THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION:

EXISTENCE, DECAY AND CONCENTRATION RESULTS

VINCENZO AMBROSIO

Abstract. In this paper we investigate the existence, decay and concentration behavior of solutions
for the following class of fractional relativistic Schrödinger equations:
(−∆ + m2 )s u + V (ε x)u = f (u) in RN ,


u ∈ H s (RN ), u > 0 in RN ,
arXiv:2003.06155v1 [math.AP] 13 Mar 2020

where ε > 0 is a small parameter, s ∈ (0, 1), m > 0, N > 2s, (−∆ + m2 )s is the fractional relativistic
Schrödinger operator, V : RN → R is a continuous potential satisfying a local condition, and f : R → R
is a continuous subcritical nonlinearity. The approach is based on a variant of the extension method
and applying appropriate variational techniques.

1. Introduction
In this paper we consider the following class of nonlinear fractional elliptic problems:
(−∆ + m2 )s u + V (ε x)u = f (u) in RN ,

(1.1)
u ∈ H s (RN ), u > 0 in RN ,
where ε > 0 is a small parameter, s ∈ (0, 1), N > 2s, m > 0, V : RN → R is a continuous potential and
f : R → R is a continuous nonlinearity. The nonlocal operator (−∆ + m2 )s appearing in (1.1) is defined
via Fourier transform by
N
(−∆ + m2 )s u := (2π)− 2 F −1 ((|k|2 + m2 )s F u)
for any u : RN → R belonging to the Schwartz space S(RN ) of rapidly decaying functions, or equivalently
(see [28, 40])
u(x) − u(y)
Z
N +2s
2 s 2s
(−∆ + m ) u(x) := m u(x) + C(N, s)m P.V. N +2s K N +2s (m|x − y|) dy, (1.2)
2

RN |x − y| 2
2

where P.V. stands for the Cauchy principal value, Kν is the modified Bessel function of the third kind (or
Macdonald function) of index ν (see [9, 26]) which satisfies the following well-known asymptotic formulas
for ν ∈ R and r > 0:
Γ(ν)  r −ν
Kν (r) ∼ as r → 0, for ν > 0, (1.3)
r2 2
π − 1 −r
Kν (r) ∼ r 2 e as r → ∞, for ν ∈ R, (1.4)
2
and C(N, s) is a positive constant whose exact value is given by
N +2s N s(1 − s)
C(N, s) := 2− 2 +1
π − 2 22s .
Γ(2 − s)
Equations involving (−∆+m2 )s arise in the study of standing waves ψ(x, t) for Schrödinger-Klein-Gordon
equations of the form
∂ψ
ı = (−∆ + m2 )s ψ − f (x, ψ), in RN × R,
∂t
√
which describe the behavior of bosons. In particular, when s = 1/2, the operator −∆ + m2 − m plays
an important role in relativistic quantum mechanics because it corresponds to the kinetic energy of a
relativistic particle with p
mass m > 0. If p is the momentum of the particle then its relativistic kinetic
energy is given by E = p2 + m2 . In the process of quantization the momentum p is replaced by the
differential operator −ı∇ and the quantum analog of the relativistic kinetic energy is the free relativistic

2010 Mathematics Subject Classification. 47G20, 35R11, 35J10, 35J20, 35J60, 35B09.
Key words and phrases. Fractional Schrödinger operator; extension method; variational methods.
1
2 V. AMBROSIO

√
Hamiltonian −∆ + m2 − m. Physical models related to this operator have been widely studied over the
past 30 years and there exists a huge literature on the spectral properties of relativistic Hamiltonians,
most of it has been strongly influenced by the works of Lieb on the stability of relativistic matter;
see [25, 35, 41, 42] for more physical background. On the other hand, there is also a deep connection
between (−∆ + m2 )s and the theory of Lévy processes. Indeed, m2s − (−∆ + m2 )s is the infinitesimal
generator of a Lévy process Xt2s,m called 2s-stable relativistic process having the following characteristic
function 2s,m 2 2 s 2s
E 0 eık·Xt = e−t[(|k| +m ) −m ] , k ∈ RN ;
we refer to [13, 16, 45] for a more detailed discussion on relativistic stable processes. When m = 0, the
previous operator boils down to the fractional Laplacian operator (−∆)s which has been extensively
studied in these last years due to its great applications in several fields of the research; see [11, 23, 43]
for an introduction on this topic. In particular, a great interest has been devoted to the existence and
multiplicity of solutions for fractional Schrödinger equations [39] like
ε2s (−∆)s u + V (x)u = f (u) in RN , (1.5)
and the asymptotic behavior as ε → 0; see for instance [2, 5, 6, 21, 29, 32] and the references therein.
When m > 0 and ε = 1 in (1.1), some interesting existence, multiplicity, and qualitative results of
solutions for (1.1) can be found in [4, 12, 17, 19, 20, 31, 36, 46], while only
√ one result [18] treats with the
semiclassical analysis ε → 0 of a fractional Hartree equation involving − ε2 ∆ + m2 .
Motivated by the above papers, in this work we focus our attention on the concentration phenomenon
of solutions to (1.1) as ε → 0. Along this paper, we suppose that V : RN → R is a continuous function
which satisfies the following conditions due to del Pino and Felmer [22]:
(V1 ) there exists V1 ∈ (0, m2s ) such that −V1 := inf x∈RN V (x),
(V2 ) there exists a bounded open set Λ ⊂ RN such that
−V0 := inf V (x) < min V (x),
x∈Λ x∈∂Λ

with V0 > 0. We also set M := {x ∈ Λ : V (x) = −V0 }. Without loss of generality, we may assume
that 0 ∈ M.
Concerning the nonlinearity f , we assume that f : R → R is continuous, f (t) = 0 for t ≤ 0, and f fulfills
the following hypotheses:
(f1 ) limt→0 f (t)
t = 0,
(f2 ) lim supt→∞ ft(t)
p < ∞ for some p ∈ (1, 2∗s − 1), where 2∗s := N2N −2s is the fractional critical exponent,
(f3 ) there exists θ ∈ (2, 2∗s ) such that 0 < θF (t) ≤ tf (t) for all t > 0,
(f4 ) f (t)
t is increasing for t > 0.
The main result of this work can be stated as follows:
Theorem 1.1. Assume that (V1 )-(V2 ) and (f1 )-(f4 ) are satisfied. Then, for every small ε > 0, there
exists a solution uε to (1.1) such that uε has a maximum point satisfying
lim dist(ε xε , M) = 0,
ε→0
and for which
0 < uε (x) ≤ Ce−c|x−xε | ∀x ∈ RN ,
for suitable constants C, c > 0. Moreover, for any sequence (εn ) with εn → 0, there exists a subsequence,
still denoted by itself, such that there exist a point x0 ∈ M with εn yεn → x0 , and a positive least energy
solution u ∈ H s (RN ) of the limiting problem
(−∆ + m2 )s u − V0 u = f (u) in RN ,
for which we have
uεn (x) = u(x − yεn ) + Rn (x)
where limn→∞ kRn kH s (RN ) = 0.
The proof of Theorem 1.1 is obtained through suitable variational techniques. Firstly, we start by
observing that (−∆ + m2 )s is a nonlocal operator and that does not scale like the fractional Laplacian
operator (−∆)s . More precisely, the first operator is not compatible with the semigroup R+ acting on
functions as t ∗ u 7→ u(t−1 x) for t > 0. This fact does not permit to adapt in a simple way the same
arguments performed to deal with (−∆)s . Nevertheless, we overcome these difficulties by using a variant
of the extension method [14] for (−∆ + m2 )s (see [19,28,49]) which permits to study via local variational
methods a degenerate elliptic equation in a half-space with a nonlinear Neumann boundary condition.
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 3

Clearly, some additional difficulties arise in the investigation of this problem because we have to handle
the trace terms of the involved functions and thus a more careful analysis will be needed.
Due to the lack of informations on the behavior of V at infinity, we carry out a penalization argument
[22] which consists in modifying appropriately the nonlinearity f outside Λ, and thus consider a modified
problem whose corresponding energy functional fulfills all the assumptions of the mountain pass theorem
[3]. Then we need to check that, for ε > 0 small enough, the solutions of the auxiliary problem are indeed
solutions of the original one. This goal will be achieved by combing an appropriate Moser iteration
argument [44] with some elliptic regularity estimates established in [28]. To our knowledge this is the
first time that the penalization trick is used to study the concentration phenomena for the fractional
relativistic Schrödinger operator (−∆ + m2 )s for all s ∈ (0, 1) and m > 0. When m = 0, namely when
(1.1) reduces to (1.5) after rescaling, we refer the interested reader to [2, 5, 6] for similar approaches.
Finally, we show that the solutions of (1.1) have an exponential decay, contrary to the case m = 0 for
which the solutions of (1.5) satisfy the power-type decay |x|−(N +2s) as |x| → ∞; see [6, 29, 30]. To
investigate the decay of solutions to (1.1), we construct a suitable comparison function and we carry out
some refined estimates which take care of an adequate estimate concerning 2s-stable relativistic density
with parameter m found in [34], and that the modified Bessel function Kν has an exponential decay
at infinity. We stress that exponential type estimates for equations like (1.1), appear in [18, 19] where
s = 21 , V is bounded, f is a Hartree type nonlinearity, and in [31] where s ∈ (0, 1), m = 1, V ≡ 0
and |f (u)| ≤ Cup for some p ∈ (1, 2∗s − 1). Anyway, our approach to obtain the decay estimate is
completely different from the above mentioned papers, it is more general and we believe that can be
applied in other situations to deal with fractional problems driven by (−∆ + m2 )s . We conclude this
introduction by pointing out that in view of the techniques developed here and the recent result in [7],
we are preparing a work [8] in which we obtain a multiplicity result for (1.1) when f is a Beresticky-Lions
type nonlinearity [10].
The paper is organized as follows. In section 2 we collect some notations and preliminary results which
will be used along the paper. In section 3 we introduce a penalty functional in order to apply suitable
variational arguments. The section 4 is devoted to the proof of Theorem 1.1. Finally, we give some
interesting results for (−∆ + m2 )s in the appendix.

2. preliminaries
2.1. Notations and functional setting. We denote the upper half-space in RN +1 by
RN+
+1
:= {(x, y) ∈ RN +1 : y > 0},
and for (x, y) ∈ RN +1
p
+ we consider the Euclidean norm |(x, y)| := |x|2 + y 2 .
Let p ∈ [1, ∞] and A ⊂ RN be a measurable set. We indicate by Lp (A) the set of measurable functions
u : RN → R such that ( R
1/p
|u|Lp (A) := A
|u|p dx < ∞ if p < ∞,
esssupx∈A |u(x)| if p = ∞.
When A = RN , we simply write |u|p instead of |u|Lp (RN ) . With kwkLp (RN +1 ) we will always denote the
+

norm of w ∈ Lp (RN +
+1
).
Let D ⊂ RN +1 be a bounded domain, that is a bounded connected open set, with boundary ∂D, we
denote by ∂ ′ D the interior of D ∩ ∂RN
+
+1
in RN , and we set ∂ ′′ D := ∂D \ ∂ ′ D. For R > 0, we put
+
BR := {(x, y) ∈ RN
+
+1
: |(x, y)| < R},
Γ0R := ∂ ′ BR
+
= {(x, 0) ∈ ∂RN
+
+1
: |x| < R},
′′ +
Γ+
R := ∂ BR = {(x, y) ∈ R
N +1
: y ≥ 0, |(x, y)| = R}.
Now, we introduce the Lebesgue spaces with weight (see [27, 37] for more details). Let D ⊂ RN +
+1
be an
r 1−2s
open set and r ∈ (1, ∞). Denote by L (D, y ) the weighted Lebesgue space of all measurable functions
v : D → R such that
Z Z  r1
1−2s r
kvkLr (D,y1−2s ) := y |v| dxdy < ∞.
D
We say that v ∈ H 1 (D, y 1−2s ) if v ∈ L2 (D, y 1−2s ) and its weak derivatives, collectively denoted by ∇v,
exist and belong to L2 (D, y 1−2s ). The norm of v in H 1 (D, y 1−2s ) is given by
ZZ
kvk2H 1 (D,y1−2s ) := y 1−2s (|∇v|2 + v 2 ) dxdy < ∞.
D
4 V. AMBROSIO

It is clear that H 1 (D, y 1−2s ) is a Hilbert space with the inner product
ZZ
y 1−2s (∇v∇w + vw) dxdy.
D

Let H (R ) be the fractional Sobolev space defined as the completion of Cc∞ (RN ) with respect to the
s N

norm
Z  12
2 2 s 2
|u|H s (RN ) := (|k| + m ) |F u(k)| dk .
RN
Then, H (R ) is continuously embedded in L (RN ) for all p ∈ [2, 2∗s ) and compactly in Lploc (RN ) for
s N p

all p ∈ [1, 2∗s ); see [1, 9, 23, 40]. Next we define X s (RN
+
+1
) := H 1 (RN
+
+1 1−2s
,y ) as the completion of
Cc∞ (RN
+
+1
) with respect to the norm
ZZ ! 21
kukX s(RN +1 ) := y 1−2s (|∇u|2 + m2 u2 ) dxdy .
+
RN
+
+1

By Lemma 3.1 in [28], we deduce that X s (RN

+
+1
) is continuously embedded in L2γ (RN
+
+1 1−2s
,y ), that is
kukL2γ (RN +1 ,y1−2s ) ≤ C∗ kukX s(RN +1 ) ∀u ∈ X s (RN
+
+1
), (2.1)
+ +

where γ := 1 + 2
N −2s , and Lr (RN
+
+1 1−2s
,y ) is the weighted Lebesgue space, with r ∈ (1, ∞), endowed
with the norm ZZ
kukLr (RN +1 ,y1−2s ) := y 1−2s |u|r dxdy.
+
RN
+
+1

Moreover, by Lemma 3.1.2 in [24], we also have that X s (RN +

+1 + 1−2s
) is compactly embedded in L2 (BR ,y )
for all R > 0. From Proposition 5 in [28], we know that there exists a linear trace operator Tr :
X s (RN
+
+1
) → H s (RN ) such that
√
σs |Tr(u)|H s (RN ) ≤ kukX s(RN +1 ) for any u ∈ X s (RN+
+1
), (2.2)
+

1−2s
where σs := 2 Γ(1 − s)/Γ(s). We also note the (2.2) and the definition of H s -norm imply that
Z ZZ
σs m2s |Tr(u)|2 dx ≤ y 1−2s (|∇u|2 + m2 u2 ) dxdy. (2.3)
RN RN
+
+1

In what follows, in order to simplify the notation, we denote Tr(u) by u(·, 0).
Since T r(X s (RN
+
+1
)) ⊂ H s (RN ) and H s (RN ) ⊂ Lq (RN ) for all q ∈ [2, 2∗s ) and the embedding is locally
compact for all q ∈ [1, 2∗s ) (see [1, 9, 23]), we obtain the following result:
Theorem 2.1. T r(X s (RN +
+1
)) is continuously embedded in Lq (RN ) for all q ∈ [2, 2∗s ) and compactly
q
embedded in Lloc (R ) for all q ∈ [1, 2∗s ).
N

In order to circumvent the nonlocal character of the pseudo-differential operator (−∆ + m2 )s , we make
use of a variant of the extension method [14] given in [28, 49]. More precisely, for any u ∈ H s (RN ) there
exists a unique function U ∈ X s (RN +
+1
) solving the following problem
− div(y 1−2s ∇U ) + m2 y 1−2s U = 0 in RN +1

+ ,
+1 ∼ N
U (·, 0) = u on ∂RN + =R .
The function U is called the s-extension of u and possesses the following properties:
(i)
∂U ∂U
:= − lim y 1−2s (x, y) = σs (−∆ + m2 )s u(x) in distribution sense,
∂ν 1−2s y→0 ∂y
√
(ii) σs |u|H s (RN ) = kU kX s(RN +1 ) ≤ kV kX s (RN +1 ) for all V ∈ X s (RN
+
+1
) such that V (·, 0) = u.
+ +

(iii) U ∈ C ∞ (RN
+
+1
) and can be expressed as
Z
U (x, y) = Ps,m (x − z, y)u(z) dz
RN
with
N +2s N +2s
Ps,m (x, y) := c′N,s y 2s m 2 |(x, y)|− 2 K N +2s (m|(x, y)|),
2
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 5

and N +2s
2 2 −1
c′N,s := pN,s ,
Γ( N +2s
2 )
where pN,s is the constant for the (normalized) Poissonp kernel with m = 0; see [49].
We note that Ps,m is the Fourier transform of k 7→ ϑ( |k|2 + m2 ) and that
Z
Ps,m (x, y) dx = ϑ(my), (2.4)
RN
where ϑ ∈ H 1 (R+ , y 1−2s ) solves the following ordinary differential equation
 ′′ 1−2s ′
ϑ + y ϑ − ϑ = 0 in R+ ,
ϑ(0) = 1.
We also recall that ϑ can be expressed via modified Bessel functions, more precisely ϑ(r) =
2 r s
Γ(s) ( 2 ) Ks (r); see [28] for more details.
Taking into account the previous facts, problem (1.1) can be realized in a local manner through the
following nonlinear boundary value problem:
− div(y 1−2s ∇w) + m2 y 1−2s w = 0 in RN +1

+ ,
∂w N (2.5)
∂ν 1−2s = σs [−Vε (x)w(·, 0) + f (w(·, 0))] in R ,
where Vε (x) := V (ε x). For simplicity of notation, we will omit the constant σs from the second equation
in (2.5). For all ε > 0, we define
 Z 
Xε := u ∈ X s (RN +
+1
) : V ε (x)u 2
(x, 0) dx < ∞
RN
endowed with the norm
 Z  12
2 2
kukε := kukX s (RN +1 ) + Vε (x)u (x, 0) dx .
+
RN
We note that k · kε is actually a norm. Indeed,
 Z  Z
2 2 2
kukε = kukX s(RN +1 ) − V1 u (x, 0) dx + (Vε (x) + V1 )u2 (x, 0) dx,
+
RN RN
and using (2.3) and (V1 ) we can see that
   
V1
Z
1 − 2s kuk2X s (RN +1 ) ≤ kuk2X s(RN +1 ) − V1 u2 (x, 0) dx ≤ kuk2X s(RN +1 )
m + +
RN +

that is Z
kuk2X s (RN +1 ) − V1 u2 (x, 0) dx
+
RN
is a norm equivalent to k·k2X s (RN +1 ) . This observation yields the required claim. Clearly, Xε ⊂ X s (RN
+
+1
),
+
and using (V1 ) we have
m2s
 
kuk2X s (RN +1 ) ≤ kuk2ε ∀u ∈ Xε . (2.6)
+ m2s − V1
2.2. Elliptic estimates. For reader’s convenience, we list some results about local Schauder estimates
for degenerate elliptic equations involving the operator
−div(y 1−2s ∇v) + m2 y 1−2s v.
Firstly we give the following definition:
2N
Definition 2.1. Let D ⊂ RN
+
+1
be a bounded domain with ∂ ′ D 6= ∅. Let f ∈ Lloc
N +2s
(∂ ′ D) and g ∈
1 ′
Lloc (∂ D). Consider
−div(y 1−2s ∇v) + m2 y 1−2s v = 0 in D,

∂v (2.7)
∂ν a = f (x)v + g(x) on ∂ ′ D.
We say that v ∈ H 1 (D, y 1−2s ) is a weak supersolution (resp. subsolution) to (2.7) in D if for any
nonnegative ϕ ∈ Cc1 (D ∪ ∂ ′ D),
ZZ Z
y 1−2s (∇v∇φ + m2 vφ) dxdy ≥ (≤) [f (x)v(·, 0) + g(x)]ϕ(·, 0) dx.
D ∂′D
6 V. AMBROSIO

We say that v ∈ H 1 (D, y 1−2s ) is a weak solution to (2.7) in D if it is both a weak supersolution and a
weak subsolution.
We denote by QR = BR × (0, R) where BR ⊂ RN is the ball in RN centered at 0 and with radius
R > 0. Then we recall the following version of De Giorgi-Nash-Moser type theorems established in [28]
(see also [27, 37] for the case m = 0).
N
Proposition 2.1. [28] Let f, g ∈ Lq (B1 ) for some q > 2s .
1 1−2s
(i) Let v ∈ H (Q1 , y ) be a weak subsolution to (2.7) in Q1 . Then
sup v + ≤ C(kv + kL2 (Q1 ,y1−2s ) + |g + |Lq (B1 ) ),
Q1/2

where v + := max{v, 0}, and C > 0 depends only on N , s, q, |f + |Lq (B1 ) .

(ii) (weak Harnack inequality) Let v ∈ H 1 (Q1 , y 1−2s ) be a nonnegative weak supersolution to (2.7) in
Q1 . Then for some p0 > 0 and any 0 < µ < τ < 1 we have
inf v + |g− |Lq (B1 ) ≥ CkvkLp0 (Qτ ,y1−2s ) ,
Q̄µ

where v− := max{−v, 0}, and C > 0 depends only on N , s, q, µ, τ , |f− |Lq (B1 ) .
(iii) Let v ∈ H 1 (Q1 , y 1−2s ) be a nonnegative weak solution to (2.7) in Q1 . Then v ∈ C 0,α (Q1/2 ) and in
addition
kvkC 0,α (Q1/2 ) ≤ C(kvkL2 (Q1 ) + |g|Lq (B1 ) ),
with C > 0, α ∈ (0, 1) depending only on N , s, p, |f |Lq (B1 ) .

3. The penalization argument

In order to find solutions to (2.5), we follow the penalization approach in [22] which permits to study
our problem via variational arguments. Fix κ > m2sV−V 1
1
and a > 0 such that f (a) V1
a = κ . Define

f (t) for t < a,
f˜(t) := V1
κ t for t ≥ a.
Let us consider the following Carathéodory function
g(x, t) := χΛ (x)f (t) + (1 − χΛ (x))f˜(t) for (x, t) ∈ RN × R,
Rt
where χΛ denotes the characteristic function of Λ. Set G(x, t) := 0 g(x, τ ) dτ . By (f1 )-(f4 ) it follows
that
(g1 ) limt→0 g(x,t)
t = 0 uniformly in x ∈ RN ,
(g2 ) g(x, t) ≤ f (t) for all x ∈ RN , t > 0,
(g3 ) (i) 0 < θG(x, t) ≤ tg(x, t) for all x ∈ Λ and t > 0,
(ii) 0 ≤ 2G(x, t) ≤ tg(x, t) ≤ Vκ1 t2 for all x ∈ Λc = RN \ Λ and t > 0,
(g4 ) for each x ∈ Λ the function t 7→ g(x,t) t is increasing in (0, ∞), and for each x ∈ Λc the function
g(x,t)
t 7→ t is increasing in (0, a).
Consider the following modified problem:
− div(y 1−2s ∇u) + m2 y 1−2s u = 0 in RN +1

+ ,
∂u N (3.1)
∂ν 1−2s = −V ε (x)u(·, 0) + g ε (·, u(·, 0)) in R ,
where we set gε (x, t) := g(ε x, t). Obviously, if uε is a positive solution of (3.1) satisfying uε (x, 0) < a for
all x ∈ Λcε , then uε is indeed a solution of (2.5). The corresponding energy functional is defined as
1
Z
2
Jε (u) := kukε − Gε (x, u(x, 0)) dx.
2 RN
Clearly, Jε ∈ C 1 (Xε , R) and its differential is given by:
ZZ Z
′ 1−2s 2
hJε (u), vi = y (∇u∇v + m uv) dxdy + Vε (x)u(x, 0)v(x, 0) dx
RN
+
+1
RN
Z
− gε (x, u(x, 0))v(x, 0) dx
RN
for all u, v ∈ Xε .
We start by proving that Jε satisfies all the assumptions of the mountain pass theorem [3].
Lemma 3.1. Jε has a mountain pass geometry, that is:
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 7

• there exist α, ρ > 0 such that Jε (u) ≥ α for all u ∈ X such that kuk = ρ,
• there exists v ∈ X such that kvk > ρ and Jε (v) < 0.

Proof. By (f1 ), (f2 ), (g1 ), (g2 ), we deduce that for all η > 0 there exists Cη > 0 such that
∗
|gε (x, t)| ≤ η|t| + Cη |t|2s −1 for all x ∈ RN , t > 0, (3.2)

and
η 2 Cη 2∗s
|Gε (x, t)| ≤ |t| + ∗ |t| for all x ∈ RN , t > 0. (3.3)
2 2s
Fix η ∈ (0, m2s − V1 ). Using (3.3), (2.3), (2.6) and Theorem 2.1 we have

1 η Cη 2∗
Jε (u) ≥ kuk2ε − |u(·, 0)|22 − ∗ |u(·, 0)|2s∗s
2 2 2s
1 η Cη 2∗
= kuk2ε − 2s
m2s |u(·, 0)|22 − ∗ |u(·, 0)|2s∗s
2 2m 2s
1 η 2∗
≥ kuk2ε − 2s
kuk2X s (RN +1 ) − Cη CkukXss (RN +1 )
2 2m + +
 
1 η 2 2∗
≥ − kukε − Cη Ckukεs
2 2(m2s − V1 )

from which we deduce the thesis. Now, let v ∈ Cc∞ (RN +

+1
) such that supp(v(·, 0)) ⊂ Λε . Then, by (f3 ),
we have
t2
Z
Jε (tv) = kvk2ε − F (tv(x, 0)) dx
2 RN
t2
Z
≤ kvk2ε − tθ |v(x, 0)|θ dx + C|Λε | → −∞ as t → ∞.
2 Λε

By Lemma 3.1 and using a variant of mountain pass lemma without Palais-Smale condition [50], we
can find a Palais-Smale sequence (un ) ⊂ Xε such that

Jε (un ) → cε Jε′ (un ) → 0, (3.4)

where
cε := inf max Jε (γ(t))
γ∈Γε t∈[0,1]

and
Γε := {γ ∈ C([0, 1], Xε ) : γ(0) = 0, Jε (γ(1)) ≤ 0}.

In view of (f4 ), it is standard to check (see [50]) that

cε = inf Jε (u) = inf sup Jε (tu),

u∈Nε u∈Xε \{0} t≥0

where
Nε := {u ∈ Xε : hJε′ (u), ui = 0}

is the Nehari manifold associated with Jε . Now we prove the following fundamental compactness result:

Lemma 3.2. Jε satisfies the Palais-Smale condition at any level c ∈ R.

Proof. Let c ∈ R and (un ) ⊂ Xε be such that

Jε (un ) → c Jε′ (un ) → 0. (3.5)

8 V. AMBROSIO

Then, using (g3 ), (2.3) and (2.6), we have

1
Jε (un ) − hJε′ (un ), un i
θ
 
1 1 1
Z
= − kun k2ε + gε (x, un (x, 0))un (x, 0) − θGε (x, un (x, 0)) dx
2 θ θ RN
   
1 1 1 1 V1
Z
≥ − kun k2ε − − u2 (x, 0) dx
2 θ 2 θ κ RN n
   
1 1 1 1 V1
Z
= − kun k2ε − − m 2s
u2n (x, 0) dx
2 θ 2 θ κm2s RN
   
1 1 1 1 V1
≥ − kun k2ε − − kun k2X s (RN +1 )
2 θ 2 θ κm2s +
  
1 1 V1
≥ − 1− kun k2ε .
2 θ κ(m2s − V1 )
Since θ > 2 and κ > m2sV−V 1
1
, we deduce that (un ) is bounded in Xε . Hence, up to a subsequence, we
may assume that un ⇀ u in Xε . Now we prove that this convergence is indeed strong. Using the fact
that g has subcritical growth and applying Theorem 2.1, it is easy to check that hJε′ (u), ϕi = 0 for all
ϕ ∈ Xε . In particular,
Z Z
2 2 2
kukX s(RN +1 ) − V1 |u(·, 0)|2 + (Vε (x) + V1 )u (x, 0) dx + [(Vε (x) + V1 )u2 (x, 0) − gε (x, u(x, 0))u(x, 0)] dx
+
Λε Λcε
Z
= f (u(x, 0))u(x, 0) dx. (3.6)
Λε

On the other hand, by hJε′ (un ), un i = on (1), we get

Z Z
2 2 2
kun kX s (RN +1 ) − V1 |un (·, 0)|2 + (Vε (x) + V1 )un (x, 0) dx + [(Vε (x) + V1 )u2n (x, 0) − gε (x, un (x, 0))un (x, 0)] dx
+
Λε Λcε
Z
= f (un (x, 0))un (x, 0) dx + on (1). (3.7)
Λε

Since Λε is bounded, by the compactness of Sobolev embeddings in Theorem 2.1 we have

Z Z
lim f (un (x, 0))un (x, 0) dx = f (u(x, 0))u(x, 0) dx, (3.8)
n→∞ Λε Λε
Z Z
lim (Vε (x) + V1 )u2n (x, 0) dx = (Vε (x) + V1 )u2 (x, 0) dx. (3.9)
n→∞ Λε Λε

In view of (3.6), (3.7), (3.8), (3.9), we obtain

Z !
lim sup kun k2X s (RN +1 ) − V1 |un (·, 0)|22 + [(Vε (x) + V1 )u2n (x, 0) − gε (x, un (x, 0))un (x, 0)] dx
n→∞ +
Λcε
Z
= kuk2X s(RN +1 ) − V1 |u(·, 0)|22 + [(Vε (x) + V1 )u2 (x, 0) − gε (x, u(x, 0))u(x, 0)] dx.
+
Λcε

On the other hand, by (g2 ) and Fatou’s Lemma, we get

Z !
2 2 2
lim inf kun kX s (RN +1 ) − V1 |un (·, 0)|2 + [(Vε (x) + V1 )un (x, 0) − gε (x, un (x, 0))un (x, 0)] dx
n→∞ +
Λcε
Z
≥ kuk2X s(RN +1 ) − V1 |u(·, 0)|22 + [(Vε (x) + V1 )u2 (x, 0) − gε (x, u(x, 0))u(x, 0)] dx.
+
Λcε

Hence,
lim kun k2X s (RN +1 ) − V1 |un (·, 0)|22 = kuk2X s (RN +1 ) − V1 |u(·, 0)|22 (3.10)
n→∞ + +

and
Z Z
lim [(Vε (x) + V1 )u2n (x, 0) − gε (x, un (x, 0))un (x, 0)] dx = [(Vε (x) + V1 )u2 (x, 0) − gε (x, u(x, 0))u(x, 0)] dx.
n→∞ Λcε Λcε
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 9

The last limit combined with definition of gε yields

Z Z
2
lim (Vε (x) + V1 )un (x, 0) dx = (Vε (x) + V1 )u2 (x, 0) dx
n→∞ Λcε Λε

which together with (3.9) implies that

Z Z
lim (Vε (x) + V1 )u2n (x, 0) dx = (Vε (x) + V1 )u2 (x, 0) dx. (3.11)
n→∞ RN RN
From (3.10) and (3.11) we have
lim kun k2ε = kuk2ε
n→∞
and since Xε turns out to be a Hilbert space we deduce that un → u in Xε . 
In light of Lemma 3.1 and Lemma 3.2, it follows from the mountain pass theorem [3] that for all ε > 0
there exists uε ∈ Xε \ {0} such that
Jε (uε ) = cε and Jε′ (uε ) = 0. (3.12)
N +1
Since f (t) = 0 for t ≤ 0, it follows from hJε′ (uε ), u− −
ε i = 0, where uε := min{uε , 0}, that uε ≥ 0 in R+ .
2s
Next, for all µ > −m , we consider the following family of autonomous problems related to (2.5),
namely
− div(y 1−2s ∇w) + m2 y 1−2s w = 0 in RN +1

+ ,
∂w N (3.13)
∂ν 1−2s = −µw(·, 0) + f (w(·, 0)) in R .
Define the energy functionals Lµ : X s (RN
+
+1
) → R given by
1 µ
Z
Lµ (u) := kuk2X s (RN +1 ) + |u(·, 0)|22 − F (u(x, 0)) dx,
2 + 2 RN

and denote by Mµ the Nehari manifold associated with Lµ , that is Mµ := {u ∈ X s (RN +

+1
) : hL′µ (u), ui =
0}. Arguing as in the case ε > 0, one can check that Lµ has a mountain pass geometry [3], so we can
find a Palais-Smale sequence (un ) ⊂ X s (RN
+
+1
) at the mountain pass level dµ of Lµ . As before, (un ) is
s N +1
bounded in X (R+ ). We also note that, by (f4 ), it holds
dµ = inf Lµ (u) = inf sup Lµ (tu).
u∈Mµ u∈X s (RN +1
)\{0} t≥0
+

Next we show the existence of a ground state solution to (3.13). We first prove some useful technical
lemmas. The first one is a vanishing Lions type result.
Lemma 3.3. Let t ∈ [2, 2∗s ) and R > 0. If (un ) ⊂ X s (RN
+
+1
) is a bounded sequence such that
Z
lim sup |un (x, 0)|t dx = 0,
n→∞ z∈RN BR (z)
q N
then un (·, 0) → 0 in L (R ) for all q ∈ (t, 2∗s ).
Proof. Take q ∈ (t, 2∗s ). Given R > 0 and z ∈ RN , by using Hölder inequality, we get
|un (·, 0)|Lq (BR (z)) ≤ |un (·, 0)|1−λ λ
Lt (BR (z)) |un (·, 0)|L2s (BR (z))
∗ ∀n ∈ N,
where
1−λ λ 1
+ ∗ = .
t 2s q
Now, covering RN by balls of radius R, in such a way that each point of RN is contained in at most N + 1
balls, we find
(1−λ)q
|un (·, 0)|qq ≤ (N + 1)|un (·, 0)|Lt (BR (z)) |un (·, 0)|λq
2∗
s
,
which combined with Theorem 2.1 and the assumptions yields
(1−λ)q
|un (·, 0)|qq ≤ C(N + 1)|un (·, 0)|Lt (BR (z)) kun kλq
X s (RN +1 ) +
(1−λ)q
≤ C(N + 1) sup |un (·, 0)|Lt (BR (z)) → 0 as n → ∞.
z∈RN

This ends the proof of lemma. 

10 V. AMBROSIO

Lemma 3.4. Let (un ) ⊂ X s (RN +

+1
) be a Palais-Smale sequence at level c ∈ R and such that un ⇀ 0 in
s N +1
X (R+ ). Then we have either
(a) un → 0 in X s (RN +
+1
), or
(b) there exists a sequence (zn ) ⊂ RN and constants R, β > 0 such that
Z
lim inf |un (x, 0)|2 dx ≥ β.
n→∞ BR (zn )

Proof. Assume that (b) does not occur. Then, for all R > 0, we have
Z
lim sup |un (x, 0)|2 dx = 0.
n→∞ z∈RN BR (z)

Using Lemma 3.3 (with t = 2), we can see that un (·, 0) → 0 in Lq (RN ) for all q ∈ (2, 2∗s ). This fact and
(f1 )-(f2 ) imply that Z
f (un (x, 0))un (x, 0) dx → 0 as n → ∞.
RN
Hence, using hL′µ (un ), un i = on (1), µ > −m2s and (2.3), we get
 µ 
1 + 2s kun k2X s (RN +1 ) ≤ kun k2X s (RN +1 ) + µ|un (·, 0)|22 ≤ on (1)
m + +

s N +1
that is un → 0 in X (R+ ) as n → ∞. 
Now we prove the following existence result for (3.13).
Theorem 3.1. Let µ > −m2s . Then (3.13) has a positive ground state solution.
Proof. Since Lµ has a mountain pass geometry [3], we can find a Palais-Smale sequence (un ) ⊂ X s (RN +
+1
)
s N +1 s N +1
at the level dµ . Thus (un ) is bounded in X (R+ ) and there exists u ∈ X (R+ ) such that un ⇀ u in
X s (RN
+
+1
). Clearly, L′µ (u) = 0. If u = 0, then we can use Lemma 3.4 to deduce that for some sequence
(zn ) ⊂ RN , vn (x, y) := un (x + zn , y) is a bounded Palais-Smale sequence at the level dµ and having a
nontrivial weak limit v. Hence, v ∈ Mµ , where Mµ denotes the Nehari manifold associated with Lµ .
Moreover, using the weak lower semicontinuity of k · kX s (RN +1 ) + µ| · |2 , (f4 ) and Fatou’s lemma, it is
+
easy to see that Lµ (v) = dµ . When u 6= 0, as before, we can deduce that u is a ground state solution to
(3.13). In conclusion, for any µ > −m2s , there exists a ground state solution w = wµ ∈ X s (RN +
+1
) \ {0}
such that
Lµ (w) = dµ and L′µ (w) = 0.
Since f (t) = 0 for t ≤ 0, we deduce that w ≥ 0 in RN +
+1
and w 6≡ 0. A Moser iteration argument (see
Lemma 4.1 below) shows that w(·, 0) ∈ Lp (RN ) for all p ∈ [2, ∞] and w ∈ L∞ (RN
+
+1
). Using Proposition
2.1-(iii) we obtain that w ∈ C 0,α (RN
+
+1
) for some α ∈ (0, 1). By Proposition 2.1-(ii) we conclude that w
is positive. 
In the next lemma we establish an important connection between cε and dV (0) = d−V0 (we remark
that V (0) = −V0 > −m2s ):
Lemma 3.5. The numbers cε and dV (0) verify the following inequality:
lim sup cε ≤ dV (0) .
ε→0

Proof. By Theorem 3.1, we know that there exists a positive ground state solution w to (3.13) with
µ = V (0). Let η ∈ Cc∞ (R) be such that 0 ≤ η ≤ 1, η = 1 in [−1, 1] and η = 0 in R \ (−2, 2). Suppose
that B2 (0) ⊂ Λ. Define wε (x, y) = η(ε |(x, y)|)w(x, y) and note that supp(wε (·, 0)) ⊂ Λε . It is easy to
prove that wε → w in X s (RN+
+1
) and LV (0) (wε ) → LV (0) (w) as ε → 0. On the other hand, by definition
of cε , we have
t2
Z
cε ≤ max Jε (twε ) = Jε (tε wε ) = ε kwε k2ε − F (twε (x, 0)) dx (3.14)
t≥0 2 RN
for some tε > 0. Recalling that w ∈ MV (0) and using (f4 ), it is easy to check that tε → 1 as ε → 0. Note
that
t2
Z
Jε (tε wε ) = LV (0) (tε wε ) + ε (Vε (x) − V (0))wε2 (x, 0) dx. (3.15)
2 RN
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 11

Since Vε (x) is bounded on the support of wε (·, 0), and Vε (x) → V (0) as ε → 0, we can apply the dominated
convergence theorem and use (3.14) and (3.15) to conclude the proof. 

Now we come back to study (3.1) and consider the mountain pass solutions uε satisfying (3.12).

Lemma 3.6. There exist R, β, ε∗ > 0 and (yε ) ⊂ RN such that

Z
u2ε (x, 0) dx ≥ β, ∀ ε ∈ (0, ε∗ ).
Br (yε )

Proof. Since uε verifies (3.12), it follows from the growth assumptions on f that there exist α > 0
independent of ε > 0 such that
kuε k2ε ≥ α ∀ ε > 0. (3.16)
Let (εn ) ⊂ (0, ∞) be such that εn → 0. If by contradiction there exists r > 0 such that
Z
lim sup u2εn (x, 0) dx = 0,
n→∞ y∈RN Br (y)

we can use Lemma 3.3 to deduce that uεn (·, 0) → 0 in Lq (RN ) for all q ∈ (2, 2∗s ). Then, (3.12) and the
growth assumptions on f imply that kuεn kεn → 0 as n → ∞ which contradicts (3.16). 

Lemma 3.7. For any εn → 0, consider the sequence (yεn ) ⊂ RN given in Lemma 3.6 and wn (x, y) =
uεn (x + yεn , y). Then there exists a subsequence of wn , still denoted by itself, and w ∈ X s (RN
+
+1
) \ {0}
such that

wn → w in X s (RN
+
+1
).

Moreover, there exists x0 ∈ Λ such that

εn yεn → x0 and V (x0 ) = −V0 .

Proof. In what follows, we denote by (yn ) and (un ), the sequences (yεn ) and (uεn ) respectively. Since
each un satisfies (3.12), we can argue as in the proof of Lemma 3.2 to deduce that (un ) is bounded in
X s (RN
+
+1
). Thus (wn ) is bounded in X s (RN
+
+1
) and there exists w ∈ X s (RN
+
+1
) \ {0} such that

wn ⇀ w in X s (RN
+
+1
) as n → ∞. (3.17)

Moreover, by Lemma 3.6, we know that

Z
w2 (x, 0) dx ≥ β > 0. (3.18)
Br (0)

Next we show that (εn yn ) is bounded in R. First of all, we prove that

dist (εn yn , Λ) → 0 as n → ∞. (3.19)

If (3.19) does not hold, there exists δ > 0 and a subsequence of (εn yn ), still denoted by itself, such that

dist (εn yn , Λ) ≥ δ ∀n ∈ N.

Then there is R > 0 such that BR (εn yn ) ⊂ Λc for all n ∈ N. By the definition of X s (RN
+
+1
) and using
s N +1
the fact that w ≥ 0, we know that there exists (ψj ) ⊂ X (R+ ) such that ψj has compact support in
RN+
+1
and ψj → w in X s (RN +
+1
) as j → ∞. Fix j ∈ N. Taking ψj as test function in hJε′ (un ), φi = 0 for
all φ ∈ Xε , we get
ZZ Z
y 1−2s (∇wn ∇ψj + m2 wn ψj ) dxdy + V (εn x + εn yn )wn (x, 0)ψj (x, 0) dx
RN +1 RN
Z+
= g(εn x + εn yn , wn (x, 0))ψj (x, 0) dx. (3.20)
RN
12 V. AMBROSIO

By the definition of gε there holds

Z Z
g(εn x + εn yn , wn (x, 0))ψj (x, 0) dx ≤ g(εn x + εn yn , wn (x, 0))ψj (x, 0) dx
RN B R
εn
Z
+ g(εn x + εn yn , wn (x, 0))ψj (x, 0) dx
B cR
εn

V1
Z Z
≤ wn (x, 0)ψj (x, 0) dx + f (wn (x, 0))ψj (x, 0) dx.
κ B R B cR
εn εn

Using (V1 ) and (3.20) we can see that

 Z
1
ZZ
y 1−2s (∇wn ∇ψj + m2 wn ψj ) dxdy − V1 1 + wn (x, 0)ψj (x, 0) dx
RN
+
+1 κ RN
Z
≤ f (wn (x, 0))ψj (x, 0) dx.
B cR
εn

Taking into account that ψj has compact support, εn → 0, the growth assumptions on f , and (3.17), we
deduce that as n → ∞
Z
f (wn (x, 0))ψj (x, 0) dx → 0,
B cR
εn

and
Z 
1
ZZ
1−2s 2
y (∇wn ∇ψj + m wn ψj ) dxdy − V1 1 + wn (x, 0)ψj (x, 0) dx
RN
+
+1 κ RN
 Z
1
ZZ
→ y 1−2s (∇w∇ψj + m2 wψj ) dxdy − V1 1 + w(x, 0)ψj (x, 0) dx.
RN
+
+1 κ RN

The previous relations of limits combined with (3.20) give

 Z
1
ZZ
1−2s 2
y (∇w∇ψj + m wψj ) dxdy − V1 1 + w(x, 0)ψj (x, 0) dx ≤ 0
RN+
+1 κ RN

and passing to the limit as j → ∞ we find

 
1
kwk2X s (RN +1 ) − V1 1 + |w(·, 0)|22 dx ≤ 0.
+ κ
V1
Thus (2.3) and κ > m2s −V1 yield
  
V1 1
0 ≤ 1 − 2s 1 + kwk2X s (RN +1 ) ≤ 0
m κ +

that is w ≡ 0 in RN and this is in contrast with (3.18). Consequently, there exist a subsequence of
(εn yn ), still denoted by itself, and x0 ∈ Λ such that εn yn → x0 as n → ∞. Next we prove that x0 ∈ Λ.
Using (g2 ) and (3.20), we know that
ZZ Z Z
1−2s 2
y (∇wn ∇ψj + m wn ψj ) dxdy + V (εn x + εn yn )wn (x, 0)ψj (x, 0) dx ≤ f (wn (x, 0))ψj (x, 0) dx.
RN
+
+1
RN RN

Letting n → ∞ and using (3.17) and the continuity of V we find

ZZ Z Z
1−2s 2
y (∇w∇ψj + m wψj ) dxdy + V (x0 )w(x, 0)ψj (x, 0) dx ≤ f (w(x, 0))ψj (x, 0) dx.
RN
+
+1
RN RN

By passing to the limit as j → ∞ we obtain

ZZ Z Z
y 1−2s (|∇w|2 + m2 w2 ) dxdy + V (x0 )w2 (x, 0) dx ≤ f (w(x, 0))w(x, 0) dx.
RN
+
+1
RN RN

Hence there exists t1 ∈ (0, 1) such that t1 w ∈ MV (x0 ) . In view of Lemma 3.5 we have
dV (x0 ) ≤ LV (x0 ) (t1 w) ≤ lim inf Jεn (un ) = lim inf cεn ≤ dV (0)
n→∞ n→∞
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 13

from which dV (x0 ) ≤ dV (0) and thus V (x0 ) ≤ V (0) = −V0 . Since −V0 = inf x∈Λ V (x), we achieve that
V (x0 ) = −V0 . Using (V2 ), we conclude that x0 ∈ Λ.
Finally, we show that wn → w in X s (RN +
+1
). Set
Λ − εn ỹn
Λ̃n :=
εn
and define

1 if x ∈ Λ̃n ,
χ̃1n (x) :=
0 if x ∈ RN \ Λ̃n ,
χ̃2n (x) := 1 − χ̃1n (x).
Let us also consider the following functions for all x ∈ RN and n ∈ N
 
1 1
h1n (x) := − (V (εn x + εn yn ) + V1 )wn2 (x, 0)χ̃1n (x)
2 θ
 
1 1
h1 (x) := − (V (x0 ) + V1 )w2 (x, 0)
2 θ
  
1 1 1
h2n (x):= − (V (εn x + εn yn ) + V1 )wn2 (x, 0) + g(εn x + εn yn , wn (x, 0))wn (x, 0) − G(εn x + εn yn , wn (x, 0)) χ̃2n (x)
2 θ θ
  
1 1 1
≥ − − (V (εn x + εn ỹn ) + V1 )wn2 (x, 0)χ̃2n (x)
2 θ κ
 
1
h3n (x) := g(εn x + εn yn , wn (x, 0))wn (x, 0) − G(εn x + εn yn , wn (x, 0)) χ̃1n (x)
θ
 
1
= (f (wn (x, 0))wn (x, 0) − F (wn (x, 0))) χ̃1n (x)
θ
1
h3 (x) := (f (w(x, 0))w(x, 0) − F (w(x, 0))) .
θ
From (f3 ) and (g3 ), we see that the above functions are nonnegative in RN . Since
wn (x, 0) → w(x, 0) a.e. x ∈ RN ,
εn yn → x0 ∈ Λ,
we get
χ̃1n (x) → 1, h1n (x) → h1 (x), h2n (x) → 0 and h3n (x) → h3 (x) a.e. x ∈ RN .
Hence, observing that k · k2X s (RN +1 ) − µ| · |22 is weakly lower semicontinuous for all µ ∈ (0, m2s ), and using
+

Fatou’s Lemma and the invariance of RN by translation, we have

 
1 ′
dV (0) ≥ lim sup cεn = lim sup Jεn (un ) − hJεn (un ), un i
n→∞ n→∞ θ
 h 
1 1 2 2
i Z
1 2 3
≥ lim sup − kwn kX s (RN +1 ) − V1 |wn (·, 0)|2 + (hn + hn + hn ) dx
n→∞ 2 θ +
RN
 h 
1 1 i Z
2 2 1 2 3
≥ lim inf − kwn kX s (RN +1 ) − V1 |wn (·, 0)|2 + (hn + hn + hn ) dx
n→∞ 2 θ +
RN
 h
1 1 i Z
≥ − kwk2X s (RN +1 ) − V1 |w(·, 0)|22 + (h1 + h3 ) dx = dV (0) .
2 θ +
RN
Accordingly,
lim kwn k2X s (RN +1 ) − V1 |wn (·, 0)|22 = kwk2X s (RN +1 ) − V1 |w(·, 0)|22 , (3.21)
n→∞ + +

and
h1n → h1 , h2n → 0 and h3n → h3 in L1 (R3 ).
Then
Z Z
lim (V (εn x + εn yn ) + V1 )w2 (x, 0) dx = (V (x0 ) + V1 )w2 (x, 0) dx,
n→∞ RN RN
14 V. AMBROSIO

which implies that

lim |wn (·, 0)|22 = |w(·, 0)|22 . (3.22)
n→∞

Putting together (3.21) and (3.22), and using the fact that X s (RN
+
+1
) is a Hilbert space, we attain
kwn − wkX s (RN +1 ) → 0 as n → ∞.
+

This ends the proof of lemma. 

4. The proof of Theorem 1.1

In this last section we give the proof of Theorem 1.1. We start by proving the following lemma which
will be crucial to study the behavior of maximum points of the solutions. The proof is based on a variant
of the Moser iteration argument [44] (see also [4, 6, 19]).
Lemma 4.1. Let (wn ) be the sequence of Lemma 3.7. Then, wn (·, 0) ∈ L∞ (RN ) and there exists M > 0
such that
|wn (·, 0)|∞ ≤ M for all n ∈ N.

Moreover, wn ∈ L∞ (RN
+
+1
) and there exists R > 0 such that
kwn kL∞ (RN +1 ) ≤ R for all n ∈ N.
+

Proof. We note that wn is a weak solution to

− div(y 1−2s ∇wn ) + m2 y 1−2s wn = 0 in RN +1

+ ,
∂wn N (4.1)
∂ν 1−2s = −V (εn x + εn yn )wn (·, 0) + g(εn x + εn yn , wn (·, 0)) in R .
2β
Let zn,L := wn wn,L where wn,L := min{wn , L}, L > 0 and β > 0 will be chosen later. Taking zL,n as
test-function in (4.1) we deduce that
ZZ ZZ
1−2s 2β 2 2 2 2β
y wn,L (|∇wn | + m wn ) dxdy + 2βy 1−2s wn,L |∇wn |2 dxdy
RN
+
+1
Dn,L
Z Z
2β 2β
=− V (εn x + εn yn )wn2 (x, 0)wn,L (x, 0) dx + g(εn x + εn yn , wn (·, 0))wn (x, 0)wn,L (x, 0) dx
RN RN
(4.2)

where Dn,L := {(x, y) ∈ RN

+
+1
: |wn (x, y)| ≤ L}. It is easy to check that
ZZ ZZ
1−2s β 2 2β
y |∇(wn wn,L )| dxdy = y 1−2s wn,L |∇wn |2 dxdy
RN
+
+1
RN
+
+1

ZZ
2β
+ (2β + β 2 )y 1−2s wn,L |∇wn |2 dxdy. (4.3)
Dn,L

Then, putting together (4.2), (4.3), (V1 ), (f1 )-(f2 ), (g1 )-(g2 ), we get
ZZ
β β 2β
kwwn,L k2X s (RN +1 ) = y 1−2s (|∇(wn wn,L )|2 + m2 wn2 wn,L ) dxdy
+
RN
+
+1

 
β
ZZ ZZ
1−2s 2β 2β
= y wn,L (|∇wn |2
+m 2
wn2 ) dxdy
+ 2β 1 + y 1−2s wn,L |∇wn |2 dxdy
RN
+
+1
Dn,L 2
"Z Z ZZ #
1−2s 2β 2 2 2 1−2s 2β 2
≤ cβ y wn,L (|∇wn | + m wn ) dxdy + 2βy wn,L |∇wn | dxdy
RN
+
+1
Dn,L
 Z Z 
2β 2β
= cβ − V (εn x + εn yn )wn2 (x, 0)wn,L (x, 0) dx + g(εn x + εn yn , wn (·, 0))wn (x, 0)wn,L (x, 0) dx
N RN
Z R 
2β 2β
≤ cβ (V1 + 1)wn2 (x, 0)wn,L (x, 0) + C1 wnp+1 (x, 0)wn,L (x, 0) dx (4.4)
RN

where
β
cβ := 1 + .
2
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 15

Now, we prove that there exist a constant c > 0 independent of n, L, β, and hn ∈ LN/2s (RN ), hn ≥ 0
and independent of L and β, such that
2β 2β 2β
(V1 + 1)wn2 (x, 0)wn,L (x, 0) + C1 wnp+1 (x, 0)wn,L (x, 0) ≤ (c + hn )wn2 (·, 0)wn,L (·, 0) on RN . (4.5)
Firstly, we notice that
2β 2β
(V1 + 1)wn2 (x, 0)wn,L (x, 0) + C1 wnp+1 (x, 0)wn,L (x, 0)
2β 2β
≤ (V1 + 1)wn2 (x, 0)wn,L (x, 0) + C1 wnp−1 (·, 0)wn2 (·, 0)wn,L (·, 0) on RN .
Moreover,
wnp−1 (·, 0) ≤ 1 + hn on RN ,
where hn := χ{|wn (·,0)|>1} wnp−1 (·, 0) ∈ LN/2s (RN ). In fact, we can observe that
wnp−1 (·, 0) = χ{|wn (·,0)|≤1} wnp−1 (·, 0) + χ{|wn (·,0)|>1} wnp−1 (·, 0) ≤ 1 + χ{|wn (·,0)|>1} wnp−1 (·, 0) on RN ,
N
and that if (p − 1) 2s < 2 then, recalling that (wn (·, 0)) is bounded in H s (RN ),
Z Z
N
χ{|wn (·,0)|>1} |wn (x, 0)| 2s (p−1) dx ≤ χ{|wn (·,0)|>1} |wn (x, 0)|2 dx ≤ C, for all n ∈ N,
RN RN
N N
while if 2 ≤ (p − 1) 2s we deduce that (p − 1) 2s ∈ [2, 2∗s ], and by Theorem 2.1 and the boundedness of
(wn ) in X s (RN
+
+1
) we find
Z
N
2s (p−1)
N
χ{|wn (·,0)|>1} |wn (x, 0)| 2s (p−1) dx ≤ Ckwn kX s (RN +1 )
≤ C,
RN +

for some C > 0 depending only on N , s and p. Taking into account (4.4) and (4.5) we obtain that
Z
β 2β
kwn wn,L k2X s (RN +1 ) ≤ cβ (c + hn (x))wn2 (x, 0)wn,L (x, 0)dx,
+
RN

and by the monotone convergence theorem ((wn,L ) is nondecreasing with respect to L) we have as L → ∞
Z Z
β+1 2 2(β+1)
k|wn | kX s (RN +1 ) ≤ ccβ |wn (x, 0)| dx + cβ hn (x)|wn (x, 0)|2(β+1) dx. (4.6)
+
RN RN

Fix M > 1 and let A1,n := {hn ≤ M } and A2,n := {hn > M }. Then, by Hölder inequality,
Z Z Z
2(β+1) 2(β+1)
hn (x)|wn (x, 0)| dx = hn (x)|wn (x, 0)| dx + hn (x)|wn (x, 0)|2(β+1) dx
RN A1,n A2,n

≤ M ||wn (·, 0)|β+1 |22 + ε(M )||wn (·, 0)|β+1 |22∗s (4.7)
Z ! 2s
N

where ε(M ) := sup hN/2s

n dx → 0 as M → ∞ due to the fact that wn (·, 0) → w(·, 0) in H s (RN ).
n∈N A2,n
In view of (4.6) and (4.7) we get
k|wn |β+1 k2X s (RN +1 ) ≤ cβ (c + M )||wn (·, 0)|β+1 |22 + cβ ε(M )||wn (·, 0)|β+1 |22∗s . (4.8)
+

We note that Theorem 2.1 yields

||wn (·, 0)|β+1 |22∗s ≤ C22∗s k|wn |β+1 k2X s (RN +1 ) . (4.9)
+

Then, choosing M large so that

1
ε(M )cβ C22∗s < ,
2
and using (4.8) and (4.9) we obtain that
||wn (·, 0)|β+1 |22∗s ≤ 2C22∗s cβ (c + M )||wn (·, 0)|β+1 |22 . (4.10)
∗
Then we can start a bootstrap argument: since wn (·, 0) ∈ L2s (RN ) and |wn (·, 0)|2∗s ≤ C for all n ∈ N,
∗ 2N 2
N
we can apply (4.10) with β1 + 1 = N −2s to deduce that wn (·, 0) ∈ L(β1 +1)2s (RN ) = L (N −2s)2 (RN ).
2N k
Applying again (4.10), after k iterations, we find wn (·, 0) ∈ L (N −2s)k (RN ), and so wn (·, 0) ∈ Lq (RN ) for
all q ∈ [2, ∞) and |wn (·, 0)|q ≤ C for all n ∈ N.
16 V. AMBROSIO

Now we prove that actually wn (·, 0) ∈ L∞ (RN ). Since wn (·, 0) ∈ Lq (RN ) for all q ∈ [2, ∞) we have
N
that hn ∈ L s (RN ) and |hn | N ≤ D for all n ∈ N. Then, by the generalized Hölder inequality and Young’s
s
inequality with λ > 0, we can see that for all λ > 0
Z
hn (x)|wn (x, 0)|2(β+1) dx ≤ |hn | N ||wn (·, 0)|β+1 |2 ||wn (·, 0)|β+1 |2∗s
s
RN
 
β+1 2 1 β+1 2
≤ D λ||wn (·, 0)| |2 + ||wn (·, 0)| |2∗s .
λ
Consequently, using (4.6) and (4.9), we deduce that
||wn (·, 0)|β+1 |22∗s ≤ C22∗s k|wn |β+1 k2X s (RN +1 )
+

cβ D
≤ cβ C22∗s (c + Dλ)||wn (·, 0)|β+1 |22 + C22∗s ||wn (·, 0)|β+1 |22∗s . (4.11)
λ
Taking λ > 0 such that
cβ DC22∗s 1
=
λ 2
we obtain that
||wn (·, 0)|β+1 |22∗s ≤ 2cβ (c + Dλ)C22∗s ||wn (·, 0)|β+1 |22 = Mβ ||wn (·, 0)|β+1 |22 ,
where
Mβ := 2cβ (c + Dλ)C22∗s .
Now we can control the dependence on β of Mβ as follows:
√
Mβ ≤ Cc2β ≤ C(1 + β)2 ≤ M02 e2 β+1
,
for some M0 > 0 independent of β, and we get
1
√1
|wn (·, 0)|2∗s (β+1) ≤ M0β+1 e β+1 |wn (·, 0)|2(β+1) .
As before, iterating this last relation and choosing β0 = 0 and 2(βj+1 + 1) = 2∗s (βj + 1) we have that
Pj 1 Pj 1
i=0 βi +1 i=0
√
|wn (·, 0)|2∗s (βj +1) ≤ M0 e βi +1
|wn (·, 0)|2(β0 +1) .
We note that  j
N
βj = − 1, (4.12)
N − 2s
so the series
∞ ∞
X 1 X 1
and √
i=0
βi + 1 i=0
βi + 1
are convergent. Recalling that |wn (·, 0)|q ≤ C for all n ∈ N and q ∈ [2, ∞), we get
|wn (·, 0)|∞ = lim |wn (·, 0)|2∗s (βj +1) ≤ M for all n ∈ N.
j→∞

This proves the L∞ -desired estimate for the trace. At this point, we prove that there exists R > 0 such
that
kwn kL∞ (RN +1 ) ≤ R for all n ∈ N. (4.13)
+

Using (4.11) with λ = 1 and that |wn (·, 0)|q ≤ C for all q ∈ [2, ∞], we deduce that
k|wn |β+1 k2X s (RN +1 ) ≤ c̃cβ C 2(β+1) for all n ∈ N,
+

for some c̃, C > 0 independent on β and n. On the other hand, from (2.1), we obtain that
β+1
! 2γ(β+1)
ZZ
y a |wn (x, y)|2γ(β+1) dxdy = k|wn |β+1 kL2γ (RN +1 ,y1−2s ) ≤ C∗ k|wn |β+1 kX s (RN +1 )
+ +
RN
+
+1

which combined with the previous inequality yields

1
kwn kL2γ(β+1) (RN +1 ,y1−2s ) ≤ C ′ (C̃∗ cβ ) 2(β+1) for all n ∈ N.
+

Since  
1 β β 1
log 1 + ≤ ≤ for all β > 0,
β+1 2 2(β + 1) 2
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 17

1
we can see that there exists C̄ > 0 such that C ′ (C̃∗ cβ ) 2(β+1) ≤ C̄ for all β > 0, and so
kwn kL2γ(β+1) (RN +1 ,y1−2s ) ≤ C̄ for all n ∈ N, β > 0.
+

Now, fix R > C̄ and define An := {(x, y) ∈ RN +

+1
: wn (x, y) > R}. Hence, for all n, j ∈ N, we have
1
ZZ ! 2γ(βj +1) γ(βj +1)−1
Z Z  2γ(β1 +1)
j
1−2s 2γ(βj +1) 1−2s 2
C̄ ≥ y |wn (x, y)| dxdy ≥R γ(βj +1)
y |wn (x, y)| dxdy
RN
+
+1
An

which yields
β +1− γ1 1
C̄ j
 Z Z  2γ
1
≥ 1 y 1−2s |wn (x, y)|2 dxdy ,
R C̄ γ An
where βj is given in (4.12). Letting j → ∞, we have that βj → ∞ and then
ZZ
y 1−2s |wn (x, y)|2 dxdy = 0 for all n ∈ N,
An

which implies that |An | = 0 for all n ∈ N. Consequently, (4.13) holds true. 
Lemma 4.2. The sequence (wn ) satisfies wn (·, 0) → 0 as |x| → ∞ uniformly in n ∈ N.

Proof. By Lemma 4.1 and Proposition 2.1-(iii), we obtain that wn is continuous in RN +

+1
. On the other
s N +1 2γ N +1 1−2s
hand, from (2.1) and wn → w in X (R+ ), we have that wn → w in L (R+ , y ). Fix x̄ ∈ RN .
Using (V1 ) and (3.2) we see that wn is a weak subsolution to
−div(y 1−2s ∇wn ) + m2 y 1−2s wn = 0 in Q1 (x̄, 0) := B1 (x̄) × (0, 1),

∂wn 2∗
s −1
∂ν 1−2s = (V1 + η)wn + Cη wn on B1 (x̄),
where η ∈ (0, m2s − V1 ) is fixed. Applying Proposition 2.1-(i) and observing that L2γ (A, y 1−2s ) ⊂
L2 (A, y 1−2s ) for any bounded set A ⊂ RN , we get
2∗ −1
0 ≤ sup wn ≤ C(kwn kL2γ (Q1 (x̄,0),y1−2s ) + |wns (·, 0)|Lq (B1 (x̄)) ) for all n ∈ N,
Q1 (x̄,0)
N
where q > 2s is fixed and C > 0 is a constant depending only on N, m, s, q, γ and independent of n ∈ N
N
and x̄. Note that q(2∗s − 1) ∈ (2, ∞) because N > 2s and q > 2s . Taking the limit as |x̄| → ∞ we infer
that wn (x̄, 0) → 0 as |x̄| → ∞ uniformly in n ∈ N. 
Now we have all tools to give the proof of the main result of this work.
Proof of Theorem 1.1. We begin by proving that there exists ε0 > 0 such that for any ε ∈ (0, ε0 ) and any
mountain pass solution uε ∈ Xε of (3.1), it holds
|uε (·, 0)|L∞ (RN \Λε ) < a. (4.14)
Assume by contradiction that for some subsequence (εn ) such that εn → 0, we can find un := uεn ∈ Xεn
such that Jεn (un ) = cεn , Jε′ n (un ) = 0 and
|un (·, 0)|L∞ (RN \Λεn ) ≥ a. (4.15)
In view of Lemma 3.7, we can find (yn ) ⊂ RN such that wn (x, y) := un (x + yn , y) → w in X s (RN +
+1
) and
εn yn → x0 for some x0 ∈ Λ such that V (x0 ) = −V0 .
Now, if we choose r > 0 such that Br (x0 ) ⊂ B2r (x0 ) ⊂ Λ, we can see that B εrn ( εxn0 ) ⊂ Λεn . Then, for
any x ∈ B εrn (yn ) it holds
   
x − x0  ≤ |x − yn | + yn − x0  < 1 (r + on (1)) < 2r for n sufficiently large.
   
 εn   ε n  εn εn
Therefore,
RN \ Λεn ⊂ RN \ B εrn (yn ) (4.16)
for any n big enough. Using Lemma 4.2 we see that
wn (x, 0) → 0 as |x| → ∞ (4.17)
uniformly in n ∈ N. Therefore there exists R > 0 such that
wn (x, 0) < a for any |x| ≥ R, n ∈ N.
18 V. AMBROSIO

Hence, un (x, 0) = wn (x − yn , 0) < a for any x ∈ RN \ BR (yn ) and n ∈ N. On the other hand, by (4.16),
there exists n0 ∈ N such that for any n ≥ n0 we have

RN \ Λεn ⊂ RN \ B εrn (yn ) ⊂ RN \ BR (yn ),

which implies that un (x, 0) < a for any x ∈ RN \ Λεn and n ≥ n0 . This is impossible according to (4.15).
Since uε ∈ Xε satisfies (4.14), by the definition of g it follows that uε is a solution of (2.5) for ε ∈ (0, ε0 ).
From the Harnack inequality we conclude that uε (x, 0) > 0 in RN .
In what follows, we study the behavior of the maximum points of solutions to problem (1.1). Take
εn → 0 and let (un ) ⊂ Xεn be a sequence of solutions to (3.1) as above. Consider the translated sequence
wn (x, y) = un (x + yn , y) where (yn ) is given by Lemma 3.7. Let us prove that there exists δ > 0 such
that

|wn (·, 0)|∞ ≥ δ for all n ∈ N. (4.18)

Assume by contradiction that |wn (·, 0)|∞ → 0. Using (f1 ) we can see that there exists ν ∈ N such that

f (|wn (·, 0)|∞ ) V1

< for all n ≥ ν.
|wn (·, 0)|∞ κ

From hJε′ n (un ), un i = 0, (g2 ) and (f4 ), we can see that for all n ≥ ν

kwn k2X s (RN +1 ) − V1 |wn (·, 0)|22 = kun k2X s (RN +1 ) − V1 |un (·, 0)|22
+ +
Z
≤ f (un (x, 0))un (x, 0) dx
N
ZR
f (|wn (·, 0)|∞ ) 2
≤ wn (x, 0) dx
R N |wn (·, 0)|∞
V1
Z
≤ w2 (x, 0) dx
κ RN n

which combined with (2.3) yields

  
V1 1
1 − 2s 1 + kwn k2X s (RN +1 ) ≤ 0.
m κ +

Since κ > m2sV−V1

1
we get kwn kX s (RN +1 ) = 0 for all n ≥ ν which is a contradiction. Therefore, if qn is a
+
global maximum point of wn (·, 0), we deduce from Lemma 4.1 and (4.18) that there exists R > 0 such that
|qn | < R for all n ∈ N. Thus xn := qn + yn is a global maximum point of un (·, 0), and εn xn → x0 ∈ M.
Using the continuity of V we deduce that

lim V (εn xn ) = V (x0 ) = −V0 .

n→∞

Finally, we prove a decay estimate for un (·, 0). Using (f1 ), the definition of g and (4.17), we can find
R1 > 1 sufficiently large such that

g(εn x + εn yn , wn (x, 0))wn (x, 0) ≤ δwn2 (x, 0) for |x| ≥ R1 , (4.19)

where δ ∈ (0, m2s − V1 ) is fixed. Pick a smooth cut-off function φ such that 0 ≤ φ ≤ 1, φ(x) = 0 for
|x| ≥ 1, and φ 6≡ 0. By Riesz representation theorem, there exists a unique function w̄ ∈ H s (RN ) such
that

(−∆ + m2 )s w̄ − (V1 + δ)w̄ = φ in RN . (4.20)

Since w̄ = B2s,m ∗ φ, for some positive kernel B2s,m whose expression is given below, we can see that
w̄ ≥ 0 in RN . Denote by W̄ the s-extension of w̄, namely W̄ (x, y) = (Ps,m (·, y) ∗ w̄)(x). Fix x ∈ RN .
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 19

Then, from Young’s inequality and (2.4), we can see that

kW̄ kL2 (B1 (x)×[0,1],y1−2s ) ≤ kW̄ kL2 (RN +1 ,y1−2s ) = kPs,m (·, y) ∗ w̄kL2 (RN +1 ,y1−2s )
+ +
Z 1  21
≤ |Ps,m (·, y) ∗ w̄|22 y 1−2s dy
0
Z 1  21
≤ |Ps,m (·, y)|21 |w̄|22 y 1−2s
dy
0
Z 1  21
2 1−2s
≤ |w̄|2 ϑ (my)y dy
0
≤ cs,m |w̄|2 ,
for some constant cs,m > 0. Therefore, by Proposition 2.1-(i) and translation invariance of (4.20) with
respect to x ∈ RN , we deduce that w̄ ∈ L∞ (RN ), and thus, by interpolation, w̄ ∈ Lq (RN ) for any
q ∈ [2, ∞]. Hence, (−∆ + m2 )s w̄ = (V1 + δ)w̄ + φ ∈ L∞ (RN ) and applying Theorem 5.4 (see also
Theorem 5.1-(v) and Proposition 2.1-(iii)) we obtain that w̄ is Hölder continuous in RN . Using Harnack’s
inequality (see Proposition 2.1-(ii)) we conclude that w̄ > 0 in RN . Moreover, we can prove that there
exist c, C > 0 such that
0 < w̄(x) ≤ Ce−c|x| for all x ∈ RN . (4.21)
Assume for the moment that (4.21) holds and we postpone the proof of it after proving the decay estimate
for ûn (·, 0). We know that
(−∆ + m2 )s w̄ − (V1 + δ)w̄ = 0 in RN \ BR1 . (4.22)
On the other hand, by (V1 ) and (4.19), we can see that
(−∆ + m2 )s wn (·, 0) − (V1 + δ)wn (·, 0) ≤ 0 in RN \ BR1 . (4.23)
Set b := inf BR1 w̄ > 0 and zn := (ℓ + 1)w̄ − bwn (·, 0), where ℓ := supn∈N |wn (·, 0)|∞ . We aim to show that
zn ≥ 0 in RN . Assume by contradiction that there exists a sequence (xj,n ) such that
inf zn (x) = lim zn (xj,n ) < 0. (4.24)
x∈RN j→∞

In view of (4.17) and (4.21), we notice that

lim w̄(x) = lim wn (x, 0) = 0
|x|→∞ |x|→∞

where the second limit is uniform in n ∈ N, and so

lim zn (x) = 0 uniformly in n ∈ N.
|x|→∞

Consequently, (xj,n ) is bounded and, up to a subsequence, we may assume that xj,n → x∗n for some
xn ∈ RN . Hence, (4.24) becomes
inf zn (x) = zn (x∗n ) < 0. (4.25)
x∈RN

Then, from the minimality of x∗n and using Theorem 5.4, we deduce that
C(N, s) N +2s 2zn (x∗n ) − zn (x∗n + y) − zn (x∗n − y)
Z
2 s ∗ 2s ∗
(−∆ + m ) zn (xn ) = m zn (xn ) + m 2
N +2s K N +2s (m|y|) dy
2 RN |y| 2 2

≤ m2s zn (x∗n ). (4.26)

Since zn ≥ 0 in BR1 , it follows from (4.24) that
x∗n ∈ RN \ BR1 . (4.27)
Gathering (4.22) and (4.23) we can see that
(−∆ + m2 )s zn − (V1 + δ)zn ≥ 0 in RN \ BR1 ,
and this is a contradiction because (4.25), (4.26), (4.27) and V1 + δ < m2s imply that
(−∆ + m2 )s zn (x∗n ) − (V1 + δ)zn (x∗n ) ≤ (m2s − (V1 + δ))zn (x∗n ) < 0.
20 V. AMBROSIO

Hence, zn ≥ 0 in RN . In the light of (4.21) we obtain that there exist c, C > 0 such that
0 ≤ wn (x) ≤ Ce−c|x| for all x ∈ RN , n ∈ N,
which combined with un (x, 0) = wn (x − yn , 0) yields
un (x, 0) = wn (x − yn , 0) ≤ Ce−c|x−yn| for all x ∈ RN , n ∈ N.
In what follows, we focus our attention on the estimate (4.21). Note that w̄ = B2s,m ∗ φ, where
B2s,m (x) = F −1 ([(|k|2 + m2 )s − (V1 + δ)]−1 ).
Since φ has compact support, the exponential decay of w̄ at infinity follows if we show the exponential
decay of B2s,m (x) for big values of |x|. After that, due to the fact that w̄ is continuous in RN , we can
deduce the exponential decay of w̄ in the whole of RN . Next we prove the exponential decay of B2s,m (x)
for |x| large. Then
1 1
Z
B2s,m (x) = N eık·x 2 2 s − (V + δ)]
dk
(2π) 2 R N [(|k| + m ) 1
Z ∞ 
1
Z
ık·x −t[(|k|2 +m2 )s −(V1 +δ)]
= N e e dt dk
(2π) 2 RN 0
Z ∞ !
1
Z
−γt ık·x −t[(|k|2 +m2 )s −m2s ]
= e N e e dk dt
0 (2π) 2 RN
Z ∞
= e−γt ps,m (x, t) dt (4.28)
0
where
γ := m2s − (V1 + δ) > 0,
and ∞
1
Z
2s |x|2 2
ps,m (x, t) := em t
e−m z ϑs (t, z) dz
N e− 4t

0 (4πz) 2

is the 2s-stable relativistic density with parameter m (see pag. 4 formula (7) in [45], and pag. 4875
formula (2.12) and Lemma 2.2 in [13]), and ϑs (t, z) is the density function of the s-stable process whose
s
Laplace transform is e−tλ . Using the scaling property ps,m (x, t) = mN ps,1 (mx, m2s t) (see pag. 4876
formula (2.15) in [13]) and Lemma 2.2 in [34], we can see that
    
mx mx
ps,m (x, t) ≤ C gm2s t √ + m2s tν m √ for all x ∈ RN , t > 0, (4.29)
2 2
where
1 − |x|
2
gt (x) := N e
4t

(4πt) 2
and ν m is the Lévy measure of relativistic process with parameter m > 0 given by
2s−N   N +2s
m 2s2 2 m 2
ν (x) := N K N +2s (m|x|)
π 2 Γ(1 − s) |x| 2

(see pag. 4877 formula (2.17) in [13]). Therefore, (4.28) and (4.29) yield
Z ∞   Z ∞  
mx mx
B2s,m (x) ≤ C e−γt gm2s t √ dt + C e−γt tν m √ dt
0 2 0 2
=: I1 (x) + I2 (x). (4.30)
We start with the estimate of I1 (x) for |x| ≥ 2. Observing that
m2−2s 2 m2−2s
γt + |x| ≥ γt + for all |x| ≥ 2, t > 0,
8t 2t
and that ab ≤ ε a2 + 41ε b2 for all a, b ≥ 0 and ε > 0 gives
m2−2s 2 m1−s √
γt + |x| ≥ √ |x| γ for all x ∈ RN , t > 0,
8t 2
we deduce that for all |x| ≥ 2 and t > 0
m2−2s 2 t m2−2s m1−s √
γt + |x| ≥ γ + + √ |x| γ.
8t 2 4t 2 2
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 21

Thus, using the definition of gt , we can see that for all |x| ≥ 2
Z ∞ −γ t √
e 2 − m2−2s − m1−s √ |x| γ
I1 (x) ≤ C N e 4t e 2 2 dt
0 t2
Z ∞ −γ t
e 2 − m2−2s
≤ Ce−c|x| N e 4t dt ≤ Ce−c|x| (4.31)
0 t2
where we used the fact that
∞
e−αt − β
Z
e t dt < ∞ ∀α, β, p > 0.
0 tp
Here C, c > 0 depend only on N, s, m. Now we estimate I2 (x) for large values of |x|. Recalling formula
(1.4) concerning the asymptotic behavior of Kν at infinity, we deduce that there exists r0 > 0 such that
Kν (mr) e−mr
ν
≤ C ν+ 1 for all r ≥ r0 ,
r r 2
and then
K N +2s (m|x|) e−m|x|
2
N +2s ≤C N +2s+1 for all |x| ≥ r0 .
|x| 2 |x| 2

Consequently, using the definition of ν m , for all |x| ≥ r0 we get

Z ∞
e−m|x| e−m|x|
I2 (x) ≤ C N +2s+1 t e−γt dt ≤ C N +2s+1 . (4.32)
|x| 2 0 |x| 2
Gathering (4.28), (4.30), (4.31), (4.32) we find that for any |x| ≥ max{r0 , 2}
e−C4 |x|
B2s,m (x) ≤ C1 e−C2 |x| + C3 N +2s+1 ≤ C5 e−C6 |x| .
|x| 2

Thus (4.21) holds true and this ends the proof of Theorem 1.1. 

Remark 4.1. When s = 21 , p 12 ,m (x, t) can be calculated explicitly (see [13, 40]) and is given by
 m  N2+1 N +1
t emt (|x|2 + t2 )−
p
p 21 ,m (x, t) = 2 4 K N +1 (m |x|2 + t2 ).
2π 2

Remark 4.2. By the definitions of B2s,m and ps,m , it follows that B2s,m is radial, positive, strictly
decreasing in |x|, and smooth on RN \ {0}.

5. Appendix: Bessel potentials

In this appendix we collect some useful results concerning Bessel potentials (with m = 1). For more
details we refer to [9, 15, 33, 48].
Definition 5.1. Let α > 0. The Bessel potential of order α of u ∈ S(RN ) is defined as
Z
α
Jα u(x) := (1 − ∆)− 2 u(x) = (Gα ∗ u)(x) = Gα (x − y)u(y) dy,
RN

where
N α
Gα (x) := (2π)− 2 F −1 ((1 + |k|2 )− 2 )(x)
is called Bessel kernel.
Remark 5.1. If α ∈ R (or α ∈ C), then we may define the Bessel potential of a temperate distribution
u ∈ S ′ (RN ) (see [15]) by setting
N α
F Jα u(k) := (2π)− 2 (1 + |k|2 )− 2 F u(k).
From definition it is evident that Gα+β = Gα ∗ Gβ . It is possible to prove (see [9]) that
1 α−N
Gα (x) = N +α−2 N
K N −α (|x|)|x| 2 .
2 2 π 2 Γ( α2 ) 2
22 V. AMBROSIO

Thus Gα (x) is positive, decreasing function of |x|, analytic except at x = 0, and for x ∈ RN \ {0}, Gα (x)
is an entire function of α. Moreover, from (1.3) and (1.4), we have
|x|−N +α
Gα (x) = + o(|x|−N +α ) as |x| → 0, if 0 < α < N,
γ(α)
Gα (x) = O(e−c|x| ) as |x| → ∞, for some c > 0,
Gα ∈ L1 (RN ) for all α > 0, and RN Gα (x) dx = 1. We also have the following integral formula (see [48]):
R
Z ∞
1 1 − π|x|
2
δ α−N dδ
Gα (x) = α α

e δ e− 4π δ 2 .
(4π) Γ 2 0
2 δ
One the most interesting facts concerning Bessel potentials is they can be employed to define the Bessel
potential spaces; see [1, 9, 15, 33, 48]. For p ∈ [1, ∞] and α ∈ R we define the Banach space
Lαp := Gα (Lp (RN )) = {u : u = Gα ∗ f, f ∈ Lp (RN )}
endowed with the norm
kukLαp := |f |p if u = Gα ∗ f.
Thus Lαp p N
is a subspace of L (R ) for all α ≥ 0. We also have the following useful result:
Theorem 5.1. [1, 15, 48]
(i) If α ≥ 0 and 1 ≤ p < ∞, then D(RN ) is dense in Lαp .
(ii) If 1 < p < ∞ and p′ its conjugate exponent, then the dual of Lαp is isometrically isomorphic to
p′
L−α .
(iii) If β < α, then Lαp is continuously embedded in Lβp .
Np N
(iv) If β ≤ α and if either 1 < p ≤ q ≤ N −(α−β)p < ∞ or p = 1 and 1 ≤ q < N −α+β , then Lαp is
q
continuously embedded in Lβ .
(v) If 0 ≤ µ ≤ α − Np < 1, then Lαp is continuously embedded in C 0,µ (RN ).
(vi) Lkp = W k,p (RN ) for all k ∈ N and 1 < p < ∞, Lα2 = W α,2 (RN ) for any α.
(vii) If 1 < p < ∞ and ε > 0, then for every α we have the following continuous embeddings:
p p
Lα+ε ⊂ W α,p (RN ) ⊂ Lα−ε .
In order to accomplish some useful regularity results for equations driven by (−∆ + m2 )s , with m > 0,
we introduce the Hölder-Zygmund (or Lipschitz) spaces Λα ; see [15, 38, 48]. If α > 0 and α ∈ / N then we
set Λα := C [α],α−[α] (RN ). If α = k ∈ N then we set Λα := Λ∗k where
( )
∗ ∞ N N |u(x + h) + u(x − h) − 2u(x)|
Λ1 := u ∈ L (R ) ∩ C(R ) : sup < ∞ if k = 1,
x,h∈RN ,|h|>0 |h|
and
Λ∗k := u ∈ C k−1 (RN ) : Dγ u ∈ Λ∗1 for all |γ| ≤ k − 1 if k ∈ N, k ≥ 2.


Then we have the following useful result:

Theorem 5.2. [15, 48] If α, β ≥ 0, (1 − ∆)−α is an isomorphism of Lβp to Lβ+2α
p
. If α ≥ 0 and β > 0,
−α
(1 − ∆) is an isomorphism of Λβ to Λβ+2α .
As a consequence of Theorem 5.2 and the definition of Hölder-Zygmund spaces, we easily deduce the
following result:
Corollary 5.1. Let s ∈ (0, 1) and 0 < α ≤ 1. Assume that f ∈ C 0,α (RN ) and that u ∈ L∞ (RN ) is a
solution to (−∆ + 1)s u = f in RN .
• If α + 2s < 1 then u ∈ C 0,α+2s (RN ).
• If 1 < α + 2s < 2 then u ∈ C 1,α+2s−1 (RN ).
• If 2 < α + 2s < 3 then u ∈ C 2,α+2s−2 (RN ).
• If α + 2s = k ∈ {1, 2} then u ∈ Λ∗k .
Bearing in mind the asymptotic estimates (1.3) and (1.4) for Kν , we are able to gain an integral
representation formula for (−∆ + m2 )s , with s ∈ (0, 1), in the spirit of [23].
Theorem 5.3. Let s ∈ (0, 1). Then, for all u ∈ S(RN ),
C(N, s) N +2s 2u(x) − u(x + y) − u(x − y)
Z
2 s 2s
(−∆ + m ) u(x) = m u(x) + m 2
N +2s K N +2s (m|y|) dy.
2 RN |y| 2 2
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 23

Proof. Choosing the substitution z = y − x in (1.2), we obtain

u(x) − u(y)
Z
N +2s
2 s 2s
(−∆ + m ) u(x) = m u(x) + C(N, s)m P.V. N +2s K N +2s (m|x − y|) dy
2

RN |x − y| 2
2

u(x) − u(x + z)
Z
N +2s
= m2s u(x) + C(N, s)m 2 P.V. N +2s K N +2s (m|z|) dz. (5.1)
R N |z| 2 2

By substituting z̃ = −z in the last term in (5.1), we get

u(x + z) − u(x) u(x − z̃) − u(x)
Z Z
P.V. N +2s K N +2s (m|z|) dz = P.V. N +2s K N +2s (m|z̃|) dz̃, (5.2)
RN |z| 2 2
RN |z̃| 2 2

and so, after relabeling z̃ as z,

u(x + z) − u(x)
Z
2P.V. N +2s K N +2s (m|z|) dz
RN |z| 2 2

u(x + z) − u(x)
Z
= P.V. N +2s K N +2s (m|z|) dz
RN |z| 2 2

u(x − z) − u(x)
Z
+ P.V. N +2s K N +2s (m|z|) dz
R N |z| 2 2

u(x + z) + u(x − z) − 2u(x)

Z
= P.V. N +2s K N +2s (m|z|) dz. (5.3)
R N |z| 2 2

Hence, if we rename z as y in (5.1) and (5.3), we can write (−∆ + m2 )s as

C(N, s) N +2s 2u(x) − u(x + y) − u(x − y)
Z
2 s 2s
(−∆ + m ) u(x) = m u(x) + m 2 P.V. N +2s K N +2s (m|y|) dy.
2 R N |y| 2 2

(5.4)
Now, by using a second order Taylor expansion, we see that
 
 2u(x) − u(x + y) − u(x − y)  |D2 u|∞
K (m|y|) ≤ K N +2s (m|y|).
 
N +2s N +2s
 |y| N +2s−4
 
 |y| 2 2 2 2

From (1.3) we deduce that

|D2 u|∞ C
N +2s−4 K N +2s (m|y|) ∼ as |y| → 0
|y| 2 2 |y|N +2s−2
which is integrable near 0. On the other hand, using (1.4), we get
 
 2u(x) − u(x + y) − u(x − y)  3|u|∞ C −m|y|
K N +2s (m|y|) ≤ N +2s K N +2s (m|y|) ∼ N +2s+1 e as |y| → ∞
 
 N +2s
 |y| 2 2  |y| 2 2
|y| 2
which is integrable near ∞. Therefore, we can remove the P.V. in (5.4). 
By Theorem 5.3 and using the fact that |z|−ν Kν (m|z|) ≤ Cν |z|−2ν (see pag. 5865 in [28]), we can
argue as in [47] and use the C k,α estimates, with k ∈ {1, 2}, for the elliptic equation −∆ + m2 u = g in
RN , to obtain Schauder-Hölder estimates for (−∆ + m2 )s . In the light of this observation, we can give
an alternative proof of Corollary 5.1 and deduce the next helpful result. Since the proofs are similar to
the ones performed in [47] for the case m = 0, we skip the details.
Theorem 5.4. Let s ∈ (0, 1) and m > 0. Assume that f ∈ L∞ (RN ) and that u ∈ L∞ (RN ) is a solution
to (−∆ + m2 )s u = f in RN .
• If 2s ≤ 1 then u ∈ C 0,α (RN ) for any α < 2s.
• If 2s > 1 then u ∈ C 1,α (RN ) for any α < 2s − 1.

References
[1] R. A. Adams, Sobolev spaces, Pure and Applied Mathematics, Vol. 65 Academic Press, New York-London, 1975.
xviii+268 pp.
[2] C. O. Alves, O. H. Miyagaki, Existence and concentration of solution for a class of fractional elliptic equation in RN
via penalization method, Calc. Var. Partial Differential Equations 55 (2016), no. 3, Art. 47, 19 pp.
[3] A. Ambrosetti, P. H. Rabinowitz, Dual variational methods in critical point theory and applications, J. Functional
Analysis 14 (1973), 349–381.
24 V. AMBROSIO

[4] V. Ambrosio, Ground states solutions for a non-linear equation involving a pseudo-relativistic Schrödinger operator,
J. Math. Phys. 57 (2016), no. 5, 051502, 18 pp.
[5] V. Ambrosio, Multiplicity of positive solutions for a class of fractional Schrödinger equations via penalization method,
Ann. Mat. Pura Appl. (4) 196 (2017), no. 6, 2043–2062.
[6] V. Ambrosio, Concentrating solutions for a class of nonlinear fractional Schrödinger equations in RN , Rev. Mat.
Iberoam. 35 (2019), no. 5, 1367–1414.
[7] V. Ambrosio, Concentration phenomena for a class of fractional Kirchhoff equations in RN with general nonlineari-
ties, Nonlinear Anal. 195 (2020), 111761.
[8] V. Ambrosio, Multiple concentrating positive solutions for a relativistic fractional Schrödinger equation with a general
nonlinearity, in preparation.
[9] N. Aronszajn, K. T. Smith, Theory of Bessel potentials. I, Ann. Inst. Fourier (Grenoble) 11 (1961), 385–475.
[10] H. Berestycki, P. -L. Lions, Nonlinear scalar field equations. I. Existence of a ground state, Arch. Ration. Mech. Anal.
82 (1983), no. 4, 313–345.
[11] C. Bucur and E. Valdinoci, Nonlocal diffusion and applications. Lecture Notes of the Unione Matematica Italiana,
20. Springer, [Cham]; Unione Matematica Italiana, Bologna, 2016. xii+155 pp. ISBN: 978-3-319-28738-6; 978-3-319-
28739-3.
[12] H. Bueno, O.H. Miyagaki, G.A. Pereira, Remarks about a generalized pseudo-relativistic Hartree equation, J. Differ-
ential Equations 266 (2019), no. 1, 876–909.
[13] T. Byczkowski, J. Malecki, M. Ryznar, Bessel potentials, hitting distributions and Green functions, Trans. Amer.
Math. Soc. 361 (2009), no. 9, 4871–4900.
[14] L. Caffarelli, L. Silvestre, An extension problem related to the fractional Laplacian, Comm. Partial Differential Equa-
tions 32 (2007), no. 7-9, 1245–1260
[15] A.-P. Calderón, Lebesgue spaces of differentiable functions and distributions, 1961 Proc. Sympos. Pure Math., Vol.
IV pp. 33–49 American Mathematical Society, Providence, R.I.
[16] R. Carmona, W.C. Masters, B. Simon, Relativistic Schrödinger operators: Asymptotic behavior of the eigenfunctions,
J. Func. Anal 91 (1990), 117–142.
[17] S. Cingolani, S. Secchi, Ground states for the pseudo-relativistic Hartree equation with external potential, Proc. Roy.
Soc. Edinburgh Sect. A 145 (2015), no. 1, 73–90.
[18] S. Cingolani, S. Secchi, Semiclassical analysis for pseudo-relativistic Hartree equations, J. Differential Equations 258
(2015), no. 12, 4156–4179.
[19] V. Coti Zelati, M. Nolasco, Existence of ground states for nonlinear, pseudo-relativistic Schrödinger equations, Atti
Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 22 (2011), 51–72.
[20] V. Coti Zelati, M. Nolasco, Ground states for pseudo-relativistic Hartree equations of critical type, Rev. Mat. Iberoam.
29 (2013), no. 4, 1421–1436.
[21] J. Dávila, M. del Pino, J. Wei, Concentrating standing waves for the fractional nonlinear Schrödinger equation, J.
Differential Equations 256 (2014), no. 2, 858–892.
[22] M. Del Pino, P.L. Felmer, Local mountain passes for semilinear elliptic problems in unbounded domains, Calc. Var.
Partial Differential Equations, 4 (1996), 121–137.
[23] E. Di Nezza, G. Palatucci, E. Valdinoci, Hitchhiker’s guide to the fractional Sobolev spaces, Bull. Sci. Math. 136
(2012), no. 5, 521–573.
[24] S. Dipierro, M. Medina, E. Valdinoci, Fractional elliptic problems with critical growth in the whole of Rn , Appunti.
Scuola Normale Superiore di Pisa (Nuova Serie) [Lecture Notes. Scuola Normale Superiore di Pisa (New Series)], 15.
Edizioni della Normale, Pisa, 2017. viii+152 pp.
[25] A. Elgart, B. Schlein, Mean field dynamics of Boson stars, Comm. Pure Appl. Math. 60 (2007) 500–545.
[26] A. Erdélyi, W. Magnus, F. Oberhettinger, F. G. Tricomi, Higher transcendental functions. Vol. II, Based on notes
left by Harry Bateman. Reprint of the 1953 original. Robert E. Krieger Publishing Co., Inc., Melbourne, Fla., 1981.
xviii+396 pp.
[27] E. B. Fabes, C. E. Kenig, R. P. Serapioni, The local regularity of solutions of degenerate elliptic equations, Comm.
Partial Differential Equations 7 (1982), no. 1, 77–116.
[28] M. M. Fall, V. Felli, Unique continuation properties for relativistic Schrödinger operators with a singular potential,
Discrete Contin. Dyn. Syst. 35 (2015), no. 12, 5827–5867.
[29] M. M. Fall, F. Mahmoudi, E. Valdinoci, Ground states and concentration phenomena for the fractional Schrödinger
equation, Nonlinearity 28 (2015), no. 6, 1937–1961.
[30] P. Felmer, A. Quaas, J. Tan, Positive solutions of the nonlinear Schrödinger equation with the fractional Laplacian,
Proc. Roy. Soc. Edinburgh Sect. A 142 (2012), no. 6, 1237–1262.
[31] P. Felmer, I. Vergara, Scalar field equation with non-local diffusion, NoDEA Nonlinear Differential Equations Appl.
22 (2015), no. 5, 1411–1428.
[32] G.M. Figueiredo, G. Siciliano, A multiplicity result via Ljusternick-Schnirelmann category and Morse theory for a
fractional Schrödinger equation in RN , NoDEA Nonlinear Differential Equations Appl. 23 (2016), art. 12, 22 pp.
[33] L. Grafakos, Modern Fourier analysis, Third edition. Graduate Texts in Mathematics, 250. Springer, New York, 2014.
xvi+624 pp.
[34] T. Grzywny, M. Ryznar, Two-sided optimal bounds for Green functions of half-spaces for relativistic α-stable process,
Potential Anal. 28 (2008), no. 3, 201–239.
[35] I. W. Herbst, Spectral theory of the operator (p2 + m2 )1/2 − Ze2 /r, Comm. Math. Phys. 53 (1977), no. 3, 285–294.
[36] N. Ikoma, Existence of solutions of scalar field equations with fractional operator, J. Fixed Point Theory Appl. 19
(2017), no. 1, 649–690.
[37] T. Jin, Y. Li, J. Xiong, On a fractional Nirenberg problem, part I: blow up analysis and compactness of solutions, J.
Eur. Math. Soc. (JEMS) 16 (2014), no. 6, 1111–1171.
 THE NONLINEAR FRACTIONAL RELATIVISTIC SCHRÖDINGER EQUATION 25

[38] S. G. Krantz, Lipschitz spaces, smoothness of functions, and approximation theory, Exposition. Math. 1 (1983), no.
3, 193–260.
[39] N. Laskin, Fractional quantum mechanics, World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2018. xv+341
pp.
[40] E. H. Lieb, M. Loss, Analysis, Graduate Studies in Mathematics, 14. American Mathematical Society, Providence,
RI, 1997. xviii+278 pp.
[41] E. H. Lieb, H. T. Yau, The Chandrasekhar theory of stellar collapse as the limit of quantum mechanics, Comm.
Math. Phys. 112 (1987), no. 1, 147–174.
[42] E. H. Lieb, H. T. Yau, The stability and instability of relativistic matter, Comm. Math. Phys., 118 (1988), no. 2,
177–213.
[43] G. Molica Bisci, V. Rădulescu, R. Servadei, Variational Methods for Nonlocal Fractional Problems, Cambridge Uni-
versity Press, 162 Cambridge, 2016.
[44] J. Moser, A new proof of De Giorgi’s theorem concerning the regularity problem for elliptic differential equations,
Comm. Pure Appl. Math., 13 (1960), 457–468.
[45] M. Ryznar, Estimate of Green function for relativistic α-stable processes, Potential Analysis, 17, (2002), 1–23.
[46] S. Secchi, On some nonlinear fractional equations involving the Bessel potential, J. Dynam. Differential Equations
29 (2017), no. 3, 1173–1193.
[47] L. Silvestre, Regularity of the obstacle problem for a fractional power of the Laplace operator, Comm. Pure Appl.
Math., 60 (2007), no. 1, 67–112.
[48] E. Stein, Singular integrals and differentiability properties of functions, Princeton Mathematical Series, No. 30 Prince-
ton University Press, Princeton, N.J. 1970 xiv+290 pp.
[49] P. R. Stinga, J. L. Torrea, Extension problem and Harnack’s inequality for some fractional operators, Comm. Partial
Differential Equations 35 (2010), no. 11, 2092–2122.
[50] M. Willem, Minimax theorems, Progress in Nonlinear Differential Equations and their Applications, 24. Birkhäuser
Boston, Inc., Boston, MA, 1996. x+162 pp.

Vincenzo Ambrosio
Dipartimento di Ingegneria Industriale e Scienze Matematiche
Università Politecnica delle Marche
Via Brecce Bianche, 12
60131 Ancona (Italy)
E-mail address: [email protected]