3rd Reading

May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Analysis and Applications (2016)

c World Scientific Publishing Company
DOI: 10.1142/S0219530516500020

Existence theorems for fractional p-Laplacian problems

Paolo Piersanti∗,†,‡,§ and Patrizia Pucci∗,¶

∗Dipartimento di Matematica e Informatica
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Università Degli Studi di Perugia

Via Vanvitelli 1, 06123 Perugia, Italy
†Department of Mathematics

City University of Hong Kong

Tat Chee Avenue, Kowloon, Hong Kong
‡[email protected]
Anal. Appl. Downloaded from www.worldscientific.com

¶[email protected]
§[email protected]

Received 13 March 2015

Accepted 5 August 2015
Published 24 May 2016

The paper focuses on the existence of nontrivial solutions of a nonlinear eigenvalue

perturbed problem depending on a real parameter λ under homogeneous boundary con-
ditions in bounded domains with Lipschitz boundary. The problem involves a weighted
fractional p-Laplacian operator. Denoting by (λk )k a sequence of eigenvalues obtained
via mini–max methods and linking structures we prove the existence of (weak) solutions
both when there exists k ∈ N such that λ = λk and when λ ∈ / (λk )k . The paper is
divided into two parts: in the first part existence results are determined when the per-
turbation is the derivative of a globally positive function whereas, in the second part,
the case when the perturbation is the derivative of a function that could be either locally
positive or locally negative at 0 is taken into account. In the latter case, it is necessary to
extend the main results reported in [A. Iannizzotto, S. Liu, K. Perera and M. Squassina,
Existence results for fractional p-Laplacian problems via Morse theory, Adv. Calc. Var.
9(2) (2016) 101–125]. In both cases, the existence of solutions is achieved via linking
methods.

Keywords: Nonlinear eigenvalue problems; nonlocal p-Laplacian; Morse theory; cohomo-

logical local splitting.

Mathematics Subject Classification 2010: 35R11, 35J60, 35P30, 35J20, 35B09, 35S15

1. Introduction
In the last years, the interest for nonlocal nonlinear boundary value problems has
grown more and more, thanks in particular to their intriguing analytical structure
which requires somehow delicate proof techniques. In the present paper we consider

1
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

2 P. Piersanti & P. Pucci

the problem


LK u = λ a(x)|u|p−2 u + g(x, u) in Ω,
(Pλ )
u=0 in RN \Ω,

where Ω is a bounded open domain of RN with Lipschitz boundary, LK is an

integro-differential nonlocal operator defined along any function ϕ ∈ C0 ∞ (Ω) by

LK ϕ(x) = − |ϕ(x) − ϕ(y)|p−2 [ϕ(x) − ϕ(y)]K(x − y)dy
RN
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

for x ∈ Ω, the coefficient a is a positive weight of class Lα (Ω), with α > N/ps,
s ∈ (0, 1), N > ps, and g is a Carathéodory function. The weight function
K : RN \{0} → R+ is positive, and is assumed throughout the paper and with-
out further mentioning to satisfy the natural property
Anal. Appl. Downloaded from www.worldscientific.com

(K ) There exist suitable numbers cK , CK , with 0 < cK ≤ CK , such that

cK ≤ K(x)|x|N +ps ≤ CK for all x ∈ RN \{0}.

Clearly, when K(x) = |x|−(N +ps) , the operator LK reduces to the more familiar
fractional p–Laplacian operator (−∆)sp , which is defined by

|ϕ(x) − ϕ(y)|p−2 [ϕ(x) − ϕ(y)]
(−∆)sp ϕ(x) = − dy,
RN |x − y|N +ps
along any function ϕ ∈ C0∞ (Ω), see [22] and also [15]. Although the fractional
Laplacian operator (−∆)s = (−∆)s2 , p = 2, and more generally pseudodifferential
operators, have been a classical topic in harmonic analysis and partial differential
equations for a long time, the interest in such operators has constantly increased
during the last years. Nonlocal operators indeed arise in continuum mechanics,
phase transition phenomena, population dynamics, game theory and images pro-
cessing, see e.g. [4, 16, 34]. Indeed, as explained proficiently in [4], the study of
nonlocal, nonlinear problems with variational structure has interesting applications
in several fields. We just list the most fruitful branches, such as nonlinear nonlocal
diffusion theory (see also [34]) applied to ocean atmosphere interactions modeled
by the quasigeostrophic equation, the theory of semi-permeable membranes, planar
fracture dynamics; statistical mechanics applied to phase transition problems with
long range interactions; material sciences polymers applied to many scales interac-
tions and images processing. In this last field we just mention [16] and the references
therein. In particular, the paper by Gilboa and Osher [16] treats variational prob-
lems in bounded domains Ω for the reconstruction of images. More specifically a
technique to recreate and restore images is developed in [16]. In passing, we empha-
size that a great advantage in revising the fractional Laplacian in terms of nonlocal
problems in bounded domains Ω, as those treated here, is that no restrictions on
the fractional parameter s ∈ (0, 1) are imposed. Furthermore, in [11] a nonlocal
diffusion operator of type LK is also studied in bounded domains. Moreover, for
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 3

application reasons in [11] a continuous Galerkin infinite element discretization of

the nonlocal weak formulation, as well as a priori error estimates, was introduced.
The several numerical examples given in [11] illustrate as the abstract existence
theorems for nonlocal problems, as those proposed in this paper, can be applied in
concrete diffusion phenomena by approximation. For further applications in con-
trol problems constrained by a nonlocal steady diffusion we refer to [12] and the
references therein.
Problem (Pλ ) arises therefore, as shown above, from several applications. We
furthermore believe that the possibility of using standard arguments of the classical
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

critical point theory makes the solvability of nonlocal problems a relatively easy
task, even if several tricky difficulties should be overcome. Indeed, the techniques
used in the proofs for (Pλ ) have several conceptual as well as technical differences
with respect to the classical case of local variational problems. From a physical point
Anal. Appl. Downloaded from www.worldscientific.com

of view, (Pλ ) takes into account long-range particle interactions with a power-law
decay. When the decay at infinity is sufficiently weak, the long-range phenomena
may prevail and the nonlocal effects persist even on large scales. The kernel K is
fairly general and is not assumed to be even in RN \{0}. Consequently, (Pλ ) is not
a mere extension problem, since K includes rather general singular kernels, even if
of course (Pλ ) recuperates the standard p-Laplacian. In the context of fractional
quantum mechanics, (Pλ ) can be viewed as a nonlinear fractional Schrödinger
equation; this was arisen by Laskin [19, 20] as a result of expanding the Feynman
path integral, from the Brownian like to the Lévy like quantum mechanical paths.
In the last years, this subject has grown a remarkable interest in the study of the
fractional nonlinear phenomena.
In this paper we prove existence of solutions of (Pλ ) in the suitable fractional
Sobolev space D̃s,p (Ω), introduced in Sec. 2. The first main result is given under
the specific conditions

(g1 ) g is a Carathéodory function on Ω × R satisfying

|g(x, u)| ≤ g0 (x) |u|p−1 + g1 (x)|u|r−1 for a.a. x ∈ Ω and all u ∈ R,

where p < r < p∗ := N p/(N − ps), g0 and g1 are measurable positive functions
on Ω, with g0 ∈ Lr̃/(r̃−p) (Ω), g1 ∈ L∞ (Ω) and g1 ≤ Cg g0 a.e. in Ω, where
Cg > 0 is a constant depending on g and r < r̃ < p∗ ;
G(x, u) G(x, u)
(g2 ) lim = 0, lim =∞
u→0 g0 (x) |u|p |u|→∞ g0 (x) |u|p
u
uniformly for a.a. x ∈ Ω, with G(x, u) = 0 g(x, t) dt;
(g3 ) there exists µ > p such that for a.a. x ∈ Ω and all u ∈ R

−γ0 (x) − γ1 (x)|u|p ≤ µG(x, u) ≤ ug(x, u) + γ0 (x) + γ1 (x)|u|p ,

where γ0 ∈ L1 (Ω), γ1 ∈ Lα (Ω) are nonnegative in Ω, and α > N/ps is the

same Lebesgue exponent associated to the main positive weight a in (Pλ ).
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

4 P. Piersanti & P. Pucci

Theorem 1.1. Let a ∈ Lα (Ω) for some and α > N/ps. Assume (g1 )–(g3 ). Then
problem (Pλ ) has a nontrivial solution u ∈ D̃s,p (Ω) in each one of the following
cases:
(i) λ ∈ {λk : k ≥ 1};
(ii) λ ∈ {λk : k ≥ 1} and G(x, u) ≥ 0 for a.a. x ∈ Ω and all u ∈ R,
where (λk )k , defined using the Z2 -cohomological index, is a suitable sequence of
eigenvalues λk  ∞ of the associated eigenvalue problem


LK u = λ a(x)|u|p−2 u in Ω,
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

u=0 in RN \Ω,
associated to (Pλ ).
The proof of Theorem 1.1 is based on a series of steps, which are proved in
Anal. Appl. Downloaded from www.worldscientific.com

Lemmas 3.1–3.7 based on abstract results recalled in the appendix.

In Sec. 4 we study (Pλ ) in the case when the perturbation G is either locally
positive or locally negative at u = 0, under the same condition (g3 ), but in the more
restrictive case in which a ∈ L∞ (Ω) and when (g1 ) and (g2 ) are replaced by the
stronger requests
(g1 ) g is a nonidentically zero Carathéodory function on Ω × R, satisfying
|g(x, u)| ≤ C(1 + |u|r−1 ) for a.a. x ∈ Ω and all u ∈ R,
where C is a positive constant and p < r < p∗ ;
g(x, u) G(x, u)
(g2 ) lim = 0, lim = ∞,
u→0 |u|p−1 |u|→∞ |u|p
uniformly a.e. in Ω.
The main result is contained in the following theorem.

Theorem 1.2. Let a ∈ L∞ (Ω). Suppose that (g1 ) , (g2 ) and (g3 ) hold. Then, for
every λ ∈ R problem (Pλ ) has a nontrivial solution u ∈ D̃s,p (Ω) if one of these
conditions holds. Either
(i) λ ∈/ {λk : k ≥ 1}, or
(ii) λ ∈ {λk : k ≥ 1} and G(x, u) ≥ 0 for a.a. x ∈ Ω and every u ∈ R, with |u| ≤ δ,
or
(iii) λ ∈ {λk : k ≥ 1} and G(x, u) ≤ 0 for a.a. x ∈ Ω and every u ∈ R, with |u| ≤ δ,
where δ > 0 is a suitable number.
The key point in the proof is Lemma 4.1 which provides a crucial a priori esti-
mate for Lr (Ω) small solutions of (Pλ ). Lemma 4.1 is of independent interest and
its argument relies somehow on previous strategies of [18, 30], but with significant
changes and improvements. Again the long proof of Theorem 1.2 is divided in a
series of preliminary results given in Lemmas 3.1, 3.4, 3.7, and 4.1–4.4, as well as on
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 5

some abstract results of the appendix which are useful to prove the main existence
theorems.

2. Preliminaries and Auxiliary Results

In this section we present the notation and properties concerning the main operator
LK and the functional space D̃s,p (Ω) for solutions of (Pλ ). Since s ∈ (0, 1) and
N > ps, we denote by p∗ = p∗ (s) the critical Sobolev exponent for W0s,p (Ω), that
is p∗ = N p/(N − ps). For the functional setting we follow the construction and
·Ω
comments given in [29]. We first recall that D0s,p (Ω) = C0∞ (Ω) , where ·Ω is
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

the standard fractional Gagliardo norm, given by

  1/p
ϕΩ = |ϕ(x) − ϕ(y)|p |x − y|−(N +ps) dx dy
Ω×Ω
Anal. Appl. Downloaded from www.worldscientific.com

for all ϕ ∈ C0∞ (Ω).

Furthermore, Ds,p (RN ) denotes the fractional Beppo–Levi
space, that is the completion of C0∞ (RN ) with respect to the norm
  1/p
−(N +ps)
us = |u(x) − u(y)| |x − y|
p
dx dy .
R2N

Moreover, by [23, Theorems 1 and 2], the following inequalities

s(1 − s)
upLp∗ (RN ) ≤ cN,p ups ,
(N − ps)p−1
 (2.1)
dx s(1 − s)
|u(x)| p
≤ cN,p ups
RN |x|ps (N − ps)p
hold for all u ∈ D s,p
(RN ), where cN,p is a positive constant depending only on N
and p. Hence
∗
Ds,p (RN ) = {u ∈ Lp (RN ) : |u(x) − u(y)| · |x − y|−s+N/p ∈ Lp (R2N )}.
Following [17, 29], we put
∗
D̃s,p (Ω) = {u ∈ Lp (Ω) : ũ ∈ Ds,p (RN )},
where ũ is the natural extension of u in the entire RN , with value 0 in RN \Ω. We
endow Ds,p (RN ) with the norm u˜s = ũs , so that
   1/p
u˜s = upΩ + 2 |u(x)|p dx |x − y|−(N +ps) dy ≥ uΩ .
Ω RN \Ω

Since here Ω is regular, an application of [17, Theorem 1.4.2.2] shows that D̃s,p (Ω) =
·˜s
C0∞ (Ω) . Finally, since Ω is bounded and regular, by (2.1) there exists a constant
cΩ > 0 such that
cΩ ũW s,p (RN ) ≤ ũs = u˜s ≤ ũW s,p (RN ) (2.2)
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

6 P. Piersanti & P. Pucci

for all u ∈ D̃s,p (Ω). Indeed, if u ∈ D̃s,p (Ω), then ũ ∈ W s,p (RN ), and moreover there
exists a constant 1/cΩ > 0 such that
  1/p
ũW s,p (RN ) = |u(x)|p dx + |ũ(x) − ũ(y)| · |x − y|−(N +ps) dxdy
Ω RN ×RN

1
≤ ũs .
cΩ
This proves the first part of (2.2). The equality in (2.2) follows from the definition
of u˜s , while the last inequality in (2.2) is just a triviality.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Moreover, by virtue of [17, Corollary 1.4.4.10], we have the main property

D̃s,p (Ω) = {u ∈ W0s,p (Ω) : ud(·, ∂Ω)−s ∈ Lp (Ω)}
= {u ∈ Ds,p (RN ) : u = 0 a.e. in RN \Ω}
Anal. Appl. Downloaded from www.worldscientific.com

= {u ∈ W s,p (Ω) : ũ ∈ W s,p (RN )},

where d(x, ∂Ω) is the distance from x to the boundary ∂Ω of Ω. It is not hard to
see that D̃s,p (Ω) is a closed subspace of Ds,p (RN ). Hence also
D̃s,p (Ω) = (D̃s,p (Ω),  · ˜s )
is a reflexive Banach space. For simplicity and abuse of notation, in the following
we still denote by u the extension of every function u ∈ D̃s,p (Ω), by setting u = 0
in RN \Ω.
From now on we endow the space D̃s,p (Ω) with the weighted Gagliardo norm
  1/p
u = |u(x) − u(y)|p K(x − y)dx dy .
R2N
By virtue of (K ), it is easy to see that the previous norm is equivalent to the norm
·˜s , with cK ups ≤ up ≤ CK ups , for all u ∈ D̃s,p (Ω).
The assumption (K ) implies at once that mK ∈ L1 (RN ), where m(x) =
min{1, |x|p }, so that in particular ϕ < ∞ for all ϕ ∈ C02 (Ω). We refer to [31,
Lemma 11] for a detailed proof. 
The natural solution space D̃s,p (Ω) = D̃s,p (Ω),  ·  of (Pλ ) is a uniformly
convex Banach space by Proposition A.13. As in [29], from now on we denote by

D̃−s,p (Ω) the dual space of D̃s,p (Ω) and by ·, · the duality pairing between D̃s,p (Ω)
 
and D̃−s,p (Ω), i.e. ·, · : D̃s,p (Ω) × D̃−s,p (Ω) → R.
In the first part of the paper we assume that the coefficient a is a positive weight
of class Lα (Ω), with α > N/ps. Let
Lp (Ω, a) = {u : Ω → R measurable : a|u|p ∈ L1 (Ω)}.
This space is endowed with the norm ·p,a which is defined by
 1/p
up,a = p
a(x)|u(x)| dx .
Ω
Also the space Lp (Ω, a) is a uniformly convex Banach space. By [13, Corollary 7.2]

the embedding D̃s,p (Ω) → Lα p (Ω) is compact, being αp < p∗ by the assumption
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 7


that α > N/ps. Moreover, the embedding Lα p (Ω) → Lp (Ω, a) is continuous, since

upp,a ≤ aα upαp for all u ∈ Lα p (Ω) by the Hölder inequality. Hence,
the embedding D̃s,p (Ω) → Lp (Ω, a) is compact. (2.3)
Let λ1 be the first eigenvalue of the problem


LK u = λa(x)|u|p−2 u in Ω,
(2.4)
u=0 in RN \Ω,
given by the Rayleigh quotient

by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

R2N |u(x) − u(y)|p K(x − y)dxdy

λ1 = inf  . (2.5)
u∈D̃s,p (Ω), u=0 Ω
a(x)|u|p dx
By [15, Lemma 2.1] (see also [22, Theorem 5] for the fractional p-Laplacian first
eigenvalue) the infimum in (2.5) is achieved and λ1 > 0, when a ≡ 1. We refer also
Anal. Appl. Downloaded from www.worldscientific.com

to [14] for the special linear case of the fractional Laplacian and a ∈ Lip(Ω). The
general Proposition 1 of [29] proves the same result when a ∈ Lα (Ω), α > N/ps,
but with an argument completely different and more direct than the one given in
[14, 15, 22]. In particular,
λ1 = min up , (2.6)
u∈D̃s,p (Ω)
up,a =1

so that
λ1 upp,a ≤ up for every u ∈ D̃s,p (Ω). (2.7)
We now construct a suitable sequence of eigenvalues (λk )k of (2.4), which will
be useful in the next sections, in order to solve problem (Pλ ) for λ ≥ λ1 . The
construction is based on a method involving the Alexander–Spanier cohomology
with Z2 -coefficients (see [27, 33] for a complete overview on the topic). Put
M = {u ∈ D̃s,p (Ω) : J (u) = 1}, J (u) = upp,a. (2.8)
The eigenvalues and the eigenfunctions of (2.4) coincide, respectively, with the
critical points and the critical values of the constrained functional Ĩ = I M with
I(u) = up . Following [3], we prove some useful relationships.
The set M in (2.8) is a closed Z2 -invariant Finsler manifold in D̃s,p (Ω) of class
1
C . Of course M = ∅, being D̃s,p (Ω) compactly embedded in Lp (Ω, a). Furthermore,
I(u) = up is even and of class C 1 (D̃s,p (Ω)) and so of class C 1 (M). Let Fk and λk
be respectively the corresponding sets and numbers defined as in Proposition A.4.
Now, by [9, Proposition 3.1] we get that
i(A) = sup{i(K) : K ⊂ A is compact and symmetric} (2.9)
for all open symmetric subsets A of M. Hence
λk = inf max I(u),
K∈Gk u∈K
(2.10)
Gk = {K ⊂ M : K is compact and symmetric, with i(K) ≥ k}.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

8 P. Piersanti & P. Pucci

To prove the validity of (2.10) we first show that Gk = ∅ for all k. Fix k ∈ N. Since
meas(Ω) > 0, where meas denotes the standard Lebesgue measure, there exist k
open balls B1 , . . . , Bk such that Bi ∩ Bj = ∅ if i = j and
meas(Ω ∩ Bi ) > 0 for all i = 1, . . . , k.
Choose ui ∈ C0∞ (RN ),
with supp ui ⊂ Ω ∩ Bi and J (ui ) > 0 for every i = 1, . . . , k.
Normalize ui so that J (ui ) = 1 and let Ek be the span of {ui }ki=1 . For every u ∈ Ek ,
k
we have u = i=1 αi ui and
 k p k  k
J (u) = S |αi |p
a(x)|ui |p dx = |αi |p .
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

a(x) αi ui dx =
k
i=1 supp ui i=1 i=1 supp ui i=1
1/p
Hence u → J (u) defines a norm on Ek which is equivalent to the norm  · |Ek ,
being Ek finite dimensional. Therefore, K = {u ∈ Ek : J (u) = 1} is compact,
Anal. Appl. Downloaded from www.worldscientific.com

symmetric and i(K) = k by (i4 ) of Definition A.2. In conclusion, K ∈ Gk = ∅, as

required.
Now, let us call the right-hand side of (2.10) by µk . Of course, µk ≥ λk . If
µk > λk , there would exist M ∈ Fk , with i(M ) ≥ k, such that
M ⊂ {u ∈ M : I(u) < µk }.
Hence, there exists a compact and symmetric subset K of {u ∈ M : I(u) < µk },
with i(K) ≥ k by (2.9). Therefore, we get the required contradiction µk < µk , and
so (2.10) holds.
Let us show that λ1 = λ̃1 , where λ1 is defined in (2.6) and λ̃1 denotes the
number in (2.10) when k = 1, that is
λ̃1 = inf I(uK ).
K∈G1

Let u1 be the first eigenfunction with eigenvalue λ1 . Hence J (u1 ) = 1. Furthermore,

using the notation above, {u ∈ E1 = span{u1 } : J (u) = 1} = {u1 } is compact,
symmetric and with cohomological index 1. Therefore, λ̃1 ≤ λ1 = I(u1 ). On the
other hand, I(uK ) ≥ I(u1 ) = λ1 for all K ∈ G1 , since K ⊂ M and u1 is the first
eigenfunction in (2.6). Therefore, λ̃1 ≥ λ1 and in turn λ̃1 = λ1 , as claimed.
Since M = J −1 (1), where J (u) = upp,a , then u ∈ M is a critical point of I M
if and only if there is some µ ∈ R such that I  (u) = µJ  (u), J  (u) = pF (u), where
F (u) = a(x)|u|p−2 u. Of course, I  (u) = 0 whenever u ∈ M, so that µ = 0. Actually
I  (u) = 0 if and only if u = 0 ∈ D̃s,p (Ω). Similarly, every λk ∈ R+ for all k.
Now, if we show that I|M verifies the (P S)c condition for all c ∈ R+ , that is
any (un )n ⊂ M, with I(un ) → c and I  (un )D̃−s,p (Ω) → 0 as n → ∞, admits a
subsequence which converges in D̃s,p (Ω), then I M
verifies the (P S) condition and
the crucial Proposition A.4 can be applied.

Proposition 2.1. The C 1 functional I M

verifies the (P S)c condition for all c,
λ1 = min I(u), λk ∈ R+ , lim λk = ∞.
u∈M k→∞
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 9

For all k, with λk < λk+1 , we have

i({u ∈ D̃s,p (Ω)\{0} : I(u) ≤ λk J (u)})
= i({u ∈ D̃s,p (Ω) : I(u) < λk+1 J (u)}) = k.

Proof. Let c ∈ R and (un )n ⊂ M be a (P S)c sequence of I M

. Hence there exists
a suitable sequence (µn )n ⊂ R such that

I(un ) → c and I  (un ) − pµn F (un ) → 0 in D̃−s,p (Ω),
with F (u) = a(x)|u|p−2 u. Clearly, (un )n is bounded in D̃s,p (Ω) and so we can
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

pass to a subsequence un  u in D̃s,p (Ω), because D̃s,p (Ω) is reflexive. Hence,

I(un ) → I(u) in L1 (Ω) by Proposition A.12 with w = a, so that up,a = 1, and
in turn u = 0 and I(u) > 0. Now,
Anal. Appl. Downloaded from www.worldscientific.com

p(I(un ) − µn ) = I  (un ), un − pµn F(un ), un → 0.

In fact, straightforward computations show that


I (un ), un = |un (x) − un (y)|p K(x − y)dx dy = pun p
R2N

and
µn F(un ), un = un pp,a .
Hence, µn → c and 0 < I(u) ≤ c by the weak lower semicontinuity of the norm
in D̃s,p (Ω) (see [29, Lemma 1; Theorem 9.2-3]). Therefore, c > 0. Taking as test
function ϕ = un − u ∈ D̃s,p (Ω), we obtain
I  (un ), un − u − pµn F(un ), un − u
= I  (un ) − I  (u), un − u − pµn F(un ) − F(u), un − u
+ I  (u), un − u − pµn F(u), un − u
for all n ∈ N. Consequently, 0 ≤ I  (un ) − I  (u), un − u → 0 as n → ∞ by
Proposition A.12. Clearly, I  (u), un − u → 0 as n → ∞, since un  u in D̃s,p (Ω).
Hence
I  (un ), un − u → 0 as n → ∞. (2.11)
Moreover, I(u) ≤ lim inf n I(un ) due to the weak lower semicontinuity of I on
D̃s,p (Ω) (see [29, Lemma 1]). On the other hand, by convexity,
I(u) + I  (un ), un − u ≥ I(un ),
so that I(u) ≥ lim supn I(un ). In other words, I(u) = limn I(un ). Thanks to the
uniform convexity of D̃s,p (Ω) by Proposition A.13, it follows that un − u → 0
as n → ∞, as required. In other words, I M verifies the (P S)c condition for all
c ∈ R+ . An application of Proposition A.4 to the even functional I shows that
λk → ∞ as k → ∞.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

10 P. Piersanti & P. Pucci

Assume now that k ≥ 1 and λk < λk+1 . Put

C = {u ∈ M : I(u) ≤ λk } and U = {u ∈ M : I(u) < λk+1 }.
From now on the proof can follow word by word the abstract proof of [9, Theo-
rem 3.2] by virtue of Proposition A.12. However, we repeat the argument for the
convenience of the reader. Clearly, i(C) ≤ k ≤ i(U ). Assume by contradiction that
i(C) ≤ k − 1. By property (i3 ) of Definition A.2 there exists a symmetric neighbor-
hood W of C with i(W ) = i(C). Moreover W is a neighborhood of the critical set
of I M at level λk . By the (P S) conditions and so by the deformation lemma there
exist ε > 0 and an odd continuous map
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

ϕ : {u ∈ M : I(u) ≤ λk + ε} → {u ∈ M : I(u) ≤ λk − ε} ∪ W = W.
Therefore i({u ∈ M : I(u) ≤ λk + ε}) ≤ k − 1, which is impossible by the
definition (2.10) of λk . In conclusion, i(C) = k as desired. Furthermore, by (i2 ) of
Anal. Appl. Downloaded from www.worldscientific.com

Definition A.2 we also get

i({u ∈ D̃s,p (Ω)\{0} : I(u) ≤ λk J (u)}) = i(C) = k.
Suppose now by contradiction that i(U ) ≥ k + 1. By (2.9) there exists a sym-
metric compact subset K of U with i(K) ≥ k + 1. But max{I(u) : u ∈ K} =
I(uK ) < λk+1 , being uK ∈ K ⊂ U . This is impossible by definition (2.10) of λk+1 .
Consequently, i(U ) = k as desired. Again by (i2 ) of Definition A.2 we have
i({u ∈ D̃s,p (Ω) : I(u) < λk+1 J (u)}) = i(U ) = k.
This concludes the proof.

3. Proof of Theorem 1.1

In this section we prove the main theorem (Theorem 1.1) for (Pλ ), when a is in
Lα (Ω), α > N/ps. The (weak) solutions of (Pλ ) coincide with the critical points
u ∈ D̃s,p (Ω) of the C 1 -functional

up λ
Φ(u) = − up,a −
p
G(x, u)dx, u ∈ D̃s,p (Ω), (3.1)
p p Ω
provided that g is assumed to be a Carathéodory function with primitive G defined
in (g2 ). Following somehow [28], we prove Theorem 1.1 with some preparatory
lemmas.

Lemma 3.1. Assumptions (g1 ) and (g2 ) imply that


u−p |G(x, u)|dx → 0 as u → 0.
Ω

Proof. Let us first introduce the auxiliary function G0 : Ω×R → R, G0 = G0 (x, u),
defined by
|G(x, u)|
G0 (x, u) = if u = 0, and G0 (x, 0) = 0.
|u|p
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 11

Clearly G0 is a Carathéodory function on Ω × R by (g1 ) and (g2 ). We claim that

there exists a real positive constant C depending on the data in (g1 ) such that
0 ≤ G0 (x, u) ≤ C(g0 (x) + |u|r−p ),
for a.a. x ∈ Ω and all u ∈ R. Indeed, by (g1 )
g0 (x) Lg1
0 ≤ G0 (x, u) ≤ + |u|r−p ,
p r
for a.a. x ∈ Ω and all u = 0, with Lg1 = g1 ∞ . The claim follows at once, setting
C = max{1/p, Lg1 /r}. Therefore, by the Hölder inequality

by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

0≤ |G(x, u(x))|dx
Ω
  (r−p)/r
= G0 (x, u(x))|u|p dx ≤ G0 (x, u)r/(r−p) dx upr ,
Anal. Appl. Downloaded from www.worldscientific.com

Ω Ω
with

G0 (x, u)r/(r−p) dx → 0 as u → 0,
Ω
by Lemma A.11(ii), with A = Ω, d = N , ω = 1, α = r, β = r/(r − p), so that the
Nemytskii operator NG0 = G0 (x, u), NG0 (0) = 0, maps continuously Lr (Ω) into
Lr/(r−p) (Ω). In conclusion,

|G(x, u)|dx = o(up ) as u → 0,
Ω

since the immersion D̃s,p (Ω) → Lr (Ω) is even compact by virtue of [13, Corol-
lary 7.2], see also [24, 29]. This completes the proof.

Lemma 3.2. Assume that (g2 ) and (g3 ) hold. Let ω be a weight on Ω such that
the embedding D̃s,p (Ω) → Lp (Ω, ω) is compact. Suppose furthermore that there
exist a sequence (un )n ⊂ D̃s,p (Ω) and a constant c > 0 independent of n such that
un  ≤ cun p,ω for all n ∈ N. If un  → ∞, then

un −p G(x, un )dx → ∞ as n → ∞. (3.2)
Ω

Proof. Let (un )n be as in the statement, so that in particular un  → ∞ as

n → ∞, and rearranging the sequence we can suppose, without loss of generality,
that un  > 0 for all n. Hence the sequence n → vn = un /un is well defined and
is contained in the unit sphere of D̃s,p (Ω). By the reflexivity of the space D̃s,p (Ω),
up to a subsequence, still denoted for simplicity by (vn )n , there is v ∈ D̃s,p (Ω) such
that vn  v in D̃s,p (Ω), vn → v in Lp (Ω, γ1 ), vn → v a.e. in Ω and |vn | ≤ h for
some h ∈ Lp (Ω, γ1 ) and vn → v in Lp (Ω, ω) by assumption. From the second part
of Proposition A.12 we get
1 ≤ c lim vn p,ω = cvp,ω ,
n→∞
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

12 P. Piersanti & P. Pucci

since the embedding D̃s,p (Ω) → Lp (Ω, ω) is compact. Hence v ≡ 0 and the set

A = {x ∈ Ω : |v(x)| > 0}

has positive measure. Of course

G(x, un (x)) G(x, un vn (x))
= g0 (x)|vn (x)|p .
un  p g0 (x)un p |vn (x)|p
Therefore, limn un −p G(x, un (x)) = ∞ a.e. in A by (g2 ). By (g3 ) for all n suffi-
ciently large
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

G(x, un (x)) γ0 (x) |un |p

≥ − − γ1 (x) ≥ −γ0 (x) − γ1 (x)|vn |p ≥ −γ0 (x) − γ1 (x)hp (x)
un p un p un p
and γ0 + γ1 hp ∈ L1 (Ω). Therefore, we get (3.2) by the generalized Fatou lemma.
Anal. Appl. Downloaded from www.worldscientific.com

This concludes the proof.

We are going to prove that the functional Φ defined in (3.1) satisfies the (P S)
condition, using [3, Lemma 4.2]; that is the following lemma.

Lemma 3.3. Assume that (g1 ) is valid. If (un )n is bounded in Lr̃ (Ω) and converges
 
a.e. in Ω to some u ∈ Lr̃ (Ω), then Ng (un ) → Ng (u) in Lr̃ (Ω) and so in D̃−s,p (Ω).

Proof. Clearly from (g1 ) we get |u|1−r̃ g(x, u) → 0 as |u| → ∞ uniformly in Ω\N
for some N , with meas(N ) = 0. We claim that for all ε > 0 there is Cε > 0 such
that

|g(x, u)| ≤ Cε g0 (x)|u|p−1 + ε|u|r̃−1 a.e. in Ω and for all u ∈ R.

Indeed, for all ε > 0 there exists Mε > 0 such that |u|1−r̃ |g(x, u)| ≤ ε for a.a. x ∈ Ω
and for all u with |u| ≥ Mε . On the other hand, by (g1 ) for a.a. x ∈ Ω and for all
u with |u| ≤ Mε

|g(x, u)| ≤ g0 (x)|u|p−1 + Cg g0 (x)|u|r−1 ≤ Cε g0 (x)|u|p−1 ,

where Cε = 1 + Cg Mεr−p . Therefore the claim is proved. Now, applying the Young
inequality on the first term of the latter summation, we get
 (r̃−1)/(r̃−p)  
r̃ − p Cε g0 (x) p − 1 (r̃−1)/(p−1)
|g(x, u)| ≤ + ε + ε |u|r̃−1 .
r̃ − 1 ε r̃ − 1
Hence, by Lemma A.11(iii), with A = Ω, d = N , w = 1, α = r̃ and β = r̃ , we get
the assertion at once, being g0 ∈ Lr̃/(r̃−p) (Ω) by (g1 ), being r < r̃ and Ω bounded.
 
Naturally the strong convergence in Lr̃ (Ω) implies the convergence in D̃−s,p (Ω),
since the embedding D̃s,p (Ω) → Lr̃ (Ω) is compact, being r̃ < p∗ .

Lemma 3.4. Assume that (g1 )–(g3 ) hold. For all λ ∈ R the functional Φ satisfies
the (P S) condition.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 13

Proof. Let (un )n ⊂ D̃s,p (Ω) be a (P S)c sequence for Φ for any fixed c ∈ R. We
claim that (un )n is bounded in D̃s,p (Ω). Assume by contradiction that un  → ∞
as n → ∞ up to a subsequence. Clearly by (g3 )
 
µ
µΦ(un ) − Φ (un ), un = − 1 (un p − λun pp,a )
p

− [µG(x, un ) − un g(x, un )]dx
Ω

≥ 2κun p − [γ0 (x) + {2κ|λ|a(x) + γ1 (x)}|un |p ]dx,
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Ω
where 2κ = µ/p − 1 > 0. Now, since un  → ∞ as n → ∞, we have
 
Φ(un )  un 1
− Φ (un ), + γ0 (x)dx → 0
un p un p un p Ω
Anal. Appl. Downloaded from www.worldscientific.com

as n → ∞. Then, for all n sufficiently large,



µΦ(un ) − Φ (un ), un + γ0 (x)dx ≤ κun p .
Ω
Combining together the last two inequalities, we get
κun p ≤ un pp,w , w(x) = 2|λ|κa(x) + γ1 (x), (3.3)
for all n large enough. Applying Lemma 3.2, with w = 2|λ|κa + γ1 and c = 1/κ, we
derive that

−p
un  G(x, un )dx → ∞ as n → ∞.
Ω
In conclusion, since (2.7) holds, we have

Φ(un ) 1 |λ| Ω
G(x, un )dx
0 = lim ≤ + − lim = −∞,
n→∞ un p p pλ1 n→∞ un p
which is the required contradiction.
Therefore, (un )n is bounded in D̃s,p (Ω) and so there exists u ∈ D̃s,p (Ω) such
that, up to a subsequence, un  u in D̃s,p (Ω), un → u a.e. in Ω and un → u in
Lr̃ (Ω), with r̃ < p∗ . Hence, as n → ∞
o(1) = Φ (un ), un − u
1 
= I (un ), un − u − λ F(un ), un − u
p

− g(x, un )(un − u)dx,
Ω
since (un )n is a (P S)c sequence. Then, by Proposition A.12 and Lemma 3.3
I  (un ), un − u → 0 as n → ∞;
i.e. (2.11) holds. Thus, using the same argument of Proposition 2.1, we get that
un → u in D̃s,p (Ω) as n → ∞. The lemma is so proved.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

14 P. Piersanti & P. Pucci

Lemma 3.5. Assume that (g1 )–(g3 ) hold and that 0 is an isolated critical point of
Φ in D̃s,p (Ω).
If λ < λ1 , then C k (Φ, 0) ≈ δk0 Z2 for all k ≥ 0.

Proof. Let λ < λ1 be fixed. By Lemma 3.4 and Theorem A.7 it is enough to show
that Φ has a local minimizer at 0 in D̃s,p (Ω). Certainly Φ(0) = 0. For sufficiently
small ρ > 0, by virtue of Lemma 3.1 and (2.7), we have
 
max{λ, 0} up
Φ(u) ≥ 1 − + o(1) ≥ 0 for all u ∈ Bρ ,
λ1 p
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

with Bρ the closed ball of center 0 and radius ρ > 0 in D̃s,p (Ω), see Definition A.5.
This completes the proof thanks to Theorem A.7.

Lemma 3.6. Assume that (g1 ) and (g2 ) hold. If 0 is an isolated critical point of
Anal. Appl. Downloaded from www.worldscientific.com

Φ in D̃s,p (Ω), then, assumed k ∈ N, C k (Φ, 0) = 0 if either

(i) λk < λ < λk+1 , or
(ii) λ = λk < λk+1 , and G(x, u) ≥ 0 for a.a. x ∈ Ω and all u ∈ R.

Proof. We show that Φ has a cohomological local splitting near 0 in dimension k

and apply Proposition A.8. Let
E− = {u ∈ D̃s,p (Ω) : up ≤ λk upp,a },
(3.4)
E+ = {u ∈ D̃s,p (Ω) : up ≥ λk+1 upp,a }.
Then (A.4) holds by Proposition 2.1, so that it only remains to show that (A.5)
holds for sufficiently small ρ > 0. Here Φ(0) = 0.
Case (i). For sufficiently small ρ > 0,
 
λ up
Φ(u) ≤ − − 1 + o(1) ≤ 0 for all u ∈ Bρ ∩ E−
λk p
and
 
λ up
Φ(u) ≥ 1 − + o(1) ≥0 for all u ∈ Bρ ∩ E+
λk+1 p
by Lemma 3.1.
Case (ii). For all u ∈ E− ,

Φ(u) ≤ − G(x, u) dx ≤ 0,
Ω
and Φ(u) ≥ 0 for all u ∈ E+ . Indeed, by Lemma 3.1, we have, for sufficiently small
ρ > 0,
 
λk up
Φ(u) ≥ 1 − + o(1) ≥0
λk+1 p
for all u ∈ Bρ ∩ E+ . The proof is now complete.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 15

Lemma 3.7. Assume (g1 )–(g3 ). Then there exists b < 0 such that
H k (D̃s,p (Ω), Φb ) = 0
for all k ≥ 0, where Φb = {u ∈ D̃s,p (Ω) : Φ(u) ≤ b}.

Proof. Fix u ∈ D̃s,p (Ω), with u = 0. We claim that for all u ∈ D̃s,p (Ω), with u = 0,
Φ(tu)
→ −∞ as t → ∞. (3.5)
|t|p
Indeed, by (g3 ), for all t ∈ R, with |t| ≥ 1,
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

G(x, tu) γ0 (x)

µ ≥ − p − γ1 (x)|u|p ≥ −γ0 (x) − γ1 (x)|u|p = −h(x), (3.6)
|t|p |t|
where h is a nonnegative function of class L1 (Ω), since the embedding D̃s,p (Ω) →
Lp (Ω, γ1 ) is compact, being α > N/ps by (g3 ). Moreover, since |u| > 0 in a subset
Anal. Appl. Downloaded from www.worldscientific.com

A of Ω, with meas(A) > 0, by (g2 ) as t → ∞

G(x, tu(x)) G(x, tu(x))
= g0 (x)|u(x)|p → ∞ a.e. in A.
|t|p g0 (x)|t|p |u(x)|p
Therefore the generalized Fatou lemma implies at once that

lim |t|−p G(x, tu(x))dx = ∞.
t→∞ Ω
In particular,

Φ(tu) 1
= (up − λupp,a ) − |t|−p G(x, tu(x))dx → −∞
|t|p p Ω
as t → ∞ and this shows the claim (3.5).
We assert that for all c ∈ R
Φ (u), u − pΦ(u)
lim sup < 0. (3.7)
u→∞, Φ(u)≤c up
Fix c ∈ R. As in [10, Lemma 3.1] we assume by contradiction that there exist
sequences (dn )n ⊂ R and (un )n ⊂ Φc such that dn → 0, un  → ∞ as n → ∞ and
Φ (un ), un − pΦ(un ) ≥ −dn un p for all n ∈ N. (3.8)
Then,
−dn un p ≤ Φ (un ), un − pΦ(un )
 
= [µG(x, un ) − un g(x, un )]dx − (µ − p) G(x, un )dx
Ω Ω
 
≤ (γ0 + γ1 |un |p )dx − (µ − p) G(x, un )dx. (3.9)
Ω Ω
Hence, we get at once that

Ω G(x, un )dx
lim sup < ∞, (3.10)
n un p
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

16 P. Piersanti & P. Pucci

since

Ω
G(x, un )dx γ0 1 un pp,γ1
(µ − p) ≤ dn + + ,
un p un p un p
µ > p, dn → 0, un  → ∞ as n → ∞ and the embedding D̃s,p (Ω) → Lp (Ω, γ1 ) is
compact by (g3 ).
Clearly vn = un /un  is in the unit sphere of D̃s,p (Ω) for all n sufficiently
large. Therefore, up to a subsequence, still denoted for simplicity by (vn )n , there is
v ∈ D̃s,p (Ω) such that vn  v in D̃s,p (Ω), vn → v in Lp (Ω, γ1 ) and vn → v a.e. in
Ω, and there exists h ∈ Lp (Ω, γ1 ), such that |vn | ≤ h a.e. in Ω and for all n ∈ N.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

On the other hand, as shown above, by (g2 ) at any point x ∈ Ω at which v(x) = 0
G(x, un (x)) G(x, un vn (x))
lim = lim g0 (x)vn (x) = ∞.
n un p n g0 (x)un p vn (x)

If |v(x)| > 0 a.e. in Ω, by (g3 ) and the generalized Fatou lemma, which is applicable,
Anal. Appl. Downloaded from www.worldscientific.com

being for n sufficiently large, say un  ≥ 1,

G(x, un )
≥ −γ0 (x) − γ1 (x)|vn |p ≥ −γ0 (x) − γ1 (x)hp ∈ L1 (Ω),
un p
we would get by (3.10)
 
G(x, un ) G(x, un )dx
∞= lim dx ≤ lim inf Ω < ∞,
Ω n un  p n un p
which is the required contradiction. Therefore the limit v is trivial in D̃s,p (Ω), that
is v = 0 a.e. in Ω.
Clearly (3.9) can be rewritten in the equivalent form
 
(µ − p) G(x, un )dx ≤ (γ0 + γ1 |un |p )dx + dn un p . (3.11)
Ω Ω
We claim that (3.11) implies
 
µ
− 1 un p ≤ (µ − p)Φ(un ) + dn un p + γ0 1
p
   
µ
+ − 1 λa(x) + γ1 (x) |un |p dx. (3.12)
Ω p
Indeed, denoting by R the right-hand side of (3.12), then R, by definition of Φ,
can be written as
  
un p λ
R = (µ − p) − un p,a −
p
G(x, un )dx + dn un p + γ0 1
p p Ω
    
µ
+ γ1 (x) + − 1 λa(x) |un |p dx
Ω p
  
µ
= − 1 un p − (µ − p) G(x, un )dx + dn un p
p Ω
  
µ
+ (γ0 (x) + γ1 (x)|un | dx ≥
p
− 1 un p
Ω p
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 17

by (3.11). This shows the claim (3.12). Dividing (3.12) by un p , we obtain
   
µ Φ(un ) γ0 1 µ
− 1 ≤ (µ − p) + d n + + − 1 λa(x) + γ 1 (x) |vn |p dx
p unp un p Ω p
   
c γ0 1 µ
≤ (µ − p) + dn + + − 1 λa(x) + γ 1 (x) |vn |p dx,
un p un p Ω p
since (un )n ⊂ Φc and µ > p by (g3 ). Now, passing to the limit as n → ∞ in both
sides of the inequality above we get µ/p − 1 ≤ 0, since vn → 0 both in Lp (Ω, a) and
in Lp (Ω, γ1 ), being the embeddings D̃s,p (Ω) → Lp (Ω, a) and D̃s,p (Ω) → Lp (Ω, γ1 )
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

compact. This contradiction completes the proof of the assertion (3.7).

Applying (3.7), with c = 0, we see that there exists b1 ∈ R such that Φ (u), u −
p Φ(u) ≤ b1 for all u ∈ Φ0 . In particular, there exists b < 0 such that
Φ (u), u < 0 for all u ∈ Φb (3.13)
Anal. Appl. Downloaded from www.worldscientific.com

holds. Indeed, it is enough to take b < min{0, −b1/p}.

Fix u ∈ D̃s,p (Ω), with u = 0. Therefore Φ(tu) ≤ b for all sufficiently large t by
(3.5). Set
T (u) = inf{t ≥ 1 : Φ(tu) ≤ b}.
Since
d 1
Φ(tu) ≤ b ⇒ (Φ(tu)) = Φ (tu), u = Φ (tu), tu < 0
dt t
by (3.13), then
Φb = {tu : u ∈ D̃s,p (Ω)\{0}, t ≥ T (u)}
and the map D̃s,p (Ω)\{0} → [1, ∞), u → T (u), is continuous.
Thus D̃s,p (Ω)\{0} is radially deformation retracted onto Φb via
(D̃s,p (Ω)\{0}) × [0, 1] → D̃s,p (Ω)\{0}, (u, τ ) → (1 − τ )u + τ T (u)u,
and hence
H k (D̃s,p (Ω), Φb ) = H k (D̃s,p (Ω), D̃s,p (Ω)\{0}) = 0 for all k ≥ 0.
This completes the proof.

We are now ready to prove our main result.

Proof of Theorem 1.1. Suppose by contradiction that 0 is the only critical point
of Φ. The functional Φ ∈ C 1 (D̃s,p (Ω)) satisfies the (P S) condition by Lemma 3.4.
Let b < 0 be as in Lemma 3.7. Since there is no critical value for Φ in [b, 0), by
[27, Lemma 3.9] the set Φb is a deformation retract of Φ0 \{0}. The set Φ0 is a
deformation retract of D̃s,p (Ω) since there is no critical value in R+ . Thus, taking
U = D̃s,p (Ω) in (A.3) we have for all k ≥ 0
C k (Φ, 0) = H k (Φ0 , Φ0 \{0}) ≈ H k (D̃s,p (Ω), Φb ) = 0.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

18 P. Piersanti & P. Pucci

However, C k (Φ, 0) = 0 for some k ≥ 0 by Lemmas 3.5 and 3.6. This contradiction
completes the proof.

4. Proof of Theorem 1.2

In this section, following [10, 18, 30, 32], we are going to prove Theorem 1.2 for
(Pλ ), when G is allowed to be possibly negative locally at 0 in the resonance case,
but requiring that the weight a is in L∞ (Ω). Again the (weak) solutions of (Pλ )
are exactly the critical points of the functional Φ defined in (3.1), where (g1 ) and
(g2 ) are replaced by the stronger assumptions (g1 ) and (g2 ) of the Introduction,
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

respectively.
Also in this section we extend several previous results, see [18, Theorem 3.1 and
Corollary 3.2; 30, Proposition 4; 32, Proposition 7], to the general Dirichlet problem


LK u = f (x, u) in Ω
Anal. Appl. Downloaded from www.worldscientific.com

(Pf )
u=0 in RN \Ω,
where f satisfies the main request
(H ) f is a Carathéodory function in Ω × R and for appropriate exponents p, r,
with p < r < p∗ , and a suitable positive constant Cf
|f (x, u)| ≤ Cf (|u|p−1 + |u|r−1 )
holds for all (x, u) ∈ Ω × R.
In what follows a crucial role is played by the numbers
 r/p
2 2 2 2Cf
σ = H −(N/rs) /r
, H = (1 + κ 2r+r s/N
)2r+r s/N
, κ= S, (4.1)
cK
where Cf and r are given in (H ), cK > 0 is the main parameter of assumption
(K ) in the Introduction, and S is the Sobolev constant of the fractional embedding
∗
of D̃s,p (Ω) into Lp (Ω). Clearly, the number σ ∈ (0, 1), being H > 1, and σ depends
on N , p, s, Ω, cK , Cf and r.
Lemma 4.1. Every solution u ∈ D̃s,p (Ω) of (Pf ), with ur < σ, is in L∞ (Ω)
and
σu∞ ≤ ur .

Proof. Let u be a solution of (Pf ), with ur < σ. We may assume that u does
not vanish identically (otherwise there is nothing to prove). Of course, there exists
an appropriate  ∈ (0, 1), such that
ur = σ. (4.2)
Set v = u/, so that v ∈ D̃ (Ω) and vr = σ. For all n ∈ N we define vn =
s,p

(v − 1 + 2−n )+ and Vn = vn rr . Clearly v0 = v + and

0 ≤ vn+1 (x) ≤ vn (x) and lim vn (x) = (v(x) − 1)+ (4.3)
n→∞
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 19

for a.a. x ∈ Ω. We now claim that the following set inclusion holds
{x ∈ Ω : vn+1 (x) > 0} ⊂ {x ∈ Ω : 0 < v(x) < (2n+1 − 1)vn (x)}
(4.4)
∩{x ∈ Ω : vn (x) > 2−n−1 }.
We first prove that
{x ∈ Ω : vn+1 (x) > 0} ⊂ {x ∈ Ω : 0 < v(x) < (2n+1 − 1)vn (x)}.
To this aim fix x ∈ Ω, with vn+1 (x) > 0. Hence (v(x) − 1 + 2−n−1 )+ > 0, that
is v(x) > 1 − 2−n−1 ≥ 1/2, so the first inequality is proved. It remains to show
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

the inequality v(x) < (2n+1 − 1)vn (x). Since vn+1 (x) > 0 and vn+1 (x) = v(x) −
1 + 2−n−1 , then v(x) > 1 − 2−n−1 , that is 2n+1 v(x) > 2n+1 − 1. Multiplying by
(1 − 2n )2−n we get
(2n+1 − 1)(1 − 2n )
Anal. Appl. Downloaded from www.worldscientific.com

(2 − 2n+1 )v(x) < .

2n
In particular,
(2n+1 − 1)(1 − 2n )
(2n+1 − 1)vn (x) = (2n+1 − 1)v(x) +
2n
(2n+1 − 1)(1 − 2n )
= v(x) − (2 − 2n+1 )v(x) +
2n
> v(x),
as stated. The claim is so proved.
It remains to show that
{x ∈ Ω : vn+1 (x) > 0} ⊂ {x ∈ Ω : vn (x) > 2−n−1 }.
To this goal fix x ∈ Ω, with vn+1 (x) > 0. Thus, as shown above, we have v(x) −
1 + 2−n−1 > 0, that is v(x) − 1 + 2−n − 2−n + 2−n−1 > 0, or
vn (x) = v(x) − 1 + 2−n > 2−n − 2−n−1 = 2−n−1 ,
as claimed. In conclusion (4.4) holds.
We next show that Vn → 0 as n → ∞. We recall that (Vn )n is a nonincreasing
sequence in [0, 1]. For brevity, put {vn > 0} = {x ∈ Ω : vn (x) > 0} for every n ∈ N.
By the Hölder and the fractional Sobolev inequalities (see [13, Theorem 6.5]) and
(4.4), we get for all n ∈ N
 (p∗ −r)/p∗  r/p∗
∗
/(p∗ −r) ∗
Vn+1 ≤ 1p dx |vn+1 |r·p /r
dx
{vn+1 >0} {vn+1 >0}
∗
= |{vn+1 > 0}|1−r/p vn+1 rp∗
  r/p
∗ |vn+1 (x) − vn+1 (y)|p
≤ S|{vn+1 > 0}|1−r/p dxdy .
R2N |x − y|N +ps
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

20 P. Piersanti & P. Pucci

Furthermore, |{vn+1 > 0}| ≤ |{vn > 2−(n+1) }| by (4.4). Hence

  r/p
|vn+1 (x) − vn+1 (y)|p
∗
Vn+1 ≤ S|{vn+1 > 0}|1−r/p dxdy
R2N |x − y|N +ps
  r/p
S −r(n+1) 1−r/p∗ |vn+1 (x) − vn+1 (y)|p cK
≤ r/p |{vn > 2
r
}| dxdy
c R2N |x − y|N +ps
K
  r/p
S −r(n+1) 1−r/p∗
≤ r/p
|{vnr >2 }| |vn+1 (x) − vn+1 (y)| K(x − y)dxdy
p
cK R2N
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

S ∗
= r/p
|{vnr > 2−r(n+1) }|1−r/p vn+1 r ,
cK

where cK is such that cK |x|−N −ps ≤ K(x) for all x ∈ RN \{0} by the main assump-
Anal. Appl. Downloaded from www.worldscientific.com

tion (K ), stated in the Introduction. Now, by the Chebyshev inequality, we have

S ∗ S 2
/p∗ )(n+1) ∗

r/p
|{vnr > 2−r(n+1) }|1−r/p vn+1 r ≤ r/p
2(r−r Vn1−r/p vn+1 r .
cK cK

It remains to estimate vn+1 . Using the elementary inequality

|ξ + − α+ |p ≤ |ξ − α|p−2 (ξ − α)(ξ + − α+ ), for all ξ, α ∈ R,

and vn+1 as test function for (Pf ) and applying also (4.4), we obtain by (H )

vn+1  ≤ |v(x) − v(y)|p−2 · [v(x) − v(y)] · [vn+1 (x) − vn+1 (y)]K(x − y)dxdy
p
Ω
 
= f (x, v)vn+1 dx ≤ Cf (vp−1 + vr−1 )vn+1 dx
Ω {vn+1 >0}

≤ Cf [(2n+1 − 1)p−1 vnp + (2n+1 − 1)r−1 vnr ]dx
{vn+1 >0}

≤ Cf [(2n+1 − 1)p−1 |{vn+1 > 0}|1−p/r vn pr + (2n+1 − 1)r−1 vn rr ]
≤ Cf [(2n+1 − 1)p−1 |{vnr > 2−r(n+1) }|1−p/r vn pr + (2n+1 − 1)r−1 vn rr ]
≤ Cf [(2n+1 − 1)p−1 (2r(n+1) vn rr )1−p/r vn pr + (2n+1 − 1)r−1 vn rr ]
= Cf [(2n+1 − 1)p−1 2(r−p)(n+1) vn rr + (2n+1 − 1)r−1 vn rr ]
≤ Cf [2(n+1)(p−1)+(r−p)(n+1) + 2(n+1)(r−1) ] · vn rr
= 2Cf 2(r−1)(n+1) vn rr = 2Cf 2(r−1)(n+1) Vn ,

where Cf is the constant in (H ). Hence, the estimate for vn+1  is

vn+1 r ≤ (2Cf · 2(r−1)(n+1) Vn )r/p = Ĉ2(r−1)(n+1)r/p Vnr/p ,

 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 21

with Ĉ = (2 Cf )r/p . Since r/p − r/p∗ = rs/N , combining the inequalities above,
we get
S 2 ∗ ∗
Vn+1 ≤ r/p Ĉ2(r−r /p )(n+1) Vn1−r/p 2(r−1)(n+1)r/p Vnr/p
cK
2
/p∗ +r 2 /p−r/p)(n+1) ∗
= κ2(r−r Vn1−r/p +r/p

2
= κ2r(1−1/p)(n+1) 2(n+1)r s/N
Vn1+rs/N
2 2
s/N (r+r 2 s/N )n
≤ κ2r(n+1) 2(n+1)r s/N
Vn1+rs/N = κ2r+r 2 Vn1+rs/N
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

2 2
≤ [(1 + κ2r+r s/N
)2r+r s/N n
] Vn1+rs/N
= H n Vnβ ,
where H is given in (4.1) and
Anal. Appl. Downloaded from www.worldscientific.com

β = 1 + rs/N > 1. (4.5)

In conclusion we obtain the main recursive formula
Vn+1 ≤ H n Vnβ . (4.6)
We assert that
Vn ≤ σ r η n , η = H −N/rs ∈ (0, 1), (4.7)
by (4.1). Indeed, the assertion is trivially automatic if n = 0, being
V0 = v + rr ≤ vrr = σ r ,
by definition of v and (4.2). Now suppose that (4.7) holds true for n ≥ 0 and let us
prove it for n + 1. Indeed, by (4.5)–(4.7)
2
Vn+1 ≤ H n σ r η n σ r s/N nrs/N
η = σ r η n+1 ,
2
since σ r s/N = η by (4.1). This proves the assertion (4.7).
Clearly (4.7) implies at once that
lim Vn = 0.
n→∞

Therefore, vn → 0 in Lr (Ω). In particular, by (4.3) and the Fatou lemma

 
0 = lim Vn ≥ lim |vn (x)|r dx = [(v(x) − 1)+ ]r dx ≥ 0.
n→∞ Ω n→∞ Ω
+
Hence (v − 1) = 0 a.e. in Ω, that is v ≤ 1 a.e. in Ω.
Replacing now v with −v in the above argument, we get similarly that −v ≤ 1
a.e. in Ω. In conclusion, v∞ ≤ 1. Thus by the definition of v and (4.2)
σu∞ = σv∞ ≤ σ = ur .
This completes the proof of the lemma.

From now on we consider the main problem (Pλ ), with a ∈ L∞ (Ω) and g
satisfying (g1 ) and (g2 ) .
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

22 P. Piersanti & P. Pucci

Clearly g(x, 0) = 0 a.e. in Ω by (g2 ) , so that (Pλ ) admits the trivial solution
for all λ ∈ R. Moreover, by (g2 ) there exists η ∈ (0, 1) such that |g(x, u)| ≤ |u|p−1
for a.a. x ∈ Ω and for all u, with |u| ≤ η. While by (g1 ) for a.a. x ∈ Ω and for all
u, with |u| ≥ η,
 p−1 
η C
|g(x, u)| ≤ C + |u|r−1
≤ p−1 (|u|p−1 + |u|r−1 ).
η p−1 η
In conclusion, for all a.e. x ∈ Ω and all u ∈ R, we have
|g(x, u)| ≤ max{1, Cη 1−p }(|u|p−1 + |u|r−1 ).
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Put f (x, u) = λa(x)|u|p−2 u + g(x, u) for all (x, u) ∈ Ω × R. Therefore, for every
λ ∈ R,
|f (x, u)| ≤ |λ| · a∞ |u|p−1 + max{1, Cη 1−p }(|u|p−1 + |u|r−1 )
Anal. Appl. Downloaded from www.worldscientific.com

≤ Cf (|u|p−1 + |u|r−1 ), (4.8)

where Cf = |λ| · a∞ + max{1, Cη 1−p }. Hence f satisfies (H ) and the previous
lemma can be applied to problem (Pλ ).
Without further mentioning, we assume from now on that a ∈ L∞ (Ω), (g1 ) , (g2 )
and (g3 ) hold, and that α > N/ps in (g3 ) is no longer related to the coefficient a.
We also fix δ > 0 in a such a way that either (ii) or (iii) of Theorem 1.2 is satisfied.
Without loss of generality we suppose that δ ∈ (0, 1).
Following [18], we are going to prove the next lemmas of the section, putting
for all t ∈ [0, 1]
Gt (x, u) = G(x, (1 − t)u + tϑ(u)),
1
where ϑ ∈ C (R, [−δ, δ]) is a nondecreasing function such that


u if |u| ≤ δ/2,
ϑ(u) = −δ if u ≤ −δ, (4.9)


δ if u ≥ δ.
We get at once that
gt (x, u) = ∂u Gt (x, u) = ∂u G(x, (1 − t)u + tϑ(u))
= g(x, (1 − t)u + tϑ(u)) · [1 − t + tϑ (u)].
In particular, for a.a. x ∈ Ω, we have for all u ∈ R


g(x, u) if |u| ≤ δ/2,
gt (x, u) = (4.10)
(1 − t)g(x, (1 − t)u + tϑ(u)) if |u| ≥ δ.

Define Φt : D̃s,p (Ω) → R for all t ∈ [0, 1] by


up λ
Φt (u) = − upp,a − Gt (x, u)dx,
p p Ω
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 23

so that Φ = Φ0 , where as usually Φ is given in (3.1). Clearly, Φt is of class

C 1 (D̃s,p (Ω)) for all t ∈ [0, 1].

Lemma 4.2. If 0 is an isolated critical point of Φ in D̃s,p (Ω), then 0 is an isolated

critical point of Φt in D̃s,p (Ω), uniformly with respect to t ∈ [0, 1], and C k (Φ, 0) =
C k (Φ1 , 0) for all k ≥ 0.
Similarly, if 0 is an isolated critical point of Φ1 in D̃s,p (Ω), then 0 is an isolated
critical point of Φt in D̃s,p (Ω), uniformly with respect to t ∈ [0, 1], and C k (Φ, 0) =
C k (Φ1 , 0) for all k ≥ 0.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Proof. By assumption there exists  > 0 so small such that the closed ball B of
D̃s,p (Ω) contains only the critical point 0 of Φ. We claim that taking  > 0 even
smaller if necessary then implies
Anal. Appl. Downloaded from www.worldscientific.com

K (Φt ) ∩ B = {0} for all t ∈ [0, 1], (4.11)

where K (Φt ) denotes the set of critical points of the functional Φt .

Assume by contradiction that (4.11) fails. Hence, there exist a sequence (tn )n
in [0, 1] and (un )n in D̃s,p (Ω)\{0} such that Φtn (un ) = 0 for all n ∈ N and un → 0
in D̃s,p (Ω). From now on we write gn instead of gtn . Using (4.9)–(4.10), we get by
(g1 ) and (g2 ) that
gn (x, u)
lim = 0 uniformly in n, |gn (x, u)| ≤ Cϑ (1 + |u|r−1 )
u→0 |u|p−1
for a.a. x ∈ Ω and u ∈ R, where Cϑ = 2C(2 + ϑ ∞ ). Hence, putting fn (x, u) =
λa|u|p−2 u + gn (x, u) for a.a. x ∈ Ω and u ∈ R, as shown in (4.8),

|fn (x, u)| ≤ Cf (|u|p−1 + |u|r−1 ),

where Cf = |λ| · a∞ + max{1, Cϑ η 1−p } and η is independent on n and has the
property that |gn (x, u)| ≤ |u|p−1 for a.a. x ∈ Ω and for all u, with |u| ≤ η, and
all n. Note that Cf is independent of n.
For all n consider the auxiliary problem


LK u = fn (x, u) in Ω,
(Pn )
u=0 in RN \Ω.

Clearly the critical points un of Φtn are exactly the solutions of (Pn ). Now, un → 0
in D̃s,p (Ω), so that un → 0 in Lr (Ω) since the embedding D̃s,p (Ω) → Lr (Ω) is
compact. Moreover, there exists σ > 0, independent of n by (4.1), such that if
un r < σ, then un ∈ L∞ (Ω) and

σun ∞ ≤ un r .

Hence un → 0 in L∞ (Ω). In particular, for all n sufficiently large

un ∈ B , un ∞ ≤ δ/2, Φ (un ) = Φtn (un ) = 0

 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

24 P. Piersanti & P. Pucci

by (4.10), being σ ∈ (0, 1) by (4.1). Therefore we reach the required contradiction,

since
un ∈ K (Φ) ∩ B \{0}.
This proves the claim.
The C 1 (D̃s,p (Ω)) functionals Φt satisfy the (P S) condition in B for all t ∈ [0, 1]
by Lemma 3.4, since each gt and Gt verify (g1 ) , (g2 ) and (g3 ) for all t ∈ [0, 1].
Furthermore the map [0, 1]  t → Φt ∈ D̃s,p (Ω) is continuous. Thus C k (Φ, 0) =
C k (Φ1 , 0) for all k ∈ N by Theorem A.6.
To show the second part of the lemma it is enough to interchange the role
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

between Φ and Φ1 .

In Lemma 4.2 we use (g3 ) in order to have the (P S) condition. We now prove
that Φ has a nontrivial kth critical group at zero for all λ ∈ R, under appropriate
Anal. Appl. Downloaded from www.worldscientific.com

conditions. We begin with small values of λ. To this aim we follow Lemma 3.5, see
also [18, Lemma 4.4].
Lemma 4.3. If either
(i) λ < λ1 , or
(ii) λ = λ1 and G(x, u) ≤ 0 for a.a. x ∈ Ω and all u ∈ R, with |u| ≤ δ,
then C k (Φ, 0) ≈ δk0 Z2 for all k ≥ 0.

Proof. It is enough to show that Φ has a local minimizer at 0 by virtue of Theo-

rem A.7(i). Clearly Φ(0) = 0.
Case (i). For sufficiently small ρ > 0, by virtue of Lemma 3.1 and (2.7), we have
 
max{λ, 0} up
Φ(u) ≥ 1 − + o(1) ≥ 0 for all u ∈ Bρ ,
λ1 p
as required.
Case (ii). By the second part of Lemma 4.2 we shall prove that 0 is a local
minimizer for Φ, by showing that 0 is a global critical point of

up λ
Φ1 (u) = − upp,a − G1 (x, u)dx
p p Ω

in D̃s,p (Ω), where G1 (x, u) = G(x, ϑ(u)). Since ϑ(u) ≤ δ for all u ∈ D̃s,p (Ω) by
(4.9), it results
 
up λ1
Φ1 (u) ≥ − up − G(x, ϑ(u))dx = − G(x, ϑ(u))dx ≥ 0
p λ1 p Ω Ω

for all u ∈ D̃s,p (Ω), as asserted. Hence C k (Φ, 0) = C k (Φ1 , 0) ≈ δk0 Z2 by Lemma 4.2
and Theorem A.7(i). The proof is now complete.

Let us now study (Pλ ) for large values of λ. To this aim we follow Lemma 3.6,
see also [18, Lemma 4.5].
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 25

Lemma 4.4. If 0 is an isolated critical point of Φ in D̃s,p (Ω) and, for all k ≥ 1,
one of the following three cases
(i) λk < λ < λk+1 , or
(ii) λ = λk < λk+1 , and G(x, u) ≥ 0 for a.a. x ∈ Ω and all u ∈ R, with |u| ≤ δ,
(iii) λk < λk+1 = λ, and G(x, u) ≤ 0 for a.a. x ∈ Ω and all u ∈ R, with |u| ≤ δ,
holds, then C k (Φ, 0) = 0.

Proof. Let 0 be an isolated critical point of Φ in D̃s,p (Ω). We prove that Φ has a
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

cohomological local splitting near 0 in dimension k. Let us denote again by E− and

E+ the symmetric closed cones introduced in (3.4). We recall that E+ ∩ E− = {0}.
By Proposition 2.1 we get
i(E− \{0}) = i(D̃s,p (Ω)\E+ ) = k.
Anal. Appl. Downloaded from www.worldscientific.com

We now prove that, for ρ > 0 small enough, (A.5) holds for the functional Φ, where
now Φ(0) = 0.
Case (i). By Lemma 3.1 we have for all u ∈ E− ∩ Bρ
 
λ up
Φ(u) ≤ 1 − + o(1) ≤ 0,
λk p
and for all u ∈ E+ ∩ Bρ
 
λ up
Φ(u) ≥ 1 − + o(1) ≥ 0.
λk+1 p
In conclusion, (A.5) holds and the assertion follows from Proposition A.8.
Case (ii). By Lemma 4.2 the vector 0 is an isolated critical point of Φ1 in D̃s,p (Ω).
Moreover, as already noted in the proof of Lemma 4.2, the function G1 satisfies
conditions (g1 ) , (g2 ) and (g3 ), so that in particular Lemma 3.1 is valid. Since
ϑ(u) ≤ δ for all u ∈ D̃s,p (Ω), in particular for all u ∈ E−
   
λ up
Φ1 (u) ≤ 1 − − G1 (x, u)dx = − G(x, ϑ(u))dx ≤ 0,
λk p Ω Ω

and for all u ∈ E+ ∩ Bρ by Lemma 3.1

  
λ up
Φ1 (u) ≥ 1 − − G1 (x, u)dx
λk+1 p Ω
  
λk up
= 1− − G(x, ϑ(u))dx
λk+1 p Ω
 
λk up
= 1− + o(1) ≥ 0.
λk+1 p
In conclusion (A.5) holds. Hence, C k (Φ, 0) = C k (Φ1 , 0) = 0 by Lemma 4.2 and
Proposition A.8 and we are done.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

26 P. Piersanti & P. Pucci

Case (iii). Repeating the previous argument as in (ii), for all u ∈ E− ∩ Bρ by

Lemma 3.1
    
λ up λk+1 up
Φ1 (u) ≤ 1 − − G1 (x, u)dx = 1 − + o(1) ≤ 0,
λk p Ω λk p
while for all u ∈ E+
   
λ up
Φ1 (u) ≥ 1 − − G1 (x, u)dx = − G(x, ϑ(u))dx ≥ 0.
λk+1 p Ω Ω

Again (A.5) holds and C k (Φ, 0) = C k (Φ1 , 0) = 0 by Lemma 4.2 and Proposi-
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

tion A.8. The proof is now complete.

Proof of Theorem 1.2. Suppose by contradiction that 0 is the only critical point
of Φ. Clearly, Φ satisfies the (P S) condition by Lemma 3.4. Let b < 0 be as in
Anal. Appl. Downloaded from www.worldscientific.com

Lemma 3.7. Since Φ has no critical value in [b, 0), by [27, Lemma 3.9] the set Φb
is a deformation retract of Φ0 \{0}. The set Φ0 is a deformation retract of D̃s,p (Ω)
since there are no critical values in R+ . Thus, taking U = D̃s,p (Ω) in (A.3) we have
for all k ∈ N
C k (Φ, 0) = H k (Φ0 , Φ0 \{0}) ≈ H k (D̃s,p (Ω), Φb ) = 0.
However, C k (Φ, 0) = 0 for some k ≥ 0 by Lemmas 4.3 and 4.4. This contradiction
completes the proof.

Appendix
Throughout the appendix X = (X, ·X ) denotes a real Banach space and X  =
(X  , ·X  ) its dual space. For basic definitions and functional analysis properties,
we refer to the extensive book [6].

A.1. Cohomological local splitting

First of all let us introduce some basic notation: with S N −1 we denote the sphere
{x = (x1 , . . . , xN ) : x = 1} and with RP N we set the real projective N -space. We
recall that S N /Z2 = RP N . Following [27] we present some general basic properties
of the Alexander–Spanier cohomology theory.

Definition A.1. A cochain complex C over Z2 is a sequence of vector spaces C k

over Z2 and linear maps δ k , called coboundary operators, such that
δ k δ k−1 = 0 for all k ∈ N. (A.1)
The elements of C k are called k-cochains, the subspaces Z k (C) = ker δ k are called
k-cocycles and the subspaces B k (C) = im δ k−1 are called k-coboundaries. The quo-
tient space
H k (C) = Z k (C)/B k (C)
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 27

is called the kth cohomology group of C. The set H ∗ (C) is called the cohomological
group of C with coefficients in Z2 and it is defined by

H ∗ (C) = H n (C).
n∈N

If B k (C) = Z k (C) then H k (C) is made up only of the null class and C is called
an exact sequence.

Definition A.2. Let M be a subset of X\{0} and let M = M/Z2 be the quotient
space of M where u and −u are identified. Let f : M → RP ∞ be the classifying
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

map of M and let f ∗ : H ∗ (RP ∞ ) → H ∗ (M ) be the induced homomorphism of the

Alexander–Spanier cohomology rings. The cohomological index of M is defined by


sup{m ≥ 1 : f ∗ (ω m−1 ) = 0} if M = ∅,
Anal. Appl. Downloaded from www.worldscientific.com

i(M ) = (A.2)
0 if M = ∅,

where ω ∈ H 1 (RP ∞ ) is the generator if the polynomial ring H ∗ (RP ∞ ) = Z2 [ω].

Here we list some useful properties of the Z2 -cohomological index (see [27]).

(i1 ) monotonicity: if A ⊂ B, then i(A) ≤ i(B);

(i2 ) invariance: if ϕ : A → B is an odd homeomorphism, then i(A) = i(B);
(i3 ) continuity: if C is a closed symmetric subset of A, then there exists a closed
symmetric neighborhood N of C in A, such that i(N ) = i(C), and also the
interior of N is a neighborhood of C in A and i(int C) = i(C);
(i4 ) neighborhood of the origin: if U is a bounded closed symmetric neighbor-
hood of the origin in X, then i(∂U ) = dim X.

Definition A.3. Let M be a real N -dimensional differentiable manifold. We

denote by T M → M the bundle of tangent vectors and by π0 : T M0 → M
the fiber bundle of nonzero tangent vectors. A Finsler structure on M is a function
F : T M → [0, ∞) with the following properties:

(i) F is differentiable on T M0 ;
(ii) F is positively homogeneous of degree one in y, i.e. F (x, λy) = λF (x, y) for
every λ > 0;
(iii) The Hessian matrix of F 2 is positive definite on T M0 .

A Finsler manifold is a pair of a differentiable manifold M and a Finsler structure

F on M.

Proposition A.4 ([27, Proposition 3.14.7]). Let M be a C 1 Finsler complete

manifold with free Z2 -action and let I ∈ C 1 (M) be even (i.e. Z2 -invariant). Set

Fk = {M ⊂ M : M is symmetric and i(M ) ≥ k} and λk = inf sup I (u).

M∈Fk u∈M
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

28 P. Piersanti & P. Pucci

Then the following statements are true:

(i) If −∞ < λk = · · · = λk+m−1 = c < ∞ and I satisfies the (P S)c condition,
that is any sequence (un )n in M, with I (un ) → c and I  (un )||X  → 0
as n → ∞, admits a convergent subsequence in M, then i(K c ) ≥ m, where
K c = {u ∈ M : I (u) = c, I  (u) = 0}.
(ii) If −∞ < λk ≤ · · · ≤ λk+m−1 < ∞ and I satisfies the (P S)c condition for
c ∈ {λk , . . . , λk+m−1 }, then all λk , . . . , λk+m−1 are critical values for I and
I has m distinct pairs of associated critical points.
(iii) If −∞ < λk < ∞ for all sufficiently large k and I satisfies the (P S) condition,
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

that is the (P S)c condition holds for all c ∈ R, then λk  ∞ as k → ∞.

A.2. Basic properties of critical groups

Anal. Appl. Downloaded from www.worldscientific.com

We present some preliminary results for a generic functional Ψ of class C 1 (X).

Recall that for every k ∈ N the cohomological kth critical group C k (Ψ, 0) at 0 of
the generic C 1 -functional Ψ is defined by
C k (Ψ, 0) = H k (Ψ0 ∩ U, Ψ0 ∩ U \{0}), k ≥ 0, (A.3)
where Ψ0 = {u ∈ X : Ψ(u) ≤ 0}, U is any neighborhood of 0, and H denotes the
Alexander–Spanier cohomology with Z2 -coefficients. C k (Ψ, 0) is independent of U
by the excision property.
In the absence of a direct sum decomposition, the main technical tool we use to
get an estimate of the kth critical group C k (Ψ, 0) at 0 is the notion of a cohomo-
logical local splitting introduced in [27], which is a variant of the homological local
linking used in [25], see also [21, 26]. The following slightly different form of this
notion was given in [10].

Definition A.5. We say that Ψ ∈ C 1 (X) has a cohomological local splitting near
0 in dimension k ≥ 1 if there are symmetric cones E± ⊂ X, with E+ ∩ E− = {0}
and ρ > 0, such that
i(E− \{0}) = i(X\E+ ) = k (A.4)
and, putting Bρ = {u ∈ X : uX ≤ ρ},
Ψ(u) ≤ Ψ(0) for all u ∈ E− ∩ Bρ , Ψ(u) ≥ Ψ(0) for all u ∈ E+ ∩ Bρ . (A.5)
The next results show that the kth critical group is invariant under homotopies
preserving isolatedness of critical points, see [5, 7].

Theorem A.6. Let Ψt : X → R, with t ∈ [0, 1], be a family of functionals of class

C 1 (X) and let u be a critical point of Ψt . If there exists ρ > 0 such that
(i) every Ψt satisfies the (P S) condition over Bρ ;
(ii) Ψt has no critical points in Bρ other than u;
 Ψt is continuous from [0, 1] into C 1 (Bρ ),
(iii) the map t →
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 29

Then, for all k ≥ 0 the cohomological kth critical group C k (Ψt , 0) is independent
of t.
Theorem A.7. Let dim X = ∞ and let Ψ ∈ C 1 (X) be a functional satisfying the
(P S) condition. If u is an isolated critical point of Ψ, then the following properties
hold.
(i) C k (Ψ, u) ≈ δk,0 Z2 for all k ≥ 0, where δ is the Kronecker symbol, whenever u
is a local minimizer of Ψ.
(ii) C k (Ψ, u) = 0 for all k ≥ 0, whenever u is a local maximizer of Ψ.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

We state below a general result that will be useful in the paper.

Proposition A.8 ([10, Proposition 2.1]). If Ψ ∈ C 1 (X) has a cohomological
local splitting near 0 in dimension k and 0 is an isolated critical point of Ψ, then
C k (Ψ, 0) = 0.
Anal. Appl. Downloaded from www.worldscientific.com

Let us first introduce a special definition useful to state the main result contained
in [9], see also [8].
Definition A.9. Let Q, R, S be three subsets of X, with R ⊂ Q and let k be a
nonnegative number. We say that (Q, R) links S cohomologically in dimension k
over Z2 , if the restriction homomorphism
H k (X, X\S; Z2 ) → H k (Q, R; Z2 )
is not identically zero and R ∩ S = ∅.
In the next statement we summarize Corollary 2.9, Theorem 2.8 and Proposi-
tion 2.4 of [9], and Theorem 6.10 of [8]. See also [8, Theorem 3.2 and Remark 3.3].

Theorem A.10. Let E− and E+ be two symmetric cones in X such that E+ is

closed in X, E− ∩ E+ = {0} and
i(E− \{0}) = i(X\E+ ) = k < ∞.
Let r− , r+ > 0 and let e ∈ X with −e ∈ E− . Define the following sets
D− = {u ∈ E− : uX ≤ r− }, S+ = {u ∈ E+ : uX = r+ },
Q = {u + te : u ∈ E− , t ≥ 0, u + teX ≤ r− },
H = {u + te : u ∈ E− , t ≥ 0, u + teX = r− }.
Then (Q, D− ∪ H) links S+ cohomologically in dimension k + 1 over Z2 .
Let Ψ ∈ C 1 (X) satisfy the (P S) condition. If furthermore
sup Ψ(x) < inf Ψ(x) and sup Ψ(x) < ∞,
x∈D− ∪H x∈S+ x∈Q

then there exists a critical point u of Ψ at some level c, with

inf Ψ(x) ≤ c ≤ sup Ψ(x).
x∈S + x∈Q
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

30 P. Piersanti & P. Pucci

A.3. Some auxiliary results

In this section we first present some properties of weighted Lebesgue spaces, fol-
lowing Lemmas 2.1 and 2.2 of [3].

Lemma A.11. Let A be a measurable subset of Rd with d ≥ 1, let α, β ∈ [1, ∞),

let ω be a weight on A and let ϕ : A × R → R be a Carathéodory function.
(i) If (un )n , u are in Lα (A, ω) and un → u in Lα (A, ω) as n → ∞, then there
exist a subsequence (unk )k ⊂ (un )n and a function h ∈ Lα (A, ω) such that a.e.
in A
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

unk → u as k → ∞ and |unk (x)| ≤ h(x) for all k ∈ N.

(ii) If for a.a. x ∈ A and all s ∈ R
Anal. Appl. Downloaded from www.worldscientific.com

|ϕ(x, s)| ≤ ϕ0 (x) + c|s|α/β ,

for some function ϕ0 ∈ Lβ (A, ω) and some constant c > 0, then the Nemyt-
skii operator Nϕ (u) = ϕ(x, u) is continuous and bounded from Lα (A, ω) into
Lβ (A, ω).
(iii) If for every ε > 0 there exists a nonnegative function aε ∈ Lβ (A, ω) such that
|ϕ(x, s)| ≤ aε (x) + ε|s|α/β for a.a. x ∈ A and all s ∈ R.
then for any sequence (un )n bounded in Lα (A, ω), with un → u a.e. in A, it
follows that Nϕ (un ) → Nϕ (u) in Lβ (A, ω).
We complete the section with a suitable version of the main results of [3] proved

here for the dual space D̃−s,p (Ω) of D̃s,p (Ω) and for the problem (2.4), cf. [3,
Proposition 2.3]. Here p denotes the conjugated Hölder exponent of p.

Proposition A.12. The map F : D̃s,p (Ω) → D̃−s,p (Ω), defined by
F (u) = a(x)|u|p−2 u, a ∈ Lα (Ω), α > N/ps,
is weak-to-strong sequentially continuous, i.e. un  u in D̃s,p (Ω) implies that
F (un ) − F(u)D̃−s,p (Ω) → 0 as n → ∞.

Let ω be a weight on Ω such that the embedding D̃s,p (Ω) → Lp (Ω, ω) is compact.
Then, also I : D̃s,p (Ω) → L1 (Ω), defined by I (u) = ω|u|p , is weak-to-strong
sequentially continuous, that is un  u in D̃s,p (Ω) implies I (un ) − I (u)1 → 0
as n → ∞.

Proof. Let (un )n ⊂ D̃s,p (Ω) be such that un  u in D̃s,p (Ω). Hence un → u in
Lp (Ω, a), since the embedding D̃s,p (Ω) → Lp (Ω, a) is compact. Clearly, un p,a →
up,a and so vn p ,a → vp ,a , where vn = |un |p−2 un and similarly v = |u|p−2 u.

We claim that vn → v in Lp (Ω, a). Indeed, fix any subsequence (vnk )k of (vn )n .
The related subsequence (unk )k of (un )n converges in Lp (Ω, a) and admits a
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 31

subsequence, say (unkj )j , converging to u a.e. in Ω by Lemma A.11. Hence, the

corresponding subsequence (vnkj )j of (vnk )k converges to v a.e. in Ω. Therefore,
being 1 < p < ∞, by the Clarkson and Mil’man theorems it follows that vnkj  v

in Lp (Ω, a), since the sequence (vn p ,a )n is bounded, and so by the Radon the-

orem we get that vnkj → v in Lp (Ω, a), since vn p ,a → vp ,a . This shows the
claim, since the subsequence (vnk )k of (vn )n is arbitrary.
By the Hölder inequality for all ϕ ∈ D̃s,p (Ω), with ϕ = 1, we have
| F(un ) − F(u), ϕ |

by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

≤ a(x)|vn − v| · |ϕ|dx ≤ vn − vp,a ϕp,a ≤ Cvn − vp,a ,

Ω
−1/p 
where C = λ1 by (2.7) and F(un ) − F(u), ϕ = Ω (F (un ) − F(u)) · ϕdx.
Consequently, we get that F (un ) − F(u)D̃−s,p (Ω) → 0 as n → ∞.
Anal. Appl. Downloaded from www.worldscientific.com

The second part of the statement is trivial, since D̃s,p (Ω) → Lp (Ω, ω) is com-
pact. Indeed, by the elementary inequality ||x|p − |y|p | ≤ Cp (|x− y|p + |y|p−1 |x− y|)
for all x, y ∈ R, where Cp is an appropriate constant depending only on p > 1, we
find
 
|I (un ) − I (u)|dx = ω(x)||un |p − |u|p |dx
Ω Ω

≤ Cp ω(x)(|un − u|p + |u|p−1 |un − u|)dx
Ω

≤ Cp (un − upp,ω + up,ω

p−1
un − up,ω ) → 0,
since un  u in D̃s,p (Ω) and so un → u in Lp (Ω, ω) as n → ∞. This completes the
proof.

A.4. Uniform convexity of D̃ s,p(Ω) = (D̃ s,p (Ω),
·
)

We are now showing that the space D̃s,p (Ω) endowed with the norm · is a uni-
formly convex Banach space following a strategy similar to the one used in [3, Propo-
sition A.2].
Proposition A.13. The Banach space D̃s,p (Ω) = (D̃s,p (Ω),  · ) defined in Sec. 2
is uniformly convex.

Proof. Let us first consider the case p ≥ 2. Fix an arbitrary ε ∈ (0, 2) and fix
u, v ∈ D̃s,p (Ω) with u = v = 1 and such that u − v ≥ ε. By inequality (35)
of [1, Lemma 2.27], with s = u(x) − u(y) and t = v(x) − v(y), we have
   
 u + v p  u − v p
  + 
 2   2 
  
|(u + v)(x) − (u + v)(y)|p |(u − v)(x) − (u − v)(y)|p
= + ·
R2N 2p 2p
× K(x − y)dxdy
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

32 P. Piersanti & P. Pucci

 
1 1
≤ |u(x) − u(y)|p K(x − y)dxdy + |v(x) − v(y)|p K(x − y)dxdy
2 R2N 2 R2N
1
= (up + vp ) = 1.
2
Hence, being u − v ≥ ε, we get
 
 u + v p
 
 2  ≤ 1 − (ε/2) = (1 − δ) ,
p p

for a suitable δ = δ(ε) ∈ (0, 1). This concludes the proof for the case p ≥ 2.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

Assume now that case 1 < p < 2, so p /p > 1, being p the Hölder conjugate
of p and fix ε ∈ (0, 2). Let u, v ∈ D̃s,p (Ω) with u = v = 1 and such that
u − v ≥ ε. By the reverse Minkowski inequality for Lebesgue weighted spaces
proved in [2, Proposition A.5], with σ = p − 1 ∈ (0, 1) and ω = K, we have
Anal. Appl. Downloaded from www.worldscientific.com

     
 u + v p  u − v p
   
 2  + 2 
  1/(p−1)
p ·(p−1)
[u(x) − u(y)] + [v(x) − v(y)]
= K(x − y)dxdy
R 2N 2
  1/(p−1)
p ·(p−1)
[u(x) − u(y)] − [v(x) − v(y)]
+ K(x − y)dxdy
R2N 2

  
p
[u(x) − u(y)] + [v(x) − v(y)]
≤
R2N 2
 1/(p−1)
[u(x) − u(y)] − [v(x) − v(y)]
p p−1 
+ K(x − y)dxdy
2 
  1/(p−1)
|u(x) − u(y)|p + |v(x) − v(y)|p
≤ K(x − y)dxdy ,
R2N 2
where in the last step we have used inequality (34) of [1, Lemma 2.27], with s =
u(x) − u(y) and t = v(x) − v(y). In conclusion,
    p  1/(p−1)
 u + v p  
  +  u − v  ≤ 1 up + 1 vp = 1.
 2   2  2 2
In particular,
  
 u + v p
  p p
 2  ≤ 1 − (ε/2) = (1 − δ) ,

taking a suitable δ = δ(ε) ∈ (0, 1). This concludes the proof.

 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

Fractional p-Laplacian problems 33

Acknowledgments
The second author was partly supported by the Italian MIUR project titled Vari-
ational and perturbative aspects of nonlinear differential problems (201274FYK7)
and is a member of the Gruppo Nazionale per l’Analisi Matematica, la Proba-
bilità e le loro Applicazioni (GNAMPA) of the Istituto Nazionale di Alta Matem-
atica (INdAM). The manuscript was realized within the auspices of the INdAM-
GNAMPA Project 2015 titled Modelli ed equazioni non-locali di tipo frazionario
(Prot 2015 000368).
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

References
[1] R. A. Adams, Sobolev Spaces, Pure and Applied Mathematics, Vol. 65 (Academic
Press, New York, 1975).
[2] G. Autuori and P. Pucci, Existence of entire solutions for a class of quasilinear elliptic
Anal. Appl. Downloaded from www.worldscientific.com

equations, NoDEA Nonlinear Differ. Eqn. Appl. 20(3) (2003) 977–1009.

[3] G. Autuori, P. Pucci and C. Varga, Existence theorems for quasilinear elliptic eigen-
value problems in unbounded domains, Adv. Differential Equations 18(1–2) (2013)
1–48.
[4] L. Caffarelli, Non-local diffusions, drifts and games, in Nonlinear Partial Differen-
tial Equations: The Abel Symposium 2010, eds. H. Holden and K. H. Karlsen, Abel
Symposia, Vol. 7 (Springer, Berlin, 2012), pp. 37–52.
[5] K. C. Chang and N. Ghoussoub, The Conley index and the critical groups via an
extension of Gromoll–Meyer theory, Topol. Methods Nonlinear Anal. 7(1) (1996) 77–
93.
[6] P. G. Ciarlet, Linear and Nonlinear Functional Analysis with Applications (SIAM,
Philadelphia, PA, 2013).
[7] J. N. Corvellec and A. Hantoute, Homotopical stability of isolated critical points of
continuous functionals, Set-Valued Anal. 10(2–3) (2002) 143–164.
[8] M. Degiovanni, On topological and metric critical point theory, J. Fixed Point Theory
Appl. 7(1) (2010) 85–102.
[9] M. Degiovanni and S. Lancelotti, Linking over cones and nontrivial solutions for p-
Laplace equations with p-superlinear nonlinearity, Ann. Inst. H. Poincaré Anal. Non
Linéaire 24(6) (2007) 907–919.
[10] M. Degiovanni, S. Lancelotti and K. Perera, Nontrivial solutions of p-superlinear p-
Laplacian problems via a cohomological local splitting, Commun. Contemp. Math.
12(3) (2010) 475–486.
[11] M. D’Elia and M. Gunzburger, The fractional Laplacian operator on bounded
domains as a special case of the nonlocal diffusion operator, Comput. Math. Appl.
66(7) (2013) 1245–1260.
[12] M. D’Elia and M. Gunzburger, Max optimal distributed control of nonlocal steady
diffusion problems, SIAM J. Control Optim. 52(1) (2014) 243–273.
[13] E. Di Nezza, G. Palatucci and E. Valdinoci, Hitchhiker’s guide to the fractional
Sobolev spaces, Bull. Sci. Math. 136(5) (2012) 521–573.
[14] A. Fiscella, R. Servadei and E. Valdinoci, A resonance problem for non-local elliptic
operators, Z. Anal. Anwend. 32(4) (2013) 411–431.
[15] G. Franzina and G. Palatucci, Fractional p-eigenvalues, Riv. Mat. Univ. Parma 5(2)
(2014) 373–386.
[16] G. Gilboa and S. Osher, Nonlocal operators with applications to image processing,
Multiscale Model. Simul. 7(3) (2008) 1005–1028.
 3rd Reading
May 20, 2016 13:15 WSPC/S0219-5305 176-AA 1650002

34 P. Piersanti & P. Pucci

[17] P. Grisvard, Elliptic Problems in Nonsmooth Domains, Monographs and Studies in

Mathematics, Vol. 24 (Pitman, Boston, MA, 1985).
[18] A. Iannizzotto, S. Liu, K. Perera and M. Squassina, Existence results for fractional
p-Laplacian problems via Morse theory, Adv. Calc. Var. 9(2) (2016) 101–125.
[19] N. Laskin, Fractional quantum mechanics and Lévy path integrals, Phys. Lett. A
268(4–6) (2000) 298–305.
[20] N. Laskin, Fractional Schrödinger equation, Phys. Rev. E 66(5) (2000) 056108, 7 pp.
[21] S. J. Li and M. Willem, Applications of local linking to critical point theory, J. Math.
Anal. Appl. 189(1) (1995) 6–32.
[22] E. Lindgren and P. Lindqvist, Fractional eigenvalues, Calc. Var. Partial Differential
Equations 49(1–2) (2013) 795–826.
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/17/16. For personal use only.

[23] V. Maz’ya and T. Shaposhnikova, On the Bourgain, Brezis, and Mironescu theorem
concerning limiting embeddings of fractional Sobolev spaces, J. Funct. Anal. 195(2)
(2002) 230–238.
[24] G. Molica Bisci, V. D. Radulescu and R. Servadei, Variational Methods for Nonlocal
Fractional Problems (Cambridge University Press, Cambridge, 2016).
Anal. Appl. Downloaded from www.worldscientific.com

[25] K. Perera, Homological local linking, Abstr. Appl. Anal. 3(1–2) (1998) 181–189.
[26] K. Perera, Nontrivial critical groups in p-Laplacian problems via the Yang index,
Topol. Methods Nonlinear Anal. 21(2) (2003) 301–309.
[27] K. Perera, R. P. Agarwal and D. O’Regan, Morse Theoretic Aspects of p-Laplacian
Type Operators, Mathematical Surveys and Monographs, Vol. 161 (American Math-
ematical Society, Providence, RI, 2010).
[28] K. Perera, P. Pucci and C. Varga, An existence result for a class of quasilinear
elliptic eigenvalue problems in unbounded domain, NoDEA Nonlinear Differ. Equ.
Appl. 21(3) (2014) 441–451.
[29] P. Pucci and S. Saldi, Multiple solutions for an eigenvalue problem involving p-
Laplacian operators, in Geometric Methods in PDE’s, eds. G. Citti, M. Manfredini,
D. Morbidelli, S. Polidoro and F. Uguzzoni, Springer INdAM Series, Vol. 11 (Springer,
2015), pp. 159–176.
[30] R. Servadei and E. Valdinoci, A Brezis–Nirenberg result for non-local critical equa-
tions in low dimension, Commun. Pure Appl. Anal. 12(6) (2013) 2445–2464.
[31] R. Servadei and E. Valdinoci, Lewy–Stampacchia type estimates for variational
inequalities driven by (non)local operators, Rev. Mat. Iberoam. 29(3) (2013) 1091–
1126.
[32] R. Servadei and E. Valdinoci, Weak and viscosity solutions of the fractional Laplace
equation, Publ. Mat. 58(1) (2014) 133–154.
[33] E. H. Spanier, Algebraic Topology (McGraw-Hill Book Co., New York, 1966).
[34] J. L. Vázquez, Recent progress in the theory of nonlinear diffusion with fractional
Laplacian operators, Discrete Contin. Dyn. Syst. Ser. S 7 (2014) 857–885.