Subcritical Approximation of the Sobolev Quotient and a Related Concentration Result
GIAMPIEROPALATUCCI
ABSTRACT- LetVbe a general, possibly non-smooth, bounded domain ofRN,N3.
Let22N=(N 2) be the critical Sobolev exponent. We study the following variational problem
Se sup Z
V
juj2 edx: Z
V
jruj2dx1;u0on@V 8<
:
9=
;;
investigating its asymptotic behavior asegoes to zero, by means ofG-con- vergence techniques. We also show that sequences of maximizersueconcentrate energy at one pointx02V.
1. Introduction.
LetVbe ageneral bounded, possible non-smooth, bounded open set in RN;N3. Consider the elliptic Dirichlet problem with nearly critical nonlinearity
Duelu2e 1 e inV;
ue>0 inV;
ue0 on@V;
8>
<
>: (1:1)
where 22N=(N 2);lN(N 2). We know that whene>0 the above problem has a solutionue, but whene0 it becomes delicate. In particular, the existence and the properties of the solutions strongly depend on the domainV (see [14, 9, 2, 5] and others).
Indirizzo dell'A.: LATP, Universite Aix-Marseille III, Faculte des Sciences et Techniques Saint-JeÂroÃme, Avenue Escadrille Normandie-NieÂmen, 13397 Marseille cedex 20, France
E-mail: giampiero.palatucci@univ-cezanne.fr
In view of this qualitative change whene0, it is interesting to analyze the asymptotic behavior of the subcritical solutionsueof (1.1) asegoes to zero.
In the case of smooth domains V, this analysis was carefully in- vestigated by Han [8] and Rey [15]. They showed that the solutionsue of (1.1), that are maximizing for the following variational problem
Se:sup Fe(u): Z
V
jruj2dx1;u0 on@V 8<
:
9= (1:2) ;
with Fe(u) Z
V
juj2 edx;
(1:3)
concentrate at exactly one pointx0 inV; i.e., the measuresjruej2dxcon- verge (up to subsequences) in the sense of measures to a Dirac mass at x02V. They also showed thatx0is acritical point of the Robin function ofV (the diagonal of the regular part of the Green function), answering to a conjecture by Brezis and Peletier in [3], in which was analyzed the same problem in the case ofVbeing aspherical domain.
In order to obtain this concentration result, even without localizing the blowing-up, the cited authors also utilize some standard elliptic regularity techniques that require to work in smooth domains.
We also stress that regularity assumptions on the domain V have a strong impact on concentration results. In fact, in the case of smooth do- mains, the Robin function has no critical points in a neighborhood of the boundary and thus in [3, 8, 15] concentration may occur only inV. In [7], Flucher, Garroni and MuÈller were able to construct an example of a non- smooth domain V, with its Robin function achieving its infimum on the~ boundary; and then Pistoia and Rey showed that concentration can occur on the boundary, analyzing the asymptotic behavior of the maximizing solutions of the elliptic Dirichlet problem (1.2) in suchV~ (see [13]).
Here we can obtain the same concentration result in general bounded domain, not depending from any strong hypotheses of regularity, as a consequence of Lions' Concentration-compactness principle ([10]) together with the convergence of maxFe to supF0 (with F0(u)R
Vjuj2dx; see Section 2).
We also investigate the asymptotic behavior ase!0 of the subcritical Sobolev quotient Se in (1.2) by means of De Giorgi's G-convergence techniques. We can find this kind of approach, that is using G-con- vergence techniques to study concentration phenomena linked to critical
Sobolev exponent, in [1], where Amar and Garroni, following Flucher and MuÈller in [6], studied the variational problem
Sge supGe(u):u2H10(V);krukL2(V)1 (1:4)
with Ge(u)e 2 Z
V
g(eu)dx;
(1:5)
where gis anon-negative upper semi-continuous function bounded from above bycjtj2:They obtained aG-convergence result, that implies con- centration phenomena arising in critical growth problems.
Although it is not possible to reduce our problem (1.2) to (1.4), theG- convergence of the functionalFeis related to theG-convergence ofF0to its upper semi-continuous envelope scF0, obtained by Amar and Garroni in [1] (see Section 3).
A fundamental key point of theG-convergence is specifying the set- ting for the limit functional, with the choice of the topology. Here, in particular, the topology has to be sufficiently weak to assure the con- vergence of maximizing sequences and sufficiently strong to allow us to find the concentration.
Due to the presence of the constraint on the Dirichlet energy, the variational problem (1.2) suggests to study every sequence ue weakly converging to some function u in the Sobolev space H10(V). This con- vergence is not sufficient to describe the problem, so we also have to study every sequence jruej2dx converging to some measure m in the sense of measures.
We will show that the asymptotic behavior of the functional (1.3), in term of G-convergence, is described by the following functional F that depends by the two variablesu, belonging toH10(V), andm, being inM(V), the set of non-negative bounded total variation Borel measures, with its atomic coefficientsmi:
F(u;m) Z
V
juj2dxSX1
i0
m2i2; (1:6)
where S is the best Sobolev constant of the embeddingH10(V),!L2(V), that is
Ssup Z
V
juj2dx: Z
V
jruj2dx1;u0 on@V 8<
:
9=
;: (1:7)
Now, a natural question arises: can we localize the blowing up? Our hope is that it will possible to prove thatx0is a``harmonic center'' ofV, that is the minimum of the Robin function ofV. We could work by means of a reasonable second order development of the G-limit, in order to obtain more informations about the asymptotic behavior of the maximizing se- quences. We surely need to look at some techniques introduced in [6] and [7], where Flucher, Garroni and MuÈller showed that the concentration at the critical points of the Robin function is a very general phenomenon.
Second, again using a similar variational approach, in [11] the author generalizes the asymptotic analysis of subcritical solutions of the analo- gous problem to (1.1) for thep-Laplacian operator.
Finally, in a work in progress with A. Pisante ([12]) we study non-local concentration phenomena in the same spirit of this paper.
The remaining of the paper is divided in two sections. Section 2 is de- voted to the concentration result. In Section 3 we state and prove theG- convergence result.
2. The concentration result.
In this section, we will analyze the asymptotic behavior of the se- quencesuethat are maximizers for the nearly critical Sobolev quotientSe, as stated in the following theorem.
THEOREM2.1. Let uebe a maximizer sequence for the Sobolev quotient Se defined in(1.2).Then ueconcentrates at some x02V, i.e.,
jruej2dx* dx0 inM(V):
This result is a consequence of Lions's Concentration-compactness principle (see forthcoming Lemma2.2) together with the convergence of Se toSasegoes to zero (see Proposition 2.4).
LEMMA2.2 ([10, Lemma I.1]). LetV RNbe an open subset and let un be a sequence in H10(V)weakly converging to u as n! 1and such that
jrunj2dx* m and junj2dx* n inM(V):
Then, either un !u in L2(V)or there exists a finite set of distinct points x0;. . .;xkinVand positive numbersn0;. . .;nksuch that we have
n juj2dxXk
i0
nidxi; n2i(S)2(S)2:
Moreover, there exist a positive measure ~m2 M(V)with spt~mV and positive numbersm1;. . .;mk such that
m jruj2dx~mXk
i0
midxi; niSm2i2:
Armed with the concentration-compactness alternative, we can de- scribe the asymptotic behavior of the optimal sequencesun of the critical Sobolev quotient S. The key of the proof is the well-known convexity argument by Lions (see also [6, Lemma 10]).
COROLLARY 2.3. Let un be a maximizing sequence for the critical Sobolev quotient S defined in (1.7). Thenun concentrates at one point x02V.
Now, we will prove that the subcritical Sobolev quotientsSeconverge to S asegoes to zero.
PROPOSITION2.4. For everye>0, letSe be defined by (1.2)and letS be defined by(1.7).Then
lime!0SeS: (2:1)
PROOF. First, using the fact thatV is bounded together with HoÈlder inequality, we can show that
lim sup
e!0 Se S: (2:2)
Takeue2H10(V) maximizers forSe; we have Se Fe(ue)
Z
V
juej2 edx
Z
V
juej2 0
@
1 A
2 e 2
jVj2e
(S)22ejVj2e:
Thus, passing to the limit asegoes to zero, it follows inequality (2.2).
The remaining inequality easily follows by the pointwise convergence of Fe to F0. This is a standard argument. For every d>0 there exists
ud2H10(V) such thatkrudk2L2(V)1 a nd F0(ud)>S d:
(2:3)
Clearly, for such functionud, we have Se Fe(ud):
Then, combining the above inequality with (2.3) and passing to the limit ase goes to zero, we get
lim inf
e!0 Se lim
e!0Fe(ud)
F0(ud) S d;
that implied the desired inequality thanks to the arbitrariness of d. The
proof is complete. p
COMPLETION OF THE PROOF OFTHEOREM2.1. The concentration result for the sequence ue of maximizers ofS is immediate. Thanks to Propo- sition 2.4 the maximizersueofSeare a maximizing sequence forF0. Hence, Corollary 2.3 ensures thatjruej2dxconverge to aDirac mass in the sense
of measures. p
3. TheG-convergence result.
The goal of the present section is to describe by means of G-con- vergence the asymptotic behavior as egoes to zero of the functionals Fe defined by
Fe(u) Z
V
juj2 edx; 8u2H10(V) such thatkruk2L2(V) 1:
(3:1)
In order to applyG-convergence, we have to analyze the asymptotic behavior of the sequence Fe(ue) for every sequence ue such that kruek2L2(V)1:The constraint on the Dirichlet energy ofue implies that there existsm2 M(V) such thatm(V)1 a ndjruej2dx* minM(V) and, by Sobolev embedding, there exists u2H10(V) such that (up to sub- sequences)ue*uinL2(V).
Thanks to the semi-continuity ofL2-norm, we havem jruj2dx, so we can always decomposemin such way:
m jruj2dx~mX1
i0
midxi;
wheremi2[0;1] andxi2Vis such thatxi6xj if i6j;m~can be view as the
``non-atomic part'' ofm jruj2dx.
In analogy with [1], this suggests the setting for the limit functional as the spaceXdefined by
XX(V):
(u;m)2H10(V) M(V):m jruj2dx;m(V)1 ; endowed with the topologytsuch that
(ue;me)!t (u;m) ,def ue*uinL2(V) me* m inM(V):
(
Note that the topologytis compact inX, then theG-convergence of functionals in this space implies the convergence of maxima.
We also note that X is the smallest space where we have to set our problem, as stated in the following proposition by Amar and Garroni.
PROPOSITION3.1 ([1, Proposition 2.3]). Let(u;m)2X, then there exists a sequence ue in H01(V) such that, for every e>0, R
Vjruej2dx1 and (ue;jruej2dx)!t (u;m)whene!0.
Once we have taken the right space, we need to extend our functionalFe
to the whole spaceX(and we keep the same symbol), in the natural way as follows
Fe(u;m):
Z
V
juj2 edx if (u;m)2X:m jruj2dx;
0 otherwise inX:
8>
><
>>
(3:2) :
We prove the existence of the G-limit of the sequenceFe, giving its characterization in the following theorem.
THEOREM 3.2. There exists the G-limit F of the sequence of func- tionals Fedefined by(3.2)and
F(u;m) Z
V
juj2dxSX1
i0
m2i2; 8(u;m)2X:
In view of the pointwise convergence ofFetoF0and the convergence of Se toS (see Proposition 2.4), Theorem 3.2 can be proved repeating the strategy in the proof of [1, Theorem 3.1].
First, we state theG-convergence result by Amar and Garroni in [1].
Letg:R![0;1) be an upper semi-continuous function such that (i) 0g(t)cjtj2; g60 in theL1-sense;
(ii) 9aconstantg0such that
t!0lim g(t)
t2 g0: (3:3)
LetSgbe the criticalgeneralized Sobolev constant(see [6, Definition 1]) such that
Sgsup Z
V
g(u)dx:u2H10(V);
Z
V
jruj2dx1 8<
:
9=
;: (3:4)
For any positivee, consider the functionalGedefined inH10(V) M(V) by
Ge(u;m) e 2 Z
V
g(eu)dx if (u;m)2X andm jruj2dx;
0 otherwise inX:
8>
><
>>
: (3:5)
TheG-limit of the functionalGewith respect to the topologytis given in the following theorem.
THEOREM 3.3 ([1, Theorem 3.1]). There exists the G-limit G of the sequence of functionals Geand
G(u;m)g0
Z
V
juj2dxSgX1
i0
m2i2; (3:6)
for every(u;m)2X.
We note that in the particular case ofgbeing the critical power 2; i.e., g(t) jtj2,8t2R;the definitions in (3.3), (3.4), (3.5) and (3.6) become re- spectively
g01; SgS; GeF0 andGF and then, applying Theorem 3.3, it follows
F0 !G scF0F;
(3:7)
whereF0 is extended to the whole spaceX in the same way as in defini- tion (3.2) and we denote by scF0the upper semi-continuous envelope ofF0 with respect to the topologyt.
Using the convexity argument by Lions together with the fact that the critical Sobolev inequality is strict in bounded domains, Amar and Garroni also provide an optimal upper bound inequality for the limit functionalG.
LEMMA3.4 ([1, Lemma 3.6]). For every(u;m)2X, we have G(u;m)Sg
(3:8)
and the equality holds if and only if(u;m)(0;dx0)for some x02V.
PROOF OFTHEOREM3.2. Let us introduce theG-lim sup and theG- lim inf ofFe, respectively defined for every (u;m)2Xby
F(u;m)supn lim sup
e!0 Fe(ue;me):(ue;me)!t (u;m)o and
F (u;m)supn lim inf
e!0 Fe(ue;me):(ue;me)!t (u;m)o :
We can write the notion ofG-convergence in terms of theG-limsup and theG-liminf (see [4] for further details). TheG-limitFexists if and only if F F and, in this case, FF F. Since F F always holds, in order to prove Theorem 3.2 it is enough to show the following inequalities
FF (3:9)
FF : (3:10)
The first inequality (3.9) easily follows by the Concentration-compactness principle by Lions.
PROPOSITION3.5. For every(u;m)2X, we have F(u;m)lim sup
e!0 Fe(ue;me);
for every(ue;me)X such that(ue;me)!t (u;m).
PROOF. Let (ue;me) be inXsuch that (ue;me)!t (u;m); i.e., ue*uinL2(V) a nd jruej2dx* m jruj2dxm~X1
i0
midxi in M(V);
By Lemma2.2, it follows that
juej2dx* n juj2dxXk
i0
nidxi; with niSm2i2: (3:11)
Using HoÈlder Inequality by arguing as in the proof of Proposition 2.4, we have
lim sup
e!0 Fe(ue;me)lim sup
e!0
Z
V
juej2dx 22e
jVj2e
!
n(V)
Z
V
juj2dxSX1
i0
m2i2 F(u;m);
where we used (3.11). p
The remaining inequality (3.10) can be obtained like in the proof of [1, Theorem 3.1]. For the sake of self-containment, we will sketch here the proof.
According to the idea of Amar and Garroni, the G-liminf inequality will be proved in two separate cases: (u;m)(u;jruj2dx~m) a nd (u;m)
0;P
i midxi
. This decomposition is the key point in the proof. In fact, we will be able to use strongL2 convergence on the pairs of the first type (see the forthcoming Proposition 3.6, in which we useFe!F0); while on the pairs with purely atomic measure part, the functionalsFeare local, so we can compute their limit on each single Dirac mass (see Proposition 3.7, in which we use Se !S). Finally, Theorem 3.2 will be recovered thanks to aunifying lemma(see Lemma3.8).
PROPOSITION3.6. Let(u;m)2X be such thatm jruj2dxm. Then~ F(u;m)lim inf
e!0 Fe(ue;me)
for every sequence(ue;me)X such that(ue;me)!t (u;m)ase!0.
PROOF. We ta ke (ue;me)X such that me jruej2dx and (ue;me)!t (u;m) a s e!0. Since the atomic part of me is zero, ue!u in L2(V) a s e!0, so (up to asubsequence) ue!u a.e. in V.
It follows that for almost everyxinV
jue(x)j2 e! ju(x)j2 as e!0:
Hence, by Lebesgue Convergence Theorem, we have:
lime!0
Z
V
juej2 edx Z
V
juj2dxF(u;m);
that gives the desired inequality. p
PROPOSITION3.7. For every(u;m)2Xsuch that(u;m) 0;Pk
i0midxi
, withxi2V, there exists theG-limit of(Fe)and we have
G lim
e!0Fe
(u;m)F(u;m):
PROOF. First, we prove the following result. For every open setAV, for every x2A and for every (u;m)2X such that (u;m)(0;dx), there exists theG-limit of the sequence (Fe) restricted toAand the following equality holds:
G lim
e!0Fe
(0;dx;A)S:
Let us fix AV and let eh be such that eh!0 if h! 1. By the compactness property of theG-convergence, it follows that there exists a subsequence (still denoted byeh) and a functionalFA:X(A)![0;1) such that
G lim
h!1Feh
(u;m;A)FA(u;m):
We also have
Seh sup
X(A)Feh !max
X(A)FA ash! 1 and then, recalling Proposition 2.4,
maxX(A)FAS: (3:12)
By Proposition 3.6 and Lemma 3.4, withg(t) jtj2andVA, it follows FA(u;m)5S; 8(u;m)6(0;dx); 8x2A:
Combining the above inequality with (3.12), it follows the existence of x2Asuch that
FA(0;dx)S
and the above equality can be proved by density for every (u;m)(0;dx) with anyx2A.
Finally, in order to deal with the casemPk
i0midxi, it suffices to consider the sequence (ue;me) such that
ueXk
i0
mi
p uie; me jruej2dx;
where (uie;jruiej2dx)!t (0;dxi) are the recovering sequences given by the pre- vious step. A detailed proof is very well written in [1, Proposition 3.7]. p In order to complete the proof of inequality (3.10), we need the following technical lemma by Amar and Garroni, applied to our functionalsFe.
LEMMA 3.8 ([1, Lemma 4.1]).If F(u;m)F (u;m) for every (u;m)2X such that
(1) m(V)51;
(2) m jruj2dxm~Pk
i0midxi; xi2V;
(3) dist
supp (juj ~m);Sk
i0fxig
>0:
Then
F(u;m)F (u;m)for every(u;m)2X:
COMPLETION OF THE PROOF OFTHEOREM3.2. In view of Lemma3.8, it remains to prove theG-liminf inequality for every pair (u;m)2Xsuch that m(V)51,m jruj2m~Pk
i0midxiand dist
supp (juj ~m);Sk
i0fxig
>0:
As ameans to do this, it suffices to consider the recovery sequence (ue;me):(uAe uBe;mAe mBe);
where (uAe;mAe) and (uBe;mBe) are given by Proposition 3.6 and Proposi- tion 3.7, respectively (see [1, pag. 17]). The proof is complete. p REMARK3.9. We can also deduce the concentration result in Theorem 2.1, by using theG-convergence ofFetoF. In fact, by Theorem 3.2 and
G-convergence properties, we have that every maximizing sequence (ue;jruej2dx) ofFetmust converge to apair (u;m)2Xmaximizer forF:
(ue;jruej2dx)!t (u;m); with F(u;m)max
X(V)F:
By Lemma3.4, with g(t) jtj2, we have max
X(V)FF(0;dx0), for some x02V.
It follows
(ue;jruej2dx)!t (0;dx0);
that is the desired concentration result. p
Acknowledgements. I would like to thank Prof. Adriana Garroni for suggesting the study of concentration problems related to critical Sobolev exponent.
REFERENCES
[1] M. AMAR- A. GARRONI,G-convergence of concentration problems, Ann. Scuola Norm. Sup. PisaCl. Sci., Vol.2(1) (2003), pp. 151-179.
[2] A. BAHRI- J. CORON,On a nonlinear elliptic equation involving the critical Sobolev exponent: the effect of the topology of the domain, Comm. Pure Appl.
Math., Vol.41(1988), pp. 253-294.
[3] H. BREZIS- L. PELETIER,Asymptotic for Elliptic Equations involving critical growth, Partial differential equations and the calculus of variations, Vol. I, Progr. Nonlinear Differential Equations Appl., BirkhaÈuser Boston, Boston, MA (1989), pp. 149-192.
[4] G. DALMASO,An introduction toG-convergence, BirkhaÈuser, Boston, 1992.
[5] W. DING,Positive solutions ofDuu(n2)=(n 2)0on a contractible domain, J. Partial Differential Equations, Vol.2(4) (1989), pp. 83-88.
[6] M. FLUCHER - S. MUÈLLER, Concentration of lowextremals, Ann. Inst. H.
Poincare Anal. Non LineÂaire, Vol.10(3) (1999), pp. 269-298.
[7] M. FLUCHER - A. GARRONI - S. MUÈLLER, Concentration of lowenergy extremals: Identification of concentration points, Calc. Var., Vol.14(2002), pp. 483-516.
[8] Z.-C. HAN,Asymptotic approach to singular solutions for nonlinear elliptic equations involving critical Sobolev exponent, Ann. Inst. Henri PoincareÂ, Vol.
8(2) (1991), pp. 159-174.
[9] J. KAZDAN- F. WARNER, Remarks on some quasilinear elliptic equations, Comm. Pure Appl. Math., Vol.38(1975), pp. 557-569.
[10] P. L. LIONS, The concentration-compactness principle in the calculus of variations. The limit case, Rev. Mat. Iberoamericana, Vol. 1 (1985), pp.
145-201.
[11] G. PALATUCCI,p-Laplacian problems with critical Sobolev exponent, Asymp- totic Analysis, to appear.
[12] G. PALATUCCI- A. PISANTE,Sobolev embeddings and concentration-compact- ness alternative for fractional Sobolev spaces, submitted paper, available online athttp://mipa.unimes.fr/preprints.html.
[13] A. PISTOIA- O. REY,Boundary blow-up for a Brezis-Peletier problem on a singular domain, Calc. Var. Partial Differential Equations, Vol.18(3) (2003), 243-251.
[14] S. POHOZAEV,Eigenfunctions of the EquationsDulf(u), Soviet Math. Dkl., Vol.6(1965), pp. 1408-1411.
[15] O. REY,Proof of the conjecture of H. Brezis and L. A. Peletier, Manuscripta math., Vol.65(1989), pp. 19-37.
Manoscritto pervenuto in redazione il 7 dicembre 2009.