# Fr ed eric Robert dedicates this work to Cl emence Recent examples

date post

05-Jul-2018Category

## Documents

view

213download

0

Embed Size (px)

### Transcript of Fr ed eric Robert dedicates this work to Cl emence Recent examples

THE LIN-NIS PROBLEM FOR MEAN CONVEX DOMAINS

OLIVIER DRUET, FREDERIC ROBERT, AND JUNCHENG WEI

Abstract. We prove some refined asymptotic estimates for postive blowing

up solutions to u+u = n(n2)un+2n2 on , u = 0 on ; being a smooth

bounded domain of Rn, n 3. In particular, we show that concentration canoccur only on boundary points with nonpositive mean curvature when n = 3 orn 7. As a direct consequence, we prove the validity of the Lin-Nis conjecturein dimension n = 3 and n 7 for mean convex domains and with boundedenergy. Recent examples by Wang-Wei-Yan [30] show that the bound on theenergy is a necessary condition.

Frederic Robert dedicates this work to Clemence Climaque

Contents

1. Introduction 12. Lbounded solutions 53. Smooth domains and extensions of solutions to elliptic equations 64. Exhaustion of the concentration points 95. A first upper-estimate 166. A sharp upper-estimate 237. Asymptotic estimates in C1 () 388. Convergence to singular harmonic functions 398.1. Convergence at general scale 398.2. Convergence at appropriate scale 449. Estimates of the interior blow-up rates 5010. Estimates of the boundary blow-up rates 6011. Proof of Theorems 1 and 2 71Appendix A: Construction and estimates on the Greens function 73Appendix B: Projection of the test functions 84References 90

1. Introduction

Let be a smooth bounded domain of Rn, n 2. In [21], Lin, Ni and Takagigot interest in solutions u C2() to the elliptic problem u+ u = n(n 2)u

q1 in u > 0 in u = 0 on

(Eq)

Date: May 31th, 2010. Mathematics Subjet Classification 2010: 35J20, 35J60.

1

2 OLIVIER DRUET, FREDERIC ROBERT, AND J.WEI

where > 0 is a parameter and q > 2. Here and in the sequel, := div() isthe Laplace operator with minus-sign convention. This problem has its origins inthe analysis of the Gierer-Meinhardt model in mathematical biology: this modelis a system of nonlinear evolution equations of parabolic type, and the stationaryproblem with infinite diffusion constant splits into two equations like (Eq). We referto the surveys [24, 31] for the justifications of the model and its simplification.

Problem (Eq) enjoys a variational structure, since its solutions are critical pointsof the functional

u 7 12

|u|2 dx+ 2

u2 dx 1q

|u|q dx,

a functional that is defined for all u H21 ()Lq(), where H21 () is the standardSobolev space of L2functions with derivatives also in L2 endowed with the norm 2 + 2. In particular, it follows from Sobolevs embedding theorem thatH21 () Lq() continuously in case 2 < q 2? where 2? := 2nn2 (we assume herethat n 3): therefore the functional above is defined on H21 () when 2 < q 2?.Moreover, the Sobolev embedding above is compact in case q < 2?.

The system (Eq) enjoys at least a solution, namely the constant solution u (

n(n2)

) 1q2

. Moreover, in case 2 < q < 2?, variational techniques and the com-pactness of the embedding imply that for small positive , the constant solution isthe sole solution to (Eq). This uniqueness result incited Lin and Ni to conjecturethe extension of this result to the critical case q = 2?:

Question (Lin-Ni [20]): Is the constant solution the only solution to (E2?) when > 0 is small?

The mathematical difficulty of this question comes from the conformal invarianceof (E2?) and its associated unstability. Indeed, for > 0 and x0 Rn, define

Ux0,(x) :=(

2 + |x x0|2

)n22

for all x Rn. (1.1)

The scalar curvature equation for the pulled back of the spherical metric via thestereographic projection (or direct computations) yields Ux0, = n(n 2)U2

?1x0,

in Rn. Therefore, there is an abundance of solutions to u = u2?1, some of thembeing peaks blowing-up to infinity since lim0 Ux0,(x0) = +: in this sense, theequation is unstable since it enjoys many solutions that are far from each other.There are no such solutions in the subcritical case q < 2? (see [5]). This conformaldynamic transfers on the Lin-Nis problem and it follows from the famous Struwedecomposition [28] that families of solutions (u)>0 to (E2?) with bounded energymay develop some peaks like (1.1) when 0: more precisely, there exists N Nsuch that for any i {1, ..., N}, there exists sequences (xi,) Rn, (i,)i R>0such that lim0 i, = 0 and, up to the extraction of a subfamily,

u =Ni=1

Uxi,,i, +R (1.2)

where lim0R = 0 in H21 (). This decomposition is refered to as the integral de-composition. When there is at least one peak, then there are nonconstant solutions.

LIN-NIS PROBLEM 3

Conversely, in case there is no peak, elliptic estimates and simple integrations byparts (see Section 2) yield the sole constant solution for small .

In the radial case, that is when is a ball and when u is radially symmetrical,Adimurthi-Yadava solved the problem in [2, 3]: when n = 3 or n 7, the answer toLin-Nis question is positive, and it is negative for n {4, 5, 6}. In the asymmetriccase, the complete answer is not known yet, but there are a few results. Whenn = 3, it was proved independantly by Zhu [33] and Wei-Xu [32] that the answerto Lin-Nis question is positive when is convex. When n = 5, Rey-Wei [25]constructed solutions to (E2?) as a sum of peaks like (1.1) for 0. In the presentpaper, we concentrate on the localization of the peaks in the general case.

Let ()N (0, 1] be a sequence such thatlim

+ = 0.

We consider a sequence (u)N C2() such that u + u = n(n 2)u2?1 in

u > 0 in u = 0 on

(1.3)

We assume that there existe > 0 such that

u2?

dx (1.4)

for all N.Definition 1. We say that x is a non-singular point of (u) if there exists > 0 and C > 0 such that

uL(B(x)) C

for all N. We say that x is a singular point if it is not a non-singularpoint.

The singular points are exactly the points where the peaks are located. In thesequel, H(x) denotes the mean curvature at x of the oriented boundary .With our sign convention, the mean curvature of the oriented boundary of the unitball is positive. We prove the following theorem:

Theorem 1. Let (u)N C2() and > 0 such that (1.3) and (1.4) hold. LetS denote the (possibly empty) set of singular points for (u). Assume that n = 3or n 7: then S is finite and

S {x /H(x) 0}.As a consequence, we get the following:

Theorem 2. [Lin-Nis conjecture for mean convex domains] Let be a smoothbounded domain of Rn, n = 3 or n 7. Assume that H(x) > 0 for all x .Then for all > 0, there existe 0(,) > 0 such that for all (0, 0(,)) andfor any u C2(), we have that

u+ u = n(n 2)u2?1 in u > 0 in u = 0 on

u2

?

dx

u (

n(n 2)

)n24

.

4 OLIVIER DRUET, FREDERIC ROBERT, AND J.WEI

The method we use to prove Theorem 1 relies on a sharp control of the solutionsto (1.3) in the spirit of Druet-Hebey-Robert [9], our first result being that (seeProposition 5 and (7.3) in Section 7)

u C

(u +

Ni=1

Uxi,,i,

)(1.5)

where u is the average of u on and the peaks are as in Struwes decomposition(1.2). In particular, we pass from an integral description to a pointwise descrip-tion. As in Druet [6] (see also Ghoussoub-Robert [12] and Druet-Hebey [8]), thispointwise description allows us to determine exactly where two peaks may interact,and to describe precisely the behavior of u there. The localization of the singularpoints then follows from a succession of Pohozaev identities.

These results appeal some remarks. In dimension n = 3, our result must be com-pared to Zhus result: in [33], no bound on the energy is assumed, but the convexityis required; in our result, we require the bound on the energy, but a weak convexityonly is needed. The assumption on the energy (1.4) may seem technical for who isfamiliar with the Yamabe equation: indeed, in general, see Druet [7], Li-Zhu [19],Schoen [27] and Khuri-Marques-Schoen [18], any solution to the Yamabe equationautomatically satisfies a bound on the energy like (1.4). For the Lin-Nis problem,this is not the case: recently, it was proved that solutions to (E2?) may accumulatewith infinite energy when the mean curvature is negative somewhere (see Wang-Wei-Yan [29]) or when is a ball (see Wang-Wei-Yan [30]), a domain with positivemean curvature: therefore, the answer to Lin-Nis question is negative if one doesnot impose the bound (1.4).

The influence of curvature is reminiscent in the asymptotic analysis of equations like(1.3). For instance, in Druet [6, 7] and in Li-Zhu [19], it is proved that for Yamabe-type equations, the peaks are located where the potential of the equation touchesthe scalar curvature; we refer to Hebey-Robert-Wen [17] and Hebey-Robert [16]for the corresponding localization for fourth-order problems. In Ghoussoub-Robert[11, 12], that is for a singular Dirichlet-type problem, the peaks are located wherethe mean curvature is nonnegative: in Theorem 1 above, that is for a Neumannproblem, we conversely prove that the peaks are located at points of nonpositivemean curvature. For Neumann-type equations like (1.3), the role of the meancurvature has been enlighted, among others, by Adimurthi-Mancini-Yadava [1],Lin-Wang-Wei [22] and Gui-Lin [15].

The present paper is devoted to the asymptotic analysis of solutions (u) of (1.3)satisfying (1.4) when n 3. In Sections 2 to 7, we prove the pointwise control(1.5). Section 8 is devoted to the convergence of the (u)s at the scale wherepeaks interact. In Sections 9 and 10, we prove an asymptotic relation mixing theheights of the peaks, the distance between peaks and the mean curvature. Finally,we prove Theorems 1 and 2 in Section 11.

Notations: in the sequel, we define Rn := {(x1, x)