Davila Juan

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

PARTIAL REGULARITY OF FINITE MORSE INDEX

SOLUTIONS TO THE LANE-EMDEN EQUATION

JUAN DÁVILA1,3 , LOUIS DUPAIGNE2,4 AND ALBERTO FARINA2,5

1
Departamento de Ingenierı́a Matemática and CMM, Universidad de Chile
Casilla 170/3 Correo 3, Santiago, Chile
2
LAMFA, UMR CNRS 6140, Université Picardie Jules Verne
33, rue St Leu, 80039 Amiens, France
3
jdavila@dim.uchile.cl
4
louis.dupaigne@math.cnrs.fr
5
alberto.farina@u-picardie.fr

Abstract. We prove regularity and partial regularity results for finite Morse
index solutions u ∈ H 1 (Ω) ∩ Lp (Ω) to the Lane-Emden equation −∆u =
|u|p−1 u in Ω.

1. Introduction
Given an open set Ω ⊂ RN , N ≥ 1, and p > 1, consider the Lane-Emden
equation
(1.1) −∆u = |u|p−1 u in Ω.
We are interested in the classical question of regularity of solutions to (1.1). Namely,
given a class of weak solutions C, we ask: what is the largest exponent p > 1, such
that
(1.2) u ∈ C =⇒ u ∈ C 2 (Ω) ?
Consider first C = Lploc (Ω) and assume (1.1) is understood in the sense of distribu-
tions. Then, as follows from a well-known bootstrap argument, (1.2) holds true for
all p < p0 (N ), where
 +∞, if 1 ≤ N ≤ 2,

p0 (N ) = N
 , if 3 ≤ N .
N −2
2
The exponent p0 (N ) is sharp. Indeed, for all p > p0 (N ), u(x) = cN,p |x|− p−1
is a singular solution belonging to C. A (radial) singular solution also exists if
p = p0 (N ), see [2, 14].
Consider next the case C = Hloc 1
(Ω) ∩ Lploc (Ω). Then (1.2) holds true for all
p ≤ pS (N ) where
 +∞, if 1 ≤ N ≤ 2,

pS (N ) = N + 2
 , if 3 ≤ N .
N −2

Date: January 4, 2011.


1
2 J. DÁVILA, L. DUPAIGNE AND A. FARINA

When p < pS (N ), the proof uses the same bootstrap argument and the extra
1
assumption that u ∈ Hloc (Ω) for the initial step. See [6] for the critical case
p = pS (N ). The Sobolev exponent pS (N ) is again sharp in the considered class,
using the same counter-example.
Restrict at last to the class C of energy solutions having finite Morse index,
1
i.e., u ∈ C if u ∈ Hloc (Ω) ∩ Lploc (Ω) and the maximal dimension of a vector space
1
X ⊂ Cc (Ω) such that
Z Z
Qu (ϕ) := |∇ϕ|2 dx − p |u|p−1 ϕ2 dx < 0, for all ϕ ∈ X \ {0}
Ω Ω

is an integer k, called the Morse index of u. If k = 0, we say that u is stable. Note


that this class of weak solutions is a natural choice, since any C 2 solution to (1.1)
is bounded on any open set ω ⊂⊂ Ω and so must have finite Morse index on ω. We
obtain the following result.
Theorem 1.1. Let u ∈ Hloc 1
(Ω) ∩ Lploc (Ω) be a solution to (1.1) of finite Morse
index. If p < pc (N ), where

 +∞, for 1 ≤ N ≤ 10,

2

pc (N ) = (N − 2) − 4N + 8 N − 1
 , for 11 ≤ N ,
(N − 2)(N − 10)

then, u ∈ C 2 (Ω).
Under the stronger assumptions that Ω is smoothly bounded, u is stable, and
u|∂Ω = 1, the smoothness of u was first proved in [7, 13]. At no surprise, the proof
of Theorem 1.1 is by bootstrap (using the additional information that u has finite
Morse index in the initial step). The Joseph-Lundgren exponent pc (N ) is again
sharp in the considered class, using the same counter-example.
In the supercritical cases (p ≥ p0 (N ), p > pS (N ), p ≥ pc (N )), solutions can be
singular at a point, as discussed earlier, but also on larger sets. E.g. if there exists
p
an integer k such that 0 ≤ k ≤ N − 1 and k < N − 2 p−1 , then
1
− p−1
(1.3) u(x) = CN,p,k x21 + . . . x2N −k
is a solution having k-dimensional singular set. See also an example in [18] where
the Hausdorff dimension of the singular set is not an integer.
Nevertheless, if solutions are assumed to be positive and stationary, the singular
set can be estimated as follows.
Theorem 1.2 ([16]). Let N ≥ 3 and p ≥ pS (N ). Let u ∈ Hloc 1
(Ω) ∩ Lp+1
loc (Ω) be
a positive weak solution to (1.1). Assume in addition that u is stationary. Then,
u ∈ C 2 (Ω \ Σ), where Σ is a closed set of Hausdorff dimension bounded above by
p+1
Hdim (Σ) ≤ N − 2 .
p−1
The precise definition of stationary solution can be found in [16]. Smooth solu-
tions (and limits thereof in the H 1 (Ω) ∩ Lp+1 (Ω) topology) are stationary.
When the solution has finite Morse index, we prove that the singular set is in fact
much smaller.
PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 3

Theorem 1.3. Let N ≥ 11 and p ≥ pc (N ). Let u ∈ Hloc 1


(Ω) ∩ Lploc (Ω) be a positive
2
solution to (1.1) of finite Morse index. Then, u ∈ C (Ω \ Σ), where Σ is a closed
set of Hausdorff dimension bounded above by
p+γ
Hdim (Σ) ≤ N − 2 ,
p−1
p
with γ = 2p + 2 p(p − 1) − 1.
Remark 1.4. The dimension of the singular set computed in Theorem 1.3 is op-
timal at least when it is an integer. Indeed, the solution given by (1.3) is stable in
RN if p > pc (N − k), while pc (N − k) solves
p+γ
N −2 = k.
p−1
Remark 1.5. If u ∈ Hloc 1
(Ω)∩Lploc (Ω) is a positive solution with finite Morse index,
we prove in the next section that for any point x ∈ Ω, there exists a ball B = B(x, r)
such that u is the limit of C 2 solutions in the H 1 (B) ∩ Lp+1 (B) topology. Hence,
u is stationary in B.
Theorem 1.3 remains valid for sign-changing solutions, provided they are sta-
tionary.
Remark 1.6. After completing this work, K. Wang [21, 22] informed us that he
obtained results similar to our Theorem 1.3 for stable positive solutions.
We discuss at last the question of universal a priori estimates.
1
Theorem 1.7. Assume 1 < p < pc (N ) and p 6= pS (N ). Assume u ∈ Hloc (Ω) ∩
p
Lloc (Ω) is a solution to (1.1) of finite Morse index m. Then, there exists a constant
C depending on N, p, m only, such that for all x ∈ Ω,
2 2
(1.4) |u(x)| + |∇u(x)| p+1 ≤ Cdist(x, ∂Ω)− p−1 .
Remark 1.8. Assume p = pS (N ). Then, (1.4) remains true for stable solutions
(see [10]). However, the estimate is false for solutions of finite Morse index, since
for λ > 0,
p ! N2−2
λ N (N − 2)
uλ (x) =
λ2 + |x|2
provides an unbounded family of solutions of constant Morse index.
The universal estimate (1.4) was first proved for positive solutions and 1 < p <
pS (N ) (see [4, 8, 11, 19]), with a constant C independent of the Morse index m.
Note however that for such p, there do exist sign-changing solutions of arbitrary
large Morse index, for which the dependance of the constant C to m must be kept
(see [1, 3, 17]). Estimate (1.4) was then proved in [10] for C 2 solutions which are
stable, for the full range 1 < p < pc (N ).
We provide at last a universal estimate for C 2 solutions of the more general
problem:
(1.5) −∆u = f (u) in Ω
where f ∈ C 1 (R, R) behaves like a power of u at infinity. More precisely,
4 J. DÁVILA, L. DUPAIGNE AND A. FARINA

Theorem 1.9. Suppose


f 0 (t)
(1.6) lim =a
t→±∞ p|t|p−1
1
for some a > 0 and 1 < p < pc (N ) and p 6= pS (N ). Let u ∈ Hloc (Ω) ∩ Lploc (Ω)
be a solution of (1.5) of finite Morse index m. Then, there exists a constant C
depending on N, f, m only, such that for all x ∈ Ω,
2 2
|u(x)| + |∇u(x)| p+1 ≤ C(1 + dist(x, ∂Ω)− p−1 ).
Theorem 1.9 was proved in [20] for positive solutions, 1 < p < pS (N ), and with a
constant C independent of the Morse index m. In the case p = pS (N ), the theorem
remains valid for stable solutions, but fails for solutions of finite Morse index (see
the counter-example in Remark 1.8). Similar statements can be derived for the
nonlinearity f (u) = eu , as we shall demonstrate in a future publication. See [22]
for recent results in this direction.

2. Preliminary results
2.1. Reduction to the case of stable solutions.
Proposition 2.1. Let u ∈ Hloc1
(Ω)∩Lploc (Ω) be a solution to (1.1) with finite Morse
index. Then, for every x0 ∈ Ω, there exists r0 > 0 such that u is stable in B(x0 , r0 ).
1
Proof. When N = 1, any function u ∈ Hloc (Ω) is locally bounded by Morrey’s
inequality. In particular, the linearized operator L = −∆ − p|u|p−1 has positive
principal eigenvalue in any sufficiently small ball, whence u is stable on such a ball.
Assume now N ≥ 2. We may always assume that B(0, 1) ⊂ Ω and it suffices to
prove that u is stable near the origin. Assume first that u has Morse index 1. Either
u is stable in B(0, 1/n) for some n ≥ 2 and we are done. Or, for all n ≥ 2, there
exists a direction ϕn ∈ Cc1 (B(0, 1/n)) such that Qu (ϕn ) < 0. Since u has index 1,
this implies that u is stable in B(0, 1) \ B(0, 1/n). This being true for all n ≥ 2,
we deduce that u is stable in B(0, 1) \ {0}. In fact, since N ≥ 2, points have zero
Newtonian capacity and so u is stable in B(0, 1). So, every solution of index 1 is
stable in a neighborhood of 0. Take now a solution u of index k ≥ 2. Working
exactly as above, we deduce that u has index k − 1 in some ball B(0, r1 ). Working
inductively on k, we deduce that u is stable in some ball B(0, rk ).

2.2. Approximation of singular stable solutions.
Lemma 2.2. Suppose u ∈ Hloc 1
(Ω) ∩ Lploc (Ω) is a nonnegative stable weak solution
to (1.1). Then, there exists a sequence of nonnegative stable solutions un ∈ C 2 (Ω)
1
to (1.1), such that un % u a.e. and in Hloc (Ω).
Proof. The proof is a refinement of a concave truncation technique found in [5].
1
Let us first observe that since u ∈ Hloc (Ω) and u solves (1.1), we have u ∈
p+1
Lloc (Ω). Take now ω ⊂⊂ Ω with smooth boundary, so that u ∈ H 1 (ω) ∩ Lp+1 (ω).
We are going to produce a sequence un converging to u in H 1 (ω). By a standard
diagonal argument, we then reach the desired conclusion.
In the sequel, we write ω = Ω for notational convenience. Given c > 0, consider
the function  1
 − p−1
φc (t) = c + t−(p−1) , defined for t > 0.
PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 5

We set also φc (0) = 0. Then, φc is increasing, concave, and smooth for t > 0. In
addition, φc (t) % t as c & 0+ , and φc (t) ≤ t, for all t ≥ 0. Also, if c > 0, then φc ,
φ0c are uniformly bounded. We have
φc (t)p
φ0c (t) = ∀t > 0.
tp
Let wc denote the unique solution to
−∆wc = 0 in Ω


wc = φc (u) on ∂Ω.
Then, wc ≥ 0, wc ∈ L∞ (Ω) ∩ H 1 (Ω). Moreover, wc is non-increasing with respect
to c. We claim that wc → w in H 1 (Ω) as c → 0, where w is the solution to
−∆w = 0 in Ω


w=u on ∂Ω.
To see this, consider the problem
(
−∆v = (v + wc )p in Ω
(2.1)
v=0 on ∂Ω.

Since wc ∈ L∞ (Ω), (2.1) has a minimal nonnegative solution vc , which can be


constructed by the method of sub and super-solutions, as follows. Note that v = 0
is a sub-solution, since wc ≥ 0. Moreover, by Kato’s inequality, v = φc (u) − wc is
a bounded super-solution:
−∆(φc (u) − wc ) = −∆φc (u) ≥ −φ0c (u)∆u = φc (u)p = (φc (u) − wc + wc )p .
In particular, (2.1) has a minimal nonnegative solution vc . This minimal solution
is bounded and by elliptic regularity, vc belongs to C 1,α (Ω). Moreover, vc is stable
in the sense that
Z Z
p−1 2
p (vc + wc ) ϕ dx ≤ |∇ϕ|2 dx, for all ϕ ∈ Cc1 (Ω).
Ω Ω
Since vc is minimal and wc is non-increasing with respect to c, we deduce that vc
is also non-increasing with respect to c. It follows that v(x) = limc→0 vc (x) is well
defined for all x ∈ Ω. Since vc ∈ C 1 (Ω), we have
Z Z Z
|∇vc |2 dx = (vc + wc )p vc dx ≤ up+1 dx.
Ω Ω Ω

In particular, vc is bounded in H01 (Ω).


It follows that vc * v weakly in H01 (Ω).

Multiplying (2.1) by ϕ ∈ Cc (Ω), integrating, and passing to the limit as c → 0,
we see that v is a weak solution to
(
−∆v = (v + w)p in Ω
(2.2)
v=0 on ∂Ω.

Let ϕk ∈ Cc0,1 (Ω) be a sequence such that ϕk → v in H01 (Ω). Since v ≥ 0 we can
assume ϕk ≥ 0. We can also assume that ϕk → v a.e. in Ω. Multiplying (2.2) by
ϕk and integrating, we obtain
Z Z
∇v∇ϕk dx = (v + w)p ϕk dx
Ω Ω
6 J. DÁVILA, L. DUPAIGNE AND A. FARINA

By Fatou’s lemma,
Z Z Z
p
(v + w) v dx ≤ lim inf ∇v∇ϕk dx = |∇v|2 dx.
Ω k→∞ Ω Ω

By monotone convergence,
Z Z Z
lim 2 p
|∇vc | dx = lim (vc + wc ) vc dx = (v + w)p v dx.
c→0 Ω c→0 Ω Ω

Hence,
Z Z Z
2 p
lim |∇vc | dx = (v + w) v dx ≤ |∇v|2 dx.
c→0 Ω Ω Ω

Since vc * v weakly in H01 (Ω), the reverse inequality


Z Z
2
|∇v| dx ≤ lim inf |∇vc |2 dx
Ω c→0 Ω

also holds, which proves that vc → v in H01 (Ω).


We claim that u = v + w, from which Lemma 2.2 follows. By construction,
v = lim vc ≤ lim(φc (u) − wc ) = u − w. We need thus only prove that u ≤ v + w.
Note that ṽ = u − w solves (2.2). Let z = ṽ − v ≥ 0. Then, z ∈ H01 (Ω), and since
u is stable,
Z Z
(2.3) p (ṽ + w)p−1 (ṽ − v)2 dx ≤ |∇(ṽ − v)|2 dx.
Ω Ω

Now, ṽ − v satisfies
Z Z
∇(ṽ − v)∇ϕ dx = ((ṽ + w)p − (v + w)p )ϕ dx, ∀ϕ ∈ Cc∞ (Ω).
Ω Ω

We would like to take ϕ = ṽ − v. First, we claim that we can take ϕ ∈ H01 (Ω) ∩
L∞ (Ω). These functions can be approximated in H01 (Ω) by functions in Cc∞ (Ω)
with a uniform bound. Then, take ϕ = min(ṽ − v, t), t > 0, which belongs to
H01 (Ω) ∩ L∞ (Ω). We get
Z Z
|∇(ṽ − v)|2 dx = ((ṽ + w)p − (v + w)p ) min(ṽ − v, t) dx.
[ṽ−v≤t] Ω

Now let t → ∞. Then,


Z Z
|∇(ṽ − v)| dx = ((ṽ + w)p − (v + w)p )(ṽ − v) dx
2
Ω Ω

Combined with (2.3) we find


Z
(ṽ − v) p(ṽ + w)p−1 (ṽ − v) − (ṽ + w)p + (v + w)p dx ≤ 0.
 

By convexity, p(ṽ + w)p−1 (ṽ − v) − (ṽ + w)p + (v + w)p ≥ 0 with strict inequality,
unless ṽ ≡ v. 
PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 7

2.3. Some well-known ingredients. Proofs of all the results in this section can
be found in [9]. We begin with a so-called ε-regularity result for weak solutions to
(1.1) in Morrey spaces. Recall the following definition.
Definition 2.3. Let Ω be a bounded open set of RN , N ≥ 1. Given p > 1 and
λ ∈ [0, N ], the Morrey space Lp,λ (Ω) is the set of functions u in Lp (Ω) such that
the following norm is finite:
Z
kukpLp,λ (Ω) = sup r−λ |u|p dx < ∞.
x0 ∈Ω, r>0 B(x0 ,r)∩Ω

Then,
p+1
Theorem 2.4 ([12, 16]). Let N ≥ 3, p > 1, and λ = N − 2 p−1 . Let also B(x0 , r0 )
be a ball. There exists ε = ε(N, p) > 0 such that for any weak solution u ∈
H 1 (B(x0 , r0 )) ∩ C(B(x0 , r0 )) to (1.1) satisfying
(2.4) kukLp+1,λ (B(x0 ,r0 )) ≤ ε,
there holds   2
4 p−1
kukL∞ (B(x0 ,r0 /2)) ≤ .
r0
Also recall the following classical result from geometric measure theory.
Theorem 2.5. Let Ω denote an open set of RN , N ≥ 1, u a function in L1loc (Ω)
and 0 ≤ s < N . Set
( Z )
−s
Es = x ∈ Ω : lim sup r |u(y)| dy > 0 .
r→0+ Br (x)

Then,
Hs (Es ) = 0,
where Hs denotes the Hausdorff measure of dimension s.
The next ingredient in the proof of Theorem 1.3 is the following monotonicity
formula.
Theorem 2.6 ([15]). Let u ∈ H 1 (Ω) ∩ Lp+1 (Ω) denote a stationary weak solution
to (1.1). For x ∈ Ω, r > 0, such that B(x, r) ⊂ Ω, consider the energy Eu (x, r)
given by
r−µ−1
Z   Z
1 1
(2.5) Eu (x, r) = r−µ |∇u|2 − |u|p+1 dx + |u|2 dσ,
B(x,r) 2 p+1 p − 1 ∂B(x,r)
where
p+1
µ=N −2 .
p−1
Then,
• Eu (x, r) is nondecreasing in r.
• Eu (x, r) is continuous in x ∈ Ω and r > 0.
Remark 2.7 ([15]). The energy Eu (x, r) can be equivalently written as
(2.6) Eu (x, r) =
  !
p − 1 −µ
Z Z
1 2 1 p+1 1 d −µ 2
r |∇u| + |u| dy + r |u| dσ .
p+3 B(x,r) 2 p+1 p + 3 dr ∂B(x,r)
8 J. DÁVILA, L. DUPAIGNE AND A. FARINA

We shall use at last the following capacitary estimate.

Proposition 2.8 ([10]). Let Ω be an open set of RN , p > 1. Let p 1


u ∈ Hloc (Ω) ∩
p
Lloc (Ω) denote a stable solution to (1.1). Then, for any γ ∈n[1, 2p+2
o p(p − 1)−1),
p+γ
any ψ ∈ Cc1 (Ω), 0 ≤ ψ ≤ 1, and any integer m ≥ max p−1 , 2 , there exists a
constant Cp,m,γ > 0 such that
Z    2  Z
2 p+γ
|∇ψ| ( p−1 ) dx.
γ−1
p+γ 2m
∇ |u| u + |u| ψ dx ≤ Cp,m,γ
2

Ω Ω

In the case where u ∈ C 2 (Ω), the proof of this result is given in [10]. This
proof can be adapted to the case u ∈ Hloc 1
(Ω) ∩ Lploc (Ω) as follows: multiply (1.1)
γ−1
with |Tk (u)| uϕ , where Tk (s) = max(−k, min(u, k)) and ϕ ∈ Cc2 (Ω) and apply
2
γ−1
stability with test function |Tk (u)| 2 uϕ.

3. Proofs of Theorems 1.1 and 1.3.


p+γ
Proof ofpTheorem 1.1. Thanks to Proposition 2.8, u ∈ Lloc (Ω) for all γ ∈
[1, 2p + 2 p(p − 1) − 1). Using elliptic estimates and a standard bootstrap argu-
ment, we deduce that u ∈ C 2 (Ω), provided N ≤ 10 or N ≥ 11 and p < pc (N ).


Proof of Theorem 1.3 . By Proposition 2.1, we may assume that u is a nonneg-


ative stable weak solution to (1.1). Given ε > 0, define
( Z )
p+1
p+1 2 N −2 p−1
Σε = x ∈ Ω : ∀r > 0 (u + |∇u| ) dx ≥ εr
B(x,r)

Step 1. There exists a fixed value of ε > 0 such that for every x 6∈ Σε , u is bounded
(hence C 2 ) in a neighborhood of x.
To see this, let x0 6∈ Σε : there exists r0 > 0 such that
Z
−µ
r0 (up+1 + |∇u|2 ) dx < ε,
B(x0 ,r0 )

p+1
where µ = N − 2 p−1 . By (2.5), for r < r0 ,

r−µ−1
Z Z
1
Eu (x0 , r) ≤ r−µ |∇u|2 dy + u2 dσ
B(x0 ,r) 2 p−1 ∂B(x0 ,r)
r−µ−1
Z Z
1
≤ r−µ |∇u|2 dy + u2 dσ
B(x0 ,r0 ) 2 p − 1 ∂B(x0 ,r)
 −µ −µ−1 Z
ε r r
≤ + u2 dσ
2 r0 p − 1 ∂B(x0 ,r)
PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 9

Integrating between r = r0 /2 and r0 , and recalling that Eu (x, r) is nondecreasing


in r, we deduce that
Z r0 Z !
r0 µ−2 1 −µ−1 2
Eu (x0 , r0 /2) ≤ 2 εr0 + r u dσ dr
2 p − 1 r0 /2 ∂B(x0 ,r)
Z
−µ−1
≤ Cεr0 + Cr0 u2 dy
B(x0 ,r0 )
2
! p+1
Z 2
N (1− p+1 )
≤ Cεr0 + Cr0−µ−1 up+1 dy r0
B(x0 ,r0 )

< Cεr0 .
Hence,
Eu (x0 , r0 /2) < Cε.
Since Eu is continuous in x, there exists r1 < r0 /2 such that Eu (x, r0 /2) < 2Cε, for
x ∈ B(x0 , r1 ). Since Eu is non-increasing in r, we deduce that for all x ∈ B(x0 , r1 )
and all r < r1 ,
(3.1) Eu (x, r) < 2Cε.
Now take an approximating sequence un given by Lemma 2.2. Integrating (2.6)
between 0 and r2 < r1 , we find
 !
p − 1 r2 −µ
Z Z 
1 2 1 p+1
r |∇un | + u dy dr+
p+3 0 B(x,r) 2 p+1 n
r−µ
Z
+ 2 u2 dσ ≤ r2 Eun (x, r2 ).
p + 3 ∂B(x,r2 ) n
It follows that
!
Z r2 Z
−µ p+1
Cr2 Eu (x, r2 ) ≥ r u dy dr
0 B(x,r)
!
Z r2 Z
−µ p+1
≥ r u dy dr
r2 /2 B(x,r)

By the fundamental theorem of calculus, we deduce that there exists r3 ∈ (r2 /2, r2 )
such that
Z Z
CEu (x, r2 ) ≥ r3−µ up+1 dy ≥ r2−µ up+1 dy
B(x,r3 ) B(x,r2 /2)

Apply now (3.1). Then,


Z
−µ
r up+1 dy ≤ Cε,
B(x,r)

for all x ∈ B(x0 , r1 ) and all r < r1 /2. Taking ε sufficiently small, it follows
from Theorem 2.4 that (un ) is uniformly bounded near x0 and so, u is C 2 in a
neighborhood of x0 .
Step 2. For all γ ≥ 1, there exists ε0 > 0 such that
( Z )
p+γ
Σ ⊆ Σ̃0 := x ∈ Ω : ∀r > 0 up+γ dx ≥ ε0 rN −2 p−1 .
B(x,r)
10 J. DÁVILA, L. DUPAIGNE AND A. FARINA

Indeed, suppose x 6∈ Σ̃ε0 . Then,


Z
p+γ
up+γ dx < ε0 rN −2 p−1
B(x,r)

for some r > 0. By Hölder’s inequality,


p+1
! p+γ
Z Z
p+1
(3.2) u p+1
dx ≤ C u p+γ
dx rN (1− p+γ )
B(x,r) B(x,r)
 p+1
 p+γ
p+γ p+1 p+1 p+1
< C ε0 rN −2 p−1 rN (1− p+γ ) = C(ε0 ) p+γ rN −2 p−1 .

Take a function ϕ ∈ Cc2 (Ω) and multiply the Lane-Emden equation (1.1) by uϕ2 .
Then,
Z Z Z
|∇u|2 ϕ2 dx + u∇u · ∇ϕ2 dx = up+1 ϕ2 dx
Ω Ω Ω

i.e.
Z Z Z
2 2 p+1 2 1
|∇u| ϕ dx = u ϕ dx + u2 ∆ϕ2 dx
Ω Ω 2 Ω

Choose now ϕ such that ϕ = 1 in B(x, r/2), ϕ = 0 outside B(x, r), and |∆ϕ2 | ≤
C/r2 . Then,
Z Z Z
C
|∇u|2 dx ≤ C up+1 dx + 2 u2 dx
B(x,r/2) B(x,r) r B(x,r)

We estimate
2
! p+γ
Z Z
1 C 2
u2 dx ≤ 2 up+γ dx r1− p+γ
r2 B(x,r) r B(x,r)
2
C  0 n−2 p+γ  p+γ 2 2 p+1
< ε r p−1 r1− p+γ = C(ε0 ) p+γ rN −2 p−1 .
r2
Using (3.2), we deduce that
Z
2 p+1
(up+1 + |∇u|2 ) dx < C(ε0 ) p+γ rN −2 p−1 .
B(x,r/2)

2
Choosing ε0 such that C(ε0 ) p+γ ≤ ε, we deduce that x 6∈ Σε . And so, Σ̃ε0 ⊇ Σε .
p+γ
p 3. By the capacitary estimate (Proposition 2.8),0 u ∈ Lloc (Ω) if γ ∈ [1, 2p +
Step
2 p(p − 1) − 1). By Theorem 2.5 it follows that for ε > 0 small,
p+γ
HN −2 p+1 (Σ̃ε0 ) = 0.
p
This being true for all γ ∈ [1, 2p + 2 p(p − 1) − 1), Theorem 1.3 follows.

PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 11

4. Proof of the a priori estimates

Proof of Theorem 1.7. The proof of (1.4) is the same as the one given in [20],
except for the use of Theorem 2 of [10] stating that there are no entire solutions of
finite Morse index if p is in the range of Theorem 1.1. 
Proof of Theorem 1.9. By Theorem 1.1, any finite Morse index solution to
(1.1) is C 2 , provided p < pc (N ). Working by contradiction, as in the proof of
Theorem 2.3 of [20], we can find a sequence (uk ) of solutions of (1.5) (with Morse
index at most m) and a sequence of points (xk ) such that by setting
p−1 p−1
λk = (|uk (xk )| 2 + |∇uk (xk )| p+1 )−1
2
vk (y) = λkp−1 uk (xk + λk y)
we have λk → 0,
−∆vk = fk (vk ) in B(0, k)
where
2p 2
− p−1
fk (v) = λkp−1 f (λk v),
and
p−1 p−1
|vk | 2 + |∇vk | p+1 ≤ 2 in B(0, k)
p−1 p−1
|vk (0)| 2 + |∇vk (0)| p+1 = 1.
Note that fk (vk ) and ∇[fk (vk )] are both uniformly bounded. Then, up to subse-
1 0,α
quence vk → v in the Cloc (RN ) topology, fk (vk ) → g in the Cloc (RN ) topology (for
some α ∈ (0, 1)) and −∆v = g in the sense of distributions. By standard elliptic
2,α
estimates v is then a classical Cloc (RN ) solution of −∆v = g in RN .
We claim that v satisfies
(4.1) −∆v = a|v|p−1 v in RN .

To this end it is enough to prove that g = a|v|p−1 v in RN . The assumption (1.6)


implies
f (t)
(4.2) lim = a.
t→±∞ |t|p−1 t
Therefore, on the open set [v 6= 0], fk (vk (x)) → a|v(x)|p−1 v(x) pointwise, hence
g = |v|p−1 v on [v 6= 0] and also on [v 6= 0] by continuity. If y 6∈ [v 6= 0] then v is
zero in a neighborhood Uy of y and hence 0 = −∆v = g in Uy , giving in particular
that g(y) = 0.
It remains to verify that v has Morse index at most m. We first prove that
limk→+∞ fk0 (vk (y)) = ap|v(y)|p−1 pointwise in RN . This is clearly true for y ∈
[v 6= 0], thanks to (1.6). On the other hand, for y ∈ [v = 0], the desired
conclusion holds true since lim supk→+∞ |fk0 (vk (y))| = 0. Indeed, let us sup-
pose the contrary, then limn→+∞ |fk0 n (vkn (y))| > 0 for a sequence kn % +∞.
2 2
− −
Since fk0 n (vkn (y)) = λ2kn f 0 (λknp−1 vkn (y)), the sequence λknp−1 |vkn (y)| must be un-
2

bounded. Hence, up to a subsequence, λknp−1 |vkn (y)| → +∞ and then, by (1.6),
|fk0 n (vkn (y))| ≤ C|vkn (y)|p−1 → 0. A contradiction.
12 J. DÁVILA, L. DUPAIGNE AND A. FARINA

To conclude we work again by contradiction. Assume there exist m + 1 linearly


independent functions ϕj ∈ Cc∞ (RN ) such that Qv (ϕj ) < 0, j = 1, . . . , m+1, where
Z
|∇ϕ|2 − ap|v|p−1 ϕ2 dx.

Qv (ϕ) =
RN

Then, since limk→+∞ fk0 (vk (y)) = ap|v(y)|p−1 pointwise in RN , we have


Z
|∇ϕ|2 − fk0 (vk )ϕ2 dx = Qv (ϕ),

lim
k→∞ B(0,k)

for any ϕ ∈ Cc∞ (RN ).


Therefore, for k large enough, uk has Morse index greater
or equal than m + 1, which is a contradiction.
We have constructed a nontrivial C 2 solution of (4.1) of finite Morse index, which
is not possible by Theorem 2 of [10] if p 6= pS (N ) (and by Theorem 1 of [10] when
p = pS (N ) and m = 0). 

Ackowledgments. J.D. was partially supported by Fondecyt 1090167, CAPDE-


Anillo ACT-125 and Fondo Basal CMM. The authors also acknowledge the support
of the MathAmSud NAPDE project (08MATH01) and ECOS-Conicyt C09E06.

References
[1] Antonio Ambrosetti and Paul H. Rabinowitz, Dual variational methods in critical point theory
and applications, J. Functional Analysis 14 (1973), 349–381.
[2] Patricio Aviles, On isolated singularities in some nonlinear partial differential equations,
Indiana Univ. Math. J. 32 (1983), 773–791.
[3] A. Bahri and P.-L. Lions, Solutions of superlinear elliptic equations and their Morse indices,
Comm. Pure Appl. Math. 45 (1992), 1205–1215.
[4] Marie-Françoise Bidaut-Véron and Laurent Véron, Nonlinear elliptic equations on compact
Riemannian manifolds and asymptotics of Emden equations, Invent. Math. 106 (1991), 489–
539.
[5] Haı̈m Brezis, Thierry Cazenave, Yvan Martel, and Arthur Ramiandrisoa, Blow up for ut −
∆u = g(u) revisited, Adv. Differential Equations 1 (1996), 73–90.
[6] Haı̈m Brézis and Tosio Kato, Remarks on the Schrödinger operator with singular complex
potentials, J. Math. Pures Appl. (9) 58 (1979), 137–151.
[7] Michael G. Crandall and Paul H. Rabinowitz, Some continuation and variational methods
for positive solutions of nonlinear elliptic eigenvalue problems, Arch. Rational Mech. Anal.
58 (1975), 207–218.
[8] E. N. Dancer, Superlinear problems on domains with holes of asymptotic shape and exterior
problems, Math. Z. 229 (1998), 475–491.
[9] Louis Dupaigne, Stable solutions to elliptic partial differential equations, CRC Press, to
appear, 2010.
[10] Alberto Farina, On the classification of solutions of the Lane-Emden equation on unbounded
domains of RN , J. Math. Pures Appl. (9) 87 (2007), 537–561. (English, with English and
French summaries)
[11] B. Gidas and J. Spruck, A priori bounds for positive solutions of nonlinear elliptic equations,
Comm. Partial Differential Equations 6 (1981), 883–901.
[12] Zongming Guo and Jiayu Li, The blow up locus of semilinear elliptic equations with super-
critical exponents, Calc. Var. Partial Differential Equations 15 (2002), 133–153.
[13] Fulbert Mignot and Jean-Pierre Puel, Sur une classe de problèmes non linéaires avec non
linéairité positive, croissante, convexe, Comm. Partial Differential Equations 5 (1980), 791–
836. (French)
[14] Wei-Ming Ni and Paul Sacks, Singular behavior in nonlinear parabolic equations, Trans.
Amer. Math. Soc. 287 (1985), 657–671.
[15] Frank Pacard, Partial regularity for weak solutions of a nonlinear elliptic equation,
Manuscripta Math. 79 (1993), 161–172.
PARTIAL REGULARITY OF FINITE MORSE INDEX SOLUTIONS 13

[16] , Convergence and partial regularity for weak solutions of some nonlinear elliptic
equation: the supercritical case, Ann. Inst. H. Poincaré Anal. Non Linéaire 11 (1994), 537–
551. (English, with English and French summaries)
[17] Paul H. Rabinowitz, Dual variational methods for nonlinear eigenvalue problems, in “Eigen-
values of Nonlinear Problems”, CIME Varenna, Edizioni Cremonese, Roma, 1974.
[18] R. Schoen and S.-T. Yau, Conformally flat manifolds, Kleinian groups and scalar curvature,
Invent. Math. 92 (1988), 47–71.
[19] James Serrin, Local behavior of solutions of quasi-linear equations, Acta Math. 111 (1964),
247–302.
[20] Peter Poláčik, Pavol Quittner, and Philippe Souplet, Singularity and decay estimates in
superlinear problems via Liouville-type theorems. I. Elliptic equations and systems, Duke
Math. J. 139 (2007), 555–579.
[21] Kelei Wang, Partial regularity of stable solutions to the supercritical equations and its appli-
cations (Preprint).
[22] , Partial regularity of stable solutions to Emden equation (Preprint).

You might also like