoftheproof oftheresultachievedbyHerschin[13],eveniftheintroductionoftheanisotropyobligesustoavoidtouse Hersch’sestimatesand toadapttheproof.
Theorem5.5. LetΩ⊂ RN be aconvexbody,then
λK1(Ω)≥ π
2
4(R
K Ω)−2.
Proof. Uptoatranslationandadilationofthedomain,wecanassumethatΩ satisfiestheassumptionsof
Lemma 5.2(noticethatinthiscase,RK
Ω = 1)byrepeatingtheconstructionexplainedinLemmas 5.2 and 5.3
wefindaconvexpolyhedraldomain T⊇ Ω withthesameanisotropicunitaryinradius.Forj = 1,. . . ,i we
call Tj the pyramidaldomain obtainedbytaking theconvexhullof theoriginandthe j-thfacet Fj ofT ,
andweset Ωj= Ω∩ Tj.
Letnowu bethefirsteigenfunction,wecompute
λK1 (Ω) = Ωh 2 K(∇u) dx Ωu2dx = j Ωjh 2 K(∇u) dx j Ωju 2dx ≥ min j Ωjh 2 K(∇u) dx Ωju 2dx = minj Tjh 2 K(∇u) dx Tju 2dx .
Asweshallsee, toconcludeitisenoughtogetanestimatefrombelowofthequantity Tjh 2 K(∇u) dx Tju 2dx .
To thisaim,as wedidinTheorem 4.9,weconsider alinearmapL such thatLνj⊥= Id andLDhK(νj)=
hK(νj)νj.ArguingasinLemma 5.4,itiseasytoshowthat
Tjh 2 K(∇u) dx Tju 2dx = LTjh 2 LK(∇f) dx LTjf 2dx ,
where f (x)= u(L−1x).MoreoveritisclearthatLTj isagraphoverFj.
Forany functionf definedonTj wehave
h2LK(∇f) ≥ h2LK(νj)(∇νjf )
2. (5.1)
Indeed,sinceDhK(νj)∈ K,thenhK(νj)νj∈ LK andthence,byrecallingthathK(νj)= hLK(νj),wehave
hLK(∇f) ≥ hK(νj)νj· ∇f = hLK(νj)∇νjf.
Werecallthat,foranytest functionv(t): [0,]→ R,suchthatv(0)= 0 it holdstruethat 0v(t) 2dt 0v(t)2dt ≥ π2 42 . (5.2)
Thanksto (5.1)and(5.2), wecanconcludebycomputing Tjh 2 K(∇u) dx Tju 2dx = LFj (y) 0 h 2 K(∇f) dy dt LFj (y) 0 f2dy dt ≥ LFj (y) 0 h 2 K(νj)∇νjf dy dt LFj (y) 0 f2dy dt = h2K(νj) ⎡ ⎢ ⎣ LFj (y) 0 ∇νjf dt (y) 0 f2dt (y) 0 f 2dt 1 dy ⎤ ⎥ ⎦ 1 LFj (y) 0 f2dy dt ≥ h2 K(νj) ⎡ ⎢ ⎣ LFj π2(y) 0 u 2dt 4(y)2 ⎤ ⎥ ⎦ 1 LFj (y) 0 u2dy dt ≥ h2 K(νj) π2 42 max ,
where max,byconstruction,isRKΩhK(νj). 2
6. Regularity
IntheprevioussectionswelimitourselvestothecaseofnormsassociatedtoaC2
+-regularconvexbodyK.
InSection3,wementionedthatthisassumptionisnecessarytomaketheFinslerLaplaceoperatoruniformly elliptic, nonetheless thedefinitionsoftheK-torsionalrigidityandtheK-principal frequencyvia(1.2)and
(1.1)makesense evenforlessregularnormshK.Moreover,severalinterestingexamplesofapplicationsfor
those inequalities in the case of non-Euclidean norms arise when we consider the norms associated to a square, ormoreingeneral,p-normoftheform
vp =
|vi|p
1/p
.
Notice that,for 1≤ p< 2 and p=∞, thefunction · 2
p isnot C2-regular, while for2< p <∞ itis
smooth butits Hessianmayvanish.Thus allthese cases arenotcovered bythe theorems thatweproved so far.
In this section we remove any regularity assumption on the convex body K. This result, contained in
Theorem 6.2,reliesonaresultbythefirstauthorandDalMaso(see[7])concerningtheΓ-convergence(we referto[8]foranexhaustivereviewonthetopic) ofsomeintegralfunctionalsoftheform
Fh(u) = Ω fh(x, u, Du) dx , toafunctional F (u) = Ω f (x, u, Du) dx
bymeansofthepointwiseconvergenceoftheintegrandfunctionsfh toafunctionf .
Forthereaderconvenienceletusstatetheabovementionedresult.
Theorem 6.1 ([7]).Assume that all fh = fh(x,u,p) areconvex in p and that the sequence (f )h converges
pointwise to a function f . Assume that there exist two increasing continuous functions μ,ν : R+ → R+, with μ(0)= ν(0)= 0 and μ concave, suchthat
|fh(x, u, p)− fh(y, v, p)| ≤ ν(|x − y|)(1 + fh(x, u, p)) + μ(|u − v|),
foreach x,y∈ Ω,u,v∈ R,p∈ Rn.Then
Γ- lim Fh(u) = F (u).
Our strategyis thefollowing: weapproximate aconvex bodyK with a sequenceof C+2-regularconvex bodiesKn,andwepasstotheΓ-limitthedesiredinequalities.
Theorem6.2. LetK andΩ be convexbodies,then (2.1)and(2.2)holdtrue.
Proof. Welimitourselvestoprove(2.2),beingtheproofof(2.1)completelyanalogous.LetKnbeasequence
ofC2
+-regularconvexbodiesconvergingtoK withrespect totheHausdorffmetric(theexistenceofsucha
sequenceisanoldtheorem byMinkowski,butwereferto [17]forashorterproof),andlet
Fn(u) =
Ω
h2Kn(∇u) dx ,
u∈ X, where X is the functional space madeof all functionsinW01,2(Ω),with L2 norm equalto 1. The
minimizersofthefunctionalsFn,sayun,satisfy
π2 4 (R Kn Ω )−2 ≤ Fn(un)≤ j 2 0(R Kn Ω )−2. (6.1) NoticethatRKn
Ω → RKΩ.Moreover,sincehKn→ hK andsincehK(v)≤ cv,forsomepositiveconstantc,
then thesecond inequalityin(6.1)givesthe equi-boundednessof theminimizersun inW1,2, thatensures
theexistence,uptoextractingasubsequence,ofaweaklimit,u.
Sincethefunctionals Fn satisfytheassumptions ofTheorem 6.1,wehavethatF (u)=
Ωh 2
K(∇u)dx is
theΓ-limitofthefunctionalsFn and,sinceminimizersconverge tominimizers,weobtain
Since both the right-handside and the left-handside of (6.1)are converging,it is possible to passto the limitthose inequalitiesto findthedesiredresult. 2
Acknowledgments
Theworkofthefirstand secondauthors ispartoftheproject2015PA5MP7“Calcolodelle Variazioni”
fundedbytheItalianMinistryofResearchandUniversity.ThethirdauthorissupportedbytheMIUR SIR-grant“GeometricVariationalProblems” (RBSI14RVEZ).TheauthorsaremembersoftheGruppoNazionale perl’AnalisiMatematica,laProbabilitàeleloroApplicazioni(GNAMPA)oftheIstitutoNazionalediAlta Matematica (INdAM).The authors wish to thank the anonymous referee for several suggestions and re-marks.
References
[1]A.Alvino,V.Ferone,P.L.Lions,G.Trombetti,Convexsymmetrizationandapplications,Ann.Inst.H.PoincaréAnal. NonLinéaire14(1997)275–293.
[2]M. van den Berg, V. Ferone,C. Nitsch, C.Trombetti, On Pólya’s inequalityfor torsional rigidity and first Dirichlet eigenvalue,IntegralEquationsOperatorTheory86 (4)(2016)579–600.
[3]L.Brasco,Ontorsional rigidity andprincipalfrequencies: an invitationtotheKohler–Jobin rearrangement technique, ESAIMControlOptim.Calc.Var.20 (1)(2014)315–338.
[4]D.Bucur, G.Buttazzo,VariationalMethodsinShapeOptimizationProblems,Progr.Nonlinear DifferentialEquations Appl.,vol. 65,Birkhäuser Verlag,Basel,2005.
[5]G.Buttazzo,Spectraloptimizationproblems,Rev.Mat.Complut.24(2011)277–322.
[6]G.Buttazzo, M. van den Berg, B.Velichkov, Optimizationproblems involving the first Dirichlet eigenvalueand the torsionalrigidity,in:NewTrendsinShapeOptimization,BirkhäuserVerlag,Basel,2015,pp. 19–41.
[7]G.Buttazzo,G.DalMaso,Γ-limitsofintegralfunctionals,J.Anal.Math.37(1980)145–185.
[8]G.DalMaso,AnIntroductiontoΓ-Convergence,Progr.NonlinearDifferentialEquationsAppl.,vol. 8,Birkhäuser-Verlag, Basel,1993.
[9]F.DellaPietra,N.Gavitone,Sharpboundsforthefirsteigenvalueandthetorsionalrigidityrelatedtosomeanisotropic operators,Math.Nachr.287 (2–3)(2014)194–209.
[10]D.Gilbarg,N.S.Trudinger,EllipticPartialDifferentialEquationsofSecondOrder,secondedition,Springer,1997. [11]A.Henrot(Ed.),ShapeOptimizationandSpectralTheory,DeGruyter,Berlin,2017,inpress.
[12]A.Henrot,M.Pierre,VariationetOptimisationdeFormes. UneAnalyseGeometrique,Math.Appl., vol. 48, Springer-Verlag,Berlin,2005.
[13]J. Hersch, Sur lafréquence fondamentale d’une membrane vibrante:evaluations par defaut etprincipe de maximum, Z. Angew.Math.Phys.11(1960)387–413.
[14]E.Makai,Ontheprincipalfrequencyofamembraneandthetorsionalrigidityofabeam,in:StudiesinMath.Analysis andRelatedTopics,StanfordUniv.Press,Stanford,1962,pp. 227–231.
[15]G.Polya,G.Szego,IsoperimetricInequalitiesinMathematicalPhysics,Ann.ofMath.Stud.,vol. 27,PrincetonUniversity Press,Princeton,1951.
[16]M.H. Protter, Lower Boundsfor the First Eigenvalueof Elliptic Equations andRelated Topics, Tech. ReportNo. 8, AFOSR,Univ.ofCalifornia,Berkeley,1958.
[17]M.Schmuckenschläger,AsimpleproofofanapproximationtheoremofH.Minkowski,Geom.Dedicata48(1993)319–324. [18]R.Schneider, Convex Bodies:TheBrunn–Minkovski Theory, second edition,CambridgeUniversity Press, Cambridge,