Extremal functions for the anisotropic Sobolev inequalities Fonctions minimales pour des inégalités de Sobolev anisotropiques
A. El Hamidi
a, J.M. Rakotoson
b,∗aLaboratoire de Mathématiques, Université de La Rochelle, Av. Michel Crépeau, 17042 La Rochelle cedex 09, France bLaboratoire de Mathématiques, U.M.R. 6086, Université de Poitiers, SP2MI, Boulevard Marie et Pierre Curie, Téléport 2,
BP 30179, 86962 Futuroscope Chasseneuil cedex, France
Received 2 September 2005; received in revised form 31 May 2006; accepted 21 June 2006 Available online 19 December 2006
Abstract
The existence of multiple nonnegative solutions to the anisotropic critical problem
− N
i=1
∂
∂xi ∂u
∂xi
pi−2∂u
∂xi
= |u|p∗−2u inRN
is proved in suitable anisotropic Sobolev spaces. The solutions correspond to extremal functions of a certain best Sobolev constant.
The main tool in our study is an adaptation of the well-known concentration-compactness lemma of P.-L. Lions to anisotropic operators. Furthermore, we show that the set of nontrival solutionsSis included inL∞(RN)and is located outside of a ball of radiusτ >0 inLp∗(RN).
©2006 Elsevier Masson SAS. All rights reserved.
Résumé
Nous montrons l’existence d’une infinité de solutions positives pour le problème anisotropique avec exposant critique. La mé- thode consiste à regarder la meilleure constante d’une inégalité du type Poincaré–Sobolev et à adapter le fameux principe de concentration-compacité de P. L. Lions. De plus, on montre que l’ensemble des solutionsS est contenu dansL∞(RN)et est localisé en dehors d’une boule de rayonτ >0 dansLp∗(RN).
©2006 Elsevier Masson SAS. All rights reserved.
Keywords:Quasilinear problems; Concentration-compactness; Anisotropic Sobolev inequalities
* Corresponding author.
E-mail address:rako@mathlabo.univ-poitiers.fr (J.M. Rakotoson).
0294-1449/$ – see front matter ©2006 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.anihpc.2006.06.003
1. Introduction
In this paper, the existence of nontrivial nonnegative solutions to the anisotropic critical problem
− N
i=1
∂
∂xi ∂u
∂xi
pi−2∂u
∂xi
= |u|p∗−2u inRN (1)
is studied, where the exponentspi andp∗satisfy the following conditions pi>1,
N
i=1
1 pi >1,
and the critical exponentp∗is defined by p∗:= N
N
i=1 1 pi−1.
In the best of our knowledge, anisotropic equations with different orders of derivation in different directions, involving critical exponents were never studied before. In the subcritical case, we can refer the reader to the recent paper by I. Fragala et al. [4].
In the special casepi=2,i∈ {1,2, . . . , N}, Problem (1) is reduced to the limiting equation arising in the famous Yamabe problem [13]:
−u=u2∗−1, u >0 inRN. (2)
Indeed, let(M, g)be aN-dimensional Riemannian manifold andSgbe the scalar curvature of the metricg. Consider a conformal metricg˜onMdefined byg˜:=uN4−2gwhose scalar curvature (which is assumed to be constant) is denoted bySg˜, whereuis a positive function inC∞(M,R). The unknown functionusatisfies then
−gu+ N−2
4(N−1)Sgu= N−2
4(N−1)Sg˜u2∗−1, u >0 inM, (3)
whereg denotes the Laplace–Beltrami operator. It is clear that, up to a scaling, the limiting problem of (3) (Eq. (3) without the subcritical term 4(NN−−21)Sgu) is exactly (2). The question of existence of minimizing solutions to (2) was completely solved by Aubin [1] and G. Talenti [9]. Their proofs are based on symmetrization theory. Notice that this theory is not relevant in our context since the radial symmetry of solutions cannot hold true because of the anisotropy of the operator.
In [5], P.-L. Lions introduced the famous concentration-compactness lemma which constitutes a powerful tool for the study of critical nonlinear elliptic equations. The concentration-compactness lemma allows an elegant and simple proof of the existence of solutions to (2) by minimization arguments. In the present work, we will adapt the concentration-compactness lemma to the anisotropic case and show that the infimum
|u|Lp∗Inf(RN )=1
N
i=1
1 pi
∂u
∂xi pi
pi
is achieved, of course, the functional space has to be specified.
The motivation of the present work is to give a new result which can provide extremal functions associated to the critical level corresponding to anisotropic problems involving critical exponents. Notice that the genuine extremal functions are obtained by minimization on the Nehari manifold associated to the problem and the critical level is nothing than the energy of these extremal functions.
The natural functional framework of Problem (1) is the anisotropic Sobolev spaces theory developed by [3,6,11,7, 8,10]. Then, letD1,p(RN)be the completion of the spaceD(RN)with respect to the norm
u1,p:=
N
i=1
∂u
∂xi
pi
.
It is well known that(D1,p(RN), · 1,p)is a reflexive Banach space which is continuously embedded inLp∗(RN).
In what follows, we will assume that p+=max{p1, p2, . . . , pN}< p∗,
thenp∗is the critical exponent associated to the operator:
N
i=1
∂
∂xi ∂
∂xi pi−2 ∂
∂xi
.
The spaceD1,p(RN)can also be seen as D1,p(RN)=
u∈Lp∗ RN
: ∂u
∂xi
∈Lpi RN
.
In the sequel, we will setp−=min{p1, p2, . . . , pN},p+=max{p1, p2, . . . , pN}andp=(p1, p2, . . . , pn). Also, the integral symbol
will denote
RN and · pi will denote the usual Lebesgue norm inLpi(RN). We denote by M(RN)(resp.M+(RN)) the space of finite measures (resp. positive finite measures) onRN, and by · its usual norm.
2. Existence of extremal functions for a Sobolev type inequality
In this paragraph, we shall prove that a certain best Sobolev constant is achieved.
Theorem 1.Under the above assumptions onpi,i=1, . . . , N,N2, there exists at least one functionu∈D1,p(RN), u0,u=0:
− N
i=1
∂
∂xi ∂u
∂xi
pi−2∂u
∂xi
=up∗−1 inD RN
.
The proof will need two fundamental lemmas, the first one is a result due to M. Troisi [10]:
Lemma 1.(Troisi [10])There is a constantT0>0depending only onpandNsuch that:
T0up∗ N
i=1
∂u
∂xi
1 N
pi
and up∗ 1 N T0
N
i=1
∂u
∂xi
pi
,
for allu∈D1,p(RN).
The second lemma is a rescaling type result ensuring the conservation of suitable norms:
Lemma 2.Letαi=pp∗i −1,i=1, . . . , N. For everyy∈RN,u∈D1,p(RN), andλ >0, if we writex=(x1, . . . , xN), y=(y1, . . . , yN),v(x)˙=uλ,y(x)=λu(λα1x1+y1, . . . , λαNxN+yN), we get
up∗= vp∗, ∂u
∂xi
pi
= ∂v
∂xi
pi
, fori=1, . . . , N, thus,u1,p= uλ,y1,p.
Proof. Noticing thatN
i=1αi =p∗, a straightforward computation with adequate changes of variables gives the result. 2
Lemma 3.Let
S= Inf
u∈D1,p(RN),up∗=1
N
i=1
1 pi
∂u
∂xi
pi
pi
.
ThenS >0.
Proof. From Lemma 1, we obtain that ifup∗=1, then N
i=1
∂u
∂xi
pi
N T0>0. (4)
Using standard argument, the infimum Inf
N
i=1
1
piapii, (a1, . . . , an)∈RN, N
i=1
aiN T0, ai0 .
=S1
is achieved and thus this minimum is positive. By relation (4), one concludes thatSS1>0. 2
Corollary 1 of Lemma 3 (Sobolev type inequality). Letp−=min(p1, . . . , pN),p+=max(p1, . . . , pN)andF be the real valued function defined by
F (σ )=
σp+ ifσ1, σp− ifσ1.
Then for everyu∈D1,p(RN), one has SF
up∗
N
i=1
1 pi
∂u
∂xi
pi
pi
˙=P (∇u).
Proof. Letube inD1,p(RN). Ifu=0 the inequality is true. Ifu=0, setw=u/up∗, then from the definition ofS one has:
N
i=1
1 pi
∂w
∂xi pi
pi
S. (5)
Sincetpitp+ ift >1 andtpitp−otherwise, the result follows from relation (5) and the definition ofF. 2 Remark 1.Along this paragraph, we only need the inequality:
Supp+∗ P (∇u) wheneverup∗1.
We shall call(P)the minimization problem (P) Inf
up∗=1
N
i=1
1 pi
∂u
∂xi pi
pi
= Inf
up∗=1
P (∇u) .
Let(un)⊂D1,p(RN)be a minimizing sequence for the problem(P). As in [5] and Willem [12], we define the Levy concentration function:
Qn(λ)= sup
y∈RN
E(y,λα1,...,λαN)
|un|p∗dx, λ >0.
HereE(y, λα1, . . . , λαN)is the ellipse defined by
z=(z1, . . . , zN)∈RN, N
i=1
(zi−yi)2 λ2αi 1
withy=(y1, . . . , yN)andαi>0 as in Lemma 2. Since for everyn, limλ→0Qn(λ)=0 andQn(λ) −→
λ→+∞1. There existsλn>0 such thatQn(λn)=12. Moreover there existsyn∈RNsuch that
E(yn,λαn1,...,λαNn )
|un|p∗dx=1 2.
Thus by a change of variables one has forvn .
=uλnn,yn:
B(0,1)
|vn|p∗dx=1 2= sup
y∈RN
B(y,1)
|vn|p∗dx.
Since
vnp∗= unp∗, ∂vn
∂xi
pi
= ∂un
∂xi
pi
, P (∇un)=P (∇vn)
we deduce that(vn)is bounded inD1,p(RN)and is also a minimizing sequence for(P). We may then assume that:
• vn vinD1,p(RN),
• |∂x∂i(vn−v)|pi μi inM+(RN),
• |vn−v|p∗ νinM+(RN),
• vn→va.e. inRN. We define:
μ= N
i=1
1 pi
μi,
μ∞= lim
R→+∞lim
n
N
i=1
1 pi
|x|>R
∂vn
∂xi
pidx, (6)
ν∞= lim
R→+∞lim
n
|x|>R
|vn|p∗dx. (7)
We start with some general lemmas. First by the Brezis–Lieb’s Lemma [2], direct computations give the following Lemma 4.
|vn|p∗|v|p∗+ν inM+ RN
.
The lemma which follows gives some reverse Hölder type inequalities connecting the measures ν, μ and μi, 1iN.
Lemma 5.Under the above statement, one has for allϕ∈Cc∞(RN)
|ϕ|p∗dν p∗1
1
T0 N
i=1
|ϕ|pidμi
1
Npi,
|ϕ|p∗dν p∗1
p
1 N+p∗1
+ μN1+p∗1 −p1+ · 1 T0
|ϕ|p+dμ 1
p+
.
Proof. Letϕ∈Cc∞(RN)and setwn=vn−v. Since
|ϕxi|pi|wn|pidx −→
n→+∞0, we then have:
limn ∂
∂xi(ϕwn)
pidx=lim
n
|ϕ|pi ∂wn
∂xi
pidx=
|ϕ|pidμi. (8)
Thus from Lemma 1, it follows that
|ϕ|p∗dν p∗1
=lim
n
|ϕwn|p∗dx p∗1
1
T0
N
i=1
|ϕ|pidμi
1
Npi. (9)
On the other hand, since
|ϕ|pidμip+
|ϕ|pidμp+μ1−ppi+
|ϕ|p+dμ ppi
+ (10)
applying the estimates (9) and (10) and knowing thatN i=1
1
pi =1+pN∗, we deduce
|ϕ|p∗dν p∗1
p
1 N+p∗1
+ μN1+p∗1 −p1+ · 1 T0
|ϕ|p+dμ 1
p+
.
This ends the proof. 2
We then havevp∗1. So ifvp∗=1 thenvis an extremal function sinceP (∇v)lim infnP (∇vn)=Sand SP (∇v). Thus, we want to show that fact, by proving that if it is not true then we have a concentration ofνat a single point and thereforev=0.
Main Lemma.
vp∗=1.
The remainder of this section is devoted to the proof of the main Lemma Lemma 6.Ifv=0then
limn vn−vpp∗∗=1− vpp∗∗<1.
Proof. From Brezis–Lieb’s Lemma we have:
limn
vnpp∗∗− vn−vpp∗∗
= vpp∗∗,
Sincevnp∗=1, we derive the result. 2 Lemma 7.
Sνpp∗+ μ.
Proof. For largen, according to Lemma 6, we have:
|vn−v|p∗dx1.
Thus for allϕ∈Cc∞(RN),|ϕ|∞1, it holds:
S
|ϕ|p∗|vn−v|p∗ pp∗+
N
i=1
1 pi
|ϕ|pi
∂(vn−v)
∂xi
pidx+on(1).
Lettingn→ +∞, one gets:
S
|ϕ|p∗dν pp∗+
N
i=1
1 pi
|ϕ|pidμiμ. (11)
Using the density ofCc∞(RN)inCc(RN), we get then S
sup
ϕ∈Cc(RN),|ϕ|∞=1
|ϕ|p∗dν pp∗+
μ,
that is the desired result. 2
Lemma 8.LetψR be inC1(R),0ψR1,ψR=1 if|x|> R+1,ψR(x)=0 if|x|< R. Then for any γi >0, i=0, . . . , N, the two equalities
ν∞= lim
R→+∞lim
n
|vn|p∗ψRγ0dx,
μ∞= lim
R→+∞lim
n
N
i=1
1
pi ∂vn
∂xi
piψRγidx
hold true, whereν∞andμ∞are defined by(6), (7).
Proof. As in Willem [12], one has:
|x|>R+1
|vn|p∗dx
|vn|p∗ψRγ0dx
|x|>R
|vn|p∗dx,
|x|>R+1
∂vn
∂xi
pidx ∂vn
∂xi
ψRγi
|x|>R
∂vn
∂xi
pidx.
We conclude with the definition ofν∞andμ∞. 2
Lemma 9.Letwn=vn−v. Then, for anyγi>0,i=0, . . . , N, we get ν∞= lim
R→∞lim
n
|wn|p∗ψRγ0dx, and
μ∞= lim
R→∞lim
n ∂wn
∂xi
piψRγidx.
Proof. Since
R→+∞lim
|v|p∗ψRγ0= lim
R→+∞ ∂v
∂xi
piψRγidx=0.
Thus
Rlim→∞lim
n
|wn|p∗ψRγ0dx= lim
R→∞lim
n
|vn|p∗ψRγ0dx=ν∞ and
Rlim→∞lim
n
N
i=1
1
pi ∂wn
∂xi
piψRγidx= lim
R→∞lim
n
N
i=1
1 pi ∂vn
∂xi
piψRγidx. 2
Lemma 10.
Sν
p+
∞p∗ μ∞.
Proof. From Lemma 6, we know that fornlarge enough, we have
ψRp∗|wn|p∗
|wn|p∗dx1.
Thus by Sobolev inequality (Corollary 4 of Lemma 3), it follows S |ψRwn|p∗dx
pp∗+
N
i=1
1 pi ∂
∂xi
(ψRwn) pi, S
R→+∞lim lim
n
|ψRwn|p∗dx pp∗+
lim
R→+∞lim
n
N
i=1
1 pi ∂
∂xi(ψRwn)
pi. (12) Since
limn
N
i=1
1
pi ∂ψR
∂xi
pi|wn|pi=0,
then
R→+∞lim lim
n
N
i=1
1 pi ∂
∂xi(ψRwn)
pi= lim
R→+∞lim
n
N
i=1
1
pi ∂wn
∂xi
piψRpi=μ∞.
Relation (12) and Lemma 9 give:
Sν
p+
∞p∗ μ∞. 2
Following again the arguments used in [12] we claim that:
Lemma 11.
1=lim
n vnpp∗∗= vpp∗∗+ ν +ν∞. Proof. From Lemma 4, we have:
|vn|p∗|v|p∗+ν.
Thus
R→+∞lim lim
n 1−ψRp∗
|vn|p∗dx=
|v|p∗dx+
dν.
Rewritingvnpp∗∗as
vnpp∗∗= 1−ψRp∗
|vn|p∗+
ψRp∗|vn|p∗, we obtain
limn vnpp∗∗= lim
R→+∞lim
n 1−ψRp∗
|vn|p∗+ lim
R→+∞lim
n
ψRp∗|vn|p∗= vpp∗∗+ ν +ν∞. 2
Next, we shall prove the following corollary:
Corollary 1 (Of Lemma 5). There exists an at most countable index set J of distinct points {xj}j∈J ⊂RN and nonnegative weightsaj andbj,j∈J, such that:
(1) ν=
j∈Jajδxj.
(2) μ
j∈Jbjδxj. (3) Sa
p+ p∗
j bj,∀j∈J.
Proof. The proof follows essentially the concentration compactness principle of P.L. Lions [5] because we have the reverse Hölder type inequalities of Lemma 5.
Indeed, the second statement of this lemma implies that for all Borelian setsE⊂RN, one has:
ν(E)cμμ(E)
p∗
p+. (13)
Since the setD= {x∈RN: μ({x}) >0}is at most countable becauseμ∈M(RN), thereforeD= {xj, j∈J}and bj=. μ({xj})satisfiesμ
j∈Jbjδxj.
Relation (13) implies thatνis absolutely continuous with respect toμ, i.e.,νμand ν(B(x, r))
μ(B(x, r))cμμ
B(x, r)p∗
p+−1
,
provided thatμ(B(x, r))=0 (remember thatp∗> p+). Thus, we have:
ν(E)=
E rlim→0
ν(B(x, r)) μ(B(x, r))dμ(x), and
Dμν(x)=lim
r→0
ν(B(x, r))
μ(B(x, r))=0, μa.e. onRN\D.
Setting aj=Dμν(xj)bj, relation (13) implies thatν has only atoms that are given by{xj}, that we have already get. 2
Letϕ∈Cc∞(RN),ϕ(xj)=1,ϕ∞=1. Then, using statement (1) of this corollary and relation (11), we have Sa
pp∗+
j S
|ϕ|p∗dν pp∗+
N
i=1
1 pi
|ϕ|pidμi. (14)
We shall consider φ ∈ Cc∞(RN), 0 φ1, support(φ)⊂B(0,1), φ(0)=1. We fix j ∈J and set xj = (xj,1, . . . , xj,N), qi =pip∗/(p∗−pi), i =1, . . . , N. Then αi .
= q1i satisfy N
k=1αk−αiqi =0. For ε >0, we define, for everyz∈RN,z=(z1, . . . , zN):
φε(z)=φ
z1−xj,1
εα1 , . . . ,zN−xj,N εαN
. (15)
Thus we have:
∂φε
∂xi
qi= ∂φ
∂xi
qi(z)dz (16) and then
∂φε
∂xi
pi|v|pi ∂φ
∂xi
qidz
1−p∗pi
B(xj,maxiε
1q i)
|v|p∗dz pip∗
−→ε→00. (17)
Lemma 12.Letxj∈Dandφεbe the function defined above associated toxj. Then:
Sa
p+ p∗
j lim
ε→0lim
n
N
i=1
1 pi
φεpi
∂vn
∂xi pidx.
Proof. Since 0φε1 then
φpε∗|vn|p∗dx1. From Corollary 4 of Lemma 3, it follows S
φεp∗|vn|p∗dx pp∗+
N
i=1
1 pi ∂
∂xi(φεvn)
pi. (18) From relation (17), we have
εlim→0 ∂φε
∂xi
pi|v|pidx=0. (19)
Since
n→+∞lim ∂φε
∂xi
pi|vn−v|pidx=0, (20)
then one has:
εlim→0lim
n
N
i=1
1 pi ∂
∂xi(φεvn)
pidx=lim
ε→0lim
n
N
i=1
1 pi ∂vn
∂xi
piφεpidx. (21)
From relations (18) and (21), knowing that|vn|p∗|v|p∗+ν(see Lemma 4), we obtain Sa
pp∗+
j lim
ε→0lim
n
N
i=1
1 pi
φεpi
∂vn
∂xi
pidx. 2
Lemma 13.Assume that N
i=1
1 pi
∂vn
∂xi
piμ˜ inM+ RN
.
Then
(1) For allj ∈J,Sa
pp∗+
j limε→0μ(support˜ φε)(one has supportφε⊂B(xj,maxiε
1 qi)).
(2) ˜μSνpp∗+ +P (∇v).
(3) S=limn→+∞P (∇vn)˙= ˜μ +μ∞P (∇v)+Sνpp∗+ +μ∞. Proof. From Lemma 12, sinceφεpiφεand
limn
N
i=1
1 pi
φεpi
∂vn
∂xi
pidx
φεdμ,˜ one obtains
Sa
pp∗+
j lim
ε→0
φεdμ˜lim
ε→0μ˜
B
xj; max
1iN
ε
1 qi
. (22)
This shows that{xj}j∈J are all atomic points ofμ˜ and sinceN
i=1 1
pi|∂x∂vi|piis orthogonal to the atomic part ofμ, one˜ deduces from relation (22) that
˜
μS
j∈J
a
pp∗+
j δxj+ N
i=1
1 pi
∂v
∂xi
pi. (23)
This implies in particular that:
˜μS
j∈J
a
p+ p∗
j +P (∇v). (24)
Since pp+∗ <1 one has
j∈J
aj pp∗+
j∈J
a
pp∗+
j . (25)
Asν=
j∈Jajδxj, it holds ν =
j∈J
aj, (26)
which means, combining relations (24) to (26), that:
˜μSνpp∗+ +P (∇v).
For the last statement, we argue as before:
S=lim
n P (∇vn)
= lim
R→+∞lim
n
RN
(1−ψR) N
i=1
1 pi
∂vn
∂xi
pidx+ lim
R→+∞lim
n
ψR
N
i=1
1 pi
∂vn
∂xi pidx,
whereψR=1 on|x|> R+1, 0ψR1,ψR=0 if|x|< R,ψR∈C(R).
By the definition ofμ, one has:˜
R→+∞lim lim
n
(1−ψR) N
i=1
1 pi
∂vn
∂xi
pidx=lim
R
(1−ψR)dμ˜ = ˜μ, and (see Lemma 8):
R→+∞lim lim
n
ψR
N
i=1
1 pi
∂vn
∂xi
pidx=μ∞,
thus, by the preceding statements:
S= ˜μ +μ∞P (∇v)+Sνpp∗+ +μ∞. 2
Lemma 14.Ifvp∗<1thenν =1,ν∞=0andv=0.
Proof. From Lemma 10, we know that Sν
pp∗+
∞ μ∞.
And by Corollary 4 of Lemma 3, we have Svpp+∗ P (∇v).
From the last statement of Lemma 13 and the above inequalities we deduce that:
SS
vpp∗∗pp∗+
+ νpp∗+ +ν
pp∗+
∞ .
Thus we obtain, due to Lemma 11, that vpp∗∗pp∗+
+ νpp∗+ +ν
p+
∞p∗
1=
vpp∗∗+ ν +ν∞pp∗+ .
Using the inequality
vpp∗∗+ ν +ν∞pp∗+
v
pp∗+
p∗ + νpp∗+ +ν
pp∗+
∞ ,