Available online at www.sciencedirect.

com

Journal of Functional Analysis 265 (2013) 153–184

www.elsevier.com/locate/jfa

Groundstates of nonlinear Choquard equations:

Existence, qualitative properties and decay asymptotics
Vitaly Moroz a , Jean Van Schaftingen b,∗
a Swansea University, Department of Mathematics, Singleton Park, Swansea, SA2 8PP, Wales, United Kingdom
b Université Catholique de Louvain, Institut de Recherche en Mathématique et Physique,
Chemin du Cyclotron 2 bte L7.01.01, 1348 Louvain-la-Neuve, Belgium
Received 29 May 2012; accepted 17 April 2013
Available online 28 April 2013
Communicated by F.-H. Lin

Abstract
We consider a semilinear elliptic problem

 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

where Iα is a Riesz potential and p > 1. This family of equations includes the Choquard or nonlinear
Schrödinger–Newton equation. For an optimal range of parameters we prove the existence of a posi-
tive groundstate solution of the equation. We also establish regularity and positivity of the groundstates
and prove that all positive groundstates are radially symmetric and monotone decaying about some point.
Finally, we derive the decay asymptotics at infinity of the groundstates.
© 2013 Elsevier Inc. All rights reserved.

Keywords: Stationary Choquard equation; Stationary nonlinear Schrödinger–Newton equation; Stationary Hartree
equation; Riesz potential; Nonlocal semilinear elliptic problem; Pohožaev identity; Existence; Symmetry;
Decay asymptotics

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
2. Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

* Corresponding author.
E-mail addresses: [email protected] (V. Moroz), [email protected] (J. Van Schaftingen).

0022-1236/$ – see front matter © 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jfa.2013.04.007
154 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

2.1. Variational characterizations of the groundstate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

2.2. Tools for the existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
2.3. Proof of the existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3. Pohožaev identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.1. Pohožaev identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.2. Proof of the nonexistence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
3.3. An integral identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4. Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5. Positivity and symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.1. Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.2. Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6. Decay asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.1. Asymptotics of the nonlocal term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.2. Superlinear case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6.3. Linear case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.4. Sublinear case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

1. Introduction

Given p ∈ (1, ∞), N ∈ N∗ = {1, 2, . . .} and α ∈ (0, N), we consider the problem
  
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,
(1.1)
u(x) → 0 as |x| → ∞,

for u : RN → R, where Iα : RN → R is the Riesz potential defined by

Γ ( N−α
2 )
Iα (x) = ,
α
Γ ( 2 )π N/2 2α |x|N−α

and Γ is the Gamma function, see [27, p. 19].

If u solves (1.1) when α = 2 and N  3, the pair (u, v) = (u, Iα ∗ |u|p ) satisfies the system

⎧
⎪ −u + u = v|u|p−2 u in RN ,
⎪
⎪
⎨ −v = |u|p in RN ,
⎪
⎪ u(x) → 0 as |x| → ∞,
⎪
⎩
v(x) → 0 as |x| → ∞.

Eq. (1.1) is usually called the nonlinear Choquard or Choquard–Pekar equation. It has several
physical origins. In the physical case N = 3, p = 2 and α = 2, the problem
  
−u + u = I2 ∗ |u|2 u in R3 ,
(1.2)
u(x) → 0 as |x| → ∞,
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 155

appeared at least as early as in 1954, in a work by S.I. Pekar describing the quantum mechanics
of a polaron at rest [26]. In 1976 P. Choquard used (1.2) to describe an electron trapped in its
own hole, in a certain approximation to Hartree–Fock theory of one component plasma [16].
In 1996 R. Penrose proposed (1.2) as a model of self-gravitating matter, in a programme in
which quantum state reduction is understood as a gravitational phenomenon [23]. In this context
equation of type (1.1) is usually called the nonlinear Schrödinger–Newton equation. If u solves
(1.1) then the function ψ defined by ψ(t, x) = eit u(x) is a solitary wave of the focusing time-
dependent Hartree equation
 
iψt = −ψ − Iα ∗ |ψ|p |ψ|p−2 ψ in R+ × RN .

In this context (1.1) is also known as the stationary nonlinear Hartree equation.
Problem (1.1) has a variational structure: the critical points of the functional Eα,p ∈
2Np 2Np
C 1 (W 1,2 (RN ) ∩ L N+α (RN ); R) defined for u ∈ W 1,2 (RN ) ∩ L N+α (RN ) by

1 1  
Eα,p (u) = |∇u|2 + |u|2 − Iα ∗ |u|p |u|p
2 2p
RN RN

are weak solutions of (1.1). This functional is well defined by the Hardy–Littlewood–Sobolev
Ns
inequality which states that if s ∈ (1, Nα ) then for every v ∈ Ls (RN ), Iα ∗ v ∈ L N−αs (RN ) and

N
Ns N−αs
|Iα ∗ v| N−αs  C |v|s , (1.3)
RN RN

where C > 0 depends only on α, N and s. Also note that by the Sobolev embedding,
2Np
W 1,2 (RN ) ⊂ L N+α (RN ) if and only if N+α
N−2
 p1  N+α
N
.
The existence of solutions was proved for p = 2 with variational methods by E.H. Lieb,
P.-L. Lions and G. Menzala [16,17,21] and with ordinary differential equations techniques [9,23,
28]. We prove the following general existence result.

Theorem 1. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). If N−2

N+α < 1
p < N
N+α , then there exists
u ∈ W 1,2 (RN ) such that
 
−u + u = Iα ∗ |u|p |u|p−2 u

weakly in RN . Moreover,

   
Eα,p (u) = inf Eα,p (v): v ∈ W 1,2 RN \ {0} and |∇v|2 + |v|2 = Iα ∗ |v|p |v|p .
RN RN

This existence result was mentioned without proof in [20, p. 457]. Here we provide a complete
proof of Theorem 1, with an emphasis on the case p < 2 and the condition p > 1 + Nα which
replaces the standard superlinearity condition p > 1 for similar local problems.
156 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

The existence result of Theorem 1 is sharp, in the sense that if p1  N+α

N−2
or p1  N+α
N
, problem
(1.1) does not have any sufficiently regular nontrivial variational solution. In particular, when
p = 2 and α  N − 4, (1.1) has no nontrivial smooth variational solution. Theorem 1 can be
extended to a general class of nonhomogeneous nonlinearities [25].
2Np
Theorem 2. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). Assume that u ∈ W 1,2 (RN ) ∩ L N+α (RN )
2Np
1,2
and that ∇u ∈ Wloc (RN ) ∩ Lloc
N+α
(RN ). If 1
p  N−2
N+α or 1
p  N
N+α and

 
−u + u = Iα ∗ |u|p |u|p−2 u

weakly in RN , then u = 0.

The proof of Theorem 2 is based on a Pohožaev identity for (1.1).

We call groundstate a function that satisfies the conclusions of Theorem 1. When N = 3,
α = 2 and p = 2, E.H. Lieb has proved that the groundstate u is radial and unique up to transla-
tions [16]. Wei Juncheng and M. Winter have shown that the groundstate is up to translations a
nondegenerate critical point [32].
The symmetry and the regularity of solutions have been proved under the assumption

p  2 and
  
2N pN (2p − 2)N (2p − 1)N (2p − 1)N
2, ∩ p, ∩ , ∩ ,∞ = ∅ (1.4)
N −2 α α+2 α+2 N +α

respectively by Ma Li and Zhao Lin [20] and by S. Cingolani, M. Clapp and S. Secchi [10].
We prove here the regularity, positivity and radial symmetry of groundstates of Choquard equa-
tion (1.1) for the optimal range of parameters.

Theorem 3. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). Assume that N−2

N+α < 1
p < N
N+α . If u ∈
W 1,2 (RN ) is a groundstate of
 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

then u ∈ L1 (RN ) ∩ C ∞ (RN ), u is either positive or negative and there exist x0 ∈ RN and a
monotone function v ∈ C ∞ (0, ∞) such that for every x ∈ RN , u(x) = v(|x − x0 |).

In order to treat the radial symmetry of the solutions in the case p < 2, we use a variational
argument based on the techniques of polarization [4,31].
Finally we study the decay asymptotics of the groundstates. It was known that if p > 2, then
 N−1 
u(x) = O |x|− 2 e−|x|

as |x| → ∞ and if p  2, then for every  > 0,

 
u(x) = O e−(1−)|x|
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 157

as |x| → ∞ [10, Proposition A.2], [17, Remark 1]. We have already studied optimal lower
bounds of the asymptotics of distributional supersolutions of Choquard equations in exterior
domains [24]. Here we derive the following sharp asymptotics for groundstates.

Theorem 4. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). Assume that N−2

N+α < 1
p < N
N+α . If u ∈
W 1,2 (RN ) is a nonnegative groundstate of
 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

then

(Iα ∗ |u|p )(x)

lim = |u|p
|x|→∞ Iα (x)
RN

and

– if p > 2,

N−1
lim u(x)|x| 2 e|x| ∈ (0, ∞);
|x|→∞

– if p = 2,

|x|
N−1 ν N−α
lim u(x)|x| 2 exp 1− ds ∈ (0, ∞),
|x|→∞ s N−α
ν

where

Γ ( N−α2 ) Γ ( N−α2 )
ν N−α = α |u|2 = α (α + 4 − N )Eα,2 (u);
Γ ( 2 )π N/2 2α Γ ( 2 )π N/2 2α
RN

– if p < 2,

 2−p N−α Γ ( N−α2 )

lim u(x) |x| = α |u|p .
|x|→∞ Γ ( 2 )π N/2 2α
RN

In the case where either p > 2 or p = 2 and α < N − 1, similarly to the local stationary
Schrödinger equation, the linear part dominates the nonlocal term and nonnegative groundstates
of Choquard equation have the same asymptotics as the fundamental solution of the Schrödinger
N−1
operator − + 1 on RN and thus u(x) |x|− 2 e−|x| .
If p < 2 the decay of the groundstates is polynomial and entirely controlled by the nonlo-
cal term in the equation. A similar polynomial decay has been observed in the limiting integral
equation for p = 1 + Nα [8]. Note that for the nonlinear Schrödinger equation with fractional
158 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Laplacian the solutions also decay polynomially [12, Theorem 1.3]; in the latter case how-
ever the polynomial decay is inherited from the linear operator whereas in our case it comes
from the nonlinearity. An interesting feature of the decay asymptotic is that it could be used
to show that even though Eα,p with p < 2 is nowhere twice Fréchet-differentiable, it is twice
Gâteaux-differentiable at a groundstate u. In particular, the question of nondegeneracy of u
makes sense.
When p = 2 and α  N − 1, the effect of the nonlocal term in the Choquard equation becomes
strong enough in order to modify the asymptotics of the groundstates. All the three terms in the
equation become balanced and each term should be taken into account in order to derive the
asymptotics of the groundstates. If p = 2 and α = N − 1, the nonlocal term creates a polynomial
N−1 ν N−α
correction to the standard exponential asymptotics, so that u(x) |x|− 2 + 2 e−|x| . If α >
α−(N−1)
N − 1, the correction to the exponential asymptotics is an exponential bounded by ec|x|
(see Remark 6.1). When p = 2 the asymptotics of Iα ∗|u|2 depend only on the groundstate energy
Eα,2 (u); this might be useful in order to prove uniqueness of the groundstate in these cases.
In the classical physical case N = 3, α = 2 and p = 2, where the uniqueness of the ground-
state is known [16], Theorem 4 gives the decay asymptotics of the groundstate with an explicit
polynomial correction to the standard exponential decay rate |x|−1 e−|x| of the fundamental solu-
tion of − + 1 in R3 . This polynomial correction seems to be missing in [32, Theorem I.1(1.7)].

Corollary 5. If u ∈ W 1,2 (RN ) is a nonnegative groundstate of

 
−u + u = I2 ∗ |u|2 u in R3 ,

then
3
lim u(x)|x|1− 8π E2,2 (u) e|x| ∈ (0, ∞).
|x|→∞

The sequel of this paper is organized as follows. In Section 2 we explain the relations between
different variational problems which were historically used in connection with Choquard equa-
tion. Then we provide a detailed proof of Theorem 1, with an emphasis on including the case
p < 2, which was not considered before in the literature. In Section 3, we derive a Pohožaev
identity for Choquard equation and prove the nonexistence result of Theorem 2. The four last
sections are devoted to the proof of Theorem 3. We begin with the regularity (Section 4), con-
tinue with the positivity (Section 5.1) and the symmetry (Section 5.2), and conclude with the
asymptotics of the solutions in Section 6.

2. Existence

2.1. Variational characterizations of the groundstate

There are several ways to construct variationally a groundstate of (1.1).

The first way historically to construct solutions consists in minimizing the quantity

1 1  
0
Eα,p (u) = |∇u|2 − Iα ∗ |u|p |u|p ,
2 2p
RN RN
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 159


under the mass constraint RN |u|2 = m, in order to obtain a weak solution u ∈ W 1,2 (RN ) of
 
−u + μu = Iα ∗ |u|p |u|p−2 u,

where μ ∈ R is a Lagrange multiplier [16,17, Section 3], [18, Section III]. Such solutions are
called normalized solutions. Another approach to the existence is to study the existence of a min-
imizer of Eα,p under the Nehari constraint Eα,p (u), u = 0 [17, Section 2]. A third construction
2Np
[18, Section III.4] consists in minimizing the functional Sα,p ∈ C 1 (W 1,2 (RN ) ∩ L N+α (RN ) \ {0})
2Np
defined for u ∈ W 1,2 (RN ) ∩ L N+α (RN ) \ {0} by

N |∇u| + |u|
2 2
Sα,p (u) =  R 1
.
( RN (Iα ∗ |u|p )|u|p ) p

One more approach to construct solutions variationally is to consider the functional associated
to the interpolation inequality associated to the problem:

 N N+α 
− − 2
N+α N−2
( RN |∇u|2 ) 2 2p ( RN |u|2 ) 2p
Wα,p (u) =  1
,
( RN (Iα ∗ |u|p )|u|p ) p

see [33] for the local analogue.

The next proposition describes the relationships between the different minimization problems.
2Np
Proposition 2.1. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). Let u ∈ W 1,2 (RN ) ∩ L N+α \ {0}. For
t ∈ R∗ , let ut : RN → R be the dilated function defined for x ∈ RN by ut (x) = u(tx). One has

1 1 p
max Eα,p (tu) = − Sα,p (u) p−1 .
t>0 2 2p

N−2 1 N
If N+α < p < N+α , then

N N+α
− 2 − 2p
1 N−2
2 p N+α
min Sα,p (ut ) = Wα,p (u).
t>0 N+α
p − (N − 2) N+α − p
N 1

N 1 N
If N+α+2 < p < N+α , then

2
 N  1 1 2p N(p−1)−2−α
0
min Eα,p t ut = −
2 −
t>0 N (p − 1) − α 2 N (p − 1) − α
2p
Wα,p (u) N(p−1)−2−α
×  N+α N−2
.
− 2
( RN |u|2 ) 2p

N
If 1
p = N
N+α+2 ,
0 (t 2 u ) = t 2 E 0 (u).
then Eα,p t α,p
160 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

N
If 1 N 0 (t 2 u ) = −∞ and
p < N+α+2 , then inft>0 Eα,p t

2
 N  1 1 2p N(p−1)−2−α
0
max Eα,p t ut =
2 −
t>0 2 N (p − 1) − α N (p − 1) − α
2p
Wα,p (u) N(p−1)−2−α
×  N+α N−2
.
− 2
( RN |u|2 ) 2p

Proof. This is proved by a direct computation. 2

This proposition shows that under some restrictions on the exponents, any infimum of one of
the variational problems described above can be written in terms of any other, and any minimizer
of one problem is up to suitable dilation and rescaling a minimizer of the otherproblems. Note
however that for p1  N+α+2N 0 (u) under the mass constraint
the energy Eα,p RN |u| = m is
2

unbounded below and hence, as it was already remarked [17, Remark 9], [18, Remark III.7] in
the case p = 2, the minimization of Sα,p works under less restrictions than the minimization of
0 among normalized functions.
Eα,p
In this rest of this section, we shall prove the existence of a minimizer for the variational
problems Sα,p .

Proposition 2.2. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). If N−2

N+α < 1
p < N
N+α , then there exists
u ∈ W 1,2 (RN ) such that

   
Sα,p (u) = inf Sα,p (v): v ∈ W 1,2 RN and v = 0 .

Observe that Theorem 1 is a straightforward consequence of Propositions 2.1 and 2.2.

There are several strategies to prove Proposition 2.2. A first strategy consists in minimizing
among radial functions and then prove with the symmetrization by rearrangement that a radial
minimizer is a global minimizer. This approach was used for normalized solutions [16,21].
The other approach is the concentration-compactness method of P.-L. Lions [18,19]. In the
sequel of this section we give a proof of Proposition 2.2 that relies on the simplest tools of
concentration-compactness.

2.2. Tools for the existence

The first tool in our proof of Proposition 2.2 is a concentration-compactness lemma of

P.-L. Lions that we reformulate as an inequality [19, Lemma I.1] (see also [34, Lemma 1.21]).

Lemma 2.3. Let q ∈ [1, ∞). If 1

2 − 1
N  1
q  1
2 then for every u ∈ W 1,2 (RN ),

1− q2
|u|q  C sup |u|q |∇u|2 + |u|2 .
x∈RN
RN B1 (x) RN
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 161

Proof. By the Gagliardo–Nirenberg–Sobolev inequality on the ball [1, Theorem 5.8], for every
x ∈ RN , one has

1− q2
|u|  C
q
sup |u| q
|∇u|2 + |u|2 .
x∈RN
B1 (x) B1 (x) B1 (x)

We reach the conclusion by integrating x over RN . 2

Our second tool is a Brezis–Lieb lemma for the nonlocal term of the functional.

Lemma 2.4. Let N ∈ N, α ∈ (0, N), p ∈ [1, N+α

2N
) and (un )n∈N be a bounded sequence in
2Np
L N+α (RN ). If un → u almost everywhere on RN as n → ∞, then

     
lim Iα ∗ |un |p |un |p − Iα ∗ |un − u|p |un − u|p = Iα ∗ |u|p |u|p .
n→∞
RN RN RN

In order to prove Lemma 2.4, we state an easy variant of the classical Brezis–Lieb lemma [5]
(see also [6, Proposition 4.7.30], [35, Theorem 4.2.7]).

Lemma 2.5. Let Ω ⊆ RN be a domain, q ∈ [1, ∞) and (un )n∈N be a bounded sequence in
Lr (Ω). If un → u almost everywhere on Ω as n → ∞, then for every q ∈ [1, r],

 r
lim |un |q − |un − u|q − |u|q  q = 0.
n→∞
Ω

Also recall that pointwise convergence of a bounded sequence implies weak convergence (see
for example [5, Proposition 4.7.12], [35, Proposition 5.4.7]).

Lemma 2.6. Let Ω ⊆ RN be a domain, q ∈ (1, ∞) and (un )n∈N be a bounded sequence in
Lq (Ω). If un → u almost everywhere on Ω as n → ∞, then un u weakly in Lq (Ω).

We now have the ingredients to prove Lemma 2.4.

Proof of Lemma 2.4. For every n ∈ N, one has

   
Iα ∗ |un |p |un |p − Iα ∗ |un − u|p |un − u|p
RN RN
   
= Iα ∗ |un |p − |un − u|p |un |p − |un − u|p
RN
  
+2 Iα ∗ |un |p − |un − u|p |un − u|p .
RN
162 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

2Np
By Lemma 2.5 with q = p and r = N+α , one has |un − u|p − |un |p → |u|p , strongly
2N
in L (RN ) as n → ∞. By the Hardy–Littlewood–Sobolev inequality (1.3), this implies
N+α
2N
that Iα ∗ (|un − u|p − |un |p ) → Iα ∗ |u|p in L N−α (RN ) as n → ∞. Since by Lemma 2.6,
2N
|un − u|p 0 weakly in L N+α (RN ) as n → ∞, we reach the conclusion. 2

2.3. Proof of the existence

We now employ Lemmas 2.3 and 2.4 in order to prove Proposition 2.2.

Proof of Proposition 2.2. Set

   
cα,p = inf Sα,p (u): u ∈ W 1,2 RN \ {0}
   
= inf Sα,p (u): u ∈ W 1,2 RN and Iα ∗ |u|p |u|p = 1 .
RN


Let (un )n∈N be a sequence in W 1,2 (RN ) such that RN (Iα ∗ |un |p )|un |p = 1 for every n ∈ N, and
limn→∞ Sα,p (un ) = cα,p . In particular, the sequence (un )n∈N is bounded in W 1,2 (RN ).
By the Hardy–Littlewood–Sobolev inequality (1.3) and by Lemma 2.3, taking into account
that 12 − N1  N+α
2Np  2 , we obtain
1

N+α
  2Np N
Iα ∗ |un | |un |  C
p p
|un | N+α

RN RN

2Np
1− N+α
Np
N+α
N
C sup |un | N+α |∇un | + |un |
2 2
.
x∈RN
B1 (x) RN

1 N
Since (un )n∈N is bounded in W 1,2 (RN ) and p < N+α , we deduce that the exists a sequence
(xn )n∈N in RN such that

2Np
inf |un | N+α > 0. (2.1)
n∈N
B1 (xn )

Since the problem is invariant by translation, we can assume that for every n ∈ N, xn = 0. Be-
cause the sequence (un )n∈N is bounded, we can also assume without loss of generality that
un u ∈ W 1,2 (RN ) weakly in W 1,2 (RN ) as n → ∞. By (2.1), u = 0. Note now that

 
|∇u|2 + |u|2 = lim |∇un |2 + |un |2 − ∇(un − u)2 + |un − u|2 ,
n→∞
RN RN RN

and therefore
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 163

 1
∗ |un |p )|un |p p
R N (Iα
Sα,p (u) = lim Sα,p (un ) 
RN (Iα ∗ |u| )|u|
n→∞ p p

 1
N (Iα ∗ |un − u| )|un − u|
p p p
− Sα,p (un − u) R 
RN (Iα ∗ |u| )|u|
p p

 p p1
N (Iα ∗ |un | )|un |
p
 cα,p lim inf R 
RN (Iα ∗ |u| )|u|
n→∞ p p

 p p1
N (Iα ∗ |un − u| )|un − u|
p
− R  .
RN (Iα ∗ |u| )|u|
p p

By Lemma 2.4, since p > 1, we reach a contradiction if

   
Iα ∗ |u|p |u|p < lim inf Iα ∗ |un |p |un |p .
n→∞
RN RN

We conclude that Sα,p (u) = cα,p which completes the proof. 2

3. Pohožaev identity

In this section we prove the nonexistence result of Theorem 2 and establish a relationships
between the energy and L2 -norm of a solution.

3.1. Pohožaev identity

We establish the following Pohožaev type identity.

2Np
Proposition 3.1. Let u ∈ W 1,2 (RN ) ∩ L N+α (RN ). If

 
−u + u = Iα ∗ |u|p |u|p−2 u

2,2 2N
and u ∈ Wloc (RN ) ∩ W 1, N+α p (RN ), then

N −2 N N +α  
|∇u|2 + |u|2 = Iα ∗ |u|p |u|p .
2 2 2p
RN RN RN

G. Menzala has used a similar identity to prove the nonexistence of solutions to the equation
−u + u − βIα (x) = (Iα ∗ |u|2 )u [22]. When N = 3, α = 2 and p = 2, Proposition 3.1 has been
proved for smooth solutions [11, Lemma 2.1].
Our proof of Proposition 3.1 follows a classical strategy of testing the equation against
x · ∇u(x) which is made rigorous by multiplying by suitable cut-off functions [15, Proposi-
tion 6.2.1], [34, Appendix B].
164 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Proof of Proposition 3.1. We take ϕ ∈ Cc1 (RN ) such that ϕ = 1 on B1 . By testing the equation
2Np
against the function vλ ∈ W 1,2 (RN ) ∩ L N+α (RN ) defined for λ ∈ (0, ∞) and x ∈ RN by

vλ (x) = ϕ(λx)x · ∇u(x),

we have

 
∇u · ∇vλ + uvλ = Iα ∗ |u|p |u|p−2 uvλ .
RN RN RN

We compute for every λ > 0,

uvλ = u(x)ϕ(λx)x · ∇u(x) dx

RN RN

|u|2
= ϕ(λx)x · ∇ (x) dx
2
RN

  |u(x)|2
=− N ϕ(λx) + λx · ∇ϕ(λx) dx.
2
RN

By Lebesgue’s dominated convergence theorem, it holds

N
lim uvλ = − |u|2 .
λ→0 2
RN RN

2,2
Next, since u ∈ Wloc (RN ) we have

|∇u|2
∇u · ∇vλ = ϕ(λx) |∇u|2 + x · ∇ (x) dx
2
RN RN

  |∇u(x)|2
=− (N − 2)ϕ(λx) + λx · ∇ϕ(λx) dx.
2
RN

By Lebesgue’s dominated convergence again, since u ∈ W 1,2 (RN ),

N −2
lim ∇u · ∇vλ = − |∇u|2 .
λ→0 2
RN RN

Finally, we have for every λ > 0,

 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 165

   
u(y)p Iα (x − y)ϕ(λx)x · ∇ |u| (x) dx dy
p
Iα ∗ |u|p |u|p−2 uvλ =
p
RN RN RN

1  p |u|p
= Iα (x − y) u(y) ϕ(λx) x · ∇ (x)
2 p
RN RN
 p |u|p
+ u(x) ϕ(λy) y · ∇ (y) dx dy
p
   
u(y)p Iα (x − y) N ϕ(λx) + x · ∇ϕ(λx) |u(x)| dx dy
p
=−
p
RN RN
N −α  
+ u(y)p Iα (x − y)
2p
RN RN
(x − y) · (xϕ(λx) − yϕ(λy))  p
× u(x) dx dy.
|x − y| 2

We can thus apply Lebesgue’s dominated convergence theorem to conclude that

  N +α  
lim Iα ∗ |u|p |u|p−2 u vλ = − Iα ∗ |u|p |u|p . 2
λ→0 2p
RN RN

3.2. Proof of the nonexistence

We now complete the proof of Theorem 2.

Proof of Theorem 2. By testing the equation against u, we obtain the identity

 
|∇u|2 + |u|2 = Iα ∗ |u|p |u|p .
RN RN RN

Hence, we have

N −2 N +α N N +α
− |∇u|2 + − |u|2 = 0.
2 2p 2 2p
RN RN

If 1
p  N−2
N+α or 1
p  N
N+α , this implies that u = 0. 2

3.3. An integral identity


Proposition 3.1 allows to express RN |u|2 in terms of Eα,p (u). We will see in Section 6 that
the asymptotics of the groundstates when p  1 depend on RN |u|p . When p = 2, we will thus
be able to express the asymptotics of a groundstate in terms of the energy of the groundstate.
166 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Proposition 3.2. Let α ∈ (0, N) and p ∈ (1, ∞) be such that N−2

N+α < 1
p < N
N+α . If u ∈ W 1,2 (RN )
is a groundstate of
 
−u + u = Iα ∗ |u|p |u|p−2 u,

then

α+2
|u|2 = − (N − 2) Eα,p (u).
p−1
RN

Proof. The proof is a direct computation. 2

4. Regularity

In this section we study the regularity of a solution of (1.1).

Proposition 4.1. Let α ∈ (0, N) and p ∈ (1, ∞) such that N−2

N+α < 1
p < N
N+α . Let u ∈ W 1,2 (RN )
be a solution of
 
−u + u = Iα ∗ |u|p |u|p−2 u.

Then u ∈ L1 (RN ) ∩ C 2 (RN ), u ∈ W 2,s (RN ) for every s > 1, and u ∈ C ∞ (RN \ u−1 ({0})).

S. Cingolani, M. Clapp and S. Secchi [10, Lemma A.1] have proved that u ∈ Ls (RN ) for every
s  2 and that u is smooth under the additional assumptions that p  2 and that (1.4) holds.
We shall split the proof of Proposition 4.1 in the proof of three separate claims.

Claim 1. For every s  1 such that

1 α 1 2
> 1− − ,
s N p N

one has u ∈ Ls (RN ) and for every r > 1 such that

1 α 1
> 1− ,
r N p

one has u ∈ W 2,r (RN ).

Proof of Claim 1. Since u ∈ W 1,2 (RN ), u ∈ Ls0 (RN ) with

2Np
s0 = .
N +α

Set s 0 = s 0 = s0 . Assume that u ∈ Ls (RN ) for every s ∈ [s n , s n ]. By the Hardy–Littlewood–

Sobolev inequality (1.3), if
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 167

1 p α
= − >0
t s N

then Iα ∗ |u|p ∈ Lt (RN ). Further, if

1 2p − 1 α
= − < 1,
r s N

then we have (Iα ∗ |u|p )|u|p−2 u ∈ Lr (RN ). Hence, u ∈ W 2,r (RN ) by the classical Calderón–
Zygmund Lp regularity estimates [14, Chapter 9].
Therefore, if

α 1 N +α
< < ,
Np s N (2p − 1)

then u ∈ W 2,r (RN ). In other words, we have thus proved that u ∈ W 2,r (RN ), for every r > 1
such that

2p − 1 α 1 2p − 1 α
− < < −
sn N r sn N

and

1 α 1
> 1− .
r N p

Finally, by the Sobolev embedding theorem, u ∈ Ls (Ω) provided that

2p − 1 α + 2 1 2p − 1 α
− < < −
sn N s sn N

and

1 α 1 2
> 1− − .
s N p N

2Np
From s n < N+α < s n , we deduce that

2p − 1 α + 2 1 2Np 1 2p − 1 α
− < < < < − .
sn N sn N + α sn sn N

2p−1 2p−1 2p−1

If sn − α+2
N  1, we are done. Otherwise, we set s n+1 = s n − N . Similarly, if s n − N 
1 α α+2

N (1 − p ) − N ,
α 1 2
then we are done. Otherwise we 1
set s̄n+1 = 2p−1
s n − N and iterate. This allows
α+2

to reach the conclusion after a finite number of steps. 2

Claim 2. For every r > 1, u ∈ W 2,r (RN ).

168 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

The proof of this claim is classical, we give a proof in the same style as the proof of Claim 1
for the sake of completeness.

N (1 − p ) − N < Np and by Claim 1, Iα ∗ |u|p ∈

1 N−2 α 1 2 α
Proof of Claim 2. Since p > N+α , we have
L∞ (RN ). If r0 ∈ (1, ∞) is defined by r10 = N (1 −
α
1, u ∈ W 2,r (RN ) for every
1
p ), by Claim
r ∈ (1, r0 ).
Assume now that u ∈ W 2,r (RN ) for every r ∈ (1, rn ). By the classical Sobolev embedding
theorem, u ∈ Ls (RN ) for every s ∈ [1, ∞) such that 1s > r1n − N2 . Hence, (Iα ∗ |u|p )|u|p−2 u ∈
Lr (RN ) for every r ∈ (1, ∞) such that p − 1 > 1r > (p − 1)( r1n − N2 ). By the classical
Calderón–Zygmund theory [14, Chapter 9], u ∈ W 2,r (RN ) for every r ∈ (1, ∞) such that
r > (p − 1)( rn − N ). If rn  2 , we are done. Otherwise set
1 1 2 N

1 1 2
= (p − 1) − .
rn+1 rn N

If p  2, then

1 1 2(p − 1)
< − ,
rn+1 rn N

N (1 − p )
1 α 1
and the conclusion is reached after a finite number of steps. If p > 2, then since rn <
and 1
p  N−2
N+α ,

1 1 α 2 2
< + (p − 1) 1− −
rn+1 rn N p N
1 α N −2 2 1
< + (p − 1) 1−2 − < ;
rn N N +α N rn

the conclusion is again reached after a finite number of steps. 2

Claim 3. For every k ∈ N, λ ∈ (0, 1) and every ball B ⊂ RN , if k + λ < p + 1 or infB |u| > 0,
k,λ
u ∈ Cloc (RN ).

Proof of Claim 3. By Claim 2 and by the Morrey–Sobolev embedding, the conclusion holds for
k ∈ {0, 1}. Let B̃ ⊂ RN be a ball such that B̄ ⊂ B̃ and infB̃ |u|  12 infB |u|. Take η ∈ Cc∞ (RN )
such that η = 1 on B1 . Write Iα = ηIα + (1 − η)Iα . Note that since u ∈ Ls (RN ) for every s  1

   
(1 − η)Iα ∗ |u|p ∈ C ∞ RN .

On the other hand, if either k + λ < p − 1 or infB |u| > 0, |u|p , |u|p−2 u ∈ C k,λ (B̃). Since ηIα ∈
L1 (RN ), (ηIα ) ∗ |u|p ∈ C k,λ (B̃). Therefore (Iα ∗ |u|p )|u|p−2 u ∈ C k,λ (B̃) and by the classical
Schauder regularity estimates [14, Theorem 4.6], u ∈ C 2,λ (B). 2
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 169

5. Positivity and symmetry

5.1. Positivity

We show that any groundstate of (1.1) is a positive function.

Proposition 5.1. Let α ∈ (0, N) and p ∈ (1, ∞) be such that N−2

N+α < 1
p < N
N+α . If u ∈ W 1,2 (RN )
is a groundstate of
 
−u + u = Iα ∗ |u|p |u|p−2 u,

then either u > 0 or u < 0.

Proof. If u is a groundstate note that |u| is also a groundstate, and therefore by Proposition 4.1
one has |u| ∈ C 2 (RN ) and
 
−|u| + |u| = Iα ∗ |u|p |u|p−1 .

By the strong maximum principle, either |u| > 0 or |u| = 0. Since u = 0, we conclude that either
u > 0 or u < 0. 2

Proposition 5.1 can also be proved without using any regularity of the groundstates. Indeed, if
u ∈ H 1 (RN ) and −|u| + |u|  0 weakly, then either u > 0, u < 0 or u = 0 almost everywhere
[31, Proposition 2.1].

5.2. Symmetry

In this section we prove that any groundstate of (1.1) is a radial and radially decreasing solu-
tion.

Proposition 5.2. Let α ∈ (0, N) and p ∈ (1, ∞) be such that N−2

N+α < 1
p < N
N+α . If u ∈ W 1,2 (RN )
is a positive groundstate of
 
−u + u = Iα ∗ |u|p |u|p−2 u,

then there exist x0 ∈ Rn and v : (0, ∞) → R a nonnegative nonincreasing function such that for
almost every x ∈ RN
 
u(x) = v |x − x0 | .

By Proposition 5.1, any groundstate of (1.1) is positive. Therefore, when p  2 and assump-
tions (1.4) holds, Proposition 5.2 follows from the result of Li Ma and Lin Zhao [20, Theorem 2].
Our proof will follow the strategy of T. Bartsch, M. Willem and T. Weth [4,31], which consists
in using the minimality property of the groundstate to deduce some relationship between the
function and its polarization. The argument is simplified here because the nonlocal term has
some strong symmetrizing effect.
170 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Let H ⊂ RN be a closed half-space. Let σH denote the reflection with respect to ∂H and
uH : RN → R be the polarization of u : RN → R defined for x ∈ RN by

max(u(x), u(σH (x))) if x ∈ H ,

uH (x) =
min(u(x), u(σH (x))) if x ∈
/H

(see [7]).
A first tool is the study of equality cases in a polarization inequality.
2N
Lemma 5.3. Let α ∈ (0, N) and u ∈ L N+α (RN ) and H ⊂ RN be a closed half-space. If u  0
and

u(x)u(y) uH (x)uH (y)

dx dy  dx dy,
|x − y|N−α |x − y|N−α
RN RN RN RN

then either uH = u or uH = u ◦ σH .

This inequality is proved without the equality cases by A. Baernstein II [3, Corollary 4] (see
also [30, Proposition 8]).

Proof. First note that the integrals are finite by the Hardy–Littlewood–Sobolev inequality. Also
note that

u(x)u(y) u(x)u(y) + u(σH (x))u(σH (y))

dx dy =
|x − y|N−α |x − y|N−α
RN RN H H

u(x)u(σH (y)) + u(σH (x))u(y)

+ dx dy
|x − σH (y)|N−α

and that a similar identity holds for uH . By an inspection of all the possible cases, we have, since
u  0 and α < N , for every x ∈ H and y ∈ H

u(x)u(y) + u(σH (x))u(σH (y)) u(x)u(σH (y)) + u(σH (x))u(y)

+
|x − y|N−α |x − σH (y)|N−α
uH (x)uH (y) + uH (σH (x))uH (σH (y))

|x − y|N−α
uH (x)uH (σH (y)) + uH (σH (x))uH (y)
+ ,
|x − σH (y)|N−α

with equality if and only if either uH (x) = u(x) and uH (y) = u(y) or uH (x) = u(σH (x))
and uH (y) = u(σH (y)). Since this holds for almost every (x, y) ∈ H 2 , we have proved the
lemma. 2

Now we show how the information that either uH = u or uH = u ◦ σH can be used to deduce
some symmetry.
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 171

Lemma 5.4. Let s  1 and u ∈ Ls (RN ). If u  0 and for every closed half-space H ⊂ RN ,
uH = u or uH = u ◦ σH , then there exist x0 ∈ RN and v : (0, ∞) → R a nondecreasing function
such that for almost every x ∈ RN , u(x) = v(|x − x0 |).

This was already proved [31, Proposition 3.15]. We give here a simpler argument that does
not rely on the approximation of symmetrizations by polarizations. In order to prove Lemma 5.4,
we recall the following result.

Lemma 5.5. Let u ∈ Ls (RN ), w ∈ Lt (RN ) with 1s + 1t = 1 to be a radial function such that
for every x, y ∈ RN with |x|  |y|, w(x)  w(y) with equality if and only if |x| = |y|, and let
H ⊂ RN be a closed half-space. If 0 is an interior point of H and

uH w  uw,
RN RN

then u = uH .

The proof is elementary and was given for example in [29, Lemma 4]. We also use a classical
characterization of functions invariant under polarizations [7, Lemma 6.3], [29, Lemma 5].

Lemma 5.6. Let u ∈ Ls (RN ). If for every closed half-space H ⊂ RN such that x0 ∈ H , uH = u,
then there exists v : (0, ∞) → R a nondecreasing function such that for almost every x ∈ RN ,
u(x) = v(|x − x0 |).

Proof of Lemma 5.4. Choose w that satisfies the assumptions of Lemma 5.5 and define for
x ∈ RN the function

W (x) = u(y)w(x − y) dy.

RN

This function is nonnegative, continuous and lim|x|→∞ W (x) = 0, and therefore admits a maxi-
mum point x0 ∈ RN .
In order to conclude with Lemma 5.6, we now claim that for every closed half-space H such
that x0 ∈ H , uH = u. Indeed, if uH = u ◦ σH one has since w is radial and by definition of x0

 
uH (y)w(x0 − y) dy = u σH (y) w(x0 − y) dy
RN RN
 
= u(y)w σH (x0 ) − y dy
RN

 u(y)w(x0 − y) dy.
RN
172 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Therefore, by Lemma 5.5, if x0 is an interior point of H , we have uH = u. The case x0 ∈ ∂H

follows by continuity. 2

Proof of Proposition 5.2. Let H ⊂ RN be a closed half-space. Note that

 H 2
∇u  = |∇u|2
RN RN

and

 H 2
u  = |u|2 .
RN RN

In view of the characterization of grounds states, we have necessarily

  p  p  
Iα ∗ uH  uH   Iα ∗ |u|p |u|p .
RN RN

By Lemma 5.3, for every closed half-space H ⊂ RN , either uH = u or uH = u◦σH . We conclude

by Lemma 5.4. 2

6. Decay asymptotics

In this last section, we study the asymptotic behavior of solutions.

6.1. Asymptotics of the nonlocal term

We first study the asymptotics of the nonlocal term. We give explicit bounds on convergence
rates that we shall need in the sequel.

Proposition 6.1. Let N ∈ N∗ , α ∈ (0, N) and p ∈ (1, ∞). If N−2

N+α < 1
p < N
N+α and u is a ground-
state of

 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

then there exists C ∈ R such that for every x ∈ RN ,

 
  C 1 1
Iα ∗ |u|p (x) − Iα (x) |u|p   + .
 |x|N−α 1 + |x| 1 + |x|N (p−1)
RN

The proof of this proposition will follow from regularity estimates together with a computa-
tion of the asymptotics of a Riesz potential.
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 173

Lemma 6.2. Let α ∈ (0, N), β ∈ (N, ∞) and f ∈ L∞ (RN ). If

 
sup f (x)|x|β < ∞,
x∈RN

then there exists C > 0 such that for every x ∈ RN

 
 f (y) 1  C 1 1
 dy − N−α f (y) dy   N−α + .
 |x − y| N−α |x| |x| 1 + |x| 1 + |x|β−N
RN RN

Similar statements have already appeared in [13, Lemma 2.1]. Here we emphasize the precise
control on the rate of decay at infinity. The reader will see in the proof that C only depends on
supx∈RN |f (x)|(1 + |x|β ).

Proof of Lemma 6.2. We need to estimate the quantity

   
 1 1    1 1 
 − dy   f (y)
 f (y)
|x − y|N−α |x|N−α  |x − y|N−α − |x|N−α  dy.
RN RN

On the one hand, there exists C ∈ R such that if x, y ∈ RN and |y|  2|x|, one has
 
 1 1  C|y|

 |x − y|N−α − |x|N−α   |x|N−α+1

and thus
   
 1 1   1 C |y| 
 f (y) − dy    N−α+1 dy 
 |x − y|N−α |x|N−α |x| 1 + |y|β 
B2|x| B2|x|

C 1 1
 + . (6.1)
|x|N−α 1 + |x| 1 + |x|β−N

On the other hand, there exists C ∈ R such that if x, y ∈ RN and |y|  2|x|,
 
 1 1  C
 −
 |x − y|N−α |x|N−α   |x|N−α ,

from which we compute that

   
 1 1   1 C 
 f (y) − N−α dy    N−α dy 
 |x − y| N−α |x| |x| |y| β
RN \B2|x| RN \B2|x|

C
 . (6.2)
|x|N−β

The conclusion follows from (6.1) and (6.2). 2

174 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

Proof of Proposition 6.1. By Proposition 4.1, u ∈ L1 (RN ) ∩ L∞ (RN ) and by Proposition 5.2,
|u| is radial and radially decreasing. Therefore for every x ∈ RN ,

  1 C
u(x)  |u|  .
|B|x| | |x|N
RN

The conclusion follows from the application of Lemma 6.2 with f = |u|p and β = Np. 2

6.2. Superlinear case

In the case p > 2, we are going to show that groundstates have classical exponential decay.

Proposition 6.3. Let N ∈ N, α ∈ (0, N) and p ∈ (2, ∞). If 1

p < N−2
N+α and u is a positive ground-
state of
 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

then
N−1
lim u(x)|x| 2 e|x| ∈ (0, ∞).
|x|→∞

Note that if α  N − 4 then the assumptions of the proposition cannot be satisfied. S. Cin-
golani, M. Clapp and S. Secchi have proved that the limit is finite [10, Lemma A.2].
The proof of this result follows the standard proof for the nonlocal problem. Our main tool
computational tool is an analysis of the decay rate of solutions of a linear Schrödinger equation.

Lemma 6.4. Let ρ  0 and W ∈ C 1 ((ρ, ∞), R). If

lim W (s) > 0

s→∞

and for some β > 0,

lim W (s)s 1+β = 0,

s→∞

then there exists a nonnegative radial function v : RN \ Bρ → R such that

−v + W v = 0

in RN \ Bρ and for some ρ0 ∈ (ρ, ∞),

|x|
N−1 √
lim v(x)|x| 2 exp W = 1.
|x|→∞
ρ0
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 175

S. Agmon has proved this result when β > 12 [2, Theorem 3.3]. The proof of Lemma 6.4
appears in a previous work by the authors [24, Proposition 6.1]. The proof is based on comparison
with functions of the form
 |x| 
− N−1
√
vτ (x) = |x| 2 exp − W + τ |x|β
ρ

for suitable values of τ ∈ R.

Proof of Proposition 6.3. By Proposition 4.1, u ∈ L1 (RN ) ∩ L∞ (RN ). Since p > 2, we have
   p−2
lim Iα ∗ |u|p (x)u(x) = 0.
|x|→∞

Hence, there exists ρ ∈ R such that in x ∈ RN and |x|  ρ,

   p−2 3
Iα ∗ |u|p (x)u(x)  .
4

We have thus in RN \ Bρ ,

1
−u + u  0.
4

Let v ∈ C 2 (RN \ Bρ , R) be such that

⎧
⎪ 1
⎪
⎪ −v + v = 0 if x ∈ RN \ Bρ ,
⎨ 4
v(x) = u(x) if x ∈ ∂Bρ ,
⎪
⎪
⎪
⎩ lim v(x) = 0.
|x|→∞

By Lemma 6.4 with W = 14 , there exists μ ∈ R such that for every x ∈ RN \ Bρ

μ |x|
v(x)  N−1
e− 2 .
|x| 2

Hence, by the comparison principle, for every x ∈ RN \ Bρ

μ |x|
u(x)  v(x)  N−1
e− 2 ,
|x| 2

which implies that there exists ν ∈ R such that for every x ∈ RN \ Bρ

   p−2
Iα ∗ |u|p (x)u(x)
p−2
 νe− 2 |x| .
176 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

We have now

−u + u  0  −u + W u,

in RN \ Bρ , where W ∈ C 1 (RN \ Bρ ) is defined for x ∈ RN \ ρ by

p−2
W (x) = 1 − νe− 2 |x| .

Define now u, u ∈ C 2 (RN \ Bρ , R) such that

⎧
⎪
⎪ −u + u = 0 if x ∈ RN \ Bρ ,
⎨
u(x) = u(x) if x ∈ ∂Bρ ,
⎪
⎪ lim u(x) = 0,
⎩
|x|→∞

and

⎧
⎪ −u + W u = 0 if x ∈ R \ Bρ ,
⎪ N
⎨
u(x) = u(x) if x ∈ ∂Bρ ,
⎪
⎪ lim u(x) = 0.
⎩
|x|→∞

By the comparison principle u  u  u in RN \ Bρ , whence in view of Lemma 6.4

N−1 N−1
0 < lim u(x)|x| 2 e|x|  lim inf u(x)|x| 2 e|x|
|x|→∞ |x|→∞
N−1 N−1
 lim sup u(x)|x| 2 e|x|  lim u(x)|x| 2 e|x| < ∞. (6.3)
|x|→∞ |x|→∞

N−1
It remains to prove that the u(x)|x| 2 e|x| has a limit as x → ∞. Since u and u, are both radial
functions, we have by the comparison principle, for every r, s ∈ (ρ, ∞) such that r  s,

u(r) u(s)
 ,
u(r) u(s)

that is, the function u/u is nondecreasing. On the other hand, by (6.3), the function u/u is
bounded. We conclude then that u/u has a finite limit at infinity and that

N−1 u(x) N−1

lim u(x)|x| 2 e|x| = lim lim u(x)|x| 2 e|x| ∈ (0, ∞). 2
|x|→∞ |x|→∞ u(x) |x|→∞
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 177

6.3. Linear case

The analysis of the case p > 2 is based on the fact that |u|p−2 decays exponentially at infinity.
When p = 2, we have to take into account the Riesz potential Iα ∗ |u|2 .

Proposition 6.5. Let N ∈ N, α ∈ (0, N) and α > N − 4. If u is a positive groundstate of

 
−u + u = Iα ∗ |u|2 u in RN ,

then

|x|
N−1 ν N−α
lim u(x)|x| 2 exp 1− ds ∈ (0, ∞),
|x|→∞ s N−α
ν

where

Γ ( N−α2 )
ν N−α = α |u|p .
Γ ( 2 )π N/2 2α
RN

Remark 6.1. The asymptotics of the groundstate are thus related to the behavior of the function

|x|
ν N−α
1− ds
s N−α
ρ

as |x| → ∞. When α < N − 1, we have

|x| 
ν N−α
1− 1− ds < ∞,
s N−α
ρ

and therefore,
N−1
lim u(x)|x| 2 e|x| ∈ (0, ∞),
|x|→∞

as in Proposition 6.3. When α < N − 12 , we have

|x| 
ν N−α ν N−α
1 − N−α − 1− ds < ∞.
2s s N−α
ρ

Therefore, in the critical case α = N − 1 which includes the physical case N = 3 and α = 2,
we have a polynomial perturbation of the previous asymptotics,
178 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

N−1−ν
lim u(x)|x| 2 e|x| ∈ (0, ∞),
|x|→∞

whereas when α ∈ (N − 1, N − 12 ),

ν N−α |x|1−(N−α)
e|x| e−
N−1
lim u(x)|x| 2 2(1−(N−α)) ∈ (0, ∞).
|x|→∞

Larger values of α ∈ [N − 12 , N) could be analyzed by taking higher-order Taylor expansions of

the square root [24, Remark 6.1].

We are now ready to establish the asymptotics of the solution in the linear case p = 2.

Proof of Proposition 6.5. By Proposition 6.1, there exists μ > 0 such that for every x ∈ RN ,

μ   ν N−α μ
−  Iα ∗ |u|2 (x) − N−α  N−α+1 .
|x|N−α+1 |x| |x|

Define now W ∈ C 1 (RN \ {0}) and W ∈ C 1 (RN \ {0}) for x ∈ RN \ {0} by

ν N−α μ
W (x) = 1 − +
|x|N−α |x|N−α+1

and

ν N−α μ
W (x) = 1 − − .
|x|N−α |x|N−α+1

One has, in RN \ {0},

−u + W u  0  −u + W u.

Define now u, u ∈ C 2 (RN \ Bρ , R) such that

⎧
⎪
⎪ −u + W u = 0 if x ∈ RN \ B1 ,
⎨
u(x) = u(x) if x ∈ ∂B1 ,
⎪
⎪ lim u(x) = 0,
⎩
|x|→∞

and
⎧
⎪
⎪ −u + W u = 0 if x ∈ RN \ B1 ,
⎨
u(x) = u(x) if x ∈ ∂B1 ,
⎪
⎪ lim u(x) = 0.
⎩
|x|→∞
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 179

One has for ρ large enough,

|x|
N−1 
lim u(x)|x| 2 W (s) ds ∈ (0, ∞)
|x|→∞
ρ

and

|x|
N−1
lim u(x)|x| 2 W (s) ds ∈ (0, ∞).
|x|→∞
ρ

Note now that

∞
 
W − W < ∞,
ρ

from which we deduce that

|x|
N−1 ν N−α
lim u(x)|x| 2 1− ds ∈ (0, ∞)
|x|→∞ s N−α
ρ

and

|x|
N−1 ν N−α
lim u(x)|x| 2 1− ds ∈ (0, ∞).
|x|→∞ s N−α
ρ

We conclude that

|x|
N−1 ν N−α
lim u(x)|x| 2 1− ds ∈ (0, ∞)
|x|→∞ s N−α
ρ

as in the proof of Proposition 6.3. 2

6.4. Sublinear case

Whereas in the case p > 2, the nonlinear term did not play any role in the asymptotics and for
p = 2 all the terms were playing a role, in the case p < 2, the asymptotics are governed by the
terms without derivatives in the equation.

Proposition 6.6. Let N ∈ N, α ∈ (0, N) and p ∈ (1, 2). If N

N+α < 1
p < N−2
N+α and u is a positive
groundstate of
180 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

 
−u + u = Iα ∗ |u|p |u|p−2 u in RN ,

then

 2−p N−α Γ ( N−α2 )

lim u(x) |x| = α |u|p .
|x|→∞ Γ ( 2 )π N/2 2α
RN

When u is merely a distributional supersolution to the equation, it was already known that
[24, Theorem 5]

(u(x))2−p
lim inf > 0.
|x|→∞ |x|N−α

In the limiting case p = N+α

N , Chen Weng, Li Congming and Ou Biao [8] have proved that if
2N
u∈L N−α (RN ) is nonnegative and satisfies the equation without the differential operator
 N+α  α
u = Iα ∗ |u| N u N ,

then

N−α N+α Cα
u(x) N = |u| N
N−α
.
(λ2 + |x|2 ) 2
RN

In the proof of Proposition 6.6, we shall use the asymptotics of the nonlocal term derived in
Proposition 6.1, together with asymptotics of solutions to some linear equations.

Lemma 6.7. Let ρ > 0, β > 0 and λ > 0 and u ∈ C 2 (RN \ Bρ ). If for every x ∈ RN \ Bρ ,

1
−u(x) + λu(x) = ,
|x|β

and

lim u(x) = 0,
|x|→∞

then

lim u(x)λ|x|β = 1.
|x|→∞

Proof. Let w ∈ C 2 (RN \ Bρ ) be a solution of −w + w = 0 such that w > 0 and

lim|x|→∞ w(x) = 0. For σ, τ ∈ R define vσ,τ ∈ C 2 (RN \ Bρ ) for every x ∈ RN \ Bρ by

1 σ
vσ,τ (x) = + β+2 + τ w(x).
λ|x|β |x|
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 181

One has, for every σ, τ ∈ R and x ∈ RN \ Bρ ,

1 1 β(N − 2 − β) (β + 2)(N − β)
−vσ,τ (x) + vσ,τ (x) = + β+2 +σ λ+ .
|x|β |x| λ |x|2

Let R ∈ (0, ∞) such that

|β + 2||N − β| λ
 .
R2 2

If we choose now σ , σ ∈ R such that

3λ β(β + 2 − N ) λ
σ  σ ,
2 λ 2

for every x ∈ RN \ BR and τ ∈ R, we have

1
−vσ ,τ (x) + λvσ ,τ (x)   −vσ ,τ (x) + λvσ ,τ (x).
|x|β

Since u is continuous, u is bounded on ∂BR and there exist τ , τ ∈ R such that vσ ,τ 

u  vσ ,τ on ∂BR . By the comparison principle, vσ ,τ  u  vσ ,τ in RN \ BR . Since
lim|x|→∞ vσ ,τ (x)λ|x|β = lim|x|→∞ vσ ,τ (x)λ|x|β = 1, we conclude that lim|x|→∞ λu(x)|x|β =
1. 2

The proof yields in fact a stronger statement that are not needed in the sequel:

1 1
u(x) = +O ,
λ|x|β |x|β+2

as |x| → ∞. By Phragmen–Lindelöf theory, the assumption β > 0 √ can be dropped and the as-
N−1
sumption lim|x|→∞ u(x) = 0 can be replaced by lim|x|→∞ |x| 2 −
e λx u(x) = 0.

Proof of Proposition 6.6. By Proposition 6.1, there exists μ ∈ R such that we have for every
x ∈ RN
 
  μ
Iα ∗ up (x) − Iα (x) |u|p   , (6.4)
 |x|N−α+δ
RN

with 0 < δ  min(1, N(p − 1)).

By Proposition 4.1, u > 0 and u ∈ C 2 (RN ). Hence, by the chain rule, u2−p ∈ C 2 (RN ) and

−u2−p = −(2 − p)u1−p u + (2 − p)(p − 1)|∇u|2

in RN . Since p ∈ (1, 2), by the equation satisfied by u and by (6.4), we have for every x ∈ RN ,
182 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

μ
−u2−p (x) + (2 − p)u2−p (x)  (2 − p)Iα (x) |u|p − .
|x|δ
RN

Let u : RN \ B1 be such that

⎧
⎪ μ
⎪
⎪ −u(x) + (2 − p)u(x) = (2 − p)Iα (x) |u|p − for x ∈ RN \ B̄1 ,
⎪
⎪ |x|δ
⎨
RN
⎪
⎪ u(x) = u(x) 2−p
for x ∈ ∂B1 ,
⎪
⎪
⎪
⎩ lim u(x) = 0.
|x|→∞

By the comparison principle, u  u2−p in RN \ B1 . We apply now Lemma 6.7 twice and use the
linearity of the operator − + 1 to write

u
lim = |u|p .
|x|→∞ Iα (x)
RN

We conclude that

(u(x))2−p
lim inf  |u|p . (6.5)
|x|→∞ Iα (x)
RN

For the converse inequality, we note that by Young’s inequality,

    1
Iα ∗ |u|p |u|p−2 u  (2 − p) Iα ∗ |u|p 2−p + (p − 1)u.

By (6.4), we have for every x ∈ RN \ B1 .

1
   1 1 C p−2
Iα ∗ |u|p (x) 2−p  Iα (x) 2−p |u|p +
|x|δ
RN
1
1 2−p ν
 Iα (x) 2−p |u|p + .
|x|δ
RN

Therefore, for every x ∈ RN \ B1 ,

1
1 2−p ν
−u(x) + (2 − p)u(x)  (2 − p)Iα (x) 2−p |u| p
+ .
|x|δ
RN
 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184 183

Define now ū ∈ C 2 (RN \ B1 ) by

⎧
⎪
1
⎪
⎪ 1 2−p ν
⎪
⎪ −ū(x) + (2 − p)ū(x) = (2 − p)Iα (x) 2−p |u| p
+ if x ∈ RN \ B̄1 ,
⎪
⎨ |x|δ
RN
⎪
⎪ ū(x) = u(x) if x ∈ ∂B1 ,
⎪
⎪
⎪
⎪
⎩ lim ū(x) = 0.
|x|→∞

By the comparison principle, we have u  ū in RN \ B1 . By Lemma 6.7, we have

1
ū(x) 2−p
lim 1
= |u| p
.
|x|→∞
Iα (x) 2−p
RN

Thus

(u(x))2−p
lim sup  |u|p , (6.6)
|x|→∞ Iα (x)
RN

and the assertion follows from the combination of (6.5) and (6.6). 2

References

[1] R.A. Adams, J.J.F. Fournier, Sobolev Spaces, 2nd ed., Pure Appl. Math., vol. 140, Elsevier/Academic Press, Ams-
terdam, 2003.
[2] S. Agmon, Bounds on exponential decay of eigenfunctions of Schrödinger operators, in: Schrödinger Operators,
Como, 1984, in: Lecture Notes in Math., vol. 1159, Springer, Berlin, 1985, pp. 1–38.
[3] A. Baernstein II, A unified approach to symmetrization, in: Partial Differential Equations of Elliptic Type, Cortona,
1992, in: Sympos. Math., vol. XXXV, Cambridge Univ. Press, 1994, pp. 47–91.
[4] T. Bartsch, T. Weth, M. Willem, Partial symmetry of least energy nodal solutions to some variational problems,
J. Anal. Math. 96 (2005) 1–18.
[5] V.I. Bogachev, Measure Theory, Springer, Berlin, 2007.
[6] H. Brézis, E. Lieb, A relation between pointwise convergence of functions and convergence of functionals, Proc.
Amer. Math. Soc. 88 (3) (1983) 486–490.
[7] F. Brock, A.Yu. Solynin, An approach to symmetrization via polarization, Trans. Amer. Math. Soc. 352 (4) (2000)
1759–1796.
[8] W. Chen, C. Li, B. Ou, Classification of solutions for an integral equation, Comm. Pure Appl. Math. 59 (3) (2006)
330–343.
[9] P. Choquard, J. Stubbe, M. Vuffray, Stationary solutions of the Schrödinger–Newton model—An ODE approach,
Differential Integral Equations 21 (7–8) (2008) 665–679.
[10] S. Cingolani, M. Clapp, S. Secchi, Multiple solutions to a magnetic nonlinear Choquard equation, Z. Angew. Math.
Phys. 63 (2) (2012) 233–248.
[11] S. Cingolani, S. Secchi, M. Squassina, Semi-classical limit for Schrödinger equations with magnetic field and
Hartree-type nonlinearities, Proc. Roy. Soc. Edinburgh Sect. A 140 (5) (2010) 973–1009.
[12] P. Felmer, A. Quaas, Jinggang Tan, Positive solutions of nonlinear Schrödinger equation with the fractional Lapla-
cian, Proc. Roy. Soc. Edinburgh Sect. A 142 (6) (2012) 1237–1262.
[13] B. Gidas, W.M. Ni, L. Nirenberg, Symmetry of positive solutions of nonlinear elliptic equations in Rn , in: Mathe-
matical Analysis and Applications, Part A, in: Adv. Math. Suppl. Stud., vol. 7, Academic Press, New York, 1981.
[14] D. Gilbarg, N.S. Trudinger, Elliptic Partial Differential Equations of Second Order, Grundlehren Math. Wiss.,
vol. 224, Springer, Berlin, 1983.
184 V. Moroz, J. Van Schaftingen / Journal of Functional Analysis 265 (2013) 153–184

[15] O. Kavian, Introduction à la théorie des points critiques et applications aux problèmes elliptiques, Math. Appl.,
vol. 13, Springer, Paris, 1993.
[16] E.H. Lieb, Existence and uniqueness of the minimizing solution of Choquard’s nonlinear equation, Stud. Appl.
Math. 57 (2) (1976/1977) 93–105.
[17] P.-L. Lions, The Choquard equation and related questions, Nonlinear Anal. 4 (6) (1980) 1063–1072.
[18] P.-L. Lions, The concentration-compactness principle in the calculus of variations. The locally compact case. I,
Ann. Inst. H. Poincaré Anal. Non Linéaire 1 (2) (1984) 109–145.
[19] P.-L. Lions, The concentration-compactness principle in the calculus of variations. The locally compact case. II,
Ann. Inst. H. Poincaré Anal. Non Linéaire 1 (4) (1984) 223–283.
[20] Li Ma, Lin Zhao, Classification of positive solitary solutions of the nonlinear Choquard equation, Arch. Ration.
Mech. Anal. 195 (2) (2010) 455–467.
[21] G.P. Menzala, On regular solutions of a nonlinear equation of Choquard’s type, Proc. Roy. Soc. Edinburgh Sect.
A 86 (3–4) (1980) 291–301.
[22] G.P. Menzala, On the nonexistence of solutions for an elliptic problem in unbounded domains, Funkcial. Ek-
vac. 26 (3) (1983) 231–235.
[23] I.M. Moroz, R. Penrose, P. Tod, Spherically-symmetric solutions of the Schrödinger–Newton equations, Classical
Quantum Gravity 15 (9) (1998) 2733–2742.
[24] V. Moroz, J. Van Schaftingen, Nonexistence and optimal decay of supersolutions to Choquard equations in exterior
domains, J. Differential Equations 254 (8) (2013) 3089–3145.
[25] V. Moroz, J. Van Schaftingen, Existence of groundstates for a class of nonlinear Choquard equations, arXiv:
1212.2027.
[26] S. Pekar, Untersuchung über die Elektronentheorie der Kristalle, Akademie Verlag, Berlin, 1954.
[27] M. Riesz, L’intégrale de Riemann–Liouville et le problème de Cauchy, Acta Math. 81 (1949) 1–223.
[28] P. Tod, I.M. Moroz, An analytical approach to the Schrödinger–Newton equations, Nonlinearity 12 (2) (1999) 201–
216.
[29] J. Van Schaftingen, Explicit approximation of the symmetric rearrangement by polarizations, Arch. Math.
(Basel) 93 (2) (2009) 181–190.
[30] J. Van Schaftingen, M. Willem, Set transformations, symmetrizations and isoperimetric inequalities, in: Nonlinear
Analysis and Applications to Physical Sciences, Springer, Milan, 2004, pp. 135–152.
[31] J. Van Schaftingen, M. Willem, Symmetry of solutions of semilinear elliptic problems, J. Eur. Math. Soc.
(JEMS) 10 (2) (2008) 439–456.
[32] Juncheng Wei, M. Winter, Strongly interacting bumps for the Schrödinger–Newton equations, J. Math. Phys. 50 (1)
(2009) 012905, 22 pp.
[33] M.I. Weinstein, Nonlinear Schrödinger equations and sharp interpolation estimates, Comm. Math. Phys. 87 (4)
(1982/1983) 567–576.
[34] M. Willem, Minimax Theorems, Progr. Nonlinear Differential Equations Appl., vol. 24, Birkhäuser, Boston, MA,
1996.
[35] M. Willem, Functional Analysis: Fundamentals and Applications, Cornerstones, vol. XIV, Birkhäuser, Basel, 2013.