Electronic Journal of Differential Equations, Monograph 07, 2006, (213 pages).

ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu

ftp ejde.math.txstate.edu (login: ftp)

EXISTENCE, MULTIPLICITY, PERTURBATION, AND
CONCENTRATION RESULTS FOR A CLASS OF

QUASI-LINEAR ELLIPTIC PROBLEMS

MARCO SQUASSINA

To my parents and to Maria and Giulia

Abstract. The aim of this monograph is to present a comprehensive sur-

vey of results about existence, multiplicity, perturbation from symmetry and

concentration phenomena for the quasi-linear elliptic equation

−
nX

i,j=1

Dxj (aij(x, u)Dxiu) +
1

2

nX
i,j=1

Duaij(x, u)DxiuDxj u = g(x, u) in Ω,

where Ω is a smooth domain of Rn, n ≥ 1. Under natural assumptions on the

coefficients aij , the above problem admits a standard variational structure,

but the associated functional f : H1
0 (Ω)→ R,

f(u) =
1

2

Z
Ω

nX
i,j=1

aij(x, u)DxiuDxj u dx−
Z
Ω

G(x, u) dx,

turns out to be merely continuous. Therefore, some tools of non-smooth critical

point theory will be employed throughout the various sections.

Contents

Preface 3
Notation 4
1. Introduction 4
2. Review of Critical Point Theory 6
2.1. Notions of non-smooth analysis 6
2.2. The case of lower semi-continuous functionals 9
2.3. Functionals of the calculus of variations 12
3. Super-linear Elliptic Problems 14
3.1. Quasi-linear elliptic systems 14
3.2. The concrete Palais-Smale condition 16

2000 Mathematics Subject Classification. 35J20, 35J25, 49J52, 58E35, 74G35.
Key words and phrases. Critical points; non-smooth functionals; weak slope; deformation the-

orem; invariant functionals, mountain pass theorem; relative category; calculus of variations, Euler
equations; quasi-linear elliptic equations; multiple solutions; quasi-linear variational inequalities;

Palais-Smale condition; lack of compactness; singularly perturbed equations; spiked solutions;

variational identities; nonexistence results.
c©2006 Texas State University - San Marcos.
Submitted September 12, 2005. Published February 16, 2006.

1



2 MARCO SQUASSINA EJDE-2006/MON. 07

3.3. Existence of multiple solutions for elliptic systems 20
3.4. Regularity of weak solutions for elliptic systems 21
3.5. Fully nonlinear scalar problems 23
3.6. The concrete Palais-Smale condition 24
3.7. Existence of a weak solution 33
3.8. Super-linear problems with unbounded coefficients 34
3.9. General setting and main results 36
3.10. Verification of the key condition 38
3.11. The variational setting 39
3.12. A compactness result for J 47
3.13. Proofs of the main Theorems 51
3.14. Summability results 54
4. Perturbation from Symmetry 57
4.1. Quasi-linear elliptic systems 57
4.2. Symmetry perturbed functionals 59
4.3. Boundedness of concrete Palais-Smale sequences 63
4.4. Compactness of concrete Palais-Smale sequences 65
4.5. Existence of multiple solutions 65
4.6. Semi-linear systems with nonhomogeneous data 68
4.7. Reduction to homogeneous boundary conditions 70
4.8. The Palais-Smale condition 75
4.9. Comparison of growths for min-max values 77
4.10. Bolle’s method for non-symmetric problems 79
4.11. Application to semi-linear elliptic systems 80
4.12. The diagonal case 83
5. Problems of Jumping Type 84
5.1. Fully nonlinear elliptic equation 84
5.2. The main result 85
5.3. The concrete Palais-Smale condition 87
5.4. Min-Max estimates 92
5.5. Proof of the main result 96
5.6. Fully nonlinear variational inequalities 97
5.7. The main result 98
5.8. The bounded Palais-Smale condition 100
5.9. The Palais-Smale condition 105
5.10. Min-Max estimates 110
5.11. Proof of the main result 111
6. Problems with Loss of Compactness 111
6.1. Positive entire solutions for fully nonlinear problems 111
6.2. The concrete Palais-Smale condition 113
6.3. Fully nonlinear problems at critical growth 123
6.4. The first solution 126
6.5. The concrete Palais-Smale condition 130
6.6. The second solution 133
6.7. One solution for a more general nonlinearity 134
6.8. Existence of one nontrivial solution 135
6.9. Problems with nearly critical growth 139
6.10. The main results 141



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 3

6.11. The weak limit 143
6.12. Proof of the main results 147
6.13. Mountain-pass critical values 150
7. The Singularly Perturbed Case, I 151
7.1. The del Pino-Felmer penalization scheme 154
7.2. Energy estimates and concentration 158
7.3. Proof of the main result 168
7.4. A few related open problems 171
8. The Singularly Perturbed Case, II 172
8.1. Penalization and compactness 174
8.2. Two consequences of the Pucci-Serrin identity 179
8.3. Energy estimates 184
8.4. Proofs of the main results 193
9. Nonexistence Results 195
9.1. A general Pucci-Serrin type identity 195
9.2. The approximation argument 196
9.3. Non-strict convexity in some particular cases 200
9.4. The splitting case 201
9.5. The one-dimensional case 203
9.6. Non-existence results 205
References 208

Preface

This monograph is an updated, expanded and restyled elaboration of the Ph.D.
thesis that the author defended at the University of Milan on January 2002. It
contains some of the author’s researches undertaken from 1997 to 2003 in the field
of variational quasi-linear elliptic partial differential equations, under the supervi-
sion of Marco Degiovanni. The author thanks him for his teaching, encouragement
and advice. The author is grateful to Lucio Boccardo and Filomena Pacella for
supporting a couple of stay at Rome University La Sapienza in 2002 and 2005.
Further thanks are due to the Managing Editors of the Electronic Journal of Dif-
ferential Equations, in particular to Professor Alfonso Castro for his kindness. The
author was supported by the MIUR research project “Variational and Topological
Methods in the Study of Nonlinear Phenomena” and by the Istituto Nazionale di
Alta Matematica “F.Severi”.

The presentation of the material is essentially self-contained. It only requires
some basic knowledge in functional analysis as well as in the theory of linear elliptic
problems. The work is arranged into nine paragraphs, and each of these is divided
into various numbered subsections. All results are formally stated as Theorems,
Propositions, Lemmas or Corollaries which are numbered by their section number
and order within that section. Throughout the manuscript formulae have double
indexing in each section, the first digit being the section number. When formulae
from another section are referred to, the number corresponding to the section is
placed first.

Marco Squassina



4 MARCO SQUASSINA EJDE-2006/MON. 07

Notation

(1) N, Z, Q, R denote the set of natural, integer, rational, real numbers;
(2) Rn (or RN ) is the usual real Euclidean space;
(3) Ω is an open set (often implicitly assumed smooth) in Rn;
(4) ∂Ω is the boundary of Ω;
(5) a.e. stands for almost everywhere;
(6) p′ is the conjugate exponent of p;
(7) Lp(Ω) is the space of u measurable with

∫
Ω
|u|pdx <∞, 1 ≤ p <∞;

(8) L∞(Ω) is the space of u measurable with |u(x)| ≤ C for a.e. x ∈ Ω;
(9) ‖ · ‖p and ‖ · ‖∞ norms of the spaces Lp and L∞;

(10) Dxi
u(x) is the i-th partial derivative of u at x;

(11) ∇u(x) stands for (Dx1u(x), . . . , Dxn
u(x));

(12) ∆u(x) stands for
∑n
i=1D

2
x2

i
u(x);

(13) H1(Ω), H1(Rn), H1
0 (Ω), W 1,p(Ω), W 1,p(Rn), W 1,p

0 (Ω) are Sobolev spaces;
(14) H−1(Ω), W−1,p′

0 (Ω) are the first duals of Sobolev spaces;
(15) W k,p(Rn), W k,p

0 (Ω) denotes higher order Sobolev spaces;
(16) ‖ · ‖1,p, ‖ · ‖k,p, ‖ · ‖−1,p norms of the Sobolev spaces;
(17) Liploc(Rn) indicate the space locally Lipschitz functions;
(18) C∞c (Ω) functions differentiable at any order with compact support;
(19) Ln(E) denotes Lebesgue measure of E;
(20) Hn−1(A) denotes the Hausdorff measure of A;
(21) H usually stands for a suitable deformation;
(22) |df |(u) stands for the weak slope of f at u;
(23) (um) denotes a sequence of scalar functions;
(24) (um) denotes a sequence of vector valued functions;
(25) u+ (resp. u−) is the positive (resp. negative) part of u;
(26) ⇀ (resp. →) stands for the weak (resp. strong) convergence;
(27) limn means the limit as n→ +∞;
(28) Br(x) or B(x, r) is the ball of center x and radius r;
(29) d(x,E) is the distance of x from E.
(30) 〈ϕ, x〉 evaluation of the linear functional ϕ at x;
(31) x · y scalar product between elements x, y ∈ Rn;
(32) δij is 1 for i = j and 0 for i 6= j;
(33) χE (or 1E) is the characteristic function of the set E;
(34) A⊕B is the direct sum between A and B.

1. Introduction

The recent years have been marked out by an evergrowing interest in the research
of solutions (and, besides, of their various qualitative behaviors) of semi-linear
elliptic problems via techniques of classical critical point theory. Readers which
are interested in these aspects may look at the following books: Aubin-Ekeland
[13], Chabrowski [39, 40], Ghoussoub [73], Mawhin-Willem [103], Rabinowitz [120],
Struwe [136], Willem [146] and Zeidler [147].

The present work aims to show how various achievements, well-established in
the semi-linear case, can be extended to a more general class of problems. More
precisely, let Ω be an open bounded subset in Rn (n ≥ 2) and f : H1

0 (Ω) → R a



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 5

functional of the form

f(u) =
1
2

∫
Ω

n∑
i,j=1

aij(x)DxiuDxju dx−
∫

Ω

G(x, u) dx.

Since the pioneering paper of Ambrosetti-Rabinowitz [5], critical point theory has
been successfully applied to the functional f , yielding several important results
(see e.g. [42, 103, 120, 136]). However, the assumption that f : H1

0 (Ω) → R is of
class C1 turns out to be very restrictive for more general functionals of calculus of
variations, like

f(u) =
∫

Ω

L (x, u,∇u) dx−
∫

Ω

G(x, u) dx,

(see e.g. [53]). In particular, if f has the form

f(u) =
1
2

∫
Ω

n∑
i,j=1

aij(x, u)DxiuDxju dx−
∫

Ω

G(x, u) dx,

we may expect f to be of class C1 only when the aij ’s are independent of u or when
n = 1. In fact, if f was locally Lipschitz continuous, for u ∈ H1

0 (Ω), we would have

sup
{
f ′(u)(v) : v ∈ C∞c (Ω), ‖v‖H1

0 (Ω) ≤ 1
}
<∞,

that is to say
n∑

i,j=1

Dsaij(x, u)Dxi
uDxj

u ∈ H−1(Ω).

The above term naturally belongs to L1(Ω), which is not included in H−1(Ω) for
n ≥ 2. On the other hand, since the papers of Chang [43] and Marino-Scolozzi
[101], techniques of critical point theory have been extended to some classes of
non-smooth functionals. In our setting, in which f is naturally continuous but not
locally Lipschitz, it turns out to be convenient to apply the theory developed in
[50, 58, 86, 87] according to the approach started by Canino [33]. Let us point out
that a different approach has been also considered in the literature. If we consider
the space H1

0 (Ω) ∩ L∞(Ω) endowed with the family of norms

‖u‖ε = ‖u‖H1
0

+ ε‖u‖L∞ , ε > 0,

then, under suitable assumptions, f is of class C1 in (H1
0 (Ω) ∩ L∞(Ω), ‖ · ‖ε) for

each ε > 0. This allows an approximation procedure by smooth problems (the
original one is obtained as a limit when ε → 0). The papers of Struwe [137] and
Arcoya-Boccardo [6, 7] follow, with some variants, this kind of approach. However,
in view of multiplicity results, it is hard to keep the multiplicity of solutions at
the limit. In particular, when f is even and satisfies assumptions of Ambrosetti-
Rabinowitz type, the existence of infinitely many solutions has been so far proved
only by the former approach. The aim of this manuscript is to present some results
concerning existence, nonexistence, multiplicity, perturbation from symmetry, and
concentration for quasi-linear problems such as

−
n∑

i,j=1

Dxj
(aij(x, u)Dxi

u) +
1
2

n∑
i,j=1

Dsaij(x, u)DxiuDxju = g(x, u) in Ω

u = 0 on ∂Ω



6 MARCO SQUASSINA EJDE-2006/MON. 07

and even for the more general class of elliptic problems

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) in Ω
u = 0 on ∂Ω,

including the case when g reaches the critical growth with respect to the Sobolev
embedding. New results have been obtained in the following situations:

Section 3: infinitely many solutions for quasi-linear problems with odd nonlinear-
ities; existence of a weak solution for a general class of Euler’s equations of multiple
integrals of calculus of variations; existence and multiplicity for quasi-linear elliptic
equations having unbounded coefficients (cf. [132, 133, 114]).

Section 4: multiplicity of solutions for perturbed symmetric quasi-linear elliptic
problems; multiplicity results for semi-linear systems with broken symmetry and
non-homogeneous boundary data (cf. [128, 129, 30, 110]).

Section 5: problems of jumping type for a general class of Euler’s equations of
multiple integrals of calculus of variations; problems of jumping type for a general
class of nonlinear variational inequalities (cf. [79, 80]).

Section 6: positive entire solutions for fully nonlinear elliptic equations; existence
of two solutions for fully nonlinear problems at critical growth with perturbations
of lower order; asymptotics of solutions for a class of nonlinear problems at nearly
critical growth (cf. [127, 134, 130, 107]).

Section 7: concentration phenomena for singularly perturbed quasi-linear ellip-
tic equations. Existence of families of solutions with a spike-like shape around a
suitable point (cf. [131]).

Section 8: multi-peak solutions for degenerate singularly perturbed elliptic equa-
tions. Existence of families of solutions with multi spike-like profile around suitable
points (cf. [74]).

Section 9: Pucci-Serrin type identities for C1 solutions of Euler’s equations and
related non-existence results (cf. [59]).

For the sake of completeness, we wish to mention a quite recent paper [35] dealing
with the variational bifurcation for quasi-linear elliptic equations (extending some
early results due to Rabinowitz in the semi-linear case [121]) and the paper [91]
regarding improved Morse index type estimates for the functional f .

2. Review of Critical Point Theory

In this section, we shall recall some results of abstract critical point theory
[36, 50, 58, 86, 87]. For the proofs, we refer to [36] or [50].

2.1. Notions of non-smooth analysis. Let X be a metric space endowed with
the metric d and let f : X → R be a function. We denote by Br(u) the open ball
of center u and radius r and we set

epi(f) = {(u, λ) ∈ X × R : f(u) ≤ λ}.

In the following, X × R will be endowed with the metric

d((u, λ), (v, µ)) =
(
d(u, v))2 + (λ− µ)2

)1/2

and epi(f) with the induced metric.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 7

Definition 2.1. For every u ∈ X with f(u) ∈ R, we denote by |df |(u) the supre-
mum of the σ′s in [0,+∞[ such that there exist δ > 0 and a continuous map

H : (Bδ(u, f(u)) ∩ epi(f))× [0, δ] → X

satisfying
d(H ((v, µ), t), v) ≤ t, f(H ((v, µ), t)) ≤ µ− σt,

whenever (v, µ) ∈ Bδ(u, f(u)) ∩ epi(f) and t ∈ [0, δ]. The extended real number
|df |(u) is called the weak slope of f at u.

Proposition 2.2. Let u ∈ X with f(u) ∈ R. If (uh) is a sequence in X with
uh → u and f(uh) → f(u), then we have |df |(u) ≤ lim infh |df |(uh).

Remark 2.3. If the restriction of f to {u ∈ X : f(u) ∈ R} is continuous, then

|df | : {u ∈ X : f(u) ∈ R} → [0,+∞]

is lower semi-continuous.

Proposition 2.4. Let f : X → R ∪ {+∞} be a function. Set

D(f) := {u ∈ X : f(u) < +∞}
and assume that f |D(f) is continuous. Then for every u ∈ D(f) we have

|df |(u) =
∣∣df |D(f)

∣∣(u)
and this value is in turn equal to the supremum of the σ’s in [0,+∞[ such that there
exist δ > 0 and a continuous map

H : (Bδ(u) ∩D(f))× [0, δ] → X

satisfying
d(H (v, t), v) ≤ t, f(H (v, t)) ≤ f(v)− σt,

whenever v ∈ Bδ(u) ∩D(f) and t ∈ [0, δ].

Definition 2.5. An element u ∈ X is said to be a (lower) critical point of f if
|df |(u) = 0. A real number c is said to be a (lower) critical value of f if there exists
a critical point u ∈ X of f such that f(u) = c. Otherwise c is said to be a regular
value of f .

Definition 2.6. Let c be a real number. The function f is said to satisfy the
Palais-Smale condition at level c ((CPS)c for short), if every sequence (uh) in X
with |df |(uh) → 0 and f(uh) → c admits a subsequence (uhk

) converging in X to
some u.

Let us also introduce some usual notations. For every b ∈ R ∪ {+∞} and c ∈ R
we set

f b = {u ∈ X : f(u) ≤ b}, Kc = {u ∈ X : |df |(u) = 0, f(u) = c}.

Theorem 2.7 (Deformation Theorem). Let c ∈ R. Assume that X is complete,
f : X → R is a continuous function which satisfies (CPS)c. Then, given ε̄ > 0, a
neighborhood U of Kc (if Kc = ∅, we allow U = ∅) and λ > 0, there exist ε > 0
and a continuous map η : X × [0, 1] → X such that for every u ∈ X and t ∈ [0, 1]
we have:

(a) d(η(u, t), u) ≤ λt;
(b) f(η(u, t)) ≤ f(u);
(c) f(u) /∈]c− ε̄, c+ ε̄[⇒ η(u, t) = u;



8 MARCO SQUASSINA EJDE-2006/MON. 07

(d) η(fc+ε \ U, 1) ⊂ fc−ε.

Theorem 2.8 (Noncritical Interval Theorem). Let a ∈ R and b ∈ R∪ {+∞} (a <
b). Assume that f : X → R is a continuous function which has no critical points u
with a ≤ f(u) ≤ b, that (CPS)c holds and fc is complete whenever c ∈ [a, b[. Then
there exists a continuous map η : X × [0, 1] → X such that for every u ∈ X and
t ∈ [0, 1] we have:

(a) η(u, 0) = u;
(b) f(η(u, t)) ≤ f(u);
(c) f(u) ≤ a⇒ η(u, t) = u;
(d) f(u) ≤ b⇒ f(η(u, 1)) ≤ a.

Theorem 2.9. Let X be a complete metric space and f : X → R ∪ {+∞} a
function such that D(f) is closed in X and f|D(f)

is continuous. Let u0, v0, v1 be in
X and suppose that there exists r > 0 such that ‖v0 − u0‖X < r, ‖v1 − u0‖X > r,
inf f(Br(u0)) > −∞, and

a′ = inf{f(u) : u ∈ X, ‖u− u0‖X = r} > max{f(v0), f(v1)}.

Let
Γ = {γ : [0, 1] → D(f) continuous with γ(0) = v0, γ(1) = v1}

and assume that Γ 6= ∅ and that f satisfies the Palais-Smale condition at the two
levels

c1 = inf f(Br(u0)), c2 = inf
γ∈Γ

max
[0,1]

(f ◦ γ).

Then −∞ < c1 < c2 < +∞ and there exist at least two critical points u1, u2 of f
such that f(ui) = ci (i = 1, 2).

We now recall the mountain pass theorem without Palais-Smale.

Theorem 2.10. Let X is a Banach space and f : X → R is a continuous func-
tional. Assume that the following facts hold:

(a) There exist η > 0 and % > 0 such that

∀u ∈ X : ‖u‖X = %⇒ f(u) > η ;

(b) f(0) = 0 and there exists w ∈ X such that:

f(w) < η and ‖w‖X > %.

Moreover, let us set

Φ = {γ ∈ C([0, 1], X) : γ(0) = 0, γ(1) = w}

and
η ≤ β = inf

γ∈Φ
max
t∈[0,1]

f(γ(t)).

Then there exists a Palais-Smale sequence for f at level β.

In the next theorem, we recall a generalization of the classical perturbation argu-
ment of Bahri, Berestycki, Rabinowitz and Struwe devised around 1980 for dealing
with problems with broken symmetry adapted to our non-smooth framework (See
[118]).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 9

Theorem 2.11. Let X be a Hilbert space endowed with a norm ‖ · ‖X and let
f : X → R be a continuous functional. Assume that there exists M > 0 such that f
satisfies the concrete Palais-Smale condition at each level c ≥M . Let Y be a finite
dimensional subspace of X and u∗ ∈ X \ Y and set

Y ∗ = Y ⊕ 〈u∗〉 , Y ∗+ = {u+ λu∗ ∈ Y ∗ : u ∈ Y, λ ≥ 0} .
Assume now that f(0) ≤ 0 and that

(a) There exists R > 0 such that

∀u ∈ Y : ‖u‖X ≥ R⇒ f(u) ≤ f(0) ;

(b) there exists R∗ ≥ R such that:

∀u ∈ Y ∗ : ‖u‖X ≥ R∗ ⇒ f(u) ≤ f(0).

Let us set

P =
{
γ ∈ C(X,X) : γ odd, γ(u) = u if max{f(u), f(−u)} ≤ 0

}
.

Then, if
c∗ = inf

γ∈P
sup
u∈Y ∗+

f(γ(u)) > c = inf
γ∈P

sup
u∈Y

f(γ(u)) ≥M,

f admits at least one critical value c ≥ c∗.

2.2. The case of lower semi-continuous functionals. Let X be a metric space
and let f : X → R ∪ {+∞} be a lower semi-continuous function. We set

dom(f) = {u ∈ X : f(u) < +∞} and epi f = {(u, η) ∈ X × R : f(u) ≤ η} .
The set epi f is endowed with the metric

d ((u, η), (v, µ)) =
(
d(u, v)2 + (η − µ)2

)1/2
.

Let us define the function Gf : epi f → R by setting

Gf (u, η) = η. (2.1)

Note that Gf is Lipschitz continuous of constant 1.
From now on we denote with B(u, δ) the open ball of center u and of radius δ.
We recall the definition of the weak slope for a continuous function introduced in
[50, 58, 86, 87].

Definition 2.12. Let X be a complete metric space, g : X → R a continuous
function, and u ∈ X. We denote by |dg|(u) the supremum of the real numbers σ in
[0,∞) such that there exist δ > 0 and a continuous map

H : B(u, δ)× [0, δ] → X,

such that, for every v in B(u, δ), and for every t in [0, δ] it results

d(H (v, t), v) ≤ t,

g(H (v, t)) ≤ g(v)− σt.

The extended real number |dg|(u) is called the weak slope of g at u.

According to the previous definition, for every lower semi-continuous function f
we can consider the metric space epi f so that the weak slope of Gf is well defined.
Therefore, we can define the weak slope of a lower semi-continuous function f by
using |dGf |(u, f(u)).

More precisely, we have the following



10 MARCO SQUASSINA EJDE-2006/MON. 07

Definition 2.13. For every u ∈ dom(f) let

|df |(u) =


∣∣Gf

∣∣(u,f(u))r
1−

∣∣Gf

∣∣(u,f(u))2
, if

∣∣Gf ∣∣(u, f(u)) < 1,

+∞, if
∣∣Gf ∣∣(u, f(u)) = 1.

The previous notion allow us to give the following concepts.

Definition 2.14. Let X be a complete metric space and f : X → R ∪ {+∞} a
lower semi-continuous function. We say that u ∈ dom(f) is a (lower) critical point
of f if |df |(u) = 0. We say that c ∈ R is a (lower) critical value of f if there exists
a (lower) critical point u ∈ dom(f) of f with f(u) = c.

Definition 2.15. Let X be a complete metric space, f : X → R ∪ {+∞} a lower
semi-continuous function and let c ∈ R. We say that f satisfies the Palais-Smale
condition at level c ((PS)c in short), if every sequence {un} in dom(f) such that

|df |(un) → 0,

f(un) → c,

admits a subsequence {unk
} converging in X.

For every η ∈ R, let us define the set

fη = {u ∈ X : f(u) < η}. (2.2)

The next result gives a criterion to obtain an estimate from below of |df |(u) (cf.
[58]).

Proposition 2.16. Let f : X → R ∪ {+∞} be a lower semi-continuous function
defined on the complete metric space X, and let u ∈ dom(f). Assume that there
exist δ > 0, η > f(u), σ > 0 and a continuous function H : B(u, δ)∩fη×[0, δ] → X
such that

d(H (v, t), v) ≤ t, ∀ v ∈ B(u, δ) ∩ fη,
f(H (v, t)) ≤ f(v)− σt, ∀ v ∈ B(u, δ) ∩ fη.

Then |df |(u) ≥ σ.

We will also use the notion of equivariant weak slope (see [36]).

Definition 2.17. Let X be a normed linear space and f : X → R ∪ {+∞} an
even lower semi-continuous function with f(0) < +∞. For every (0, η) ∈ epi f we
denote by |dZ2Gf |(0, η) the supremum of the numbers σ in [0,∞) such that there
exist δ > 0 and a continuous map

H = (H1,H2) : (B((0, η), δ) ∩ epi f)× [0, δ] → epi f

satisfying

d(H ((w, µ), t), (w, µ)) ≤ t, H2((w, µ), t) ≤ µ− σt,

H1((−w, µ), t) = −H1((w, µ), t),

for every (w, µ) ∈ B((0, η), δ) ∩ epi f and t ∈ [0, δ].

To compute |dGf |(u, η), the next result will be useful (cf. [58]).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 11

Proposition 2.18. Let X be a normed linear space, J : X → R ∪ {+∞} a lower
semi-continuous functional, I : X → R a C1 functional and let f = J + I. Then
the following facts hold:

(a) For every (u, η) ∈ epi(f) we have

|dGf |(u, η) = 1 ⇐⇒ |dGJ |(u, η − I(u)) = 1 ;

(b) if J and I are even, for every η ≥ f(0), we have

|dZ2Gf |(0, η) = 1 ⇐⇒ |dZ2GJ |(0, η − I(0)) = 1 ;

(c) if u ∈ dom(f) and I ′(u) = 0, then |df |(u) = |dJ |(u).

Proof. Assertions (a) and (c) follow by arguing as in [58]. Assertion (b) can be
reduced to (a) after observing that, since I is even, it results I ′(0) = 0. �

In [50, 58] variational methods for lower semi-continuous functionals are devel-
oped. Moreover, it is shown that the following condition is fundamental in order
to apply the abstract theory to the study of lower semi-continuous functions

∀(u, η) ∈ epi f : f(u) < η ⇒
∣∣Gf ∣∣(u, η) = 1. (2.3)

In the next section we will prove that the functional f satisfies (2.3). The next
result gives a criterion to verify condition (2.3) (cf. [60, Corollary 2.11]).

Theorem 2.19. Let (u, η) ∈ epi(f) with f(u) < η. Assume that, for every % > 0,
there exist δ > 0 and a continuous map

H : {w ∈ B(u, δ) : f(w) < η + δ} × [0, δ] → X

satisfying

d(H (w, t), w) ≤ %t , f(H (w, t)) ≤ (1− t)f(w) + t(f(u) + %)

whenever w ∈ B(u, δ), f(w) < η + δ and t ∈ [0, δ]. Then |dGf |(u, η) = 1. In addi-
tion, if f is even, u = 0 and H (−w, t) = −H (w, t), then we have |dZ2Gf |(0, η) = 1.

Let us now recall from [50] the following result.

Theorem 2.20. Let X be a Banach space and f : X → R ∪ {+∞} a lower semi-
continuous function satisfying (2.3). Assume that there exist v0, v1 ∈ X and r > 0
such that ‖v1 − v0‖ > r and

inf{f(u) : u ∈ X, ‖u− v0‖ = r} > max{f(v0), f(v1)}. (2.4)

Let us set

Γ = {γ : [0, 1] → dom(f), γ continuous, γ(0) = v0 and γ(1) = v1} ,
and assume that

c1 = inf
γ∈Γ

sup
[0,1]

f ◦ γ < +∞

and that f satisfies the Palais-Smale condition at the level c1. Then, there exists a
critical point u1 of f such that f(u1) = c1. If, moreover,

c0 = inf f(Br(v0)) > −∞,

and f satisfies the Palais-Smale condition at the level c0, then there exists another
critical point u0 of f with f(u0) = c0.

In the equivariant case we shall apply the following result (see [102]).



12 MARCO SQUASSINA EJDE-2006/MON. 07

Theorem 2.21. Let X be a Banach space and f : X → R ∪ {+∞} a lower
semi-continuous even function. Let us assume that there exists a strictly increasing
sequence (Wh) of finite dimensional subspaces of X with the following properties:

(a) There exist ρ > 0, γ > f(0) and a subspace V ⊂ X of finite codimension
with

∀u ∈ V : ‖u‖ = ρ ⇒ f(u) ≥ γ;

(b) there exists a sequence (Rh) in (ρ,∞) such that

∀u ∈Wh : ‖u‖ ≥ Rh ⇒ f(u) ≤ f(0);

(c) f satisfies (PS)c for any c ≥ γ and f satisfies (2.3);
(d) |dZ2Gf |(0, η) 6= 0 for every η > f(0).

Then there exists a sequence {uh} of critical points of f such that f(uh) → +∞.

2.3. Functionals of the calculus of variations. Let Ω be a bounded open subset
of Rn, n ≥ 3 and let f : W 1,p

0 (Ω; RN ) → R (N ≥ 1) be a functional of the form

f(u) =
∫

Ω

L (x, u,∇u)dx. (2.5)

The associated Euler’s equation is formally given by the quasi-linear problem

−div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u) = 0 in Ω
u = 0 on ∂Ω.

(2.6)

Assume that L : Ω×RN×RnN → R is measurable in x for all (s, ξ) ∈ RN×RnN and
of class C1 in (s, ξ) for a.e. x ∈ Ω. Moreover, assume that there exist a0 ∈ L1(Ω),
b0 ∈ R, a1 ∈ L1

loc(Ω) and b1 ∈ L∞loc(Ω) such that for a.e. x ∈ Ω and for all
(s, ξ) ∈ RN × RnN we have

|L (x, s, ξ)| ≤ a0(x) + b0|s|np/(n−p) + b0|ξ|p, (2.7)

|∇sL (x, s, ξ)| ≤ a1(x) + b1(x)|s|np/(n−p) + b1(x)|ξ|p, (2.8)

|∇ξL (x, s, ξ)| ≤ a1(x) + b1(x)|s|np/(n−p) + b1(x)|ξ|p. (2.9)

Conditions (2.8) and (2.9) imply that for every u ∈W 1,p
0 (Ω,RN ) we have

∇ξL (x, u,∇u) ∈ L1
loc(Ω; RnN ),

∇sL (x, u,∇u) ∈ L1
loc(Ω; RN ).

Therefore, for every u ∈W 1,p
0 (Ω,RN ) we have

−div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u) ∈ D ′(Ω; RN ).

Definition 2.22. We say that u is a weak solution of (2.6), if u ∈ W 1,p
0 (Ω,RN )

and
−div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u) = 0

in D ′(Ω; RN ).

If the integrand L is subjected to suitable restrictive conditions, it turns out
that f is of class C1 and

−div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u) ∈W−1,p′(Ω,RN )



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 13

for every u ∈W 1,p
0 (Ω,RN ). In this regular setting, we have that f satisfies condition

(PS)c, if and only of every sequence (uh) in W 1,p
0 (Ω,RN ) with f(uh) → c and

−div (∇ξL (x, uh,∇uh)) +∇sL (x, uh,∇uh) → 0

strongly in W−1,p′(Ω,RN ) has a strongly convergent subsequence in W 1,p
0 (Ω,RN ).

Now, a condition of this kind can be formulated also in our general context,
without any reference to the differentiability of the functional f .

Definition 2.23. Let c ∈ R. A sequence (uh) in W 1,p
0 (Ω,RN ) is said to be a

concrete Palais-Smale sequence at level c ((CPS)c-sequence, in short) for f , if
f(uh) → c,

−div (∇ξL (x, uh,∇uh)) +∇sL (x, uh,∇uh) ∈W−1,p′(Ω,RN )

eventually as h→∞ and

−div (∇ξL (x, uh,∇uh)) +∇sL (x, uh,∇uh) → 0

strongly in W−1,p′(Ω,RN ).
We say that f satisfies the concrete Palais-Smale condition at level c ((CPS)c in

short), if every (CPS)c-sequence for f admits a strongly convergent subsequence
in W 1,p

0 (Ω,RN ).

The next result allow us to connect these “concrete” notions with the abstract
critical point theory.

Theorem 2.24. The functional f is continuous and for all u ∈W 1,p
0 (Ω,RN ),

|df |(u) ≥ sup
{∫

Ω

(
∇ξL (x, u,∇u)·∇v+∇sL (x, u,∇u)·v

)
dx : v ∈ C∞c , ‖v‖1,p ≤ 1

}
.

Therefore, if |df |(u) < +∞ it follows

−div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u) ∈W−1,p′(Ω,RN )

and ∥∥− div (∇ξL (x, u,∇u)) +∇sL (x, u,∇u)
∥∥

1,p′
≤ |df |(u).

Corollary 2.25. Let u ∈ W 1,p
0 (Ω,RN ), c ∈ R and let (uh) be a sequence in

W 1,p
0 (Ω,RN ). Then the following facts hold:

(a) If u is a (lower) critical point of f , then u is a weak solution of (2.6));
(b) if (uh) is a (PS)c-sequence for f , then (uh) is a (CPS)c-sequence for f ;
(c) if f satisfies (CPS)c, then f satisfies (PS)c.

By means of the previous result, it is easy to deduce some versions of the Moun-
tain Pass Theorem adapted to the functional f .

Theorem 2.26. Let (D,S) be a compact pair, let ψ : S → W 1,p
0 (Ω,RN ) be a

continuous map and let

Φ =
{
ϕ ∈ C(D,W 1,p

0 (Ω,RN )) : ϕ|S = ψ
}
.

Assume that there exists a closed subset A of W 1,p
0 (Ω,RN ) such that

inf
A
f ≥ max

ψ(S)
f, A ∩ ψ(S) = ∅, A ∩ ϕ(D) 6= ∅ ∀ϕ ∈ Φ.



14 MARCO SQUASSINA EJDE-2006/MON. 07

If f satisfies the concrete Palais-Smale condition at level c = infϕ∈Φ maxϕ(D) f ,
then there exists a weak solution u of (2.6) with f(u) = c. Furthermore, if infA f ≥
c, then there exists a weak solution u of (2.6) with f(u) = c and u ∈ A.

Theorem 2.27. Suppose that

L (x,−s,−ξ) = L (x, s, ξ)

for a.e. x ∈ Ω and every (s, ξ) ∈ RN × RnN . Assume also that
(a) There exist ρ > 0, α > f(0) and a subspace V ⊂ W 1,p

0 (Ω,RN ) of finite
codimension with

∀u ∈ V : ‖u‖ = ρ ⇒ f(u) ≥ α;

(b) for every finite dimensional subspace W ⊂W 1,p
0 (Ω,RN ), there exists R > 0

with
∀u ∈W : ‖u‖ ≥ R ⇒ f(u) ≤ f(0);

(c) f satisfies (CPS)c for any c ≥ α.
Then there exists a sequence (uh) ⊂ W 1,p

0 (Ω,RN ) of weak solutions of (2.6) with
limh f(uh) = +∞.

3. Super-linear Elliptic Problems

We refer the reader to [132, 133]. Some parts of these publications have been
slightly modified to give the monograph a more uniform appearance.

3.1. Quasi-linear elliptic systems. Many papers have been published on the
study of multiplicity of solutions for quasi-linear elliptic equations via non-smooth
critical point theory; see e.g. [6, 8, 9, 33, 32, 36, 49, 112, 137]. However, for the
vectorial case only a few multiplicity results have been proven: [137, Theorem 3.2]
and recently [9, Theorem 3.2], where systems with multiple identity coefficients
are treated. In this section, we consider the following diagonal quasi-linear elliptic
system, in an open bounded set Ω ⊂ Rn with n ≥ 3,

−
n∑

i,j=1

Dj(akij(x, u)Diuk) +
1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u)DiuhDjuh = Dsk

G(x, u) in Ω ,

(3.1)
for k = 1, . . . , N , where u : Ω → RN and u = 0 on ∂Ω. To prove the existence of
weak solutions, we look for critical points of the functional f : H1

0 (Ω,RN ) → R,

f(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx−
∫

Ω

G(x, u) dx. (3.2)

This functional is not locally Lipschitz if the coefficients ahij depend on u; however,
as pointed out in [6, 33], it is possible to evaluate f ′,

f ′(u)(v) =
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjvh dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · vDiuhDjuh dx−

∫
Ω

DsG(x, u) · v dx

for all v ∈ H1
0 (Ω,RN ) ∩ L∞(Ω,RN ).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 15

To prove our main result and to provide some regularity of solutions, we consider
the following assumptions.

•
(
ahij(·, s)

)
is measurable in x for every s ∈ RN , and of class C1 in s for a.e.

x ∈ Ω with ahij = ahji. Furthermore, we assume that there exist ν > 0 and
C > 0 such that for a.e. x ∈ Ω, all s ∈ RN and ξ ∈ RnN

n∑
i,j=1

N∑
h=1

ahij(x, s)ξ
h
i ξ
h
j ≥ ν|ξ|2,

∣∣ahij(x, s)∣∣ ≤ C,
∣∣Dsa

h
ij(x, s)

∣∣ ≤ C (3.3)

and
n∑

i,j=1

N∑
h=1

s ·Dsa
h
ij(x, s)ξ

h
i ξ
h
j ≥ 0. (3.4)

• there exists a bounded Lipschitz function ψ : R → R, such that for a.e.
x ∈ Ω, for all ξ ∈ RnN , σ ∈ {−1, 1}N and r, s ∈ RN

n∑
i,j=1

N∑
h=1

(
1
2
Dsa

h
ij(x, s) · expσ(r, s) + ahij(x, s)Dsh

(expσ(r, s))h

)
ξhi ξ

h
j ≤ 0 (3.5)

where (expσ(r, s))i := σi exp[σi(ψ(ri)− ψ(si))] for each i = 1, . . . , N .
• the function G(x, s) is measurable in x for all s ∈ RN and of class C1 in s

for a.e. x ∈ Ω, with G(x, 0) = 0. Moreover for a.e. x ∈ Ω we will denote
with g(x, ·) the gradient of G with respect to s.

• for every ε > 0 there exists aε ∈ L2n/(n+2)(Ω) such that

|g(x, s)| ≤ aε(x) + ε|s|(n+2)/(n−2) (3.6)

for a.e. x ∈ Ω and all s ∈ RN and that there exist q > 2, R > 0 such that
for all s ∈ RN and for a.e. x ∈ Ω

|s| ≥ R⇒ 0 < qG(x, s) ≤ s · g(x, s). (3.7)

• there exists γ ∈ (0, q − 2) such that for all ξ ∈ RnN , s ∈ RN and a.e. in Ω

n∑
i,j=1

N∑
h=1

s ·Dsa
h
ij(x, s)ξ

h
i ξ
h
j ≤ γ

n∑
i,j=1

N∑
h=1

ahij(x, s)ξ
h
i ξ
h
j . (3.8)

Under these assumptions we will prove the following result.

Theorem 3.1. Assume that for a.e. x ∈ Ω and for each s ∈ RN

ahij(x,−s) = ahij(x, s), g(x,−s) = −g(x, s).

Then there exists a sequence (um) ⊂ H1
0 (Ω,RN ) of weak solutions to (3.1) such

that f(um) → +∞ as m→∞.

The above result is well known for the semi-linear scalar problem

−
n∑

i,j=1

Dj(aij(x)Diu) = g(x, u) in Ω

u = 0 on ∂Ω.



16 MARCO SQUASSINA EJDE-2006/MON. 07

Ambrosetti and Rabinowitz in [5, 120] studied this problem using techniques of
classical critical point theory. The quasi-linear scalar problem

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju = g(x, u) in Ω

u = 0 on ∂Ω ,

was studied in [32, 33, 36] and in [112] in a more general setting. In this case the
functional

f(u) =
1
2

∫
Ω

n∑
i,j=1

aij(x, u)DiuDju dx−
∫

Ω

G(x, u) dx

is continuous under appropriate conditions, but it is not locally Lipschitz. Con-
sequently, techniques of non-smooth critical point theory have to be applied. In
the vectorial case, to my knowledge, problem (3.1) has only been considered in
[137, Theorem 3.2] and recently in [9, Theorem 3.2] for coefficients of the type
ahkij (x, s) = δhkαij(x, s).

3.2. The concrete Palais-Smale condition. The first step for the (CPS)c to
hold is the boundedness of (CPS)c sequences.

Lemma 3.2. For all c ∈ R each (CPS)c sequence of f is bounded in H1
0 (Ω,RN ).

Proof. Let a0 ∈ L1(Ω) be such that for a.e. x ∈ Ω and all s ∈ RN

qG(x, s) ≤ s · g(x, s) + a0(x).

Now let (um) be a (CPS)c sequence for f and let wm → 0 in H−1(Ω,RN ) such
that for all v ∈ C∞c (Ω,RN ),

〈wm, v〉 =
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Djvh dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · vDiu
m
h Dju

m
h dx−

∫
Ω

g(x, um) · v.

Taking into account the previous Lemma, for every m ∈ N we obtain

− ‖wm‖H−1(Ω,RN )‖um‖H1
0 (Ω,RN )

≤
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · umDiu
m
h Dju

m
h dx−

∫
Ω

g(x, um) · um dx

≤
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx+

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · umDiu
m
h Dju

m
h dx− q

∫
Ω

G(x, um) dx+
∫

Ω

a0 dx.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 17

Taking into account the expression of f and assumption (3.8), we have that for
each m ∈ N,

− ‖wm‖H−1(Ω,RN )‖um‖H1
0 (Ω,RN )

≤ −
(q

2
− 1

) ∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · umDiu
m
h Dju

m
h dx+ qf(um) +

∫
Ω

a0 dx

≤ −
(q

2
− 1− γ

2

) ∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

+ qf(um) +
∫

Ω

a0 dx.

Because of (3.3), for each m ∈ N,

ν(q − 2− γ)‖Dum‖22 ≤ (q − 2− γ)
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

≤ 2‖wm‖H−1(Ω,RN )‖um‖H1
0 (Ω,RN ) + 2qf(um) + 2

∫
Ω

a0dx.

Since wm → 0 in H−1(Ω,RN ), (um) is a bounded sequence in H1
0 (Ω,RN ). �

Lemma 3.3. If condition (3.6) holds, then the map

H1
0 (Ω,RN ) −→ L2n/(n+2)(Ω,RN )

u 7−→ g(x, u)

is completely continuous.

The statement of the above lemma is a direct consequence of [36, Theorem 2.2.7].
The next result is crucial for the (CPS)c condition to hold for our elliptic system.

Lemma 3.4. Let (um) be a bounded sequence in H1
0 (Ω,RN ), and set

〈wm, v〉 =
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Djvh dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · vDiu
m
h Dju

m
h dx

for all v ∈ C∞c (Ω,RN ). If (wm) is strongly convergent to some w in H−1(Ω,RN ),
then (um) admits a strongly convergent subsequence in H1

0 (Ω,RN ).

Proof. Since (um) is bounded, we have um ⇀ u for some u up to a subsequence.
Each component umk satisfies (2.5) in [22], so we may suppose that Diu

m
k → Diuk



18 MARCO SQUASSINA EJDE-2006/MON. 07

a.e. in Ω for all k = 1, . . . , N (see also [54]). We first prove that∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · uDiuhDjuh dx = 〈w, u〉.

(3.9)

Let ψ be as in assumption (3.5) and consider the following test functions

vm = ϕ(σ1 exp[σ1(ψ(u1)− ψ(um1 ))], . . . , σN exp[σN (ψ(uN )− ψ(umN ))]),

where ϕ ∈ C∞c (Ω), ϕ ≥ 0 and σl = ±1 for all l. Therefore, since we have

Djv
m
k = (σkDjϕ+ (ψ′(uk)Djuk − ψ′(umk )Dju

m
k )ϕ) exp[σk(ψ(uk)− ψ(umk ))],

we deduce that for all m ∈ N,∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h (σhDjϕ+ ψ′(uh)Djuhϕ) exp[σh(ψ(uh)− ψ(umh ))] dx

+
∫

Ω

n∑
i,j=1

N∑
h,l=1

σl
2
Dsl

ahij(x, u
m) exp[σl(ψ(ul)− ψ(uml ))]Diu

m
h Dju

m
h ϕdx

−
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h ψ

′(umh ) exp[σh(ψ(uh)− ψ(umh ))]ϕdx

= 〈wm, vm〉.

Let us study the behavior of each term of the previous equality as m→∞. First
of all, if v = (σ1ϕ, . . . , σNϕ), we have that vm ⇀ v implies

lim
m
〈wm, vm〉 = 〈w, v〉. (3.10)

Since um ⇀ u, by Lebesgue’s Theorem we obtain

lim
m

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h (Dj(σhϕ) (3.11)

+ ϕψ′(uh)Djuh) exp[σh(ψ(uh)− ψ(umh ))] dx (3.12)

=
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)Diuh(Djvh + ϕψ′(uh)Djuh) dx. (3.13)

Finally, note that by assumption (3.5) we have
n∑

i,j=1

N∑
h=1

( N∑
l=1

σl
2
Dsl

ahij(x, u
m) exp[σl(ψ(ul)− ψ(uml ))]

−ahij(x, um)ψ′(umh ) exp[σh(ψ(uh)− ψ(umh ))]
)
Diu

m
h Dju

m
h ≤ 0.

Hence, we can apply Fatou’s Lemma to obtain

lim sup
m

{1
2

∫
Ω

n∑
i,j=1

N∑
h,l=1

Dsl
ahij(x, u

m) exp[σl(ψ(ul)− ψ(uml ))]Diu
m
h Dju

m
h (σlϕ) dx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 19

−
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h ψ

′(umh ) exp[σh(ψ(uh)− ψ(umh ))]ϕdx
}

≤ 1
2

∫
Ω

n∑
i,j=1

N∑
h,l=1

Dsl
ahij(x, u)DiuhDjuh(σlϕ) dx

−
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuhψ
′(uh)ϕdx ,

which, together with (3.10) and (3.12), yields∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjvh dx+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · vDiuhDjuh dx

≥ 〈w, v〉

for all test functions v = (σ1ϕ, . . . , σNϕ) with ϕ ∈ C∞c (Ω,RN ), ϕ ≥ 0. Since we
may exchange v with −v we get∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjvh dx+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · vDiuhDjuh dx

= 〈w, v〉

for all test functions v = (σ1ϕ, . . . , σNϕ), and since every function v ∈ C∞c (Ω,RN )
can be written as a linear combination of such functions, we infer (3.9). Now, let
us prove that

lim sup
m

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx ≤

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx.

(3.14)
Because of (3.4), Fatou’s Lemma implies that∫

Ω

n∑
i,j=1

N∑
h=1

u ·Dsa
h
ij(x, u)DiuhDjuh dx

≤ lim inf
m

∫
Ω

n∑
i,j=1

N∑
h=1

um ·Dsa
h
ij(x, u

m)Diu
m
h Dju

m
h dx.

Combining this fact with (3.9), we deduce that

lim sup
m

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

= lim sup
m

[
− 1

2

∫
Ω

n∑
i,j=1

N∑
h=1

um ·Dsa
h
ij(x, u

m)Diu
m
h Dju

m
h dx+ 〈wm, um〉

]

≤ −1
2

∫
Ω

n∑
i,j=1

N∑
h=1

u ·Dsa
h
ij(x, u)DiuhDjuh dx+ 〈w, u〉



20 MARCO SQUASSINA EJDE-2006/MON. 07

=
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx ,

so that (3.14) is proved. Finally, by (3.3) we have

ν‖Dum −Du‖22

≤
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m) (Diu

m
h Dju

m
h − 2Diu

m
h Djuh +DiuhDjuh) dx.

Hence, by (3.14) we obtain

lim sup
m

‖Dum −Du‖2 ≤ 0

which proves that um → u in H1
0 (Ω,RN ). �

We now come to the (CPS)c condition for system (3.1).

Theorem 3.5. f satisfies (CPS)c condition for each c ∈ R.

Proof. Let (um) be a (CPS)c sequence for f . Since (um) is bounded in H1
0 (Ω,RN ),

from Lemma 3.3 we deduce that, up to a subsequence, (g(x, um)) is strongly con-
vergent in H−1(Ω,RN ). Applying Lemma 3.4, we conclude the present proof. �

3.3. Existence of multiple solutions for elliptic systems. We now prove the
main result, which is an extension of theorems of [33, 36] and a generalization of
[9, Theorem 3.2] to systems in diagonal form.

Proof of Theorem 3.1. We want to apply [36, Theorem 2.1.6]. First of all, because
of Theorem 3.5, f satisfies (CPS)c for all c ∈ R. Whence, (c) of [36, Theorem
2.1.6] is satisfied. Moreover we have

ν

2

∫
Ω

|Du|2 dx−
∫

Ω

G(x, u) dx ≤ f(u) ≤ 1
2
nNC

∫
Ω

|Du|2 dx−
∫

Ω

G(x, u) dx.

We want to prove that assumptions (a) and (b) of [36, Theorem 2.1.6] are also
satisfied. Let us observe that, instead of (b) of [36, Theorem 2.1.6], it is enough
to find a sequence (Wn) of finite dimensional subspaces with dim(Wn) → +∞
satisfying the inequality of (b) (see also [102, Theorem 1.2]). Let W be a finite
dimensional subspace of H1

0 (Ω; RN ) ∩ L∞(Ω,RN ). From (3.7) we deduce that for
all s ∈ RN with |s| ≥ R

G(x, s) ≥
G

(
x,R s

|s|

)
Rq

|s|q ≥ b0(x)|s|q,

where
b0(x) = R−q inf{G(x, s) : |s| = R} > 0

a.e. x ∈ Ω. Therefore, there exists a0 ∈ L1(Ω) such that

G(x, s) ≥ b0(x)|s|q − a0(x) (3.15)

a.e. x ∈ Ω and for all s ∈ RN . Since b0 ∈ L1(Ω), we may define a norm ‖ · ‖G on
W by

‖u‖G =
( ∫

Ω

b0|u|q dx
)1/q

.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 21

Since W is finite dimensional and q > 2, from (3.15) it follows

lim
‖u‖G→+∞, u∈W

f(u) = −∞

and condition (b) of [36, Theorem 2.1.6] is clearly fulfilled too for a sufficiently large
R > 0. Let now (λh, uh) be the sequence of eigenvalues and eigenvectors for the
problem

∆u = −λu in Ω
u = 0 on ∂Ω.

Let us prove that there exist h0, α > 0 such that

∀u ∈ V + : ‖Du‖2 = 1 ⇒ f(u) ≥ α,

where V + = span
{
uh ∈ H1

0 (Ω,RN ) : h ≥ h0

}
. In fact, given u ∈ V + and ε > 0,

we find
a(1)
ε ∈ C∞c (Ω), a(2)

ε ∈ L2n/(n+2)(Ω),

such that ‖a(2)
ε ‖2n/(n+2) ≤ ε and

|g(x, s)| ≤ a(1)
ε (x) + a(2)

ε (x) + ε|s|(n+2)/(n−2).

If u ∈ V +, it follows that

f(u) ≥ ν

2
‖Du‖22 −

∫
Ω

G(x, u) dx

≥ ν

2
‖Du‖22 −

∫
Ω

((
a(1)
ε + a(2)

ε

)
|u|+ n− 2

2n
ε|u|2n/(n−2)

)
dx

≥ ν

2
‖Du‖22 − ‖a(1)

ε ‖2‖u‖2 − c1‖a(2)
ε ‖2n/(n+2)‖Du‖2 − εc2‖Du‖2n/(n−2)

2

≥ ν

2
‖Du‖22 − ‖a(1)

ε ‖2‖u‖2 − c1ε‖Du‖2 − εc2‖Du‖2n/(n−2)
2 .

Then if h0 is sufficiently large, from the fact that (λh) diverges, for all u ∈ V +,
‖Du‖2 = 1 implies

‖a(1)
ε ‖2‖u‖2 ≤

ν

6
.

Hence, for ε > 0 small enough, ‖Du‖2 = 1 implies that f(u) ≥ ν/6.

Finally, set V − = span
{
uh ∈ H1

0 (Ω,RN ) : h < h0

}
, we have the decomposition

H1
0 (Ω; RN ) = V + ⊕ V −.

Therefore, since the hypotheses for [36, Theorem 2.1.6] are fulfilled, we can find a
sequence (um) of weak solution of system (3.1) such that limm f(um) = +∞. The
proof is complete. �

3.4. Regularity of weak solutions for elliptic systems. Consider the nonlinear
elliptic system∫

Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x, u)DiuhDjvk dx =
∫

Ω

b(x, u,Du) · v dx (3.16)

for all v ∈ H1
0 (Ω; RN ). For l = 1, . . . , N , we choose

bl(x, u,Du) =
{
−

n∑
i,j=1

N∑
h,k=1

Dsl
ahkij (x, u)DiuhDjuk + gl(x, u)

}
.



22 MARCO SQUASSINA EJDE-2006/MON. 07

Assume that there exist c > 0 and q < n+2
n−2 such that for all s ∈ RN and a.e. in Ω

|g(x, s)| ≤ c (1 + |s|q) . (3.17)

Then it follows that for every M > 0, there exists C(M) > 0 such that for a.e.
x ∈ Ω, for all ξ ∈ RnN and s ∈ RN with |s| ≤M

|b(x, s, ξ)| ≤ c(M)
(
1 + |ξ|2

)
. (3.18)

A nontrivial regularity theory for quasi-linear systems (see, [75, Chapter VI]) yields
the following

Theorem 3.6. For every weak solution u ∈ H1(Ω,RN )∩L∞(Ω,RN ) of the system
(3.1) there exist an open subset Ω0 ⊆ Ω and s > 0 such that

∀p ∈ (n,+∞) : u ∈ C0,1−n
p (Ω0; RN ),

Hn−s(Ω\Ω0) = 0.

For the proof of the above theorem, see [75, Chapter VI]. We now consider the
particular case when ahkij (x, s) = αij(x, s)δhk, and provide an almost everywhere
regularity result.

Lemma 3.7. Assume that (3.18) holds. Then the weak solutions u ∈ H1
0 (Ω,RN )

of the system∫
Ω

n∑
i,j=1

N∑
h=1

aij(x, u)DiuhDjvh dx+

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsaij(x, u) · vDiuhDjuh dx =
∫

Ω

g(x, u) · v dx (3.19)

for all v ∈ C∞c (Ω,RN ), belong to L∞(Ω,RN ).

Proof. By [137, Lemma 3.3], for each (CPS)c sequence (um) there exist u ∈ H1
0 ∩

L∞ and a subsequence (umk) with umk ⇀ u. Then, given a weak solution u,
consider the sequence (um) such that each element is equal to u and the assertion
follows. �

We can finally state a partial regularity result for our system.

Theorem 3.8. Assume condition (3.18) and let u ∈ H1
0 (Ω,RN ) be a weak solution

of the system∫
Ω

n∑
i,j=1

N∑
h=1

aij(x, u)DiuhDjvh dx+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsaij(x, u) · vDiuhDjuh dx

=
∫

Ω

g(x, u) · v dx

for all v ∈ C∞c (Ω,RN ). Then there exist an open subset Ω0 ⊆ Ω and s > 0 such
that

∀p ∈ (n,+∞) : u ∈ C0,1−n
p (Ω0; RN ), Hn−s(Ω\Ω0) = 0.

To prove the above theorem, it suffices to combine the previous Lemma with
Theorem 3.6.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 23

3.5. Fully nonlinear scalar problems. Recently, some results for the quite gen-
eral problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) in Ω
u = 0 on ∂Ω ,

(3.20)

have been considered in [6, 7] and [112]. The goal of this section is to extend some of
the results of [6, 112]. To solve (3.20), we shall look for critical points of functionals
f : W 1,p

0 (Ω) → R given by

f(u) =
∫

Ω

L (x, u,∇u) dx−
∫

Ω

G(x, u) dx. (3.21)

In general, f is continuous but not even locally Lipschitz unless L does not depend
on u or L is subjected to some very restrictive growth conditions. Then, again we
shall refer to non-smooth critical point theory.

We assume that L : Ω×R×Rn → R is measurable in x for all (s, ξ) ∈ R×Rn,
of class C1 in (s, ξ) for a.e. x ∈ Ω, the function L (x, s, ·) is strictly convex and
L (x, s, 0) = 0 for a.e. x ∈ Ω. Furthermore, we will assume that:

• there exist a ∈ L1(Ω) and b0, ν > 0 such that

ν|ξ|p ≤ L (x, s, ξ) ≤ a(x) + b0|s|p + b0|ξ|p , (3.22)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn;
• for each ε > 0 there exists aε ∈ L1(Ω) such that

|DsL (x, s, ξ)| ≤ aε(x) + ε|s|p
∗

+ b1|ξ|p , (3.23)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R × Rn, with b1 ∈ R independent of ε.
Furthermore, there exists a1 ∈ Lp

′
(Ω) such that

|∇ξL (x, s, ξ)| ≤ a1(x) + b1|s|
p∗
p′ + b1|ξ|p−1, (3.24)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn;
• there exists R > 0 such that

|s| ≥ R⇒ DsL (x, s, ξ)s ≥ 0 , (3.25)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn;
• G : Ω× R → R is a Carathéodory function such that

G(x, s) ≤ d(x)|s|p + b|s|p
∗

(3.26)

lim
s→0

G(x, s)
|s|p

= 0 (3.27)

for a.e. x ∈ Ω and all s ∈ R, where d ∈ L
n
p (Ω) and b ∈ R. Moreover,

G(x, s) =
∫ s

0

g(x, τ) dτ

and there exist c1, c2 > 0 such that

|g(x, s)| ≤ c1 + c2|s|σ

for a.e. x ∈ Ω and each s ∈ R, where σ < p∗ − 1.



24 MARCO SQUASSINA EJDE-2006/MON. 07

• there is q > p and R′ > 0 such that for each ε > 0 there is aε ∈ L1(Ω) with

0 < qG(x, s) ≤ g(x, s)s , (3.28)

qL (x, s, ξ)−∇ξL (x, s, ξ) · ξ −DsL (x, s, ξ)s ≥ ν|ξ|p − aε(x)− ε|s|p (3.29)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn with |s| ≥ R′.

Under the previous assumptions, the following is our main result.

Theorem 3.9. The boundary value problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) in Ω
u = 0 on ∂Ω

has at least one nontrivial weak solution u ∈W 1,p
0 (Ω).

This result is an extension of [6, Theorem 3.3], since instead of assuming that

∀s ∈ R : qL (x, s, ξ)−∇ξL (x, s, ξ) · ξ −DsL (x, s, ξ)s ≥ ν|ξ|p ,

for a.e. x ∈ Ω and for all ξ ∈ Rn, we only request condition (3.29). In this way
the proof of Lemma (3.12) becomes more difficult. The key-point, to deal with the
more general assumption, is constituted by Lemma (3.11).

Similarly, in [112, Theorem 1], a multiplicity result for (3.20) is proved, assuming
that

∀s ∈ R : DsL (x, s, ξ)s ≥ 0 ,

∀s ∈ R : qL (x, s, ξ)−∇ξL (x, s, ξ) · ξ −DsL (x, s, ξ)s ≥ ν|ξ|p ,

for a.e. x ∈ Ω and for all ξ ∈ Rn, which are both stronger than (3.25) and (3.29).
In particular, the first inequality above and the more general condition (3.25) are
involved in Theorem 3.10.

Finally, let us point out that the growth conditions (3.22) - (3.24) are a relaxation
of those of [6, 112], where it is assumed that

ν|ξ|p ≤ L (x, s, ξ) ≤ β|ξ|p, |DsL (x, s, ξ)| ≤ γ|ξ|p,
|∇ξL (x, s, ξ)| ≤ a1(x) + b1|s|p−1 + b1|ξ|p−1,

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.

3.6. The concrete Palais-Smale condition. Let us point out that as a conse-
quence of assumption (3.22) and convexity of L (x, s, ·), we can find M > 0 such
that for each ε > 0 there is aε ∈ L1(Ω) with

∇ξL (x, s, ξ) · ξ ≥ ν|ξ|p − a(x)− b0|s|p, (3.30)

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ + aε(x) + ε|s|p
∗
, (3.31)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.
We now come to a local compactness property, which is crucial for the (CPS)c

condition to hold. This result improves [112, Lemma 2], since (3.29) relaxes condi-
tion (8) in [112].



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 25

Theorem 3.10. Let (uh) be a bounded sequence in W 1,p
0 (Ω) and set

〈wh, v〉 =
∫

Ω

∇ξL (x, uh,∇uh) · ∇v dx+
∫

Ω

DsL (x, uh,∇uh)v dx , (3.32)

for all v ∈ C∞c (Ω). If (wh) is strongly convergent to some w in W−1,p′(Ω), then
(uh) admits a strongly convergent subsequence in W 1,p

0 (Ω).

Proof. Since (uh) is bounded in W 1,p
0 (Ω), we find a u in W 1,p

0 (Ω) such that, up to
a subsequence,

∇uh ⇀ ∇u in Lp(Ω), uh → u in Lp(Ω), uh(x) → u(x) for a.e. x ∈ Ω.

By [22, Theorem 2.1], up to a subsequence, we have

∇uh(x) → ∇u(x) for a.e. x ∈ Ω. (3.33)

Therefore, by (3.24) we deduce that

∇ξL (x, uh,∇uh) ⇀ ∇ξL (x, u,∇u) in Lp
′
(Ω,Rn).

We now want to prove that u solves the equation

∀v ∈ C∞c (Ω) : 〈w, v〉 =
∫

Ω

∇ξL (x, u,∇u) ·∇v dx+
∫

Ω

DsL (x, u,∇u)v dx. (3.34)

To this aim, let us test equation (3.32) with the functions

vh = ϕ exp{−M(uh +R)+} , ϕ ∈W 1,p
0 (Ω) ∩ L∞(Ω) , ϕ ≥ 0.

It results that for each h ∈ N,∫
Ω

∇ξL (x, uh,∇uh) · ∇ϕ exp{−M(uh +R)+} dx−
〈
wh, ϕ exp{−M(uh +R)+}

〉
+

∫
Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+

]
ϕ exp{−M(uh +R)+} dx = 0.

Of course, for a.e. x ∈ Ω, we obtain

ϕ exp{−M(uh +R)+} → ϕ exp{−M(u+R)+}.
Since by inequality (3.31) and (3.25) for each ε > 0 and h ∈ N we have[

DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+
]

×ϕ exp{−M(uh +R)+} − ε|uh|p
∗
ϕ ≤ aε(x)ϕ ,

Fatou’s Lemma implies that for each ε > 0,

lim sup
h

∫
Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+

]
ϕe−M(uh+R)+

− ε|uh|p
∗
ϕdx

≤
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+

]
ϕe−M(u+R)+ − ε|u|p

∗
ϕdx.

Since (uh) is bounded in Lp
∗
(Ω), we find c > 0 such that for each ε > 0

lim sup
h

∫
Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+

]
ϕe−M(uh+R)+dx

≤
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+

]
ϕe−M(u+R)+dx− cε.



26 MARCO SQUASSINA EJDE-2006/MON. 07

Letting ε→ 0, the previous inequality yields

lim sup
h

∫
Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+

]
× ϕe−M(uh+R)+ dx

≤
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+

]
ϕe−M(u+R)+dx.

Note that we have also

ϕ exp{−M(uh +R)+}⇀ ϕ exp{−M(u+R)+} in W 1,p
0 (Ω).

Moreover,

∇ϕ exp{−M(uh +R)+} → ∇ϕ exp{−M(u+R)+} in Lp(Ω,Rn),

so that ∫
Ω

∇ξL (x, uh,∇uh) · ∇ϕ exp{−M(uh +R)+}dx

→
∫

Ω

∇ξL (x, u,∇u) · ∇ϕ exp{−M(u+R)+}dx.

Therefore, we conclude that∫
Ω

∇ξL (x, u,∇u) · ∇ϕ exp{−M(u+R)+} dx−
〈
w,ϕ exp{−M(u+R)+}

〉
+

∫
Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+

]
ϕ exp{−M(u+R)+}dx

≥ 0.

Consider now the test functions

ϕk := ϕH(
u

k
) exp{M(u+R)+} , ϕ ∈ C∞c (Ω) , ϕ ≥ 0 ,

where H ∈ C1(R), H = 1 in [− 1
2 ,

1
2 ] and H = 0 in ]−∞,−1] ∪ [1,+∞[. It follows

that ∫
Ω

∇ξL (x, u,∇u) · ∇ϕk exp{−M(u+R)+} dx− 〈w,ϕH(
u

k
)〉

+
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+

]
ϕH(

u

k
) dx ≥ 0.

Furthermore, standard computations yield

∇ϕk = exp{M(u+R)+}
[
∇ϕH(

u

k
) +H ′(

u

k
)
ϕ

k
∇u+M∇(u+R)+ϕH(

u

k
)
]
.

Since ϕH(uk ) goes to ϕ in W 1,p
0 (Ω), as k → +∞ we have〈
w,ϕH(

u

k
)
〉
→ 〈w,ϕ〉 .

By the properties of H and the growth conditions on ∇ξL , letting k → +∞ yields∫
Ω

∇ξL (x, u,∇u) · ∇ϕH(
u

k
) →

∫
Ω

∇ξL (x, u,∇u) · ∇ϕdx ,∫
Ω

∇ξL (x, u,∇u) · ∇uH ′(
u

k
)
ϕ

k
dx→ 0 ,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 27∫
Ω

M∇ξL (x, u,∇u) · ∇(u+R)+ϕH(
u

k
) →

∫
Ω

M∇ξL (x, u,∇u) · ∇(u+R)+ϕ.

Whence, we conclude that for all ϕ ∈ C∞c (Ω),

ϕ ≥ 0 ⇒ 〈w,ϕ〉 ≤
∫

Ω

∇ξL (x, u,∇u) · ∇ϕdx+
∫

Ω

DsL (x, u,∇u)ϕdx.

Choosing now as test functions

vh := ϕ exp{−M(uh −R)−} ,
where as before ϕ ≥ 0, we obtain the opposite inequality so that (3.34) is proven.

In particular, taking into account the Brezis-Browder type results, we immedi-
ately obtain

〈w, u〉 =
∫

Ω

∇ξL (x, u,∇u) · ∇u dx+
∫

Ω

DsL (x, u,∇u)u dx. (3.35)

The final step is to show that (uh) goes to u in W 1,p
0 (Ω). Consider the function

ζ : R → R defined by

ζ(s) =


Ms if 0 < s < R

MR if s ≥ R

−Ms if −R < s < 0
MR if s ≤ −R ,

(3.36)

and let us prove that

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uh exp{ζ(uh)} dx

≤
∫

Ω

∇ξL (x, u,∇u) · ∇u exp{ζ(u)} dx.

Since uh exp{ζ(uh)} are admissible test functions for (3.32), we have∫
Ω

∇ξL (x, uh,∇uh) · ∇uh exp{ζ(uh)} dx− 〈wh, uh exp{ζ(uh)}〉+

+
∫

Ω

[DsL (x, uh,∇uh) + ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh]uh exp{ζ(uh)} dx = 0.

Let us observe that (3.33) implies that

∇ξL (x, uh,∇uh) · ∇uh → ∇ξL (x, u,∇u) · ∇u for a.e. x ∈ Ω.

Since by inequality (3.31) for each ε > 0 and h ∈ N we have

[−DsL (x, uh,∇uh)− ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh]

× uh exp{ζ(uh)} −R exp{MR}ε|uh|p
∗

≤ R exp{MR}aε(x) ,
Fatou’s Lemma yields

lim sup
h

∫
Ω

[
−DsL (x, uh,∇uh)− ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh

]
× uhe

ζ(uh) −ReMRε|uh|p
∗
dx

≤
∫

Ω

[
−DsL (x, u,∇u)− ζ ′(u)∇ξL (x, u,∇u) · ∇u

]
u exp{ζ(u)}



28 MARCO SQUASSINA EJDE-2006/MON. 07

−ReMRε|u|p
∗
dx.

Therefore, since (uh) is bounded in Lp
∗
(Ω), we find c > 0 such that for all ε > 0

lim sup
h

∫
Ω

[−DsL (x, uh,∇uh)− ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh]uheζ(uh)dx

≤
∫

Ω

[−DsL (x, u,∇u)− ζ ′(u)∇ξL (x, u,∇u) · ∇u]ueζ(u)dx− cε.

Taking into account that ε is arbitrary, we conclude that

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uh exp{ζ(uh)} dx

= lim sup
h

{∫
Ω

[−DsL (x, uh,∇uh)− ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh]uheζ(uh)dx

+ 〈wh, uh exp{ζ(uh)}〉
}

≤
∫

Ω

[−DsL (x, u,∇u)− ζ ′(u)∇ξL (x, u,∇u) · ∇u]ueζ(u)dx+ 〈w, u exp{ζ(u)}〉

=
∫

Ω

∇ξL (x, u,∇u) · ∇ueζ(u) dx.

In particular, we have∫
Ω

∇ξL (x, u,∇u) · ∇ueζ(u) dx ≤ lim inf
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uheζ(uh)dx

≤ lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uheζ(uh) dx

≤
∫

Ω

∇ξL (x, u,∇u) · ∇ueζ(u)dx,

; namely,

lim
h

∫
Ω

∇ξL (x, uh,∇uh)·∇uh exp{ζ(uh)}dx =
∫

Ω

∇ξL (x, u,∇u)·∇u exp{ζ(u)}dx.

Therefore, by (3.30), generalized Lebesgue’s theorem yields

lim sup
h

∫
Ω

|∇uh|pdx ≤
∫

Ω

|∇u|pdx,

that implies the strong convergence of (uh) to u in W 1,p
0 (Ω). �

Lemma 3.11. Let c ∈ R and let (uh) be a (CPS)c-sequence in W 1,p
0 (Ω). Then for

each ε > 0 and % > 0 there exists K%,ε > 0 such that, for all h ∈ N,∫
{|uh|≤%}

∇ξL (x, uh,∇uh) · ∇uh dx

≤ ε

∫
{%<|uh|<K%,ε}

∇ξL (x, uh,∇uh) · ∇uh dx+K%,ε

and ∫
{|uh|≤%}

|∇uh|p dx ≤ ε

∫
{%<|uh|<K%,ε}

|∇uh|p dx+K%,ε.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 29

Proof. Let σ, ε > 0 and % > 0. For all v ∈W 1,p
0 (Ω), we set

〈wh, v〉 =
∫

Ω

∇ξL (x, uh,∇uh) · ∇v dx

+
∫

Ω

DsL (x, uh,∇uh)v dx−
∫

Ω

g(x, uh)v dx.
(3.37)

Let us now consider ϑ1 : R → R given by

ϑ1(s) =


s if |s| < σ

−s+ 2σ if σ ≤ s < 2σ
−s− 2σ if − 2σ < s ≤ −σ
0 if |s| ≥ 2σ.

(3.38)

Then, testing (3.37) with ϑ1(uh) ∈ L∞([−2σ, 2σ]), we obtain∫
Ω

∇ξL (x, uh,∇uh) · ∇ϑ1(uh) dx+
∫

Ω

DsL (x, uh,∇uh)ϑ1(uh) dx

≤
∫

Ω

g(x, uh)ϑ1(uh) dx+ ‖wh‖−1,p′‖ϑ1(uh)‖1,p.

Then, it follows that∫
{|uh|≤σ}

∇ξL (x, uh,∇uh) · ∇uh dx−
∫
{σ<|uh|≤2σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+
∫
{|uh|≤σ}

DsL (x, uh,∇uh)uh dx+
∫
{σ<|uh|≤2σ}

DsL (x, uh,∇uh)ϑ1(uh) dx

≤
∫

Ω

(
c1 + c2|2σ|

n(p−1)+p
n−p

)
σ dx+

4
p′
p

p′p
p′
p ν

p′
p

‖wh‖p
′

−1,p′ +
ν

4
‖ϑ1(uh)‖p1,p.

Let K0 > 0 be such that ‖wh‖−1,p′ ≤ K0. Then, since by (3.30) we have

ν‖ϑ1(uh)‖p1,p

≤
∫
{|uh|≤σ}

ν|∇uh|p dx+
∫
{σ<|uh|≤2σ}

ν|∇uh|p dx

≤
∫
{|uh|≤σ}

∇ξL (x, uh,∇uh) · ∇uh dx+
∫
{σ<|uh|≤2σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+
∫
{|uh|≤σ}

a(x) dx+ b0

∫
{|uh|≤σ}

|uh|p dx+
∫
{σ<|uh|≤2σ}

a(x) dx

+ b0

∫
{σ<|uh|≤2σ}

|uh|p dx ,

taking into account (3.31), we get for a sufficiently small value of σ > 0,(
1− σM − 1

4
) ∫

{|uh|≤σ}
∇ξL (x, uh,∇uh) · ∇uh dx

≤
(
1 + σM +

1
4
) ∫

{σ<|uh|≤2σ}
∇ξL (x, uh,∇uh) · ∇uh dx

+
∫

Ω

(
c1 + c2|2σ|

n(p−1)+p
n−p

)
σ dx+

4
p′
p

p′p
p′
p ν

p′
p

Kp′

0



30 MARCO SQUASSINA EJDE-2006/MON. 07

+
∫

Ω

aε(x) dx+
[
b0(2p + 1)σp + εσp

∗+1
]
Ln(Ω).

Whence, we have shown an inequality of the type∫
{|uh|≤σ}

∇ξL (x, uh,∇uh) · ∇uh dx

≤ K1

∫
{σ<|uh|≤2σ}

∇ξL (x, uh,∇uh) · ∇uh dx+K2.

Let us now define for each k ≥ 1 the functions ϑ2k , ϑ2k−1 : R → R by setting

ϑ2k(s) =



0 if |s| ≤ kσ

s− kσ if kσ < s < (k + 1)σ
s+ kσ if − (k + 1)σ < s < −kσ
−s+ (k + 2)σ if (k + 1)σ ≤ s < (k + 2)σ
−s− (k + 2)σ if − (k + 2)σ < s ≤ −(k + 1)σ
0 if |s| ≥ (k + 1)σ ,

and

ϑ2k−1(s) =



s
k if |s| ≤ kσ

−s+ (k + 1)σ if kσ < s < (k + 1)σ
s− (k + 1)σ if (k + 1)σ < s < (k + 2)σ
−s− (k + 1)σ if − (k + 1)σ ≤ s < −kσ
s+ (k + 1)σ if − (k + 2)σ < s ≤ −(k + 1)σ
0 if |s| ≥ (k + 1)σ.

Therefore, by iterating on k, we obtain the k-th inequality∫
{|uh|≤kσ}

∇ξL (x, uh,∇uh) · ∇uh dx

≤ K1(k)
∫
{kσ<|uh|≤(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx+K2(k).
(3.39)

Let now choose k ≥ 1 such that kσ ≥ % and kσ ≥ R. Take 0 < δ < 1 and let
ϑδ : R → R be the function defined by setting

ϑδ(s) =



0 if |s| ≤ kσ

s− kσ if kσ < s < (k + 1)σ
s+ kσ if − (k + 1)σ < s < −kσ
−δs+ σ + δ(k + 1)σ if (k + 1)σ ≤ s < (k + 1)σ + σ

δ

−δs− σ − δ(k + 1)σ if − (k + 1)σ − σ
δ < s ≤ −(k + 1)σ

0 if |s| ≥ (k + 1)σ + σ
δ .

As before, we get∫
Ω

∇ξL (x, uh,∇uh) · ∇ϑδ(uh) dx+
∫

Ω

DsL (x, uh,∇uh)ϑδ(uh) dx

≤
∫

Ω

g(x, uh)ϑδ(uh) dx+
1

p′p
p′
p δ

p′
p

‖wh‖p
′

−1,p′ + δ‖ϑδ(uh)‖p1,p.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 31

Taking into account (3.25), by computations, we deduce that∫
Ω

DsL (x, uh,∇uh)ϑδ(uh) dx ≥ 0.

Moreover, we have as before

‖ϑδ(uh)‖p1,p

≤
∫
{kσ<|uh|≤(k+1)σ}

|∇uh|p dx+
∫
{|uh|≥(k+1)σ}

|∇uh|p dx

≤ 1
ν

∫
{kσ<|uh|≤(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+
1
ν

∫
{|uh|≥(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+
1
ν

∫
{kσ<|uh|≤(k+1)σ}

a(x) dx+
b0
ν

∫
{kσ<|uh|≤(k+1)σ}

|uh|p dx

+
1
ν

∫
{|uh|≥(k+1)σ}

a(x) dx+
b0
ν

∫
{(k+1)σ+ σ

δ ≥|uh|≥(k+1)σ}
|uh|p dx ,

so that (
1− δ

ν

) ∫
{kσ<|uh|≤(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

≤
(
δ +

δ

ν

) ∫
{|uh|>(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+
∫

Ω

(
c1 + c2

∣∣(k + 1)σ +
σ

δ

∣∣n(p−1)+p
n−p

)
σ dx+

1

p′p
p′
p δ

p′
p

Kp′

0

+
2
ν

∫
Ω

a(x) dx+
b0
ν

[
(k + 1)p +

(
(k + 1) +

1
δ

)p]
σpLn(Ω).

Therefore, we get∫
{kσ<|uh|≤(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

≤ ≤ νδ + δ

ν − δ

∫
{|uh|≥(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx+K3(k, δ).

Combining this inequality with (3.39) we conclude that∫
{|uh|≤%}

∇ξL (x, uh,∇uh) · ∇uh dx

≤
∫
{|uh|≤kσ}

∇ξL (x, uh,∇uh) · ∇uh dx

≤ K1(k)
∫
{kσ<|uh|≤(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx+K2(k)

≤ K1(k)
νδ + δ

ν − δ

∫
{|uh|>(k+1)σ}

∇ξL (x, uh,∇uh) · ∇uh dx

+K1(k)K3(k, δ) +K2(k) ≤



32 MARCO SQUASSINA EJDE-2006/MON. 07

≤ ε

∫
{|uh|>%}

∇ξL (x, uh,∇uh) · ∇uh dx+K%,ε,

where we have fixed δ > 0 in such a way that K1(k)νδ+δν−δ ≤ ε. �

The next result is an extension of [112, Lemma 1], since (3.29) relaxes [112,
condition (9)].

Lemma 3.12. Let c ∈ R. Then each (CPS)c-sequence for f is bounded in W 1,p
0 (Ω).

Proof. First of all, we can find a0 ∈ L1(Ω) such that for a.e. x ∈ Ω and all s ∈ R

qG(x, s) ≤ sg(x, s) + a0(x).

Now, let (uh) be a (CPS)c-sequence for f and let for all v ∈ C∞c (Ω)

〈w, v〉 =
∫

Ω

∇ξL (x, u,∇u) · ∇v dx+
∫

Ω

DsL (x, u,∇u)v dx−
∫

Ω

g(x, uh)v dx.

According to Lemma 3.11, for each ε > 0 we have

− ‖wh‖−1,p′‖uh‖1,p

≤
∫

Ω

∇ξL (x, uh,∇uh) · ∇uh dx+
∫

Ω

DsL (x, uh,∇uh)uh dx−
∫

Ω

g(x, uh)uh dx

≤
∫

Ω

∇ξL (x, uh,∇uh) · ∇uh dx+
∫

Ω

DsL (x, uh,∇uh)uh dx

− q

∫
Ω

G(x, uh) dx+
∫

Ω

a0 dx

≤ (1 + ε)
∫
{|uh|>R′}

∇ξL (x, uh,∇uh) · ∇uh dx+
∫

Ω

DsL (x, uh,∇uh)uh dx

− q

∫
Ω

L (x, uh,∇uh) dx+ qf(uh) +
∫

Ω

a0 dx+KR′,ε.

On the other hand, from Lemma 3.11 and (3.29), for each ε > 0 we obtain∫
Ω

DsL (x, uh,∇uh)uh dx

=
∫
{|uh|≤R′}

DsL (x, uh,∇uh)uh dx+
∫
{|uh|>R′}

DsL (x, uh,∇uh)uh dx

≤ εMR′
∫
{KR′,ε>|uh|>R′}

∇ξL (x, uh,∇uh) · ∇uh dx

−
∫
{|uh|>R′}

∇ξL (x, uh,∇uh) · ∇uh dx+ q

∫
Ω

L (x, uh,∇uh) dx+
∫

Ω

aε(x) dx

+ ε

∫
{|uh|>R′}

|uh|p dx− ν

∫
{|uh|>R′}

|∇uh|p dx+KR′,ε.

Taking into account Poincaré and Young’s inequalities, by (3.24) we find c > 0 and
CR′,ε > 0 with∫

Ω

DsL (x, uh,∇uh)uh dx

≤ εc

∫
{|uh|>R′}

|∇uh|p dx−
∫
{|uh|>R′}

∇ξL (x, uh,∇uh) · ∇uh dx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 33

+ q

∫
Ω

L (x, uh,∇uh) dx+
∫

Ω

aε(x) dx− ν

∫
{|uh|>R′}

|∇uh|p dx+ CR′,ε.

Therefore, for a sufficiently small ε > 0, there exists ϑε > 0 with

ϑε

∫
{|uh|>R′}

|∇uh|p dx

≤ ‖wh‖−1,p′‖uh‖1,p + qf(uh) +
∫

Ω

a0 dx+
∫

Ω

aε dx+KR′,ε + CR′,ε.

Moreover, it satisfies∫
Ω

|∇uh|p dx ≤ (1 + ε)
∫
{|uh|>R′}

|∇uh|p dx+KR′,ε.

Since wh → 0 in W−1,p′(Ω), the assertion follows. �

3.7. Existence of a weak solution.

Lemma 3.13. Under assumptions (3.27) we have∫
Ω
G(x, uh) dx
‖uh‖p1,p

→ 0 as h→ +∞ ,

for each (uh) that goes to 0 in W 1,p
0 (Ω).

Proof. Let (uh) ⊆ W 1,p
0 (Ω) with uh → 0 in W 1,p

0 (Ω). We can find (%h) ⊆ R and a
sequence (wh) ⊆ W 1,p

0 (Ω) such that uh = %hwh, %h → 0 and ‖wh‖1,p = 1. Taking
into account (3.27), it follows

lim
h

G(x, uh(x))
‖uh‖p1,p

= 0 for a.e. x ∈ Ω.

Moreover, for a.e. x ∈ Ω,

G(x, uh(x))
‖uh‖p1,p

≤ d|wh|p + b%
p2/(n−p)
h |wh|p

∗
.

If w is the weak limit of (wh), since d|wh|p → d|w|p in L1(Ω) and b%p
2/(n−p)
h |wh|p

∗ →
0 in L1(Ω), (a variant of) Lebesgue’s Theorem concludes the proof. �

We conclude with the proof of the main result of this section.

Proof of Theorem 3.9. From Lemma (3.12) and Theorem (3.10) it follows that f
satisfies the (CPS)c condition for each c ∈ R. By (3.22) and (3.28) it easily follows
that

∀u ∈W 1,p
0 (Ω)\{0} : lim

t→+∞
f(tu) = −∞.

From Lemma (3.13) and (3.22), we deduce that 0 is a strict local minimum for f .
From Theorem (2.26) the assertion follows. �

Remark 3.14. As proved by Arcoya and Boccardo in [6], each weak solution
of (3.20) belongs to W 1,p

0 (Ω) ∩ L∞(Ω) provided that L and g satisfy suitable
conditions. Then, some nice regularity results hold for various classes of integrands
L (see [90]).



34 MARCO SQUASSINA EJDE-2006/MON. 07

3.8. Super-linear problems with unbounded coefficients. The aim of this
section is to prove existence and multiplicity results of unbounded critical points
for a class of lower semi-continuous functionals (cf. [114]). Let us consider a bounded
open set Ω ⊂ RN (N ≥ 3) and define the functional f : H1

0 (Ω) → R ∪ {+∞} by

f(u) =
∫

Ω

j(x, u,∇u)−
∫

Ω

G(x, u),

where j(x, s, ξ) : Ω × R × RN → R is a measurable function with respect to x for
all (s, ξ) ∈ R × RN , and of class C1 with respect to (s, ξ) for a.e. x ∈ Ω. We also
assume that for almost every x in Ω and every s in R

the mapping
{
ξ 7→ j(x, s, ξ)

}
is strictly convex . (3.40)

Moreover, we suppose that there exist a constant α0 > 0 and a positive increasing
function α ∈ C(R) such that the following hypothesis is satisfied for almost every
x ∈ Ω and for every (s, ξ) ∈ R× RN

α0|ξ|2 ≤ j(x, s, ξ) ≤ α(|s|)|ξ|2. (3.41)

The functions js(x, s, ξ) and jξ(x, s, ξ) denote the derivatives of j(x, s, ξ) with re-
spect of the variables s and ξ respectively. Regarding the function js(x, s, ξ), we
assume that there exist a positive increasing function β ∈ C(R) and a positive
constant R such that the following conditions are satisfied almost everywhere in Ω
and for every ξ ∈ RN :∣∣js(x, s, ξ)∣∣ ≤ β(|s|)|ξ|2, for every s in R, (3.42)

js(x, s, ξ)s ≥ 0, for every s in R with |s| ≥ R. (3.43)

Note that, from (3.40) and (3.41), it follows that jξ(x, s, ξ) satisfies the following
growth condition (see Remark 3.24 for more details)∣∣jξ(x, s, ξ)∣∣ ≤ 4α(|s|)|ξ|. (3.44)

The function G(x, s) is the primitive with respect to s such that G(x, 0) = 0 of
a Carathéodory (i.e. measurable with respect to x and continuous with respect to
s) function g(x, s). We will study two different kinds of problems, according to
different nonlinearities g(x, s), that have a main common feature. Indeed, in both
cases we cannot expect to find critical points in L∞(Ω). To be more precise, let us
consider a first model example of nonlinearity and suppose that there exists p such
that

g1(x, s) = a(x)arctgs+ |s|p−2s, 2 < p <
2N
N − 2

, (3.45)

where a(x) ∈ L
2N

N+2 (Ω) and a(x) > 0. Notice that from hypotheses (3.41) and
(3.45) it follows that f is lower semi-continuous on H1

0 (Ω). We will also assume
that

lim
|s|→∞

α(|s|)
|s|p−2

= 0. (3.46)

Condition (3.46), together with (3.41), allows f to be unbounded from below, so
that we cannot look for a global minimum. Moreover, notice that g(x, s) is odd with
respect to s, so that it would be natural to expect, if j(x,−s,−ξ) = j(x, s, ξ), the
existence of infinitely many solutions as in the semi-linear case (see [6]). Unfortu-
nately, we cannot apply any of the classical results of critical point theory, because
our functional f is not of class C1 on H1

0 (Ω). Indeed, notice that
∫
Ω
j(x, v,∇v) is

not differentiable. More precisely, since jξ(x, s, ξ) and js(x, s, ξ) are not supposed



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 35

to be bounded with respect to s, the terms jξ(x, u,∇u) · ∇v and js(x, u,∇u)v may
not be L1(Ω) even if v ∈ C∞0 (Ω). Notice that if js(x, s, ξ) and jξ(x, s, ξ) were sup-
posed to be bounded with respect to s, f would be Gateaux derivable for every u
in H1

0 (Ω) and along any direction v ∈ H1
0 (Ω) ∩ L∞(Ω) (see [6, 33, 36, 112, 132]

for the study of this class of functionals). On the contrary, in our case, for every
u ∈ H1

0 (Ω), f ′(u)(v) does not even exist along directions v ∈ H1
0 (Ω) ∩ L∞(Ω).

To deal with the Euler equation of f let us define the following subspace of
H1

0 (Ω) for a fixed u in H1
0 (Ω)

Wu =
{
v ∈ H1

0 (Ω) : jξ(x, u,∇u) · ∇v ∈ L1(Ω) and js(x, u,∇u)v ∈ L1(Ω)
}
. (3.47)

We will see thatWu is dense inH1
0 (Ω). We give the definition of generalized solution

Definition 3.15. Let Λ ∈ H−1(Ω) and assume (3.40), (3.41), (3.42). We say that
u is a generalized solution to

−div(jξ(x, u,∇u)) + js(x, u,∇u) = Λ, in Ω,
u = 0, on ∂Ω,

if u ∈ H1
0 (Ω) and it results

jξ(x, u,∇u) · ∇u ∈ L1(Ω), js(x, u,∇u)u ∈ L1(Ω),∫
Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v = 〈Λ, v〉, ∀ v ∈Wu.

Theorem 3.16. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.45), (3.46).
Moreover, suppose that there exist R′ > 0 and δ > 0 such that

|s| ≥ R′ ⇒ pj(x, s, ξ)− js(x, s, ξ)s− jξ(x, s, ξ) · ξ ≥ δ|ξ|2, (3.48)

for a.e. x ∈ Ω and all (s, ξ) ∈ R×RN . If j(x,−s,−ξ) = j(x, s, ξ), then there exists
a sequence {uh} ⊂ H1

0 (Ω) of generalized solutions of
−div(jξ(x, u,∇u)) + js(x, u,∇u) = g1(x, u), in Ω,

u = 0, on ∂Ω
(3.49)

such that f(uh) → +∞.

In the nonsymmetric case we consider a different class of nonlinearities g(x, s).
A simple model example can be the following

g2(x, s) = d(x)arctg(s2) + |s|p−2s, 2 < p <
2N
N − 2

, (3.50)

where d(x) ∈ LN
2 (Ω) and d(x) > 0. We will prove the following result.

Theorem 3.17. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.46), (3.48),
(3.50). Then there exists a nontrivial generalized solution of the problem

−div(jξ(x, u,∇u)) + js(x, u,∇u) = g2(x, u), in Ω,
u = 0, on ∂Ω.

(3.51)

Since the functions α(|s|) and β(|s|) in (3.41) and (3.42) are not supposed to be
bounded, we are dealing with integrands j(x, s, ξ) which may be unbounded with
respect to s. This class of functionals has also been treated in [7], [21] and [23].
In these papers the existence of a nontrivial solution u ∈ L∞(Ω) is proved when
g(x, s) = |s|p−2s. Note that, in this case it is natural to expect solutions in L∞(Ω).
To prove the existence result, in [21] and [23], a fundamental step is to prove that



36 MARCO SQUASSINA EJDE-2006/MON. 07

every cluster point of a Palais-Smale sequence belongs to L∞(Ω). That is, to prove
that u is bounded before knowing that it is a solution. Notice that if u is in L∞(Ω)
and v ∈ C∞0 (Ω) then jξ(x, u,∇u) · ∇v and js(x, u,∇u)v are in L1(Ω). Therefore, if
g(x, s) = |s|p−2s, it would be possible to define a solution as a function u ∈ L∞(Ω)
that satisfies the equation associated to (P1) (or (P2)) in the distributional sense.
In our case the function a(x) in (3.45) belongs to L2N/(N+2)(Ω), so that we can only
expect to find solutions in H1

0 (Ω). In the same way, the function d(x) in (3.50) is in
LN/2(Ω) and also in this case the solutions are not expected to be in L∞(Ω). For
these reasons, we have given a definition of solution weaker than the distributional
one and we have considered the subspace Wu as the space of the admissible test
functions. Notice that if u ∈ H1

0 (Ω) is a generalized solution of problem (P1) (resp.
(P2)) and u ∈ L∞(Ω), then u is a distributional solution of (P1) (resp. (P2)).

We want to stress that we have considered here particular nonlinearities (i.e.
g1 and g2) just to present - in a simple case - the main difficulties we are going
to tackle. Indeed, Theorems 3.16 and 3.17 will be proved as consequences of two
general results (Theorems 3.18 and 3.20). To prove these general results we will use
an abstract critical point theory for lower semi-continuous functionals developed in
[50, 58, 60]. So, firstly, we will show that the functional f can be studied by means
of this theory (see Theorem 3.23). Then, we will give a definition of a Palais-Smale
sequence {un} suitable to this situation (Definition 3.37), and we will prove that
un is compact in H1

0 (Ω) (Theorems 3.34 and 3.43). To do this we will follow the
arguments of [33, 36, 112, 132] where the case in which α(s) and β(s) are bounded
is studied. In our case we will have to modify the test functions used in these papers
in order to get the compactness result. Indeed, here the main difficulty is to find
suitable approximations of un that belong to Wun , in order to choose them as test
functions. For this reason a large amount of work (Theorems 3.30, 3.31, 3.32 and
3.33) is devoted to find possible improvements of the class of allowed test functions.

3.9. General setting and main results. Let us consider Ω a bounded open set
in RN (N ≥ 3). Let us define the functional J : H1

0 (Ω) → R ∪ {+∞} by

J(v) =
∫

Ω

j(x, v,∇v), (3.52)

where j(x, s, ξ) satisfies hypotheses (3.40), (3.41), (3.42), (3.43). We will prove
existence and multiplicity results of generalized solutions (see Definition 3.15) of
the problem

−div(jξ(x, u,∇u)) + js(x, u,∇u) = g(x, u), in Ω,
u = 0, on ∂Ω.

(3.53)

To do this, we will use variational methods, so that we will study the functional
f : H1

0 (Ω) → R ∪ {+∞} defined as

f(v) = J(v)−
∫

Ω

G(x, v),

where G(x, s) =
∫ s
0
g(x, t)dt is the primitive of the function g(x, s) with G(x, 0) = 0.

To state our multiplicity result let us suppose that g(x, s) satisfies the following
conditions. Assume that for every ε > 0 there exists aε ∈ L2N/(N+2)(Ω) such that

|g(x, s)| ≤ aε(x) + ε|s|
N+2
N−2 , (3.54)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 37

for a.e. x ∈ Ω and every s ∈ R. Moreover, there exist p > 2 and functions
a0(x), a(x) ∈ L1(Ω), b0(x), b(x) ∈ L

2N
N+2 (Ω) and k(x) ∈ L∞(Ω) with k(x) > 0

almost everywhere, such that

pG(x, s) ≤ g(x, s)s+ a0(x) + b0(x)|s|, (3.55)

G(x, s) ≥ k(x)|s|p − a(x)− b(x)|s|, (3.56)

for a.e. x ∈ Ω and every s ∈ R (the constant p is the same as the one in (3.48)).
In this case we will prove the following result.

Theorem 3.18. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.46), (3.48),
(3.54), (3.55), (3.56). Moreover, let

j(x,−s,−ξ) = j(x, s, ξ) and g(x,−s) = −g(x, s), (3.57)

for a.e. x ∈ Ω and every (s, ξ) ∈ R × RN . Then there exists a sequence {uh} ⊂
H1

0 (Ω) of generalized solutions of problem (3.53) with f(uh) → +∞.

Remark 3.19. In the classical results of critical point theory different conditions
from (3.54), (3.55) and (3.56) are usually supposed. Indeed, as a growth condition
on g(x, s), it is assumed that

|g(x, s)| ≤ a(x) + b|s|σ−1, 2 < σ <
2N
N − 2

, b ∈ R+, a(x) ∈ L
2N

N+2 (Ω). (3.58)

Note that (3.58) implies (3.54). Indeed, suppose that g(x, s) satisfies (3.58), then
Young inequality implies that (3.54) is satisfied with aε(x) = a(x) +C(b, ε). More-
over, as a superlinearity condition, it is usually assumed that there exist p > 2 and
R > 0 with

0 < pG(x, s) ≤ g(x, s)s, for every s in R with |s| ≥ R. (3.59)

Note that this condition is stronger than conditions (3.55), (3.56). Indeed, suppose
that g(x, s) satisfies (3.59) and notice that this implies that there exists a0 ∈ L1(Ω)
such that

pG(x, s) ≤ g(x, s)s+ a0(x), for every s in R.
Then (3.55) is satisfied with b0(x) ≡ 0. Moreover, from (3.59) we deduce that there
exists a(x) ∈ L1(Ω) such that

G(x, s) ≥ 1
Rp

min{G(x,R), G(x,−R), 1}|s|p − a(x),

so that also (3.56) is satisfied.

To state our existence result in the nonsymmetric case, assume that the function
g satisfies the following condition

|g(x, s)| ≤ a1(x)|s|+ b|s|σ−1, (3.60)

2 < σ <
2N
N − 2

, a1(x) ∈ L
N
2 (Ω), b ∈ R+.

We will prove the following

Theorem 3.20. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.46), (3.48),
(3.55), (3.56), (3.60). Also, let

lim
s→0

g(x, s)
s

= 0, a.e. in Ω. (3.61)



38 MARCO SQUASSINA EJDE-2006/MON. 07

Then there exists a nontrivial generalized solution of the problem (3.53). In ad-
dition, there exist ε > 0 such that for every Λ ∈ H−1(Ω) with ‖Λ‖−1,2 < ε the
problem

−div(jξ(x, u,∇u)) + js(x, u,∇u) = g(x, u) + Λ, in Ω,
u = 0, on ∂Ω,

(3.62)

has at least two generalized solutions u1, u2 with f(u1) ≤ 0 < f(u2).

Remark 3.21. Notice that, in order to have g(x, v)v ∈ L1(Ω) for every v ∈ H1
0 (Ω),

the function a1(x) has to be in L
N
2 (Ω). Nevertheless, also in this case we cannot

expect to find bounded solution of problem (3.53). The situation is even worse in
problem (3.62), indeed in this case we can only expect to find solutions that belong
to H1

0 (Ω) ∩ dom(J).

Remark 3.22. Notice that condition (3.60) implies (3.54). Indeed, suppose that
g(x, s) satisfies (3.60). Then Young inequality implies that, for every ε > 0, we
have

|g(x, s)| ≤ β(ε)(a1(x))
N+2

4 + ε|s|
N+2
N−2 + γ(ε, b),

where β(ε) and γ(ε, b) are positive constants depending on ε and b. Now, since we
have a1(x) ∈ L

N
2 (Ω), there holds

aε(x) =
(
β(ε)(a1(x))

N+2
4 + γ(ε, b)

)
∈ L

2N
N+2 (Ω),

which yields (3.54).

3.10. Verification of the key condition. Let us now set X = H1
0 (Ω) and con-

sider the functional J : H1
0 (Ω) → R ∪ {+∞} defined in (3.52). From hypothesis

(3.41), we immediately obtain that J is lower semicontinuous. We will now prove
that J satisfies (2.3). To this aim, for every k ≥ 1, we define the truncation
Tk : R → R at height k, defined as

Tk(s) = s if |s| ≤ k, Tk(s) = k
s

|s|
if |s| ≥ k. (3.63)

Theorem 3.23. Assume conditions (3.40), (3.41), (3.43). Then, for every (u, η) ∈
epi J with J(u) < η, there holds

|dGJ |(u, η) = 1.

Moreover, if j(x,−s,−ξ) = j(x, s, ξ), ∀ η > J(0)(= 0) it results |dZ2GJ |(0, η) = 1.

Proof. Let (u, η) ∈ epi J with J(u) < η and let % > 0. Then, there exists δ ∈ (0, 1],
δ = δ(%), and k ≥ 1, k = k(%), such that k ≥ R (where R is as in (3.43)) and

‖Tk(v)− v‖1,2 < %, for every v ∈ B(u, δ). (3.64)

From (3.41) we have

j(x, v,∇Tk(v)) ≤ α(k)|∇v|2.
Then, up to reducing δ, we get the following inequalities∫

Ω

j(x, v,∇Tk(v)) <
∫

Ω

j(x, u,∇Tk(u)) + % ≤
∫

Ω

j(x, u,∇u) + %, (3.65)

for each v ∈ B(u, δ). We now prove that, for every t ∈ [0, δ] and v ∈ B(u, δ), there
holds

J((1− t)v + tTk(v)) ≤ (1− t)J(v) + t(J(u) + %). (3.66)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 39

From (3.40) and since j(x, s, ξ) is of class C1 with respect to the variable s, there
exists θ ∈ [0, 1] such that

j(x, (1− t)v + tTk(v), (1− t)∇v + t∇Tk(v))− j(x, v,∇v)
= j(x, (1− t)v + tTk(v), (1− t)∇v + t∇Tk(v))− j(x, v, (1− t)∇v + t∇Tk(v))

+ j(x, v, (1− t)∇v + t∇Tk(v))− j(x, v,∇v)
≤ tjs(x, v + θt(Tk(v)− v), (1− t)∇v + t∇Tk(v))(Tk(v)− v)

+ t (j(x, v,∇Tk(v)) − j(x, v,∇v) ) .

Notice that

v(x) ≥ k ⇒ v(x) + θt(Tk(v(x))− v(x)) ≥ k ≥ R,

v(x) ≤ −k ⇒ v(x) + θt(Tk(v(x))− v(x)) ≤ −k ≤ −R.

Then, in light of (3.43) one has

js(x, v + θt(Tk(v)− v), (1− t)∇v + t∇Tk(v))(Tk(v)− v) ≤ 0.

It follows that

j(x, (1− t)v + tTk(v), (1− t)∇v + t∇Tk(v))
≤ (1− t)j(x, v,∇v) + tj(x, v,∇Tk(v)).

Therefore, from (3.65) one gets (3.66). To apply Theorem 2.19 we define

H : {v ∈ B(u, δ) : J(v) < η + δ} × [0, δ] → H1
0 (Ω)

by setting
H(v, t) = (1− t)v + tTk(v).

Hence, taking into account (3.64) and (3.66), it results

d(H(v, t), v) ≤ %t and J(H(v, t)) ≤ (1− t)J(v) + t(J(u) + %),

for v ∈ B(u, δ), J(v) < η + δ and t ∈ [0, δ]. The first assertion now follows from
Theorem 2.19. Finally, since H(−v, t) = H(v, t) one also has |dZ2GJ |(0, η) = 1,
whenever j(x,−s,−ξ) = j(x, s, ξ). �

3.11. The variational setting. This section concerns the relationship between
|dJ |(u) and the directional derivatives of the functional J . Moreover, we will obtain
some Brezis-Browder (see [27]) type results.

First of all, we make a few observations.

Remark 3.24. It is readily seen that hypothesis (3.40) and the right inequality of
(3.41) imply that there exists a positive increasing function α(|s|) such that∣∣jξ(x, s, ξ)∣∣ ≤ α(|s|)|ξ|, (3.67)

for a.e. x ∈ Ω and every (s, ξ) ∈ R× RN . Indeed, from (3.40) one has

∀v ∈ RN : |v| ≤ 1 ⇒ j(x, s, ξ + |ξ|v) ≥ j(x, s, ξ) + jξ(x, s, ξ) · v|ξ|.
This, and (3.41) yield

jξ(x, s, ξ) · v|ξ| ≤ 4α(|s|)|ξ|2.
From the arbitrariness of v, (3.67) follows. On the other hand, if (3.67) holds we
have

|j(x, s, ξ)| ≤
∫ 1

0

|jξ(x, s, tξ) · ξ|dt ≤
1
2
α(|s|)|ξ|2.



40 MARCO SQUASSINA EJDE-2006/MON. 07

As a consequence, it is not restrictive to suppose that the functions in the right
hand side of (3.41) and (3.67) are the same. Notice that, in particular, there holds
jξ(x, s, 0) = 0.

Remark 3.25. It is not restrictive to suppose that the functions α(s) and β(s) are
both increasing. Indeed, if this is not the case, we can consider the functions

Ar(|s|) = sup
|s|≤r

α(|s|) and Br(|s|) = sup
|s|≤r

β(|s|),

which are increasing.

Remark 3.26. The assumption of strict convexity on the function {ξ → j(x, s, ξ)}
implies that, for almost every x in Ω and for every s in R, we have

[jξ(x, s, ξ)− jξ(x, s, ξ∗)] · (ξ − ξ∗) > 0, (3.68)

for every ξ, ξ∗ ∈ RN , with ξ 6= ξ∗. Moreover, hypotheses (3.40) and (3.41) imply
that,

jξ(x, s, ξ) · ξ ≥ α0|ξ|2. (3.69)
Indeed, we have

0 = j(x, s, 0) ≥ j(x, s, ξ) + jξ(x, s, ξ) · (0− ξ),

so that inequality (3.69) follows by virtue of (3.41).

Now, for every u ∈ H1
0 (Ω), we define the subspace

Vu =
{
v ∈ H1

0 (Ω) ∩ L∞(Ω) : u ∈ L∞({x ∈ Ω : v(x) 6= 0})
}
. (3.70)

As proved in [61], Vu is a vector space dense in H1
0 (Ω). Since Vu ⊂ Wu, also Wu

(see the introduction) is dense in H1
0 (Ω). In the following proposition we study the

conditions under which we can compute the directional derivatives of J .

Proposition 3.27. Assume conditions (3.41), (3.42), (3.44). Then there exists
J ′(u)(v) for every u ∈ dom(J) and v ∈ Vu. Furthermore, we have

js(x, u,∇u)v ∈ L1(Ω) and jξ(x, u,∇u) · ∇v ∈ L1(Ω),

and
J ′(u)(v) =

∫
Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v.

Proof. Let u ∈ dom(J) and v ∈ Vu. For every t ∈ R and a.e. x ∈ Ω, we set

F (x, t) = j(x, u(x) + tv(x),∇u(x) + t∇v(x)).
Since v ∈ Vu and by using (3.41), it follows that F (x, t) ∈ L1(Ω). Moreover, it
results

∂F

∂t
(x, t) = js(x, u+ tv,∇u+ t∇v)v + jξ(x, u+ tv,∇u+ t∇v) · ∇v.

From hypotheses (3.42) and (3.44) we get that for every x ∈ Ω with v(x) 6= 0, it
results ∣∣∂F

∂t
(x, t)

∣∣ ≤ ‖v‖∞β(‖u‖∞ + ‖v‖∞)(|∇u|+ |∇v|)2

+ α(‖u‖∞ + ‖v‖∞)(|∇u|+ |∇v|)|∇v|.

Since the function in the right hand side of the previous inequality belongs to L1(Ω),
the assertion follows. �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 41

In the sequel we will often use the cut-off function H ∈ C∞(R) given by

H(s) = 1 on [−1, 1], H(s) = 0 outside [−2, 2], |H ′(s)| ≤ 2. (3.71)

Now we can prove a fundamental inequality regarding the weak slope of J .

Proposition 3.28. Assume conditions (3.41), (3.42), (3.44). Then

|d(J − w)|(u) ≥ sup
{∫

Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v − 〈w, v〉 :

v ∈ Vu, ‖v‖1,2 ≤ 1
}

for every u ∈ dom(J) and every w ∈ H−1(Ω).

Proof. If |d(J − w)|(u) = ∞, or if

sup
{∫

Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v − 〈w, v〉 : v ∈ Vu, ‖v‖1,2 ≤ 1
}

= 0,

then the inequality holds. Otherwise, let u ∈ dom(J) and let η ∈ R+ be such that
J(u) < η. Moreover, let us consider σ > 0 and v ∈ Vu such that ‖v‖1,2 ≤ 1 and∫

Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v − 〈w, v〉 < −σ. (3.72)

Let us fix ε > 0 and let us prove that there exists k0 ≥ 1 such that∥∥H( u
k0

)
v
∥∥

1,2
< 1 + ε (3.73)

and∫
Ω

js(x, u,∇u)H
( u

k0

)
v +

∫
Ω

jξ(x, u,∇u) · ∇
(
H

( u

k0

)
v
)
−

〈
w,H

( u

k0

)
v
〉
< −σ.

(3.74)
Let us set vk = H(u/k)v, whereH(s) is defined as in (3.71). Since v ∈ Vu we deduce
that vk ∈ Vu for every k ≥ 1 and vk converges to v in H1

0 (Ω). This, together with
the fact that ‖v‖1,2 ≤ 1, implies (3.73). Moreover, Proposition 3.27 implies that
we can consider J ′(u)(vk). In addition, as k goes to infinity, we have

js(x, u(x),∇u(x))vk(x) → js(x, u(x),∇u(x))v(x), for a.e. x ∈ Ω,

jξ(x, u(x),∇u(x)) · ∇vk(x) → jξ(x, u(x),∇u(x)) · ∇v(x), for a.e. x ∈ Ω.

Moreover, we get ∣∣∣js(x, u,∇u)H(
u

k
)v

∣∣∣ ≤ |js(x, u,∇u)v| ,

|jξ(x, u,∇u) · ∇vk| ≤ |jξ(x, u,∇u)||∇v|+ 2|v||jξ(x, u,∇u) · ∇u|.

Since v ∈ Vu and by using (3.42) and (3.44), we can apply Lebesgue Dominated
Convergence Theorem to obtain

lim
k→∞

∫
Ω

js(x, u,∇u)vk =
∫

Ω

js(x, u,∇u)v ,

lim
k→∞

∫
Ω

jξ(x, u,∇u) · ∇vk =
∫

Ω

jξ(x, u,∇u) · ∇v ,



42 MARCO SQUASSINA EJDE-2006/MON. 07

which, together with (3.72), implies (3.74). Since we want to apply Proposition
2.16, let us consider Jη as defined in (2.2). Let us now show that there exists
δ1 > 0 such that ∥∥H (

z

k0

)
v
∥∥ ≤ 1 + ε, (3.75)

as well as∫
Ω

jξ(x, z,∇z) · ∇
(
H

( z

k0

)
v
)

+
∫

Ω

js(x, z,∇z)H
( z

k0

)
v −

〈
w,H

( z

k0

)
v
〉
< −σ,

(3.76)
for every z ∈ B(u, δ1) ∩ Jη. Indeed, take un ∈ Jη such that un → u in H1

0 (Ω) and
set

vn = H

(
un
k0

)
v.

We have that vn → H(u/k0)v in H1
0 (Ω), so that (3.75) follows from (3.73). More-

over, note that vn ∈ Vun , so that from Proposition 3.27 we deduce that we can
consider J ′(un)(vn). From (3.42) and (3.44) it follows

|js(x, un,∇un)vn| ≤ β(2k0)‖v‖∞|∇un|2,

|jξ(x, un,∇un) · ∇vn| ≤ α(2k0)|∇un|
[

2
k0
‖v‖∞|∇un|+ |∇v|

]
.

Then we obtain

lim
n→∞

∫
Ω

js(x, un,∇un)vn =
∫

Ω

js(x, u,∇u)H
( u
k0

)
v ,

lim
n→∞

∫
Ω

jξ(x, un,∇un) · ∇vn =
∫

Ω

jξ(x, u,∇u) · ∇
[
H

( u

k0

)
v
]
,

which, combined with (3.74), immediately implies (3.76). Now, observe that (3.76)
is equivalent to say that J ′(z)

(
H

(
z
k

)
v
)
− 〈w,H

(
z
k

)
v〉 < −σ. Thus, there exists

δ < δ1 with

J
(
z +

t

1 + ε
H

( z

k0

)
v
)
− J(z)−

〈
w,

t

1 + ε
H

( z
k

)
v
〉
≤ − σ

1 + ε
t, (3.77)

for every t ∈ [0, δ] and z ∈ B(u, δ) ∩ Jη. Finally, let us define the continuous
function H : B(u, δ) ∩ Jη × [0, δ] → H1

0 (Ω) given by

H(z, t) = z +
t

1 + ε
H

( z
k0

)
v.

¿From (3.75) and (3.77) we deduce thatH satisfies all the hypotheses of Proposition
2.16. Then, |d(J −w)|(u) > σ

1+ε , and the conclusion follows from the arbitrariness
of ε. �

The next Lemma will be useful in proving two Brezis-Browder type results for
J .

Lemma 3.29. Assume conditions (3.40), (3.41), (3.42), (3.43) and let u ∈ dom(J).
Then ∫

Ω

jξ(x, u,∇u) · ∇u+
∫

Ω

js(x, u,∇u)u ≤ |dJ |(u)‖u‖1,2. (3.78)

In particular, if |dJ |(u) <∞, then

jξ(x, u,∇u) · ∇u ∈ L1(Ω) and js(x, u,∇u)u ∈ L1(Ω).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 43

Proof. First, notice that if u is such that |dJ |(u) = ∞, or∫
Ω

jξ(x, u,∇u) · ∇u+
∫

Ω

js(x, u,∇u)u ≤ 0,

then the conclusion holds. Otherwise, let k ≥ 1, u ∈ dom(J) with |dJ |(u) < ∞,
and σ > 0 be such that∫

Ω

jξ(x, u,∇u) · ∇Tk(u) +
∫

Ω

js(x, u,∇u)Tk(u) > σ‖Tk(u)‖1,2,

where Tk(s) is defined in (3.63). We will prove that |dJ |(u) ≥ σ. Fixed ε > 0, we
first want to show that there exists δ1 > 0 such that

‖Tk(w)‖1,2 ≤ (1 + ε)‖Tk(u)‖1,2, (3.79)∫
Ω

jξ(x,w,∇w) · ∇Tk(w) +
∫

Ω

js(x,w,∇w)Tk(w) > σ‖Tk(u)‖1,2, (3.80)

for every w ∈ H1
0 (Ω) with ‖w − u‖1,2 < δ1. Indeed, take wn ∈ H1

0 (Ω) such that
wn → u in H1

0 (Ω). Then, (3.79) follows directly. Moreover, notice that from (3.42)
and (3.43) there holds

js(x,wn(x),∇wn(x))wn(x) ≥ −Rβ(R)|∇wn(x)|2.

Since wn → u in H1
0 (Ω), from (3.69) and by applying Fatou Lemma we get

lim inf
n→∞

[∫
Ω

jξ(x,wn,∇wn) · ∇Tk(wn) +
∫

Ω

js(x,wn,∇wn)Tk(wn)
]

≥
∫

Ω

jξ(x, u,∇u) · ∇Tk(u) +
∫

Ω

js(x, u,∇u)Tk(u) > σ‖Tk(u)‖1,2,

which yields (3.80). Consider now the continuous map H : B(u, δ1) × [0, δ1] →
H1

0 (Ω) defined as

H(w, t) = w − t

‖Tk(u)‖1,2(1 + ε)
Tk(w).

From (3.79) and (3.80) we deduce that there exists δ < δ1 such that

d(H(w, t), w) ≤ t,

J(H(w, t))− J(w) ≤ − σ

1 + ε
,

for every t ∈ [0, δ] and w ∈ H1
0 (Ω) with ‖w−u‖1,2 < δ and J(w) < J(u)+ δ. Then,

the arbitrariness of ε yields |dJ |(u) ≥ σ. Therefore, for every k ≥ 1 we get∫
Ω

js(x, u,∇u)Tk(u) +
∫

Ω

jξ(x, u,∇u) · ∇Tk(u) ≤ |dJ |(u)‖Tk(u)‖1,2.

Taking the limit as k →∞, the Monotone Convergence Theorem yields (3.78). �

Notice that a generalized solution u (see Definition 3.15) is not, in general, a
distributional solution. This, because a test function v ∈ Wu may not belong to
C∞0 . Thus, it is natural to study the conditions under which it is possible to enlarge
the class of admissible test functions. This kind of argument was introduced in [27].
More precisely, suppose we have a function u ∈ H1

0 (Ω) such that∫
Ω

jξ(x, u,∇u) · ∇z +
∫

Ω

js(x, u,∇u)z = 〈w, z〉, ∀z ∈ Vu, (3.81)



44 MARCO SQUASSINA EJDE-2006/MON. 07

where Vu is defined in (3.70) and w ∈ H−1(Ω). A natural question is whether or
not we can take as test function v ∈ H1

0 (Ω) ∩ L∞(Ω). The next result gives an
answer to this question.

Theorem 3.30. Assume that conditions (3.40), (3.41), (3.42) hold. Let w ∈
H−1(Ω) and u ∈ H1

0 (Ω) that satisfies (3.81). Moreover, suppose that jξ(x, u,∇u) ·
∇u ∈ L1(Ω) and there exist v ∈ H1

0 (Ω) ∩ L∞(Ω) and η ∈ L1(Ω) such that

js(x, u,∇u)v + jξ(x, u,∇u) · ∇v ≥ η. (3.82)

Then jξ(x, u,∇u) · ∇v + js(x, u,∇u)v ∈ L1(Ω) and∫
Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v = 〈w, v〉.

Proof. Since v ∈ H1
0 (Ω) ∩ L∞(Ω), then H(uk )v ∈ Vu. From (3.81) we have∫

Ω

jξ(x, u,∇u) · ∇
[
H(

u

k
)v

]
+

∫
Ω

js(x, u,∇u)H(
u

k
)v =

〈
w,H(

u

k
)v

〉
, (3.83)

for every k ≥ 1. Note that∫
Ω

∣∣∣jξ(x, u,∇u) · ∇uH ′(
u

k
)
v

k

∣∣∣ ≤ 2
k
‖v‖∞

∫
Ω

jξ(x, u,∇u) · ∇u.

Since jξ(x, u,∇u) · ∇u ∈ L1(Ω), the Lebesgue Dominated Convergence Theorem
yields

lim
k→∞

∫
Ω

jξ(x, u,∇u) · ∇uH ′(
u

k
)
v

k
= 0,

lim
k→∞

〈
w,H(

u

k
)v

〉
= 〈w, v〉.

As far as the remaining terms in (3.83) concerns, note that from (3.82) it follows

[js(x, u,∇u)v + jξ(x, u,∇u) · ∇v]H(
u

k
) ≥ H(

u

k
)η ≥ −η− ∈ L1(Ω).

Thus, we can apply Fatou Lemma and obtain∫
Ω

js(x, u,∇u)v + jξ(x, u,∇u) · ∇v ≤ 〈w, v〉 .

The previous inequality and (3.82) imply that

js(x, u,∇u)v + jξ(x, u,∇u) · ∇v ∈ L1(Ω). (3.84)

Now, notice that∣∣∣[js(x, u,∇u)v + jξ(x, u,∇u) · ∇v]H(
u

k
)
∣∣∣ ≤ |js(x, u,∇u)v + jξ(x, u,∇u) · ∇v| .

From (3.84) we deduce that we can use Lebesgue Dominated Convergence Theorem
to pass to the limit in (3.83) and obtain the conclusion. �

In the next result we find the conditions under which we can use v ∈ H1
0 (Ω)

in (3.81). Moreover, we prove, under suitable hypotheses, that if u satisfies (3.81)
then u is a generalized solution (see Definition 3.15) of the corresponding problem.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 45

Theorem 3.31. Assume that conditions (3.40), (3.41), (3.42), (3.43) hold. Let
w ∈ H−1(Ω), and let u ∈ H1

0 (Ω) be such that (3.81) is satisfied. Moreover, suppose
that jξ(x, u,∇u) · ∇u ∈ L1(Ω), and that there exist v ∈ H1

0 (Ω) and η ∈ L1(Ω) such
that

js(x, u,∇u)v ≥ η and jξ(x, u,∇u) · ∇v ≥ η. (3.85)

Then js(x, u,∇u)v ∈ L1(Ω), jξ(x, u,∇u) · ∇v ∈ L1(Ω) and∫
Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v = 〈w, v〉. (3.86)

In particular, it results js(x, u,∇u)u, js(x, u,∇u) ∈ L1(Ω) and∫
Ω

jξ(x, u,∇u) · ∇u+
∫

Ω

js(x, u,∇u)u = 〈w, u〉.

Moreover, u is a generalized solution of the problem

−div(jξ(x, u,∇u)) + js(x, u,∇u) = w, in Ω,
u = 0, on ∂Ω.

(3.87)

Proof. Let k ≥ 1 be fixed. For every v ∈ H1
0 (Ω) we have that Tk(v) ∈ H1

0 (Ω) ∩
L∞(Ω) and −v− ≤ Tk(v) ≤ v+. Then, from (3.85), we get

js(x, u,∇u)Tk(v) ≥ −η− ∈ L1(Ω). (3.88)

Moreover,

jξ(x, u,∇u) · ∇Tk(v) ≥ −[jξ(x, u,∇u) · ∇Tk(v)]− ≥ −η− ∈ L1(Ω). (3.89)

Then, we can apply Theorem 3.30 and obtain∫
Ω

js(x, u,∇u)Tk(v) +
∫

Ω

jξ(x, u,∇u) · ∇Tk(v) = 〈w, Tk(v)〉 (3.90)

for every k ≥ 1. By using again (3.88) and (3.89) and by arguing as in Theorem
3.30 we obtain

js(x, u,∇u)v ∈ L1(Ω) and jξ(x, u,∇u) · ∇v ∈ L1(Ω).

Thus, we can use Lebesgue Dominated Convergence Theorem to pass to the limit
in (3.90) and get (3.86). In particular, by (3.42), (3.43) and (3.69) we can choose
v = u. Finally, since

js(x, u,∇u) = js(x, u,∇u)χ{|u|<1} + js(x, u,∇u)χ{|u|≥1}

and ∣∣js(x, u,∇u)χ{|u|≥1}
∣∣ ≤ |js(x, u,∇u)u|,

by (3.42) it results also js(x, u,∇u) ∈ L1(Ω). Finally, notice that if v ∈Wu we can
take η = jξ(x, u,∇u) · ∇v and η = js(x, u,∇u)v, so that (3.86) is satisfied. Thus,
u is a generalized solution to Problem (3.87). �

We point out that the previous result readily implies that, if u ∈ H1
0 (Ω) satisfies

(3.81) and jξ(x, u,∇u) · ∇u ∈ L1(Ω), it results that js(x, u,∇u) ∈ L1(Ω), then
js(x, u,∇u)v ∈ L1(Ω) for every v ∈ C∞0 (Ω). Instead, the term which has not a
distributional interpretation in (3.81) is jξ(x, u,∇u). In the next result we show
that if we multiply jξ(x, u,∇u) by a suitable sequence of C1

c functions, we obtain,
passing to the limit, a distributional interpretation of (3.81).



46 MARCO SQUASSINA EJDE-2006/MON. 07

Theorem 3.32. Assume conditions (3.40), (3.41), (3.42), (3.43). Let w ∈ H−1(Ω)
and u ∈ H1

0 (Ω) be such that (3.81) is satisfied. Let (ϑh) be a sequence in C1
c (R)

with

sup
h≥1

‖ϑh‖∞ <∞, sup
h≥1

‖ϑ′h‖∞ <∞,

lim
h→∞

ϑh(s) = 1, lim
h→∞

ϑ′h(s) = 0.

If jξ(x, u,∇u) · ∇u ∈ L1(Ω), the sequence

div
[
ϑh(u)jξ(x, u,∇u)

]
is strongly convergent in W−1,q(Ω) for every 1 < q < N

N−1 , and

lim
h→∞

{
−div

[
ϑh(u)jξ(x, u,∇u)

]}
+ js(x, u,∇u) = w in W−1,q(Ω).

Proof. Let w = −divF with F ∈ L2(Ω,RN ) and v ∈ C∞c (Ω). Then ϑh(u)v ∈ Vu
and we can take v as test function in (3.81). It results∫

Ω

jξ(x, u,∇u)ϑh(u) · ∇v = −
∫

Ω

jξ(x, u,∇u)ϑ′h(u) · ∇u v −
∫

Ω

js(x, u,∇u)ϑh(u)v

+
∫

Ω

Fϑ′h(u)∇u v +
∫

Ω

Fϑh(u)∇v.

Then u is a solution of the following equation

−div
[
ϑh(u)jξ(x, u,∇u)

]
= ξh in D′(Ω),

where

ξh = −
[
ϑ′h(u)(jξ(x, u,∇u)− F ) · ∇u+ ϑh(u)js(x, u,∇u)

]
− div(ϑh(u)F ).

Now, notice that
ϑh(u)F → F, strongly in L2(Ω).

Then, div(ϑh(u)F ) is a convergent sequence in H−1(Ω). Since the embedding of
H−1(Ω) in W−1,q(Ω) is continuous, we get the desired convergence. Moreover,
Theorem 3.31 implies that js(x, u,∇u) ∈ L1(Ω). Then, the remaining terms in
ξh converge strongly in L1(Ω). Thus, we get the conclusion by observing that the
embedding of L1(Ω) in W−1,q(Ω) is continuous. �

Consider the case j(x, s, ξ) = a(x, s)|ξ|2 with a(x, s) measurable with respect
to x, continuous with respect to s and such that hypotheses (3.40), (3.41), (3.42),
(3.43), (3.46) are satisfied. The next result proves that, in particular, if there exists
u ∈ H1

0 (Ω) that satisfies (3.81) and if a(x, u)|∇u|2 ∈ L1(Ω), then u satisfies (3.81)
in the sense of distribution.

Theorem 3.33. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.46). Let w ∈
H−1(Ω) and u ∈ H1

0 (Ω) that satisfies (3.81). Moreover, suppose that jξ(x, u,∇u) ·
∇u ∈ L1(Ω) and that

j(x, s, ξ) = ĵ(x, s, |ξ|). (3.91)
Then jξ(x, u,∇u) ∈ L1(Ω) and u is a distributional solution to

−div(jξ(x, u,∇u)) + js(x, u,∇u) = w, in Ω,
u = 0, on ∂Ω.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 47

Proof. It is readily seen that, in view of (3.40) and (3.91), it results

|ξ||jξ(x, s, ξ)| ≤
√

2jξ(x, s, ξ) · ξ.

for a.e. x ∈ Ω, every s ∈ R and ξ ∈ RN . Then

jξ(x, u,∇u)χ{|∇u|>1} ∈ L1(Ω).

Moreover, we take into account (3.46), and we observe that (3.44) implies that there
exists a positive constant C such that

|ξ| ≤ 1 ⇒ |jξ(x, s, ξ)| ≤ 4α(|s|) ≤ C(|s|p−2 + 1),

which, by the Sobolev embedding, implies also that jξ(x, u,∇u)χ{|∇u|≤1} ∈ L1(Ω).
Then jξ(x, u,∇u) ∈ L1(Ω). Moreover, from (3.42) and (3.43) we have

js(x, u,∇u)u ≥ js(x, u,∇u)uχ{x: |u(x)|<R} ∈ L1(Ω).

Then Theorem 3.31 implies that js(x, u,∇u)u ∈ L1(Ω). Finally, again Theorem
3.31 yields the conclusion. �

3.12. A compactness result for J . In this section we will prove the following
compactness result for J . We will follow an argument similar to the one used in
[36] and in [132].

Theorem 3.34. Assume conditions (3.40), (3.41), (3.42), (3.43). Let {un} ⊂
H1

0 (Ω) be a bounded sequence with jξ(x, un,∇un) · ∇un ∈ L1(Ω) and let {wn} ⊂
H−1(Ω) be such that

∀v ∈ Vun
:
∫

Ω

js(x, un,∇un)v + jξ(x, un,∇un) · ∇v = 〈wn, v〉. (3.92)

If wn is strongly convergent in H−1(Ω), then, up to a subsequence, un is strongly
convergent in H1

0 (Ω).

Proof. Let w be the limit of {wn} and let L > 0 be such that

‖un‖1,2 ≤ L, for every n ≥ 1. (3.93)

From (3.93) we deduce that there exists u ∈ H1
0 (Ω) such that, up to a subsequence,

un ⇀ u, weakly in H1
0 (Ω). (3.94)

Step 1. Let us first prove that u is such that∫
Ω

jξ(x, u,∇u) · ∇ψ +
∫

Ω

js(x, u,∇u)ψ = 〈w,ψ〉, ∀ψ ∈ Vu. (3.95)

First of all, from Rellich Compact Embedding Theorem, up to a subsequence,

un → u, in Lq(Ω), ∀ q ∈ [1, 2N/(N − 2)),

un(x) → u(x), for a.e. x ∈ Ω.
(3.96)

We now want to prove that, up to a subsequence,

∇un(x) → ∇u(x), for a.e. x ∈ Ω. (3.97)

Let h ≥ 1. For every v ∈ C∞c (Ω) we have that H
(
un

h

)
v ∈ Vun

(where H is again
the function defined in (3.71)), then∫

Ω

H
(un
h

)
jξ(x, un,∇un) · ∇v



48 MARCO SQUASSINA EJDE-2006/MON. 07

= −
∫

Ω

[
H

(un
h

)
js(x, un,∇un) +H ′

(un
h

)
jξ(x, un,∇un) ·

∇un
h

]
v

+
〈
wn,H

(un
h

)
v
〉
.

Let wn = −div(Fn), with (Fn) strongly convergent in L2(Ω,RN ). Then it follows
that ∫

Ω

H
(un
h

)
jξ(x, un,∇un) · ∇v

=
∫

Ω

[
H ′

(un
h

)
(Fn − jξ(x, un,∇un)) ·

∇un
h

−H
(un
h

)
js(x, un,∇un)

]
v

+
∫

Ω

H
(un
h

)
Fn · ∇v.

For the square bracket is bounded in L1(Ω) and (H
(
un

h

)
Fn) is strongly convergent

in L2(Ω,RN ) we can apply [54, Theorem 5] with

bn(x, ξ) = H

(
un(x)
h

)
jξ(x, un(x), ξ) and E = Eh = {x ∈ Ω : |u(x)| ≤ h}

and deduce (3.97) by the arbitrariness of h ≥ 1. Notice that, by virtue of Theo-
rem 3.31, for every n we have∫

Ω

jξ(x, un,∇un) · ∇un +
∫

Ω

js(x, un,∇un)un = 〈wn, un〉.

Then, in view of (3.43), one has

sup
n≥1

∫
Ω

jξ(x, un,∇un) · ∇un <∞. (3.98)

Let now k ≥ 1, ϕ ∈ C∞c (Ω), ϕ ≥ 0 and consider

v = ϕe−Mk(un+R)+H
(un
k

)
, where Mk =

β(2k)
α0

. (3.99)

Note that v ∈ Vun and

∇v = ∇ϕe−Mk(un+R)+H
(un
k

)
−Mkϕe

−Mk(un+R)+∇(un +R)+H
(un
k

)
+ ϕe−Mk(un+R)+H ′

(un
k

) ∇un
k

.

Taking v as test function in (3.92), we obtain∫
Ω

jξ(x, un,∇un) · e−Mk(un+R)+H
(un
k

)
∇ϕ

+
∫

Ω

[
js(x, un,∇un)−Mkjξ(x, un,∇un) · ∇(un +R)+

]
ϕe−Mk(un+R)+H

(un
k

)
=

∫
Ω

jξ(x, un,∇un) · ϕe−Mk(un+R)+H ′
(un
k

) ∇un
k

+
〈
wn, ϕe

−Mk(un+R)+H
(un
k

)〉
.

(3.100)
Observe that[

js(x, un,∇un)−Mkjξ(x, un,∇un) · ∇(un +R)+
]
ϕe−Mk(un+R)+H

(un
k

)
≤ 0.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 49

Indeed, the assertion follows from (3.43), for almost every x such that un(x) ≤ −R
while, for almost every x in {x : −R ≤ un(x) ≤ 2k} from (3.44), (3.69) and (3.99)
we get[
js(x, un,∇un)−Mkjξ(x, un,∇un) · ∇(un +R)+

]
≤ (β(2k)− α0Mk)|∇un|2 ≤ 0.

Moreover, from (3.44), (3.93), (3.96) and (3.97) we have∫
Ω

jξ(x, un,∇un) · e−Mk(un+R)+H
(un
k

)
∇ϕ

→
∫

Ω

jξ(x, u,∇u) · e−Mk(u+R)+H(
u

k
)∇ϕ,

and 〈
wn, ϕe

−Mk(un+R)+H
(un
k

)〉
→

〈
w,ϕe−Mk(u+R)+H

(u
k

)〉
,

as n → ∞. We take into account (3.98) and deduce that there exists a positive
constant C such that∣∣ ∫

Ω

jξ(x, un,∇un) · ϕe−Mk(un+R)+H ′
(un
k

) ∇un
k

∣∣ ≤ C

k
.

We take the limit superior in (3.100) and we apply Fatou Lemma to obtain∫
Ω

jξ(x, u,∇u) · e−Mk(u+R)+H(
u

k
)∇ϕ +

∫
Ω

js(x, u,∇u)ϕe−Mk(u+R)+H(
u

k
)

−Mk

∫
Ω

jξ(x, u,∇u) · ∇u+ϕe−Mk(u+R)+H(
u

k
)

≥ −C
k

+
〈
w,ϕe−Mk(u+R)+H(

u

k
)
〉

(3.101)
for every ϕ ∈ C∞c (Ω) with ϕ ≥ 0. Then, the previous inequality holds for every
ϕ ∈ H1

0 ∩L∞(Ω) with ϕ ≥ 0. We now choose in (3.101) the admissible test function

ϕ = eMk(u+R)+ψ, ψ ∈ Vu, ψ ≥ 0.

It results∫
Ω

jξ(x, u,∇u)·H(
u

k
)∇ψ +

∫
Ω

js(x, u,∇u)H(
u

k
)ψ ≥ −C

k
+

〈
w,H(

u

k
)ψ

〉
. (3.102)

Notice that ∣∣∣jξ(x, u,∇u) ·H(
u

k
)∇ψ

∣∣∣ ≤ |jξ(x, u,∇u)| |∇ψ|,∣∣∣js(x, u,∇u)H(
u

k
)ψ

∣∣∣ ≤ |js(x, u,∇u)ψ| .

Since ψ ∈ Vu and from (3.42) and (3.44) we deduce that we can pass to the limit
in (3.102) as k →∞, and we obtain∫

Ω

jξ(x, u,∇u) · ∇ψ +
∫

Ω

js(x, u,∇u)ψ ≥ 〈w,ψ〉, ∀ψ ∈ Vu, ψ ≥ 0.

To show the opposite inequality, we can take v = ϕe−Mk(un−R)−H
(
un

k

)
as test

function in (3.92) and we can repeat the same argument as before. Thus, (3.95)
follows.



50 MARCO SQUASSINA EJDE-2006/MON. 07

Step 2. In this step we will prove that un → u strongly in H1
0 (Ω). From (3.69),

(3.98) and Fatou Lemma, we have

0 ≤
∫

Ω

jξ(x, u,∇u) · ∇u ≤ lim inf
n

∫
Ω

jξ(x, un,∇un) · ∇un <∞

so that jξ(x, u,∇u) · ∇u ∈ L1(Ω). Therefore, by Theorem 3.31 we deduce∫
Ω

jξ(x, u,∇u) · ∇u+
∫

Ω

js(x, u,∇u)u = 〈w, u〉. (3.103)

To prove that un converges to u strongly in H1
0 (Ω) we follow the argument of [132,

Theorem 3.2] and we consider the function ζ : R → R defined by

ζ(s) =


Ms, if 0 < s < R,M = β(R)

α0
,

MR, if s ≥ R,

−Ms, if −R < s < 0,
MR, if s ≤ −R.

(3.104)

We have that vn = une
ζ(un) belongs to H1

0 (Ω), and conditions (3.42), (3.43) and
(3.44) imply that hypotheses of Theorem 3.31 are satisfied. Then, we can use vn
as test function in (3.92). It results∫

Ω

jξ(x, un,∇un) · ∇uneζ(un)

= 〈wn, vn〉 −
∫

Ω

[js(x, un,∇un) + jξ(x, un,∇un) · ∇unζ ′(un)] vn

Note that vn converges to ueζ(u) weakly in H1
0 (Ω) and almost everywhere in Ω.

Moreover, conditions (3.42), (3.43) and (3.104) allow us to apply Fatou Lemma
and get that

lim sup
h

∫
Ω

jξ(x, un,∇un) · ∇uneζ(un)

≤ 〈w, ueζ(u)〉 −
∫

Ω

[js(x, u,∇u) + jξ(x, u,∇u) · ∇uζ ′(u)]ueζ(u).

(3.105)

On the other hand (3.103) and (3.104) imply that

jξ(x, u,∇u) · ∇
[
ueζ(u)

]
+ js(x, u,∇u)ueζ(u) ∈ L1(Ω),

jξ(x, u,∇u) · ∇
[
ueζ(u)

]
∈ L1(Ω).

(3.106)

Therefore, from Theorem 3.31,∫
Ω

jξ(x, u,∇u) · ∇
[
ueζ(u)

]
+

∫
Ω

js(x, u,∇u)ueζ(u) = 〈w, ueζ(u)〉. (3.107)

Thus, (3.105) and (3.107) imply∫
Ω

jξ(x, u,∇u) · ∇u eζ(u) ≤ lim inf
n→∞

∫
Ω

jξ(x, un,∇un) · ∇uneζ(un)

≤ lim sup
n→∞

∫
Ω

jξ(x, un,∇un) · ∇uneζ(un)

≤
∫

Ω

jξ(x, u,∇u) · ∇ueζ(u).

Then (3.69) implies that un → u strongly in H1
0 (Ω). �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 51

3.13. Proofs of the main Theorems. In this section we give the definition of
a Concrete Palais-Smale sequence, we study the relation between a Palais-Smale
sequence and a Concrete Palais-Smale sequence, and we prove that f satisfies the
(PS)c for every c ∈ R. Finally, we conclude by giving the proofs of Theorems 3.18
and 3.20.

Let us consider the functional I : H1
0 (Ω) → R defined by

I(v) = −
∫

Ω

G(x, v)− 〈Λ, v〉,

where Λ ∈ H−1(Ω), G(x, s) =
∫ s
0
g(x, t) dt and g : Ω × R → R is a Carathéodory

function satisfying assumption (3.54). Then (3.41) implies that the functional f :
H1

0 (Ω) → R ∪ {+∞} defined by f(v) = J(v) + I(v) is lower semi-continuous. To
apply the abstract theory, it is crucial to have the following result.

Theorem 3.35. Assume conditions (3.40), (3.41), (3.43), (3.54). Then, for every
(u, η) ∈ epi f with f(u) < η, it results

|dGf |(u, η) = 1.

Moreover, if j(x,−s,−ξ) = j(x, s, ξ), g(x,−s) = −g(x, s) and Λ = 0, for every
η > f(0) one has |dZ2Gf |(0, η) = 1.

Proof. Since G is of class C1, Theorem 3.23 and Proposition 2.18 imply the result.
�

Furthermore, since G a C1 functional, as a consequence of Proposition 3.28 one
has the following

Proposition 3.36. Assume conditions (3.41), (3.42), (3.44), (3.54) and consider
u ∈ dom(f) with |df |(u) <∞. Then there exists w ∈ H−1(Ω) such that ‖w‖−1,2 ≤
|df |(u) and

∀v ∈ Vu :
∫

Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v −
∫

Ω

g(x, u)v − 〈Λ, v〉 = 〈w, v〉.

Proof. Given u ∈ dom(f) with |df |(u) <∞, let

Ĵ(v) = J(v)−
∫

Ω

g(x, u)v − 〈Λ, v〉,

Î(v) = I(v) +
∫

Ω

g(x, u)v + 〈Λ, v〉.

Then, since Î is of class C1 with Î ′(u) = 0, by (c) of Proposition 2.18 we get
|df |(u) = |dĴ |(u). By Proposition 3.28, there exists w ∈ H−1(Ω) with ‖w‖−1,2 ≤
|df |(u) and

∀v ∈ Vu :
∫

Ω

jξ(x, u,∇u) · ∇v +
∫

Ω

js(x, u,∇u)v −
∫

Ω

g(x, u)v − 〈Λ, v〉 = 〈w, v〉,

and the assertion is proved. �

We can now give the definition of the Concrete Palais-Smale condition.

Definition 3.37. Let c ∈ R. We say that {un} is a Concrete Palais-Smale sequence
for f at level c ((CPS)c-sequence for short) if there exists wn ∈ H−1(Ω) with
wn → 0 such that jξ(x, un,∇un) · ∇un ∈ L1(Ω) for every n ≥ 1, and

f(un) → c, (3.108)



52 MARCO SQUASSINA EJDE-2006/MON. 07∫
Ω

jξ(x, un,∇un) · ∇v +
∫

Ω

js(x, un,∇un)v −
∫

Ω

g(x, un)v − 〈Λ, v〉 (3.109)

= 〈wn, v〉, ∀ v ∈ Vun
.

We say that f satisfies the Concrete Palais-Smale condition at level c ((CPS)c for
short) if every (CPS)c-sequence for f admits a strongly convergent subsequence in
H1

0 (Ω).

Proposition 3.38. Assume conditions (3.41), (3.42), (3.43), (3.44), (3.54). If
u ∈ dom(f) satisfies |df |(u) = 0, then u is a generalized solution to

−div(jξ(x, u,∇u)) + js(x, u,∇u) = g(x, u) + Λ, in Ω,
u = 0, on ∂Ω.

Proof. It is sufficient to combine Lemma 3.29, Proposition 3.36, and Theorem 3.31.
�

The following result concerns the relation between the (PS)c condition and the
(CPS)c condition.

Proposition 3.39. Assume conditions (3.41), (3.42), (3.43), (3.44), (3.54). Then
if f satisfies the (CPS)c condition, it satisfies the (PS)c condition.

Proof. Let {un} ⊂ dom(f) that satisfies the Definition 2.15. From Lemma 3.29 and
Proposition 3.36 we get that un satisfies the conditions in Definition 3.37. Thus,
there exists a subsequence, which converges in H1

0 (Ω). �

We now want to prove that f satisfies the (CPS)c condition at every level c. To
do this, let us consider a (CPS)c-sequence {un} ∈ dom(f).

From Theorem 3.34 we deduce the following result.

Proposition 3.40. Assume that conditions (3.40), (3.41), (3.42), (3.43), (3.54)
are satisfied. Let {un} be a (CPS)c-sequence for f , bounded in H1

0 (Ω). Then {un}
admits a strongly convergent subsequence in H1

0 (Ω).

Proof. Let {un} ⊂ dom(f) be a concrete Palais-Smale sequence for f at level c.
Taking into account that, as known, by (3.54) the map {u 7→ g(x, u)} is compact
fromH1

0 (Ω) toH−1(Ω), it suffices to apply Theorem 3.92 to see that {un} is strongly
compact in H1

0 (Ω). �

Proposition 3.41. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.48), (3.54),
(3.55). Then every (CPS)c-sequence {un} for f is bounded in H1

0 (Ω).

Proof. Condition (3.43) and (3.69) allow us to apply Theorem 3.31 to deduce that
we may choose v = un as test functions in (3.109). Taking into account conditions
(3.48), (3.54), (3.59), (3.108), the boundedness of {un} in H1

0 (Ω) follows by arguing
as in [132, Lemma 4.3]. �

Remark 3.42. Note that we use condition (3.48) only in Proposition 3.41.

We can now state the following result.

Theorem 3.43. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.48), (3.54),
(3.55). Then the functional f satisfies the (PS)c condition at every level c ∈ R.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 53

Proof. Let {un} ⊂ dom(f) be a concrete Palais-Smale sequence for f at level c.
From Proposition 3.41 it follows that {un} is bounded in H1

0 (Ω). By Proposi-
tion 3.40 f satisfies the Concrete Palais-Smale condition. Finally Proposition 3.39
implies that f satisfies the (PS)c condition. �

Proof of Theorem 3.18. This theorem will as a consequence of Theorem 2.21. First,
note that (3.41) and (3.54) imply that f is lower semi-continuous. Moreover,
from (3.57) we deduce that f is an even functional, and from Theorem 3.23 we
deduce that (2.3) and condition (d) of Theorem 2.21 are satisfied. Hypothe-
ses (3.56) implies that condition (b) of Theorem 2.21 is verified (see the subse-
quent proof of Theorem 3.20). Let now (λh, ϕh) be the sequence of solutions of
−∆u = λu with homogeneous Dirichlet boundary conditions. Moreover, let us con-
sider V + = span{ϕh ∈ H1

0 (Ω) : h ≥ h0} and note that V + has finite codimension.
To prove (a) of Theorem 2.21 it is enough to show that there exist h0, γ > 0 such
that for all u ∈ V + with ‖∇u‖2 = 1 there holds f(u) ≥ γ. First, note that condition
(3.54) implies that, for every ε > 0, we find a(1)

ε ∈ C∞c (Ω) and a(2)
ε ∈ L2N/(N+2)(Ω)

with ‖a(2)
ε ‖2N/(N+2) ≤ ε and

|g(x, s)| ≤ a(1)
ε (x) + a(2)

ε (x) + ε|s|
N+2
N−2 .

Now, let u ∈ V + and notice that there exist two positive constants c1, c2 such that

f(u) ≥ α0‖∇u‖22 −
∫

Ω

G(x, u)

≥ α0‖∇u‖22 −
∫

Ω

((
a(1)
ε + a(2)

ε

)
|u|+ N − 2

2N
ε|u|

2N
N−2

)
≥ α0‖∇u‖22 − ‖a(1)

ε ‖2‖u‖2 − c1‖a(2)
ε ‖ 2N

N+2
‖∇u‖2 − εc2‖∇u‖

2N
N−2
2

≥ α0‖∇u‖22 − ‖a(1)
ε ‖2‖u‖2 − c1ε‖∇u‖2 − εc2‖∇u‖

2N
N−2
2 .

Then if h0 is sufficiently large, since λh → +∞, for all u ∈ V +, ‖∇u‖2 = 1 implies
‖a(1)
ε ‖2‖u‖2 ≤ α0/2. Thus, for ε > 0 small enough, ‖∇u‖2 = 1 implies f(u) ≥ γ

for some γ > 0. Then also (a) of Theorem 2.21 is satisfied. Theorem 3.43 implies
that f satisfies (PS)c condition at every level c, so that we get the existence of
a sequence of critical points {uh} ⊂ H1

0 (Ω) with f(uh) → +∞. Proposition 3.38
yields the assertion. �

Let us conclude this section with the following proof.

Proof of Theorem 3.20. We will prove this theorem as a consequence of Theorem
2.20. To do this, let us notice that, from (3.41) and (3.60), f is lower semi-
continuous on H1

0 (Ω). Moreover, Theorem 3.23 implies that condition (2.3) is
satisfied. From Theorem 3.43 we deduce that f satisfies (PS)c condition at every
level c. It is left to show that f satisfies the geometrical assumptions of Theorem
2.20.

Let us first consider the case in which Λ = 0. Notice that conditions (3.41),
(3.60) and (3.61) imply that there exist γ > 0 and r > 0 such that for ‖u‖1,2 = r
there holds f(u) ≥ γ. Conditions (3.41) and (3.56) imply that there holds

f(v) ≤
∫

Ω

α(|v|)|∇v|2 −
∫

Ω

k(x)|v|p + ‖ā‖1 + C0‖b̄‖ 2N
N+2

‖v‖1,2. (3.110)



54 MARCO SQUASSINA EJDE-2006/MON. 07

Now, let us consider a finite dimensional subspace W of H1
0 (Ω) such that W ⊂

L∞(Ω). Condition (3.46) implies that, for every ε > 0, there exists R > r, w ∈W ,
with ‖w‖∞ > R and a positive constant Cε such that∫

Ω

α(|w|)|∇w|2 ≤ εCW ‖w‖p1,2 + Cε‖w‖21,2, (3.111)

where CW is a positive constant depending on W . Then, by suitably choosing ε,
(3.110) and (3.111) yield condition (2.4) for a suitable v1 ∈ H1

0 (Ω) and for v0 = 0.
Thus, we can apply Theorem 2.20 and deduce the existence of a nontrivial critical
point u of f . From Proposition 3.38, u is a generalized solution of Problem (3.53).

Now, let us consider the case in which Λ 6= 0. Let ϕ1 be the first eigenfunction
of the Laplace operator with homogeneous Dirichlet boundary conditions and set
v0 = t0ϕ1 for t0 > 0. Then, if t0 sufficiently small, thanks to (3.41) and (3.60), we
get f(v0) < 0. As before, (3.41), (3.60) and (3.61) imply that there exist ε > 0,
r = r(ε) > 0 and γ > 0 such that, for every Λ ∈ H−1(Ω) with ‖Λ‖−1 < ε, there
holds

f(u) ≥ γ, for every u with ‖u− v0‖1,2 = r.

Moreover, we use condition (3.41), (3.46) and (3.56) and we argue as before to
deduce the existence of v1 ∈ H1

0 (Ω) with ‖v1 − v0‖ > r and f(v1) < 0. Condition
(2.4) is thus fulfilled. Then, we can apply Theorem 2.20 getting the existence of
two distinct nontrivial critical points of f . Finally, Proposition 3.38 yields the
conclusion. �

Remark 3.44. Notice that Theorems 3.16 and 3.17 are an easy consequence of
Theorems 3.18 and 3.20 respectively. Indeed, consider for example g1(x, s) =
a(x)arctgs + |s|p−2s. To prove Theorem 3.16, it is left to show that g1(x, s) sat-
isfies conditions (3.54), (3.55) and (3.56). First, notice that Young inequality im-
plies that, for every ε > 0, there exists a positive constant β(ε) such that (3.54)
holds with aε(x) = β(ε) + a(x). Moreover, (3.55) is satisfied with a0(x) = 0 and
b0(x) = π/2(p − 1). Finally, (3.56) is verified with k(x) = 1/p, a(x) = 0 and
b(x) = (π/2 + C) a(x) where C ∈ R+ is sufficiently large. Theorem 3.17 can be
obtained as a consequence of Theorem 3.20 in a similar fashion.

3.14. Summability results. In this section we suppose that g(x, s) satisfies the
following growth condition

|g(x, s)| ≤ a(x) + b|s|
N+2
N−2 , a(x) ∈ Lr(Ω), b ∈ R+. (3.112)

Note that (3.54) implies (3.112). Let us set 2∗ = 2N/(N − 2). We prove the
following

Theorem 3.45. Assume conditions (3.40), (3.41), (3.42), (3.43), (3.112). Let
u ∈ H1

0 (Ω) be a generalized solution of problem (P ). Then the following conclusions
hold:

(a) If r ∈ (2N/(N+2), N/2), then u belongs to Lr
∗∗

(Ω), where r∗∗ = Nr/(N−
2r);

(b) if r > N/2, then u belongs to L∞(Ω).

The above theorem will be proved as a consequence of the following result.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 55

Lemma 3.46. Let us assume that conditions (3.40), (3.41), (3.42), (3.43) are
satisfied. Let u ∈ H1

0 (Ω) be a generalized solution of the problem

−div(jξ(x, u,∇u)) + js(x, u,∇u) + c(x)u = f(x), in Ω,
u = 0, on ∂Ω.

(3.113)

Then the following conclusions hold:

(i) If c ∈ LN
2 (Ω) and f ∈ Lr(Ω), with r ∈ (2N/(N + 2), N/2), then u belongs

to Lr
∗∗

(Ω), where r∗∗ = Nr/(N − 2r);
(ii) if c ∈ Lt(Ω) with t > N/2 and f ∈ Lq(Ω), with q > N/2, then u belongs to

L∞(Ω).

Proof. Let us first prove conclusion (i). For every k > R (where R is defined in
(3.43)), let us define the function ηk(s) : R → R such that ηk ∈ C1, ηk is odd and

ηk(s) =


0, if 0 < s < R,

(s−R)2γ+1, if R < s < k,

bks+ ck, if s > k,

(3.114)

where bk and ck are constant such that ηk is C1. Since u is a generalized solution of
(3.113), v = ηk(u) belongs to Wu. Then we can take it as test function, moreover,
js(x, u,∇u)ηk(u) ≥ 0. Then from (3.43) and (3.69) we get

α0

∫
Ω

η′k(u)|∇u|2 ≤
∫

Ω

f(x)ηk(u)−
∫

Ω

c(x)uηk(u). (3.115)

Now, let us consider the odd function ψk(s) : R → R defined by

ψk(s) =
∫ s

0

√
η′k(t) dt. (3.116)

The following properties of the functions ψk and ηk can be deduced from (3.114)
and (3.116) by easy calculations

[ψ′k(s)]
2 = η′k(s), (3.117)

0 ≤ ηk(s)(s−R) ≤ C0ψk(s)2, (3.118)

|ηk(s)| ≤ C0|ψk(s)|
2γ+1
γ+1 , (3.119)

where C0 is a positive constant. Notice that for every ε > 0 there exist c1(x) ∈
L

N
2 (Ω), with ‖c1‖N

2
≤ ε and c2 ∈ L∞(Ω) such that c(x) = c1(x) + c2(x). From

(3.115), (3.117), (3.118) and Hölder inequality, we deduce

α0

∫
Ω

|∇(ψk(u))|2

≤ C0‖c1(x)‖N
2

[ ∫
Ω

|ψk(u)|2
∗
]2/2∗

+
∫

Ω

|f(x)−Rc1(x)− c2(x)u||ηk(u)|.

We fix ε = (α0S)/(2C0), where S is the Sobolev constant. We obtain[ ∫
Ω

|ψk(u)|2
∗
]2/2∗

≤ C

∫
Ω

|f(x)−Rc1(x)− c2(x)u||ηk(u)|. (3.120)

Now, let us define the function

h(x) = |f(x)−Rc1(x)− c2(x)u(x)|, (3.121)



56 MARCO SQUASSINA EJDE-2006/MON. 07

and note that h(x) belongs to Lt(Ω) with

t = min{r, 2∗}. (3.122)

Let us consider first the case in which t = r, then from (3.119) and (3.120), we get[ ∫
Ω

|ψk(u)|2
∗
]2/2∗

≤ C‖h‖r
[ ∫

Ω

|ψk(u)|r
′ 2γ+1

γ+1

]1/r′

.

Since 2N/(N + 2) < r < N/2 we can define γ ∈ R+ by

γ =
r(N + 2)− 2N

2(N − 2r)
⇒ 2∗(γ + 1) = r′(2γ + 1) = r∗∗. (3.123)

Moreover, since r < N/2 we have that 2/2∗ > 1/r′, then[ ∫
Ω

|ψk(u)|2
∗
] 2

2∗−
1
r′ ≤ C‖h‖r. (3.124)

Note that |ψk(u)| → C(γ)|u − R|γ+1χ{x:|u(x)|>R} almost everywhere in Ω. Then
Fatou Lemma implies that |u − R|γ+1χ{x:|u(x)|>R} belongs to L2∗(Ω). Thus, u
belongs to L2∗(γ+1)(Ω) = Lr

∗∗
(Ω) and the conclusion follows. Consider now the

case in which t = 2∗ and note that this implies that N > 6. In this case we get[ ∫
Ω

|ψk(u)|2
∗
]2/2∗

≤ C‖h‖2∗
[ ∫

Ω

s|ψk(u)|(2
∗)′ 2γ+1

γ+1

] 1
(2∗)′

.

Since N > 6 it results 2/2∗ > 1/(2∗)′. Moreover, we can choose γ such that

2∗(γ + 1) = (2∗)′(2γ + 1).

Thus, we follow the same argument as in the previous case and we deduce that u
belongs to Ls1(Ω) where

s1 =
2∗N

N − 2 2∗
.

If it still holds s1 < r we can repeat the same argument to gain more summability
on u. In this way for every s ∈ [2∗, r) we can define the increasing sequence

s0 = 2∗, sn+1 =
Nsn

N − 2sn
,

and we deduce that there exists n such that sn−1 < r and sn ≥ r. At this step from
(3.122) we get that t = r and then u ∈ Lr∗∗(Ω), that is the maximal summability
we can achieve.

Now, let us prove conclusion (ii). First, note that since f ∈ Lq(Ω), with q > N/2,
f belongs to Lr(Ω) for every r > (2N)/(N + 2). Then, conclusion (i) implies that
u ∈ Lσ(Ω) for every σ > 1. Now, take δ > 0 such that t−δ > N/2, since u ∈ L t

δ (Ω)
it results ∫

Ω

|c(x)u(x)|t−δ ≤ ‖c(x)‖t−δt

[ ∫
Ω

|u(x)|t/δ
] δ

t

<∞.

Then, the function d(x) = f(x) − c(x)u(x) belongs to Lr(Ω) with r = min{q, t −
δ} > N/2. Let us take k > R (R is defined in (3.43)) and consider the function
v = Gk(u) = u− Tk(u) (where Tk(s) is defined in (3.63)). Since u is a generalized
solution of (3.113) we can take v as test function. From (3.43) and (3.69) it results

α0

∫
Ω

|∇Gk(u)|2 ≤
∫

Ω

|d(x)||Gk(u)|.

The conclusion follows from Theorem 4.2 of [135]. �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 57

Remark 3.47. In classical results of this type (see e.g. [104] or [26]) it is usually
considered as test function v = |u|2γu. Note that this type of function cannot be
used here for it does not belong to the space Wu. Moreover, the classical truncation
Tu seems not to be useful because of the presence of c(x)u. Then, we have chosen
a suitable truncation of u in order to manage also the term c(x)u.

Proof of Theorem 3.45. This theorem will be proved as a consequence of Lemma
3.46. So, consider u a generalized solution of Problem (3.53), we have to prove
that u is a generalized solution of Problem 3.113 for suitable f(x) and c(x). This
is shown in Theorem 2.2.5 of [36], then we will give here a sketch of the proof of
[36] just for clearness. We set

g0(x, s) = min{max{g(x, s),−a(x)}, a(x)},
g1(x, s) = g(x, s)− g0(x, s).

It follows that g(x, s) = g0(x, s) + g1(x, s) and |g0(x, s)| ≤ a(x) so that we can set
f(x) = g0(x, u(x)). Moreover, we define

c(x) =

{
− g1(x,u(x))

u(x) , if u(x) 6= 0,

0, if u(x) = 0.

Then |c(x)| ≤ b|u(x)|
4

N−2 , so that c(x) ∈ LN
2 (Ω). Lemma 3.46 implies that conclu-

sion (a) holds. Now, if r > N/2 we have that f(x) ∈ Lr(Ω) with r > N/2. More-
over, conclusion (a) implies that u ∈ Lt(Ω) for every t < ∞, so that c(x) ∈ Lt(Ω)
with t > N/2. Then Lemma 3.46 implies that u ∈ L∞(Ω). �

Remark 3.48. When dealing with quasi-linear equations (i.e. j(x, s, ξ) = a(x, s)ξ ·
ξ), a standard technique, to prove summability results, is to reduce the problem to
the linear one and to apply the classical result (see e.g. [135]). Note that here this
is not possible due to the general form of j.

4. Perturbation from Symmetry

We refer the reader to [128, 129, 110, 92, 30]. Some parts of these publications
have been slightly modified to give this collection a more uniform appearance.

4.1. Quasi-linear elliptic systems. In critical point theory, an open problem
concerning existence, is the role of symmetry in obtaining multiple critical points
for even functionals. Around 1980, the semi-linear scalar problem

−
n∑

i,j=1

Dj(aij(x)Diu) = g(x, u) + ϕ in Ω

u = 0 on ∂Ω ,

with g super-linear and odd in u and ϕ ∈ L2(Ω), has been object of a very careful
analysis by A. Bahri and H. Berestycki in [15], M. Struwe in [137], G-C. Dong and
S. Li in [66] and by P.H. Rabinowitz in [118] via techniques of classical critical point
theory. Around 1990, A. Bahri and P.L. Lions in [17, 18] improved the previous
results via a Morse-Index type technique. Later on, since 1994, several efforts have
been devoted to study existence for quasi-linear scalar problems of the type

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju = g(x, u) in Ω



58 MARCO SQUASSINA EJDE-2006/MON. 07

u = 0 on ∂Ω.

We refer the reader to [9, 33, 32, 36, 137] and to [6, 112, 132] for a more general
setting. In this case the associated functional f : H1

0 (Ω,RN ) → R given by

f(u) =
1
2

∫
Ω

n∑
i,j=1

aij(x, u)DiuDju dx−
∫

Ω

G(x, u) dx ,

is not even locally Lipschitz unless the aij ’s do not depend on u or n = 1. Conse-
quently, techniques of non-smooth critical point theory have to be applied. It seems
now natural to ask whether some existence results for perturbed even functionals
still hold in a quasi-linear setting, both scalar (N = 1) and vectorial (N ≥ 2). In
[133] it has recently been proved that diagonal quasi-linear elliptic systems of the
type (k = 1, . . . , N)

−
n∑

i,j=1

Dj(akij(x, u)Diuk) +
1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u)DiuhDjuh = Dsk

G(x, u) in Ω ,

(4.1)
possess a sequence (um) of weak solutions inH1

0 (Ω,RN ) under suitable assumptions,
including symmetry, on coefficients ahij and G. To prove this result, we looked for
critical points of the functional f0 : H1

0 (Ω,RN ) → R defined by

f0(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx−
∫

Ω

G(x, u) dx. (4.2)

In this section we want to investigate the effects of destroying the symmetry of
system (4.1) and show that for each ϕ ∈ L2(Ω,RN ) the perturbed problem

−
n∑

i,j=1

Dj(akij(x, u)Diuk) +
1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u)DiuhDjuh

= Dsk
G(x, u) + ϕk in Ω ,

(4.3)

still has infinitely many weak solutions. Of course, to this aim, we shall study the
associated functional

f(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx−
∫

Ω

G(x, u) dx−
∫

Ω

ϕ · u dx. (4.4)

In the next, Ω will denote an open and bounded subset of Rn. To adapt the
perturbation argument of [118], we shall consider the following assumptions: - the
matrix

(
ahij(x, s)

)
is measurable in x for each s ∈ RN and of class C1 in s for a.e.

x ∈ Ω with
ahij(x, s) = ahji(x, s).

Moreover, there exist ν > 0 and C > 0 such that
n∑

i,j=1

N∑
h=1

ahij(x, s)ξ
h
i ξ
h
j ≥ ν|ξ|2,

∣∣ahij(x, s)∣∣ ≤ C,

∣∣Dsa
h
ij(x, s)

∣∣ ≤ C ,
n∑

i,j=1

N∑
h=1

s ·Dsa
h
ij(x, s)ξ

h
i ξ
h
j ≥ 0 ,

(4.5)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 59

for a.e. x ∈ Ω and for all s ∈ RN and ξ ∈ RnN ; - (if N ≥ 2) there exists a bounded
Lipschitz function ψ : R → R such that

n∑
i,j=1

N∑
h=1

(
1
2
Dsa

h
ij(x, s) · expσ(r, s) + ahij(x, s)Dsh

(expσ(r, s))h

)
ξhi ξ

h
j ≤ 0 , (4.6)

for a.e. x ∈ Ω, for all ξ ∈ RnN , σ ∈ {−1, 1}N and r, s ∈ RN , where

(expσ(r, s))i := σi exp[σi(ψ(ri)− ψ(si))] ,

for each i = 1, . . . , N . - the function G(x, s) is measurable in x for all s ∈ RN , of
class C1 in s for a.e. x ∈ Ω with G(x, 0) = 0 and g(x, ·) denotes the gradient of G
with respect of s. - there exist q > 2 and R > 0 such that

|s| ≥ R⇒ 0 < qG(x, s) ≤ s · g(x, s) , (4.7)

for a.e. x ∈ Ω and all s ∈ RN ; - there exists γ ∈]0, q − 2[ such that
n∑

i,j=1

N∑
h=1

s ·Dsa
h
ij(x, s)ξ

h
i ξ
h
j ≤ γ

n∑
i,j=1

N∑
h=1

ahij(x, s)ξ
h
i ξ
h
j , (4.8)

for a.e. x ∈ Ω and for all s ∈ RN and ξ ∈ RnN . Under the previous assumptions,
the following is our main result.

Theorem 4.1. Assume that there exists σ in
]
1, qn+(q−1)(n+2)

qn+(q−1)(n−2)

[
such that

|g(x, s)| ≤ a+ b|s|σ, (4.9)

with a, b ∈ R and that for a.e. x ∈ Ω and for each s ∈ RN

ahij(x,−s) = ahij(x, s) , g(x,−s) = −g(x, s).

Then there exists a sequence (um) ⊆ H1
0 (Ω,RN ) of solutions to the system

−
n∑

i,j=1

Dj(akij(x, u)Diuk) +
1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u)DiuhDjuh

= Dsk
G(x, u) + ϕk in Ω

such that f(um) → +∞ as m→∞.

This is clearly an extension of the results of [15, 66, 118, 137] to the quasi-linear
case, both scalar (N = 1) and vectorial (N ≥ 2).

Let us point out that in the case N = 1 a stronger version of the previous result
can be proven. Indeed, we may completely drop assumption (b) and replace Lemma
3.4 with [36, Lemma 2.2.4]. To the best of our knowledge, in the case N > 1 only
very few multiplicity results have been obtained so far via non-smooth critical point
theory (see [9, 133, 137]).

4.2. Symmetry perturbed functionals. Given ϕ ∈ L2(Ω,RN ), we shall now
consider the functional f : H1

0 (Ω,RN ) → R defined by

f(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx−
∫

Ω

G(x, u) dx−
∫

Ω

ϕ · u dx.

If ϕ 6≡ 0, clearly f is not even. Note that by (4.7) we find c1, c2, c3 > 0 such that
1
q
(s · g(x, s) + c1) ≥ G(x, s) + c2 ≥ c3|s|q. (4.10)



60 MARCO SQUASSINA EJDE-2006/MON. 07

Lemma 4.2. Assume that u ∈ H1
0 (Ω,RN ) is a weak solution to (4.3). Then there

exists σ > 0 such that∫
Ω

(G(x, u) + c2) dx ≤ σ
(
f(u)2 + 1

)1/2
.

Proof. If u ∈ H1
0 (Ω,RN ) is a weak solution to (4.3), taking into account (4.8), we

deduce that

f(u) = f(u)− 1
2
f ′(u)(u)

=
∫

Ω

[
1
2
g(x, u) · u−G(x, u)− 1

2
ϕ · u

]
dx

− 1
4

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · uDiuhDjuh dx

≥
(1
2
− 1
q

) ∫
Ω

(g(x, u) · u+ c1) dx−
1
2
‖ϕ‖2‖u‖2

− γ

4

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx− c4

≥
(q

2
− 1− γ

2

) ∫
Ω

(G(x, u) + c2) dx−
γ

2
f(u)− ε‖u‖qq − β(ε)‖ϕ‖q

′

2 − c5

with ε→ 0 and β(ε) → +∞. Choosing ε > 0 small enough, by (4.10) we have

σf(u) ≥
∫

Ω

(G(x, u) + c2) dx− c6,

where σ = 2+γ
q−2−γ , and the assertion follows as in [118, Lemma 1.8]. �

We now want to introduce the modified functional, which is the main tool in
order to obtain our result. Let us define χ ∈ C∞(R) by setting χ = 1 for s ≤ 1,
χ = 0 for s ≥ 2 and −2 < χ′ < 0 when 1 < s < 2, and let for each u ∈ H1

0 (Ω,RN )

φ(u) = 2σ
(
f(u)2 + 1

)1/2
, ψ(u) = χ

(
φ(u)−1

∫
Ω

(G(x, u) + c2) dx
)
.

Finally, we define the modified functional by

f̃(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx+

−
∫

Ω

G(x, u) dx− ψ(u)
∫

Ω

ϕ · u dx.

(4.11)

The Euler’s equation associated to the previous functional is given by

−
n∑

i,j=1

Dj(akij(x, u)Diuk) +
1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u)DiuhDjuh = g̃(x, u) in Ω ,

(4.12)
where we set

g̃(x, u) = g(x, u) + ψ(u)ϕ+ ψ′(u)
∫

Ω

ϕ · u dx.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 61

Note that taking into account the previous Lemma, if u ∈ H1
0 (Ω,RN ) is a weak

solution to (4.3), we have that ψ(u) = 1 and therefore f̃(u) = f(u). In the next
result, we measure the defect of symmetry of f̃ , which turns out to be crucial in
the final comparison argument.

Lemma 4.3. There exists β > 0 such that for all u ∈ H1
0 (Ω,RN )

|f̃(u)− f̃(−u)| ≤ β
(
|f̃(u)|1/q + 1

)
.

Proof. Note first that if u ∈ supp(ψ) then∣∣∣ ∫
Ω

ϕ · u dx
∣∣∣ ≤ α

(
|f(u)|1/q + 1

)
, (4.13)

where α > 0 depends on ‖ϕ‖2. Indeed, by (4.10) we have∣∣ ∫
Ω

ϕ · u dx
∣∣ ≤ ‖u‖2‖ϕ‖2 ≤ c‖u‖q ≤ ĉ

( ∫
Ω

(G(x, u) + c2) dx
)1/q

,

and since u ∈ supp(ψ),∫
Ω

(G(x, u) + c2) dx ≤ 4σ
(
f(u)2 + 1

)1/2 ≤ c̃(|f(u)|+ 1),

inequality (4.13) easily follows. Now, since of course

|f(u)| ≤ |f̃(u)|+ 2
∣∣∣ ∫

Ω

ϕ · u dx
∣∣∣,

by (4.13) we immediately get for some b > 0

ψ(u)
∣∣∣ ∫

Ω

ϕ · u dx
∣∣∣ ≤ bψ(u)

(
|f̃(u)|1/q +

∣∣∣ ∫
Ω

ϕ · u dx
∣∣∣1/q + 1

)
.

Using Young’s inequality, for some b1, b2 > 0 we have that

ψ(u)
∣∣∣ ∫

Ω

ϕ · u dx
∣∣∣ ≤ b1

(
|f̃(u)|1/q + 1

)
,

and

ψ(−u)
∣∣∣∣∫

Ω

ϕ · u dx
∣∣∣∣ ≤ b2

(
|f̃(u)|1/q + 1

)
,

and since
|f̃(u)− f̃(−u)| = (ψ(u) + ψ(−u))

∣∣∣ ∫
Ω

ϕ · u dx
∣∣∣,

the assertion follows. �

Theorem 4.4. There exists M > 0 such that if u ∈ H1
0 (Ω,RN ) is a weak solution

to (4.12) with f̃(u) ≥M then u is a weak solution to (4.3) and f̃(u) = f(u).

Proof. Let us first prove that there exist M̃ > 0 and α̃ > 0 such that

∀M ∈ [M̃,+∞[: f̃(u) ≥M, u ∈ supp(ψ) ⇒ f(u) ≥ α̃M. (4.14)

Since we have
f(u) ≥ f̃(u)−

∣∣∣ ∫
Ω

ϕ · u
∣∣∣,

by 4.13 we deduce that

f(u) + α |f(u)|1/q ≥ f̃(u)− α ≥ M

2



62 MARCO SQUASSINA EJDE-2006/MON. 07

for M ≥ M̃, with M̃ large enough. Now, if it was f(u) ≤ 0, we would obtain

αq
′

q′
+

1
q
|f(u)| ≥ α |f(u)|1/q ≥ M

2
+ |f(u)|,

which is not possible if we take M̃ > 2αq
′
(q′)−1. Therefore it is f(u) > 0 and

f(u) >
M

4
or f(u) ≥

(
M

4α

)q
,

and (4.14) is proven. Of course, taking into account the definition of ψ, to prove
the Lemma it suffices to show that if M > 0 is sufficiently large and u ∈ H1

0 (Ω,RN )
is a weak solution to (4.12) with f̃(u) ≥M, then

φ(u)−1

∫
Ω

(G(x, u) + c2) dx ≤ 1.

If we set

ϑ(u) = φ(u)−1

∫
Ω

(G(x, u) + c2) dx,

it follows that

ψ′(u)(u) = χ′(ϑ(u))φ(u)−2
[
φ(u)

∫
Ω

g(x, u) · u dx− (2σ)2ϑ(u)f(u)f ′(u)(u)
]
.

Define now T1, T2 : H1
0 (Ω,RN ) → R by setting

T1(u) = χ′(ϑ(u))(2σ)2ϑ(u)φ(u)−2f(u)
∫

Ω

ϕ · u dx,

T2(u) = χ′(ϑ(u))φ(u)−1

∫
Ω

ϕ · u dx+ T1(u).

Then we obtain

f̃ ′(u)(u) = (1 + T1(u))
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u)DiuhDjuh dx

+
1
2
(1 + T1(u))

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u) · uDiuhDjuh dx

− (1 + T2(u))
∫

Ω

g(x, u) · u dx− (ψ(u) + T1(u))
∫

Ω

ϕ · u dx.

Consider now the term

f̃(u)− 1
2(1 + T1(u))

f̃ ′(u)(u).

If ψ(u) = 1 and T1(u) = 0 = T2(u), the assertion follows from Lemma 4.2.
Otherwise, since 0 ≤ ψ(u) ≤ 1, if T1(u) and T2(u) are both small enough the
computations we have made in Lemma 4.2 still hold true with σ replaced by (2−ε)σ,
for a small ε > 0, and again assertion follows as in Lemma 4.2.

It then remains to show that if M →∞, then T1(u), T2(u) → 0. We may assume
that u ∈ supp(ψ), otherwise Ti(u) = 0, for i = 1, 2. Therefore, taking into account
(4.13), there exists c > 0 with

|T1(u)| ≤ c
|f(u)|1/q + 1

|f(u)|
.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 63

Finally, by (4.14) we deduce |T1(u)| → 0 as M →∞. Similarly, |T2(u)| → 0. �

4.3. Boundedness of concrete Palais-Smale sequences.

Definition 4.5. Let c ∈ R. A sequence (um) ⊆ H1
0 (Ω,RN ) is said to be a concrete

Palais-Smale sequence at level c ((CPS)c-sequence, in short) for f̃ , if f̃(um) → c,
n∑

i,j=1

N∑
h=1

Dsk
ahij(x, u

m)Diu
m
h Dju

m
h ∈ H−1(Ω,RN )

eventually as m→∞ and

−
n∑

i,j=1

Dj(akij(x, u
m)Diu

m
k ) +

1
2

n∑
i,j=1

N∑
h=1

Dsk
ahij(x, u

m)Diu
m
h Dju

m
h − g̃k(x, um) ,

approaches zero strongly in H−1(Ω,RN ), where

g̃(x, u) = g(x, u) + ψ(u)ϕ+ ψ′(u)
∫

Ω

ϕ · u dx.

We say that f̃ satisfies the concrete Palais-Smale condition at level c, if every
(CPS)c sequence for f̃ admits a strongly convergent subsequence in H1

0 (Ω,RN ).

Lemma 4.6. There exists M > 0 such that each (CPS)c-sequence (um) for f̃ with
c ≥M is bounded in H1

0 (Ω,RN ).

Proof. Let M > 0 and (um) be a (CPS)c-sequence for f̃ with c ≥M in H1
0 (Ω,RN )

such that, eventually as m→ +∞

M ≤ f̃(um) ≤ K.

for some K > 0. Taking into account [133, Lemma 3], we have f̃ ′(um)(um) → 0 as
m→ +∞. Therefore, for large m ∈ N and any % > 0, it follows

%‖um‖1,2 +K ≥ f̃(um)− %f̃ ′(um)(um)

=
(1
2
− %(1 + T1(um))

) ∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

− %

2
(1 + T1(um))

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · umDiu
m
h Dju

m
h dx

+ %(1 + T2(um))
∫

Ω

g(x, um) · um dx

−
∫

Ω

G(x, um) dx+ [%(ψ(um) + T1(um))− ψ(um)]
∫

Ω

ϕ · um dx ≥

≥
(

1
2
− %(1 + T1(um))− %γ

2
(1 + T1(um))

) ∫
Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Dju

m
h dx

+ %(1 + T2(um))
∫

Ω

g(x, um) · um dx

−
∫

Ω

G(x, um) dx+ [%(ψ(um) + T1(um))− ψ(um)]
∫

Ω

ϕ · um dx



64 MARCO SQUASSINA EJDE-2006/MON. 07

≥ ν

2
(1− % (2 + γ) (1 + T1(um))) ‖um‖21,2 + (q%(1 + T2(um))− 1)

∫
Ω

G(x, um) dx

− [%(1 + T1(um)) + 1]‖ϕ‖2‖um‖2.

If we choose M sufficiently large, we find ε > 0, η > 0 and % ∈
]

1+η
q , 1−ε

γ+2

[
such

that uniformly in m ∈ N
(1− % (2 + γ) (1 + T1(um))) > ε, (q%(1 + T2(um))− 1) > η.

Hence we obtain

%‖um‖1,2 +K ≥ νε

2
‖um‖21,2 + bη‖um‖qq − c‖um‖1,2,

which implies that the sequence (um) is bounded in H1
0 (Ω,RN ). �

Lemma 4.7. Let c ∈ R. Then there exists M > 0 such that for each bounded
(CPS)c sequence (um) for f̃ with c ≥M, the sequence (g̃(x, um)) admits a conver-
gent subsequence in H−1(Ω,RN ).

Proof. Let (um) be a bounded (CPS)c-sequence for f̃ with c ≥M . We may assume
that (um) ⊆ supp(ψ), otherwise ψ(um) = 0 and ψ′(um) = 0. Recall that

g̃(x, um) = g(x, um) + ψ(um)ϕ+ ψ′(um)
∫

Ω

ϕ · um dx.

Since by [36, Theorem 2.2.7] the maps

H1
0 (Ω,RN ) −→ H−1(Ω,RN )
u 7−→ g(x, u)

and
H1

0 (Ω,RN ) −→ H−1(Ω,RN )
u 7−→ ψ(u)ϕ,

are completely continuous, the sequences (g(x, um)) and (ψ(um)ϕ) admit a conver-
gent subsequence in H−1(Ω,RN ). Now, we have

ψ′(um) =
[
χ′(ϑ(um))φ(um)−1

]
g(x, um)+

−
[
4σ2χ′(ϑ(um))φ(um)−2ϑ(um)f(um)

]
f ′(um).

On the other hand,

f ′(um) = f̃ ′(um) +
[∫

Ω

ϕ · um dx
]
ψ′(um) + [ψ(um)− 1]ϕ.

Therefore,[
1 +

[
4σ2χ′(ϑ(um))φ(um)−2ϑ(um)f(um)

∫
Ω

ϕ · um dx
]]
ψ′(um)

=
[
χ′(ϑ(um))φ(um)−1

]
g(x, um)

−
[
4σ2χ′(ϑ(um))φ(um)−2ϑ(um)f(um)

]
f̃ ′(um)

−
[
4σ2χ′(ϑ(um))φ(um)−2ϑ(um)f(um)(ψ(um)− 1)

]
ϕ.

(4.15)

By assumption we have f̃ ′(um) → 0 in H−1(Ω,RN ). Taking into account the
definition of χ, φ and ϑ, all of the square brackets in equation (4.15) are bounded
in R for some M > 0 and we conclude that also (ψ′(um)) admits a convergent
subsequence in H−1(Ω,RN ). The assertion is now proven. �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 65

4.4. Compactness of concrete Palais-Smale sequences. The next result is
the crucial property for Palais-Smale condition to hold.

Lemma 4.8. Let (um) be a bounded sequence in H1
0 (Ω,RN ) and set

〈wm, v〉 =
∫

Ω

n∑
i,j=1

N∑
h=1

ahij(x, u
m)Diu

m
h Djvh dx

+
1
2

∫
Ω

n∑
i,j=1

N∑
h=1

Dsa
h
ij(x, u

m) · vDiu
m
h Dju

m
h dx

for all v ∈ C∞c (Ω,RN ). Then, if (wm) is strongly convergent to some w in
H−1(Ω,RN ), (um) admits a strongly convergent subsequence in H1

0 (Ω,RN ).

For the proof of the above lemma, sew [133, Lemma 6].

Theorem 4.9. There exists M > 0 such that f̃ satisfies (CPS)c-condition for
c ≥M .

Proof. Let (um) be a (CPS)c sequence for f with c ≥ M , where M > 0 is as in
Lemma 4.6. Therefore, (um) is bounded in H1

0 (Ω,RN ) and from Lemma 4.7 we
deduce that, up to subsequences, (g̃(x, um)) is strongly convergent in H−1(Ω,RN ).
Therefore, the assertion follows from Lemma 4.8. �

4.5. Existence of multiple solutions. Let (λh, uh) be the sequence of eigenvalues
and eigenvectors for the problem

∆u = −λu in Ω
u = 0 on ∂Ω,

and set
Vk = span

{
u1, . . . , uk ∈ H1

0 (Ω,RN )
}
.

We deduce that for all s ∈ RN

|s| ≥ R⇒ G(x, s) ≥
G

(
x,R s

|s|

)
Rq

|s|q ≥ b0(x)|s|q,

where
b0(x) = R−q inf{G(x, s) : |s| = R} > 0.

Then it follows that for each k ∈ N there exists Rk > 0 such that for all u ∈ Vk
‖u‖1,2 ≥ Rk ⇒ f̃(u) ≤ 0.

Definition 4.10. For each k ∈ N set

Dk = Vk ∩B(0, Rk) ,

Γk =
{
γ ∈ C(Dk,H

1
0 ) : γis odd and γ|∂B(0,Rk)

= Id
}
,

bk = inf
γ∈Γk

max
u∈Dk

f̃(γ(u)).

Lemma 4.11. For each k ∈ N, % ∈]0, Rk[ and γ ∈ Γk

γ(Dk) ∩ ∂B(0, %) ∩ V ⊥k−1 6= ∅.

For the proof of the above lemma, wee, [118, Lemma 1.44].



66 MARCO SQUASSINA EJDE-2006/MON. 07

Lemma 4.12. There exist β > 0 and k0 ∈ N such that

∀k ≥ k0 : bk ≥ βk
(n+2)−(n−2)σ

n(σ−1) .

Proof. Let γ ∈ Γk and % ∈]0, Rk[. By previous Lemma there exists

w ∈ γ(Dk) ∩ ∂B(0, %) ∩ V ⊥k−1 ,

and therefore
max
u∈Dk

f̃(γ(u)) ≥ f̃(w) ≥ inf
u∈∂B(0,%)∩V ⊥k−1

f̃(u). (4.16)

Given u ∈ ∂B(0, %) ∩ V ⊥k−1, by (4.9) we find α1, α2, α3 > with

f̃(u) ≥ 1
2
%2 − α1‖u‖σ+1

σ+1 − α2‖ϕ‖2‖u‖2 − α3.

Now, By Gagliardo-Nirenberg inequality, there is α4 > 0 such that

‖u‖σ+1 ≤ α4‖u‖ϑ1,2‖u‖1−ϑ2 ,

where ϑ = n(σ−1)
2(σ+1) . As is well known, it is

‖u‖2 ≤
1√
λk−1

‖u‖1,2,

so that we obtain

f̃(u) ≥ 1
2
%2 − α1λ

− (1−ϑ)(σ+1)
2

k %σ+1 − α2‖ϕ‖2λ
− 1

2
k %− α3.

Choosing now

% = cλ
− (1−ϑ)

2
(σ+1)
(σ−1)

k ,

yields

f̃(u) ≥ 1
4
%2
k − α2‖ϕ‖2λ

− 1
2

k %k − α3.

Now, as is shown in [52], there exists α5 > 0 such that for large k, λk ≥ α5k
2
n .

Therefore. we find β > 0 with

f̃(u) ≥ βk
(n+2)−(n−2)σ

n(σ−1) ,

and by (4.16) the Lemma is proved. �

Definition 4.13. For each k ∈ N set

Uk =
{
ξ = tuk+1 + w : t ∈ [0, Rk+1], w ∈ B(0, Rk+1) ∩ Vk, ‖ξ‖1,2 ≤ Rk+1

}
,

Λk =
{
λ ∈ C(Uk,H1

0 ) : λ|Dk
∈ Γk+1 and

λ|∂B(0,Rk+1)∪((B(0,Rk+1)\B(0,Rk))∩Vk) = Id
}

ck = inf
λ∈Λk

max
u∈Uk

f̃(λ(u)).

We now come to the our main existence tool. Of course, differently from the
proof of [118, Lemma 1.57], in this non-smooth framework, we shall apply [36,
Theorem 1.1.13] instead of the classical Deformation Lemma [118, Lemma 1.60].



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 67

Lemma 4.14. Assume that ck > bk ≥ M, where M is as in Theorem 4.9. If
δ ∈]0, ck − bk[ and

Λk(δ) =
{
λ ∈ Λk : f̃(λ(u)) ≤ bk + δ for u ∈ Dk

}
,

set
ck(δ) = inf

λ∈Λk(δ)
max
u∈Uk

f̃(λ(u)).

Then ck(δ) is a critical value for f̃ .

Proof. Let ε = 1
2 (ck − bk − δ) > 0 and assume by contradiction that ck(δ) is not

a critical value for f̃ . Therefore, taking into account Lemma 4.9, by [36, Theorem
1.1.13], there exists ε > 0 and a continuous map

η : H1
0 (Ω,RN )× [0, 1] → H1

0 (Ω,RN )

such that for each u ∈ H1
0 (Ω,RN ) and t ∈ [0, 1],

f̃(u) 6∈]ck(δ)− ε, ck(δ) + ε[ ⇒ η(u, t) = u, (4.17)

η(f̃ck(δ)+ε, 1) ⊆ f̃ck(δ)−ε. (4.18)

Choose λ ∈ Λk(δ) so that

max
u∈Uk

f̃(λ(u)) ≤ ck(δ) + ε (4.19)

and consider η(λ(·), 1) : Uk → H1
0 (Ω,RN ).

Observe that if u ∈ ∂B(0, Rk+1) or u ∈ (B(0, Rk+1)\B(0, Rk))∩Vk, by definition
f̃(λ(u)) = f̃(u). Hence, by (4.17), it is η(λ(u), 1) = u. We conclude that η(λ(·), 1) ∈
Λk. Moreover, by our choice of ε > 0 and δ > 0 we obtain

∀u ∈ Dk : f̃(λ(u)) ≤ bk + δ ≤ ck − ε ≤ ck(δ)− ε.

Therefore (4.17) implies that η(λ(·), 1) ∈ Λk(δ). On the other hand, again by (4.18)
and (4.19)

max
u∈Uk

f̃(η(λ(u), 1)) ≤ ck(δ)− ε, (4.20)

which is not possible, by definition of ck(δ). �

It only remains to prove that we cannot have ck = bk for k sufficiently large.

Lemma 4.15. Assume that ck = bk for all k ≥ k1. Then, there exist γ > 0 and
k̃ ≥ k1 with

bek ≤ γk̃
q

q−1 .

Proof. Choose k ≥ k1, ε > 0 and a λ ∈ Λk such that

max
u∈Uk

f̃(λ(u)) ≤ bk + ε.

Define now λ̃ : Dk+1 → H1
0 such that

λ̃(u) =

{
λ(u) if u ∈ Uk

−λ(−u) if u ∈ −Uk.

Since λ̃|B(0,Rk+1)∩Vk
is continuous and odd, it follows λ̃ ∈ Γk+1. Then

bk+1 ≤ max
u∈Dk+1

f̃(λ̃(u)).



68 MARCO SQUASSINA EJDE-2006/MON. 07

By Lemma 4.3 we have

max
u∈−Uk

f̃(λ̃(u)) ≤ bk + ε+ β
(
|bk + ε|1/q + 1

)
,

and since Dk+1 = Uk ∪ (−Uk), we get

∀ε > 0 : bk+1 ≤ bk + ε+ β
(
|bk + ε|1/q + 1

)
,

that yields
∀k ≥ k1 : bk+1 ≤ bk + β

(
|bk|1/q + 1

)
.

The assertion now follows recursively as in [120, Proposition 10.46]. �

We finally come to the proof of the main result, which extends the theorems of
[15, 66, 118, 137] to the quasi-linear case, both scalar and vectorial.

Proof of Theorem 4.1. Observe that the inequality

1 < σ <
qn+ (q − 1)(n+ 2)
qn+ (q − 1)(n− 2)

,

implies
q

q − 1
<

(n+ 2)− σ(n− 2)
n(σ − 1)

.

Therefore, combining Lemma 4.12 and Lemma 4.15 we deduce ck > bk so that we
may apply Lemma 4.14 and obtain that (ck(δ)) is a sequence of critical values for
f̃ . By Theorem 4.4 we finally conclude that f has a diverging sequence of critical
values. �

4.6. Semi-linear systems with nonhomogeneous data. Since the early seven-
ties, many authors have widely investigated existence and multiplicity of solutions
for semi-linear elliptic problems with Dirichlet boundary conditions, especially by
means of variational methods (see [136] and references therein). In particular, if ϕ
is a real L2 function on a bounded domain Ω ⊂ Rn, p > 2 and p < 2∗ if n ≥ 3
(here, 2∗ = 2n

n−2 ), the following model problem (P0,ϕ,1)

−∆u = |u|p−2u+ ϕ in Ω
u = 0 on ∂Ω,

(4.21)

has been extensively studied, even when the nonlinear term is more general.
If ϕ ≡ 0, the problem is symmetric, so multiplicity results have been achieved

via the equivariant Lusternik-Schnirelman theory and the notion of genus for Z2-
symmetric sets (see [120] and references therein).

On the contrary, if ϕ 6≡ 0, the problem loses its Z2-symmetry and a natural
question is whether the infinite number of solutions persists under perturbation of
the odd equation. In this case, a detailed analysis was carried on by Rabinowitz in
[118], Struwe in [137], Bahri and Berestycki in [15], Dong and Li in [66] and Tanaka
in [139]: the existence of infinitely many solutions was obtained via techniques of
classical critical point theory provided that a suitable restriction on the growth of
the exponent p is assumed.

Furthermore, Bahri and Lions have improved some of such results via a technique
based on Morse theory (see [17, 18]); while, more recently, Paleari and Squassina
have extended some of the above mentioned achievements to the quasi-linear case
by means of techniques of non-smooth critical point theory (see [110]).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 69

Other perturbation results were obtained by Bahri and Berestycki in [15] and by
Ambrosetti in [2] when p > 2 is any but subcritical: in particular, they proved that
for each ν ∈ N there exists ε > 0 such that (P0,ϕ,1) has at least ν distinct solutions
provided that ‖ϕ‖2 < ε.

The success in looking for solutions of a non-symmetric problem as (P0,ϕ,1)
made quite interesting to study the problem (Pχ,ϕ,1)

−∆u = |u|p−2u+ ϕ in Ω
u = χ on ∂Ω

(4.22)

where, in general, the boundary condition χ is different from zero. Some multiplicity
results for (4.22) have been proved in [29] provided that

2 < p < 2
n+ 1
n

, χ ∈ C(∂Ω,R) ∩H1/2(∂Ω,R), ϕ ∈ L2(Ω,R).

The upper bound to p seems to be a natural extension of the assumption 2 < p < 4
considered by Ekeland, Ghoussoub and Tehrani in [67] in order to solve such a
problem when n = 1 (in this case, the range p < 2 was covered by Clarke and
Ekeland in a previous paper [47]).

We stress that an improvement of the results in [29, 67] has been reached with a
different technique by Bolle in [24] and Bolle, Ghoussoub and Tehrani in [25]. From
one hand, they prove that if Ω ⊂ Rn is a C2 bounded domain and

2 < p <
2n
n− 1

, χ ∈ C2(∂Ω,R), ϕ ∈ C(Ω,R),

then (Pχ,ϕ,1) has infinitely many classical solutions. On the other hand, they
show that in the case n = 1 it suffices to assume p > 2, namely the result becomes
optimal.

It remains open, even for χ ≡ 0, the problem of whether (Pχ,ϕ,1) has an infinite
number of solutions for p all the way up to 2∗. For χ ≡ 0, the most satisfactory
result remains the one contained in the celebrated paper [18] of Bahri and Lions
where they prove that this fact is true for a subset of ϕ dense in L2(Ω,R).

Let us fix N ≥ 1. The purpose of this section is to show the multiplicity of
solutions for the following semi-linear elliptic system (Pχ,ϕ,N )

−
n∑

i,j=1

N∑
h=1

Dj(ahkij (x)Diuh) = b(x)|u|p−2uk + ϕk(x) in Ω

u = χ on ∂Ω
k = 1, . . . , N

(4.23)

taken any χ ∈ H1/2(∂Ω,RN ). Clearly, (4.23) reduces to the problem (4.22) if
N = 1, ahkij = δhkij and b(x) ≡ 1.

To the best of our knowledge no other result can be found in the literature about
multiplicity for systems of semi-linear elliptic equations with non-homogeneous
boundary conditions; on the contrary, some multiplicity results are known in the
case of Dirichlet boundary conditions (see [46] for the semi-linear case and [110, 133]
for some extensions to the quasi-linear case).



70 MARCO SQUASSINA EJDE-2006/MON. 07

It is well known that the functional f : Mχ → R associated with (4.23) is given
by

f(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx−
1
p

∫
Ω

b(x)|u|p dx−
∫

Ω

ϕ · u dx

where Mχ =
{
u ∈ H1(Ω,RN ) : u = χ a.e. on ∂Ω

}
.

In the next, Ω will denote a Lipschitz bounded domain of Rn with n ≥ 3 while
we shall always assume that the coefficients ahkij and b belong to C(Ω,R) with
ahkij = akhji and b > 0. Moreover, there exists ν > 0 such that

n∑
i,j=1

N∑
h,k=1

ahkij (x)ξiξjηhηk ≥ ν|ξ|2|η|2 (4.24)

for all x ∈ Ω and (ξ, η) ∈ Rn × RN (Legendre-Hadamard condition).
Here, we state our main results.

Theorem 4.16. Let p ∈]2, 2n+1
n [. Then for each ϕ in L2(Ω,RN ) and χ in the space

H1/2(∂Ω,RN ) the system (4.23) admits a sequence (um)m of solutions in Mχ such
that f(um) → +∞.

To prove Theorem 4.16, we use some perturbation arguments developed in [15,
118, 137]; so the condition p < 2 n+1

n is quite natural.
An improvement of such a “control” can be obtained by means of the Bolle’s

techniques, but more assumptions need. In fact, all the weak solutions must be
regular and the system has to be diagonal, i.e. ahkij = δhkij .

More precisely, we can prove the following theorem.

Theorem 4.17. Let p ∈]2, 2n
n−1 [, ∂Ω is of class C2, χ in C2(∂Ω,RN ), ϕ in

C0,α(Ω,RN ) for some α ∈]0, 1[ and ahkij = δhkij . Then (4.23) has a sequence (um)m
of classical solutions such that f(um) → +∞.

Clearly, Theorems 4.16 and 4.17 extend the results of [29] and [25] to semilinear
elliptic systems. We underline that (4.24) is weaker than the strong ellipticity
condition.

Let us point out that, in general, whereas De Giorgi’s famous example of an
unbounded weak solution of a linear elliptic system shows (cf. [57]), we can not
hope to find everywhere regular solutions for coefficients ahkij ∈ L∞(Ω,R). Anyway,
if ahkij ∈ C(Ω,R) and (4.24) holds we have that if u solves (Pχ,ϕ,N ) then

u ∈ C0,α(Ω,RN )

for each α ∈]0, 1[ (see [75]); but if we look for classical solutions, namely u of class
C2 on Ω, the coefficients ahkij have to be sufficiently smooth while ϕ ∈ C0,α(Ω,RN )
for some α ∈]0, 1[ and χ ∈ C2(∂Ω,RN ) (see [90] and references therein).

4.7. Reduction to homogeneous boundary conditions. As a first step, let us
reduce (4.23) to a Dirichlet type problem. To this aim, let us denote by φ ∈ Mχ



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 71

the only solution of the linear system

−
n∑

i,j=1

N∑
h=1

Dj(ahkij (x)Diφh) = 0 in Ω

φ = χ on ∂Ω
k = 1, . . . , N

(4.25)

Since p < 2∗, it results φ ∈ Lp(Ω,RN ).
From now on, we shall assume that b ≡ 1. Taking into account that there exist

two positive constants mb and Mb such that

mb ≤ b(x) ≤Mb for all x ∈ Ω,

the general case can be covered by slight modifies of some lemmas proved in the
next sections.

It is easy to show that the following fact holds.

Proposition 4.18. u ∈ Mχ is a solution of (Pχ,ϕ,N ) if and only if z ∈ H1
0 (Ω,RN )

solves

−
n∑

i,j=1

N∑
h=1

Dj(ahkij (x)Dizh) = |z + φ|p−2(zk + φk) + ϕk(x) in Ω

z = 0 on ∂Ω
k = 1, . . . , N,

where u(x) = z(x) + φ(x) for a.e. x ∈ Ω.

Therefore, in order to find solutions of our problem it is enough looking for
critical points of the C1-functional fχ : H1

0 (Ω,RN ) → R given by

fχ(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx−
1
p

∫
Ω

|u+ φ|p dx−
∫

Ω

ϕ · u dx

(we refer the reader to [120, 136] for some recalls of classical critical point theory).

Lemma 4.19. There exists A > 0 such that if u ∈ H1
0 (Ω,RN ) is a critical point of

fχ, then ∫
Ω

|u+ φ|p dx ≤ pA
(
f2
χ(u) + 1

)1/2
.

Proof. By Young’s inequality, for each ε > 0 there exist αε, βε > 0 such that

|u+ φ|p−1|φ| ≤ ε|u+ φ|p + αε|φ|p, |u+ φ||ϕ| ≤ ε|u+ φ|p + βε|ϕ|p
′
, (4.26)

with 1
p + 1

p′ = 1. Therefore, if u is a critical point of fχ, we get

fχ(u)

= fχ(u)− 1
2
f ′χ(u)[u]

=
(1
2
− 1
p

) ∫
Ω

|u+ φ|p dx− 1
2

∫
Ω

|u+ φ|p−2(u+ φ) · φdx− 1
2

∫
Ω

ϕ · u dx

≥ p− 2
2p

∫
Ω

|u+ φ|p dx− 1
2

∫
Ω

|u+ φ|p−1|φ| dx− 1
2

∫
Ω

(|u+ φ||ϕ|+ |ϕ||φ|) dx



72 MARCO SQUASSINA EJDE-2006/MON. 07

≥
(p− 2

2p
− ε

) ∫
Ω

|u+ φ|p dx− 1
2

(
αε‖φ‖pp + βε‖ϕ‖p

′

p′ + ‖ϕ‖2‖φ‖2
)
.

Choosing ε such that p− 2− 2pε > 0, i.e., ε ∈ ]0, 1
2 −

1
p [ , we get

pMεfχ(u) ≥
∫

Ω

|u+ φ|p dx− pMεγε(p, φ, ϕ),

where Mε = 2
p−2−2pε and

γε(p, φ, ϕ) =
1
2

(
αε‖φ‖pp + βε‖ϕ‖p

′

p′ + ‖ϕ‖2‖φ‖2
)
.

At this point, the assertion follows by A ≥
√

2Mε max{1, γε(p, φ, ϕ)}. �

Now, let η ∈ C∞(R,R) be a cut function such that η(s) = 1 for s ≤ 1, η(s) = 0
for s ≥ 2 while −2 < η′(s) < 0 when 1 < s < 2. For each u ∈ H1

0 (Ω,RN ) let us
define

ζ(u) = 2pA
(
f2
χ(u) + 1

)1/2
, ψ(u) = η

(
ζ(u)−1

∫
Ω

|u+ φ|p dx
)
, (4.27)

where A is as in Lemma 4.19. Finally, we introduce the modified functional f̃χ :
H1

0 (Ω,RN ) → R in order to apply the techniques used in [29]:

f̃χ(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx−
1
p

∫
Ω

|u|p dx− ψ(u)
∫

Ω

Θ(x, u) dx ,

with

Θ(x, u) =
|u+ φ|p

p
− |u|p

p
+ ϕ · u .

Let us provide an estimate for the loss of symmetry of f̃χ.

Lemma 4.20. There exists β > 0 such that∣∣∣f̃χ(u)− f̃χ(−u)
∣∣∣ ≤ β

(
|f̃χ(u)|

p−1
p + 1

)
for all u ∈ supp(ψ)

(here, supp(ψ) is the support of ψ).

Proof. First of all, let us show that there exist c1, c2 > 0 such that there results∣∣ ∫
Ω

(|u+ φ|p − |u|p) dx
∣∣ ≤ c1|fχ(u)|

p−1
p + c2, (4.28)∣∣ ∫

Ω

(|u− φ|p − |u|p) dx
∣∣ ≤ c1|fχ(u)|

p−1
p + c2, (4.29)∣∣ ∫

Ω

ϕ · u dx
∣∣ ≤ c1|fχ(u)|

p−1
p + c2 (4.30)

for all u ∈ supp(ψ). In fact, taken any u ∈ H1
0 (Ω,R) it is easy to see that

||u+ φ|p − |u|p| ≤ p2p−2|u+ φ|p−1|φ|+ p2p−2|φ|p, (4.31)

||u− φ|p − |u|p| ≤ p2p−2|u+ φ|p−1|φ|+ p22p−3|φ|p. (4.32)

Hence, by (4.31) we get∣∣ ∫
Ω

(|u+ φ|p − |u|p) dx
∣∣ ≤ p2p−2‖φ‖p

( ∫
Ω

|u+ φ|p dx
) p−1

p

+ p2p−2‖φ‖pp,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 73

while (4.32) implies∣∣∣ ∫
Ω

(|u− φ|p − |u|p) dx
∣∣∣ ≤ p2p−2‖φ‖p

( ∫
Ω

|u+ φ|p dx
) p−1

p

+ p22p−3‖φ‖pp.

Moreover, by Hölder and Young’s inequalities it results∣∣ ∫
Ω

ϕ · u dx
∣∣ ≤ ( ∫

Ω

|u+ φ|p dx
) p−1

p

+ (p− 2)
(‖ϕ‖p′
p− 1

) p−1
p−2

+ ‖ϕ‖2‖φ‖2.

If, furthermore, we assume u ∈ supp(ψ), it follows∫
Ω

|u+ φ|p dx ≤ 4pA(|fχ(u)|+ 1)

which implies (4.28), (4.29) and (4.30). Then, again by Young’s inequality, simple
calculations and (4.28), (4.30) give

|fχ(u)| ≤ a1|f̃χ(u)|+ a2, (4.33)

for suitable a1, a2 > 0. The assertion follows by combining inequalities (4.28),
(4.29), (4.30) and (4.33). �

Now, we want to link the critical points of f̃χ to those ones of fχ. To this aim we
need more information about f̃ ′χ. Taken u ∈ H1

0 (Ω,RN ), by direct computations
we get

f̃ ′χ(u)[u] =(1 + T1(u))
∫

Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx

− (1− ψ(u))
∫

Ω

|u|p dx − (ψ(u) + T1(u))
∫

Ω

ϕ · u dx

− (ψ(u) + T2(u))
∫

Ω

|u+ φ|p−2(u+ φ) · u dx

(4.34)

where T1, T2 : H1
0 (Ω,RN ) → R are defined by setting

T1(u) = 4p2A2η′(δ(u))δ(u)ζ(u)−2fχ(u)
∫

Ω

Θ(x, u) dx,

T2(u) = pη′(δ(u))ζ(u)−1

∫
Ω

Θ(x, u) dx+ T1(u) ,

with δ(u) = ζ(u)−1
∫
Ω
|u+ φ|p dx.

Remark 4.21. To point out some properties of the maps T1 and T2 defined above,
let us remark that by (4.28) and (4.30) there exist b1, b2 > 0 such that for all
u ∈ supp(ψ) it is

|Ti(u)| ≤ b1|fχ(u)|−
1
p + b2|fχ(u)|−1 for both i = 1, 2.

Therefore, arguing as in [118] (see also [29, Lemma 2.9]), there exist α0,M0 > 0
such that if M ≥M0 then

f̃χ(u) ≥M, u ∈ supp(ψ) ⇒ fχ(u) ≥ α0M ;

whence, it results |Ti(u)| → 0 as M → +∞ for i = 1, 2 (trivially, it is T1(u) =
T2(u) = 0 if u /∈ supp(ψ)).



74 MARCO SQUASSINA EJDE-2006/MON. 07

Theorem 4.22. There exists M1 > 0 such that if u is a critical point of f̃χ and
f̃χ(u) ≥M1 then u is a critical point of fχ and fχ(u) = f̃χ(u).

Proof. Let u ∈ H1
0 (Ω,RN ) be a critical point of f̃χ. By the definition of ψ it suffices

to show that, if f̃χ(u) ≥M1 for a large enough M1, then δ(u) < 1, i.e.,

ζ(u)−1

∫
Ω

|u+ φ|p dx < 1 .

By (4.34) we have

fχ(u) = fχ(u)− 1
2(1 + T1(u))

f̃ ′χ(u)[u]

= −1
p

∫
Ω

|u+ φ|p dx−
∫

Ω

ϕ · u dx+
1− ψ(u)

2(1 + T1(u))

∫
Ω

|u|p dx

+
ψ(u) + T1(u)
2(1 + T1(u))

∫
Ω

ϕ · u dx+
ψ(u) + T2(u)
2(1 + T1(u))

∫
Ω

|u+ φ|p−2(u+ φ) · u dx

=
(1
2
− 1
p

) ∫
Ω

|u+ φ|p dx− T1(u)− T2(u)
2(1 + T1(u))

∫
Ω

|u|p dx

+
1
2

(
ψ(u) + T2(u)

1 + T1(u)
− 1

) ∫
Ω

(|u+ φ|p − |u|p) dx

− ψ(u) + T2(u)
2(1 + T1(u))

∫
Ω

|u+ φ|p−2(u+ φ) · φdx

−
(

1− ψ(u) + T1(u)
2(1 + T1(u))

) ∫
Ω

ϕ · u dx.

Then, by Remark 4.21 it is possible to choose M1 > 0 so large that∣∣∣∣ 1− ψ(u)
1 + T1(u)

∣∣∣∣ ≤ 2 ,
∣∣∣∣ψ(u) + T1(u)

1 + T1(u)

∣∣∣∣ ≤ 2 ,∣∣∣∣ψ(u) + T2(u)
1 + T1(u)

− 1
∣∣∣∣ ≤ 2 ,

∣∣∣∣ψ(u) + T2(u)
1 + T1(u)

∣∣∣∣ ≤ 2 ;

so we deduce that for each ε > 0 there exist hε, γ̃ε(p, φ, ϕ) > 0 such that

fχ(u) ≥
(
p− 2
2p

− 2p−2

∣∣∣∣T2(u)− T1(u)
1 + T1(u)

∣∣∣∣− hε

) ∫
Ω

|u+ φ|p dx− γ̃ε(p, φ, ϕ)

where hε → 0 as ε → 0. At this point, choosing a priori ε and M1 in such a way
that

2p−2

∣∣∣∣T2(u)− T1(u)
1 + T1(u)

∣∣∣∣ + hε ≤
p− 2
4p

,

we obtain

fχ(u) ≥ p− 2
4p

∫
Ω

|u+ φ|p dx− γ̃ε(p, φ, ϕ) ,

which completes the proof if, as in Lemma 4.19, the constant A taken in the defi-
nition (4.27) is large enough. �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 75

4.8. The Palais-Smale condition. Let us point out that, in the check of the
Palais-Smale condition for semi-linear elliptic systems under the assumption (4.24),
an important role is played by the so called G̊arding’s inequality.

Lemma 4.23. Let (um)m be a bounded sequence in H1
0 (Ω,RN ) and let (wm)m be

a strongly convergent sequence in H−1(Ω,RN ) such that∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)Diu
m
h Djvk dx = 〈wm, v〉 for all v ∈ H1

0 (Ω,RN ).

Then (um)m has a subsequence (umk)k strongly convergent in H1
0 (Ω,RN ).

Proof. First of all, in our setting the following G̊arding type inequality holds: taken
ν as in (4.24) for each ε ∈]0, ν[ there exists cε ≥ 0 such that∫

Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx ≥ (ν − ε)‖Du‖22 − cε‖u‖22

for all u ∈ H1
0 (Ω,RN ) (see [105, Theorem 6.5.1]). Therefore, fixed ε > 0, we have

〈wl − wm, ul − um〉 =
∫

Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)Di(ulh − umh )Dj(ulk − umk ) dx

≥ (ν − ε)‖Dul −Dum‖22 − cε‖ul − um‖22
for all m, l ∈ N. Since um → u in L2(Ω,RN ), up to subsequences, we can conclude
that Dum → Du in L2(Ω,RN ). �

Now, let d ≥ 0 be such that∫
Ω

( n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk + d|u|2
)
dx ≥ ν

2
‖Du‖22 (4.35)

for all u ∈ H1
0 (Ω,RN ).

Lemma 4.24. There exists M2 > 0 such that if (um)m is a (PS)c-sequence of f̃χ
with c ≥M2, then (um)m is bounded in H1

0 (Ω,RN ).

Proof. Let M2 > 0 be fixed and consider (um)m, a (PS)c-sequence of f̃χ, with
c ≥M2, such that

M2 ≤ f̃χ(um) ≤ K ,

for a certain K > M2.
First of all, let us remark that if there exists a subsequence (umk)k such that

umk /∈ supp(ψ) for all k ∈ N, then it is a Palais-Smale sequence for the symmetric
functional

f0(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx−
1
p

∫
Ω

|u|p dx

in H1
0 (Ω,RN ). Whence, it is easier to prove that such a subsequence is bounded.

So, we can assume um ∈ supp(ψ) for all m ∈ N. For m ∈ N large enough and any
% > 0, taken d as in (4.35) by (4.34) it results

K + %‖Dum‖2



76 MARCO SQUASSINA EJDE-2006/MON. 07

≥ f̃χ(um)− %f̃ ′χ(um)[um]

=
1
2

(1− 2% (1 + T1(um)))
∫

Ω

( n∑
i,j=1

N∑
h,k=1

ahkij (x)Diu
m
h Dju

m
k + d|um|2

)
dx

− d

2
(1− 2% (1 + T1(um))) ‖um‖22 +

(
%(1− ψ(um))− 1

p

) ∫
Ω

|um|p dx

+ % (ψ(um) + T2(um))
∫

Ω

|um + φ|p−2(um + φ) · um dx

+ % (ψ(um) + T1(um))
∫

Ω

ϕ · um dx− ψ(um)
∫

Ω

Θ(x, um) dx.

Since it is p > 2, we can fix, a priori, a constant ~ ∈]1, p2 [ such that, taken
µ ∈]0, 1 − 2~

p [ , % ∈]~
p ,

1−µ
2 [ and µ̄ ∈]0, %(1 − 1

~ )[ , by Remark 4.21 if M2 is large
enough for all m ∈ N we have

|T1(um)| < min
{
1,

1− µ

2%
− 1

}
, |T2(um)| < 1− 1

~
− µ̄

%

and then

µ < 1− 2%(1 + T1(um)) ≤ 1, (4.36)

µ̄ ≤ %(1 + T2(um))− 1
p
. (4.37)

So, by (4.35) and (4.36) we obtain

K + %‖Dum‖2

≥ νµ

4
‖Dum‖22 −

d

2
‖um‖22 +

(
%(1− T2(um))− 1

p

) ∫
Ω

|um|p dx

− (% (1 + |T1(um)|) + 1)
∫

Ω

|ϕ||um| dx− % (1 + |T2(um)|)
∫

Ω

|um + φ|p−1|φ| dx

+
(
% (ψ(um) + T2(um))− ψ(um)

p

) ∫
Ω

(|um + φ|p − |um|p) dx .

Hence, fixed any ε > 0, by (4.26), (4.37) and a suitable choice of the positive
constants a1 and aε2 there results

K + % ‖Dum‖2 +
d

2
‖um‖22

≥ νµ

4
‖Dum‖22 + (µ̄− εa1) ‖um‖pp

+
(
% (ψ(um) + T2(um))− ψ(um)

p

) ∫
Ω

(|um + φ|p − |um|p) dx − aε2.

Let us point out that, as um ∈ supp(ψ), (4.28) and (4.33) imply(∫
Ω

(|um + φ|p − |um|p) dx
)
m∈N

is bounded. Whence, p > 2 and a suitable choice of ε small enough allow to
complete the proof. �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 77

Lemma 4.25. Let M2 be as in Lemma 4.24 and c ≥M2. Then, taken any (PS)c-
sequence (um)m for f̃χ, the sequence

ĝ(x, um) = |um|p−2um + ψ(um)Θ′(x, um) + ψ′(um)
∫

Ω

Θ(x, um) dx

admits a convergent subsequence in H−1(Ω,RN ).

The proof of the above lemma follows the steps in [110, Lemma 3.3].

Theorem 4.26. The functional f̃χ satisfies the Palais-Smale condition at each
level c ∈ R with c ≥M2, where M2 is as in Lemma 4.24 .

Proof. Let (um)m be a Palais-Smale sequence for f̃χ at level c ≥ M2. There-
fore, (um)m is bounded in H1

0 (Ω,RN ) and by Lemma 4.25, up to a subsequence,
(ĝ(x, um))m is strongly convergent in H−1(Ω,RN ). Hence, the assertion follows by
Lemma 4.23 applied to wm = ĝ(x, um)+f̃ ′χ(um) where, by assumption, f̃ ′χ(um) → 0
in H−1(Ω,RN ). �

4.9. Comparison of growths for min-max values. In this section we shall
build two min-max classes for f̃χ and then we compare the growth of the associated
min-max values.

Let (λl, ul)l be a sequence in R×H1
0 (Ω,RN ) such that

−∆ulk = λlulk in Ω

ul = 0 on ∂Ω,
k = 1, . . . N,

with (ul)l orthonormalized. Let us consider the finite dimensional subspaces

V0 :=
〈
u0

〉
; Vl+1 := Vl ⊕ Rul+1 for any l ∈ N.

Fixed l ∈ N it is easy to check that some constants β1, β2, β3, β4 > 0 exist such that

f̃χ(u) ≤ β1‖u‖21,2 − β2‖u‖p1,2 − β3‖u‖1,2 − β4 , for all u ∈ Vl.
Then, there exists Rl > 0 such that

u ∈ Vl, ‖u‖1,2 ≥ Rl ⇒ f̃χ(u) ≤ f̃χ(0) ≤ 0.

Definition 4.27. For any l ≥ 1 we set Dl = Vl ∩B(0, Rl),

Γl =
{
γ ∈ C(Dl,H

1
0 (Ω,RN )) : γ odd and γ|∂B(0,Rl)

= Id
}
,

and
bl = inf

γ∈Γl

max
u∈Dl

f̃χ(γ(u)).

To prove some estimates on the growth of the levels bl, a result due to Tanaka
(cf. [139]) implies the following lemma.

Lemma 4.28. There exist β > 0 and l0 ∈ N such that

bl ≥ β l
2p

n(p−2) for all l ≥ l0.

Proof. By (4.35) and simple calculations a1, a2 > 0 exist such that

f̃χ(u) ≥ ν

4
‖Du‖22 − a1‖u‖pp − a2 for all u ∈ ∂B(0, Rl) ∩ V ⊥l−1.

Then, it is enough to follow the proof of [139, Theorem 1]. �



78 MARCO SQUASSINA EJDE-2006/MON. 07

Now, let us introduce a second class of min-max values to be compared with bl.

Definition 4.29. Taken l ∈ N, define

Ul = {ξ = tul+1 + w : 0 ≤ t ≤ Rl+1, w ∈ B(0, Rl+1) ∩ Vl, ‖ξ‖1,2 ≤ Rl+1}
and

Λl =
{
λ ∈ C(Ul,H1

0 (Ω,RN )) : λ|Dl
∈ Γl and

λ|∂B(0,Rl+1)∪((B(0,Rl+1)\B(0,Rl))∩Vl) = Id
}
.

Assume
cl = inf

λ∈Λl

max
u∈Ul

f̃χ(λ(u)) .

The following result is the concrete version of Theorem 2.11.

Lemma 4.30. Assume cl > bl ≥ max{M1,M2}. Taken δ ∈]0, cl − bl[ , let us set

Λl(δ) =
{
λ ∈ Λl : f̃χ(λ(u)) ≤ bl + δ for all u ∈ Dl

}
,

cl(δ) = inf
λ∈Λl(δ)

max
u∈Ul

f̃χ(λ(u)) .

Then, cl(δ) is a critical value for f̃χ.

The proof of the above lemma can be obtained by arguing as in [118, Lemma
1.57]. Now, we prove that the situation cl = bl can not occur for all large l.

Lemma 4.31. Assume that cl = bl for all l ≥ l1. Then there exists γ > 0 with
bl ≤ γ lp.

Proof. Working as in [118, Lemma 1.64] it is possible to prove that

bl+1 ≤ bl + β
(
|bl|

p−1
p + 1

)
for all l ≥ l1.

The assertion follows by [15, Lemma 5.3] . �

Proof of Theorem 4.16. Observe that the inequality 2 < p < 2n+1
n implies

p <
2p

n(p− 2)
.

Therefore, by Lemmas 4.28 and 4.31 it follows that there exists a diverging sequence
(ln)n ⊂ N such that cln > bln for all n ∈ N, then Lemma 4.30 implies that (cln(δ))n
is a sequence of critical values for f̃χ. Whence, by Theorem 4.22 the functional fχ
has a diverging sequence of critical values. �

Remark 4.32. When p goes all the way up to 2∗, in a similar fashion, one can
prove that for each ν ∈ N there exists ε > 0 such that (Pεχ,εϕ,N ) has at least ν
distinct solutions in Mεχ. This is possible since there exists β > 0 such that∣∣∣f̃εχ(u)− f̃εχ(−u)

∣∣∣ ≤ εβ
(
|f̃εχ(u)|

p−1
p + 1

)
,

for each ε > 0 and u ∈ supp(ψ), where f̃εχ : H1
0 (Ω,RN ) → R is defined by

f̃εχ(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx+

− 1
p

∫
Ω

|u|p dx− ψε(u)
∫

Ω

Θε(x, u) dx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 79

with

Θε(x, u) =
|u+ εφ|p

p
− |u|p

p
+ εϕ · u, ψε(u) = η

(
ζ(u)−1

∫
Ω

|u+ εφ|p dx
)
;

for more details in the scalar case, see [2, 15].

4.10. Bolle’s method for non-symmetric problems. In this section we briefly
recall from [24] the theory devised by Bolle for dealing with problems with broken
symmetry.

The idea is to consider a continuous path of functionals starting from the sym-
metric functional f0 and to prove a preservation result for min-max critical levels
in order to get critical points also for the end-point functional f1.

Let X be a Hilbert space equipped with the norm ‖ · ‖ and f : [0, 1]×X → R
a C2-functional. Set fθ = f(θ, ·) if θ ∈ [0, 1].

Assume that X = X−⊕X+ and let (el)l≥1 be an orthonormal base of X+ such
that we can define an increasing sequence of subspaces as follows:

X0 := X−, Xl+1 := Xl ⊕ Rel+1 if l ∈ N.

Provided that dim(X−) < +∞, let us set

K = {ζ ∈ C(X ,X ) : ζ is odd and ζ(u) = u if ‖u‖ ≥ R}

for a fixed R > 0 and
cl = inf

ζ∈K
sup
u∈Xl

f0(ζ(u)) .

Assume that
• f satisfies a kind of Palais-Smale condition in [0, 1]×X : any ((θm, um))m

such that

(f(θm, um))m is bounded and f ′θm(um) → 0 as m→ +∞ (4.38)

converges up to subsequences;
• for any b > 0 there exists Cb > 0 such that

|fθ(u)| ≤ b ⇒
∣∣∣∣ ∂∂θ f(θ, u)

∣∣∣∣ ≤ Cb(‖f ′θ(u)‖+ 1)(‖u‖+ 1)

for all (θ, u) ∈ [0, 1]×X ;
• there exist two continuous maps η1, η2 : [0, 1]×R → R which are Lipschitz

continuous with respect to the second variable and such that η1 ≤ η2.
Suppose

η1(θ, fθ(u)) ≤
∂

∂θ
f(θ, u) ≤ η2(θ, fθ(u)) (4.39)

at each critical point u of fθ;
• f0 is even and for each finite dimensional subspace W of X it results

lim
‖u‖→+∞, u∈W

sup
θ∈[0,1]

f(θ, u) = −∞ .

Taken for i = 1, 2, let us denote by ψi : [0, 1]× R → R the solutions of

∂

∂θ
ψi(θ, s) = ηi(θ, ψi(θ, s))

ψi(0, s) = s .



80 MARCO SQUASSINA EJDE-2006/MON. 07

Note that ψi(θ, ·) are continuous, non-decreasing on R and ψ1 ≤ ψ2. Set

η1(s) = sup
θ∈[0,1]

η1(θ, s), η2(s) = sup
θ∈[0,1]

η2(θ, s) .

In this framework, the following abstract result can be proved.

Theorem 4.33. There exists C ∈ R such that if l ∈ N then

(a) either f1 has a critical point c̃l with ψ2(1, cl) < ψ1(1, cl+1) ≤ c̃l,
(b) or we have cl+1 − cl ≤ C (η1(cl+1) + η2(cl) + 1).

For the proof of the above theorem, see [24, Theorem 3] and [25, Theorem 2.2].

4.11. Application to semi-linear elliptic systems. In this section we want to
prove Theorem 4.16 in a simpler fashion by means of the arguments introduced in
Section 6.

For θ ∈ [0, 1], let us consider the functional fθ : H1
0 (Ω,RN ) → R defined as

fθ(u) =
1
2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx−
1
p

∫
Ω

|u+ θφ|p dx− θ

∫
Ω

ϕ · u dx .

It can be proved that all the previous assumptions are satisfied.

Lemma 4.34. Let ((θm, um))m ⊂ [0, 1]×H1
0 (Ω,RN ) be such that condition (4.38)

holds. Then ((θm, um))m converges up to subsequences.

Proof. Let ((θm, um))m be such that (4.38) holds. For a suitable K > 0 and any
% > 0 it is

K + % ‖Dum‖2 ≥ fθm(um)− %f ′θm(um)[um]

=
(1
2
− %

) ∫
Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)Diu
m
h Dju

m
k dx+

(
%− 1

p

) ∫
Ω

|um + θmφ|pdx

− θm%

∫
Ω

|um + θmφ|p−2(um + θmφ) · φdx

for all m large enough. Then, fixed any ε > 0 and taken d as in (4.35), (4.26) and
simple computations imply

% ‖Dum‖2 +
(1
2
− %)d‖um‖22 ≥

(1
2
− %

)ν
2
‖Dum‖22 +

1
2p−1

(
%(1− ε)− 1

p

)
‖um‖pp − aε

for a certain aε > 0. Hence, if we fix % ∈] 1p ,
1
2 [ and ε ∈]0, 1 − 1

%p [ , by this last
inequality it follows that (um)m has to be bounded in H1

0 (Ω,RN ).
So, if we assume wm = f ′θm(um) + |um + θmφ|p−2(um + θmφ) + θmϕ it is easy to
prove that (wm)m strongly converges in H−1(Ω,RN ), up to subsequences. Whence,
Lemma 4.23 implies that (um)m has a converging subsequence in H1

0 (Ω,RN ). �

Lemma 4.35. For each b > 0 there exists Cb > 0 such that

|fθ(u)| ≤ b ⇒
∣∣ ∂
∂θ
f(θ, u)

∣∣ ≤ Cb(‖f ′θ(u)‖+ 1)(‖u‖1,2 + 1)

for all (θ, u) ∈ [0, 1]×H1
0 (Ω,RN ).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 81

Proof. Fix b > 0. The condition |fθ(u)| ≤ b is equivalent to∣∣∣ ∫
Ω

(1
2

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk − 1
p
|u+ θφ|p − θϕ · u

)
dx

∣∣∣ ≤ b (4.40)

which implies that

θ

∫
Ω

ϕ · u dx ≥ p

2

∫
Ω

n∑
i,j=1

N∑
h,k=1

ah,kij (x)DiuhDjuk dx

−
∫

Ω

|u+ θφ|p dx− (p− 1) θ
∫

Ω

ϕ · u dx− p b.

(4.41)

So, taken d as in (4.35), we have

−f ′θ(u)[u] = −
∫

Ω

n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk dx

+
∫

Ω

|u+ θφ|p−2(u+ θφ) · u dx+ θ

∫
Ω

ϕ · u dx

≥
(p

2
− 1

) ∫
Ω

( n∑
i,j=1

N∑
h,k=1

ahkij (x)DiuhDjuk + d |u|2
)
dx

−
(p

2
− 1

)
d ‖u‖22 −

∫
Ω

|u+ θφ|p−2(u+ θφ) · θφ dx

− (p− 1) θ
∫

Ω

ϕ · u dx − p b

≥ (p− 2)
ν

4
‖Du‖22 −

(p
2
− 1

)
d ‖u‖22

−
∫

Ω

|u+ θφ|p−2(u+ θφ) · θφ dx− (p− 1) θ
∫

Ω

ϕ · u dx− p b.

By Hölder inequality there exist c1, c2, c3 > 0 such that∣∣∣ ∫
Ω

|u+ θφ|p−2(u+ θφ) · θφ dx
∣∣∣ ≤ c1‖u+ θφ‖p−1

p , (4.42)∣∣∣ ∫
Ω

ϕ · u dx
∣∣∣ ≤ c2‖u+ θφ‖p + c3 ; (4.43)

while (4.40) implies
‖u+ θφ‖pp ≤ c4‖Du‖22 + c5(b) (4.44)

for suitable c4, c5(b) > 0. Then, since Young’s inequality yields

c1‖u+ θφ‖p−1
p ≤ ε‖u+ θφ‖pp + c̃1(ε),

c2‖u+ θφ‖p ≤ ε‖u+ θφ‖pp + c̃2(ε),
(4.45)

for all ε > 0 and certain c̃1(ε), c̃2(ε) > 0, it can be proved that c6, c7(ε, b) > 0 exist
such that

−f ′θ(u)[u] ≥
(
(p− 2)

ν

4
− εc6

)
‖Du‖22 − c7(ε, b).

So, if ε is small enough, some c̃6, c̃7(b) > 0 can be find such that

c̃6 ‖Du‖22 − c̃7(b) ≤ −f ′θ(u)[u]. (4.46)



82 MARCO SQUASSINA EJDE-2006/MON. 07

On the other hand, since

∂

∂θ
f(θ, u) = −

∫
Ω

|u+ θφ|p−2(u+ θφ) · φdx−
∫

Ω

ϕ · u dx

by (4.42) and (4.43) it follows∣∣ ∂
∂θ
f(θ, u)

∣∣ ≤ c8‖u+ θφ‖p−1
p + c9 (4.47)

and then by (4.45) ∣∣ ∂
∂θ
f(θ, u)

∣∣ ≤ ε‖u+ θφ‖pp + c10(ε)

for any ε > 0 and c8, c9, c10(ε) > 0 suitable constants. So, for all ε > 0 and a
certain c11(ε, b) > 0, (4.44) implies∣∣ ∂

∂θ
f(θ, u)

∣∣ ≤ εc4 ‖Du‖22 + c11(ε, b). (4.48)

Hence, the proof follows by (4.46), (4.48) and a suitable choice of ε. �

Lemma 4.36. If u ∈ H1
0 (Ω,RN ) is a critical point of fθ, there exists σ > 0 such

that ∫
Ω

|u+ θφ|p dx ≤ σ
(
f2
θ (u) + 1

)1/2
.

For the proof of the above lemma it suffices to argue as in Lemma 4.19.

Lemma 4.37. At each critical point u of fθ the inequality (4.39) holds if η1, η2 are
defined in (θ, s) ∈ [0, 1]× R as

−η1(θ, s) = η2(θ, s) = C
(
s2 + 1

) p−1
2p (4.49)

for a suitable constant C > 0.

For the proof of this lemma, it is sufficient to combine (4.47) and Lemma 4.36.

New proof of Theorem 4.16. Clearly, f0 is an even functional. Moreover, by Lem-
mas 4.34, 4.35 and 4.37 all the hypotheses of the existence theorem are fulfilled.
Now, consider (Vl)l, the sequence of subspaces of H1

0 (Ω,RN ) introduced in the
previous sections. Defined the set of maps K with X = H1

0 (Ω,RN ), assume

cl = inf
ζ∈K

sup
u∈Vl

f0(ζ(u)) .

Simple computations allow to prove that, taken any finite dimensional subspace W
of H1

0 , some constants β1, β2, β3 > 0 exist such that

fθ(u) ≤ β1‖u‖21,2 − β2‖u‖p1,2 − β3 for all u ∈ W .

Then
lim

‖u‖1,2→+∞, u∈W
sup
θ∈[0,1]

fθ(u) = −∞.

Hence, Theorem 4.33 applies and, by the choice made in (4.49), the condition (b)
implies that there exists C̃ > 0 such that

|cl+1 − cl| ≤ C̃
(
(cl)

p−1
p + (cl+1)

p−1
p + 1

)
, (4.50)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 83

which implies cl ≤ γ̃ lp for some γ̃ > 0 in view of [15, Lemma 5.3]. Taking into
account Lemma 4.28 we conclude that (4.50) can not hold provided that

2p
n(p− 2)

> p,

namely p ∈]2, 2 n+1
n [. Whence, the assertion follows by (a) of Theorem 4.33. �

4.12. The diagonal case. Now, we want to prove Theorem 4.17. To this aim let
us point out that we deal with the problem

−∆uk = |u|p−2uk + ϕk(x) in Ω
u = χ on ∂Ω
k = 1, . . . , N

(4.51)

and want to prove that (4.51) has an infinite number of solutions if p ∈
]
2, 2n

n−1

[
.

In this case, the functional fθ defined in the previous section becomes

fθ(u) =
1
2

∫
Ω

|∇u|2 dx− 1
p

∫
Ω

|u+ θφ|p dx− θ

∫
Ω

ϕ · u dx

where φ solves the system (4.25) with ahkij = δhkij .
By the regularity assumptions we made on ∂Ω, χ and ϕ the following lemma can

be proved.

Lemma 4.38. There exists c > 0 such that if u is a critical point of fθ, then∣∣∣ ∫
∂Ω

(1
2
|∇w|2 −

∣∣∂w
∂n

∣∣2)dσ∣∣∣ ≤ c

∫
Ω

(
|∇w|2 + |w|p + 1

)
dx

where w = u+ θφ.

Proof. If u ∈ H1
0 (Ω,RN ) is such that f ′θ(u) = 0, then some regularity theorems

imply that u is a classical solution of the problem

−∆uk = |u+ θφ|p−2(uk + θφ) + θϕk in Ω
u = 0 on ∂Ω
k = 1, . . . , N,

then w = u+ θφ ∈ C2(Ω,RN ) solves the elliptic system

−∆wk = |w|p−2wk + θϕk in Ω
wk = θφk on ∂Ω
k = 1, . . . , N.

(4.52)

Taken δ > 0, let us consider a cut function η̃ ∈ C∞(R,R) such that η̃(s) = 1 for
s ≤ 0 and η̃(s) = 0 for s ≥ δ. Moreover, taken any x ∈ RN , let d(x, ∂Ω) be the
distance of x from the boundary of Ω. Let us point out that, since Ω is smooth
enough, δ can be chosen in such a way that d(·, ∂Ω) is of class C2 on

Ω ∩ {x ∈ Rn : d(x, ∂Ω) < δ},

and n̂(x) = ∇d(x, ∂Ω) coincides on ∂Ω with the inner normal. So, defined g : RN →
R as g(x) = η̃(d(x, ∂Ω)), for each k = 1, . . . , N let us multiply the k-th equation in



84 MARCO SQUASSINA EJDE-2006/MON. 07

(4.52) by g(x)∇wk · n̂(x). Hence, working as in [25, Lemma 4.2] and summing up
with respect to k, we get

N∑
k=1

∫
Ω

−∆wk g(x)∇wk · n̂ dx =
∫
∂Ω

(1
2
|∇w|2 −

∣∣∣∣∂w∂n
∣∣∣∣2 )

dσ +O
(
‖∇w‖22

)
,

N∑
k=1

∫
Ω

|w|p−2wk g(x)∇wk · n̂ dx =
θp

p

∫
∂Ω

|φ|p dσ +O
(
‖w‖pp

)
,

N∑
k=1

∫
Ω

θϕk(x) g(x)∇wk · n̂ dx = θ2
∫
∂Ω

ϕ · φ dσ +O (‖w‖p) .

Whence, the proof follows by putting together these identities. �

With the stronger assumptions we made in this section, the estimates in Lemma
4.37 can be improved.

Lemma 4.39. At each critical point u of fθ the inequality (4.39) holds if η1, η2 are
defined in (θ, s) ∈ [0, 1]× R as

−η1(θ, s) = η2(θ, s) = C
(
s2 + 1

)1/4

for a suitable constant C > 0.

Proof. Let u be a critical point of fθ. Then,
∂

∂θ
f(θ, u) =

∫
∂Ω

∂u

∂n
· φ dσ +

∫
Ω

ϕ · (θφ− u) dx ;

so, taking into account Lemma 4.38, it is enough to argue as in [25, Lemma 4.3]. �

Proof of Theorem 4.17. Arguing as in the proof of Theorem 4.16, we have that the
proof of Theorem 4.17 follows by Theorem 4.33 since also in this case the condition
(b) can not occur. Let us point out that, by Lemma 4.39, the incompatibility
condition is 2p

n(p−2) > 2, i.e. p ∈
]
2, 2n

n−1

[
. �

5. Problems of Jumping Type

We refer the reader to [79, 80]. Some parts of these publications have been
slightly modified to give this collection a more uniform appearance.

5.1. Fully nonlinear elliptic equation. Let us consider the semi-linear elliptic
problem

−
n∑

i,j=1

Dj(aij(x)Diu) = g(x, u) + ω in Ω

u = 0 on ∂Ω ,

(5.1)

where Ω is a bounded domain in Rn, ω ∈ H−1(Ω) and g : Ω× R → R satisfies

lim
s→−∞

g(x, s)
s

= α , lim
s→+∞

g(x, s)
s

= β. (5.2)

Let us denote by (µh) the eigenvalues of the linear operator on H1
0 (Ω)

u 7→ −
n∑

i,j=1

Dj(aij(x)Diu).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 85

Since 1972, this jumping problem has been widely investigated in the case when
some eigenvalue µh belongs to the interval ]β, α[ (see e.g. [98, 100, 124] and refer-
ences therein), starting from the pioneering paper [4] of Ambrosetti and Prodi.

On the other hand, since 1994, several efforts have been devoted to study exis-
tence of weak solutions of the quasi-linear problem

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju = g(x, u) + ω in Ω

u = 0 on ∂Ω ,

(5.3)

via techniques of non-smooth critical point theory (see e.g. [8, 32, 36, 49, 137]).
In particular, a jumping problem for the previous equation has been treated

in [31]. More recently, existence for the Euler’s equations of multiple integrals of
calculus of variations

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) + ω in Ω
u = 0 on ∂Ω ,

(5.4)

have also been considered in [6] and in [112, 132] via techniques developed in [36].
In this section we see how the results of [31] may be extended to the more general
elliptic problem (5.4). We shall approach the problem from a variational point of
view, that is looking for critical points for continuous functionals f : W 1,p

0 (Ω) → R
of type

f(u) =
∫

Ω

L (x, u,∇u) dx−
∫

Ω

G(x, u) dx− 〈ω, u〉.

We point out that, in general, these functionals are not even locally Lipschitzian,
so that classical critical point theory fails. Then we shall refer to non-smooth
critical point theory, In our main result (Theorem 5.1) we shall prove existence of
at least two solutions of the problem by means of a classical min-max theorem in
its non-smooth version.

5.2. The main result. We assume that Ω is a bounded domain of Rn, 1 < p < n,
ω ∈W−1,p′(Ω) and L : Ω×R×Rn → R is measurable in x for all (s, ξ) ∈ R×Rn
and of class C1 in (s, ξ) a.e. in Ω. Moreover, the function L (x, s, ·) is strictly
convex and for each t ∈ R L (x, s, tξ) = |t|pL (x, s, ξ) for a.e. x ∈ Ω and for all
(s, ξ) ∈ R× Rn. Furthermore, we assume that:

• There exist ν > 0 and b1 ∈ R such that:

ν|ξ|p ≤ L (x, s, ξ) ≤ b1|ξ|p , (5.5)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn;
• there exist b2, b3 ∈ R such that:

|DsL (x, s, ξ)| ≤ b2|ξ|p,

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn and

|∇ξL (x, s, ξ)| ≤ b3|ξ|p−1, (5.6)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn ;
• there exist R > 0 and a bounded Lipschitzian function ϑ : R → [0,+∞[

such that:
|s| ≥ R⇒ sDsL (x, s, ξ) ≥ 0 , (5.7)



86 MARCO SQUASSINA EJDE-2006/MON. 07

sDsL (x, s, ξ) ≤ sϑ′(s)∇ξL (x, s, ξ) · ξ , (5.8)

for a.e. x ∈ Ω and s ∈ R and for all ξ ∈ Rn. Without loss of generality, we
may take assume that ϑ(s) → ϑ as s→ ±∞ ;

• g(x, s) is a Carathéodory function and G(x, s) =
∫ s
0
g(x, τ) dτ . We assume

that there exist a ∈ Lnp/(n(p−1)+p)(Ω) and b ∈ Ln/p(Ω) such that:

|g(x, s)| ≤ a(x) + b(x)|s|p−1 , (5.9)

for a.e. x ∈ Ω and all s ∈ R. Moreover, there exist α, β ∈ R such that

lim
s→−∞

g(x, s)
|s|p−2s

= α , lim
s→+∞

g(x, s)
|s|p−2s

= β , (5.10)

for a.e. x ∈ Ω.
Let us now suppose that

lim
s→+∞

L (x, s, ξ) = lim
s→−∞

L (x, s, ξ)

(both limits exist by (5.7)) and denote by L∞(x, ξ) the common value, that we
shall assume to be of the form a(x)|ξ|p with a ∈ L∞(Ω). Moreover, assume that

sh → +∞, ξh → ξ ⇒ ∇ξL (x, sh, ξh) → ∇ξL∞(x, ξ). (5.11)

Let

λ1 = min
{
p

∫
Ω

L∞(x,∇u) dx : u ∈W 1,p
0 (Ω),

∫
Ω

|u|p dx = 1
}
, (5.12)

be the first eigenvalue of {u 7−→ −div (∇ξL∞(x,∇u))}.
Observe that by [6, Lemma 1.4] the first eigenfunction φ1 belongs to L∞(Ω) and

by [142, Theorem 1.1] is strictly positive.
Under the previous assumptions, we consider problem (5.4) in the case ω =

tφp−1
1 + ω0, with ω0 ∈W−1,p′(Ω) and t ∈ R. The following is our main result.

Theorem 5.1. If β < λ1 < α then there exist t ∈ R and t ∈ R such that the
problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) + tφp−1
1 + ω0 in Ω

u = 0 on ∂Ω
(5.13)

has at least two weak solutions in W 1,p
0 (Ω) for t > t and no solutions for t < t.

This result extends [31, Corollary 2.3] dealing with the case p = 2 and

L (x, s, ξ) =
1
2

n∑
i,j=1

aij(x, s)ξiξj −G(x, s)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.
In this particular case, existence of at least three solutions has been recently

proved in [34] assuming β < µ1 and α > µ2 where µ1 and µ2 are the first and
second eigenvalue of the operator

u 7→ −
n∑

i,j=1

Dj(AijDiu).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 87

In our general setting we only have existence of the first eigenvalue λ1 and it is not
clear how to define higher order eigenvalues λ2, λ3, . . . . Therefore in our case the
comparison of α and β with such eigenvalues is still not possible.

5.3. The concrete Palais-Smale condition. The following result is one of the
main tool of the section.

Lemma 5.2. Let (uh) be a sequence in W 1,p
0 (Ω) and (%h) ⊆ ]0,+∞[ with %h → +∞

be such that
vh =

uh
%h

⇀ v in W 1,p
0 (Ω).

Let γh ⇀ γ in Ln/p(Ω) with |γh(x)| ≤ c(x) for some c ∈ Ln/p(Ω). Moreover, let

µh → µ in Lnp
′/(n+p′)(Ω), δh → δ in W−1,p′(Ω)

be such that for each ϕ ∈ C∞c (Ω):∫
Ω

∇ξL (x, uh,∇uh) · ∇ϕdx+
∫

Ω

DsL (x, uh,∇uh)ϕdx

=
∫

Ω

γh|uh|p−2uhϕdx+ %p−1
h

∫
Ω

µhϕdx+ 〈δh , ϕ〉.
(5.14)

Then, the following facts hold:
(a) (vh) is strongly convergent to v in W 1,p

0 (Ω) ;
(b) (γh|vh|p−2vh) is strongly convergent to γ|v|p−2v in W−1,p′(Ω) ;
(c) there exist η+, η− ∈ L∞(Ω) such that:

η+(x) =

{
exp

{
−ϑ

}
if v(x) > 0

exp{MR} if v(x) < 0 ,

exp{−ϑ} ≤ η+(x) ≤ exp{MR} if v(x) = 0 ,
and

η−(x) =

{
exp

{
−ϑ

}
if v(x) < 0

exp{MR} if v(x) > 0 ,

exp{−ϑ} ≤ η−(x) ≤ exp{MR} if v(x) = 0 ,
and such that for every ϕ ∈W 1,p

0 (Ω) with ϕ ≥ 0:∫
Ω

η+∇ξL∞(x,∇v) · ∇ϕdx ≥
∫

Ω

γη+|v|p−2vϕ dx+
∫

Ω

µη+ϕdx ,∫
Ω

η−∇ξL∞(x,∇v) · ∇ϕdx ≤
∫

Ω

γη−|v|p−2vϕ dx+
∫

Ω

µη−ϕdx.

Proof. Arguing as in [31, Lemma 3.1], (b) immediately follows. Let us now prove
(a). Up to a subsequence, vh(x) → v(x) for a.e. x ∈ Ω. Consider now the function
ζ : R → R defined by

ζ(s) =


Ms if 0 < s < R

MR if s ≥ R

−Ms if −R < s < 0
MR if s ≤ −R ,

(5.15)

where M ∈ R is such that for a.e. x ∈ Ω, each s ∈ R and ξ ∈ Rn

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ. (5.16)



88 MARCO SQUASSINA EJDE-2006/MON. 07

By [132, Proposition 3.1], we may choose in (5.14) the functions ϕ = vh exp{ζ(uh)}
yielding∫

Ω

∇ξL (x, uh,∇uh) · ∇vh exp{ζ(uh)} dx

+
∫

Ω

[DsL (x, uh,∇uh) + ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh] vh exp{ζ(uh)} dx

=
∫

Ω

γh|uh|p−2uhvh exp{ζ(uh)} dx+ %p−1
h

∫
Ω

µhvh exp{ζ(uh)} dx

+ 〈δh, vh exp{ζ(uh)}〉.

Therefore, taking into account conditions (5.7) and (5.16), we have

%p−1
h

∫
Ω

∇ξL (x, uh,∇vh) · ∇vh exp{ζ(uh)} dx

≤ %p−1
h

∫
Ω

γh|vh|p exp{ζ(uh)} dx+ %p−1
h

∫
Ω

µhvh exp{ζ(uh)} dx

+ 〈δh, vh exp{ζ(uh)}〉 .

After division by %p−1
h , using the hypotheses on γh, µh and δh, we obtain

lim sup
h

∫
Ω

∇ξL (x, uh,∇vh) · ∇vh exp{ζ(uh)} dx

≤ exp{MR}
(∫

Ω

γ|v|p dx+
∫

Ω

µv dx

)
.

(5.17)

Now, let us consider the function ϑ1 : R → R given by

ϑ1(s) =


ϑ(s) if s ≥ 0
Ms if−R ≤ s ≤ 0
−MR if s ≤ −R ,

(5.18)

with ϑ satisfying (5.8). Considering in (5.14) the functions (v+ ∧ k) exp{−ϑ1(uh)}
with k ∈ N, we obtain∫

Ω

∇ξL (x, uh,∇vh) · ∇(v+ ∧ k) exp{−ϑ1(uh)} dx

+
1

%p−1
h

∫
Ω

[DsL (x, uh,∇uh)− ϑ′1(uh)∇ξL (x, uh,∇uh) · ∇uh]

× (v+ ∧ k) exp{−ϑ1(uh)} dx

=
∫

Ω

γh|vh|p−2vh(v+ ∧ k) exp{−ϑ1(uh)} dx+
∫

Ω

µh(v+ ∧ k) exp{−ϑ1(uh)} dx

+
1

%hp−1

〈
δh, (v+ ∧ k) exp{−ϑ1(uh)}

〉
.

(5.19)
By (5.7), (5.8) and (5.16) it results that for each h ∈ N

[DsL (x, uh,∇uh)− ϑ′1(uh)∇ξL (x, uh,∇uh) · ∇uh] (v+ ∧ k) exp{−ϑ1(uh)} ≤ 0.

Taking into account assumptions (5.11) and (5.6), we may apply [54, Theorem 5]
and deduce that

a.e. in Ω \ {v = 0} : ∇vh(x) → ∇v(x).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 89

Being uh(x) → +∞ a.e. in Ω \ {v = 0}, again recalling (5.11), we have

a.e. in Ω \ {v = 0} : ∇ξL (x, uh(x),∇vh(x)) → ∇ξL∞(x,∇v(x)).
By combining this pointwise convergence with (5.6), we obtain

∇ξL (x, uh,∇vh) ⇀ ∇ξL∞(x,∇v) in Lp
′
(Ω). (5.20)

Therefore, for each k ∈ N we have

lim
h
∇ξL (x, uh,∇vh) · ∇(v+ ∧ k) exp{−ϑ1(uh)}

= ∇ξL∞(x,∇v) · ∇(v+ ∧ k) exp{−ϑ} ,

strongly in L1(Ω),

lim
h

(v+ ∧ k) exp{−ϑ1(uh)} = (v+ ∧ k) exp{−ϑ} ,

weakly in W 1,p
0 (Ω), using (b)

lim
h
γh|vh|p−2vh(v+ ∧ k) exp{−ϑ1(uh)} = γ|v|p−2v(v+ ∧ k) exp{−ϑ} ,

strongly in L1(Ω) and

lim
h

1
%p−1
h

(v+ ∧ k) exp{−ϑ1(uh)} = 0 ,

weakly in W 1,p
0 (Ω). Therefore, letting h→ +∞ in (5.19), for each k ∈ N we get∫

Ω

∇ξL∞(x,∇v) · ∇(v+ ∧ k) exp{−ϑ} dx

≥
∫

Ω

γ|v|p−2v(v+ ∧ k) exp{−ϑ} dx+
∫

Ω

µ(v+ ∧ k) exp{−ϑ} dx.

Finally, if we let k → +∞, after division by exp{−ϑ}, we have∫
Ω

∇ξL∞(x,∇v+) · ∇v+ dx ≥
∫

Ω

γ|v|p−2(v+)2 dx+
∫

Ω

µv+ dx. (5.21)

Analogously, if we define a function ϑ2 : R → R by

ϑ2(s) =


ϑ(s) if s ≤ 0
−Ms if 0 ≤ s ≤ R

−MR if s ≥ R,

and consider in (5.14) the test functions (v− ∧ k) exp{−ϑ2(uh)} with k ∈ N, we
obtain ∫

Ω

∇ξL∞(x,∇v) · ∇v−dx ≤ −
∫

Ω

γ|v|p−2(v−)2 dx+
∫

Ω

µv−dx. (5.22)

Thus, combining (5.21) and (5.22) yields∫
Ω

∇ξL∞(x,∇v) · ∇v dx ≥
∫

Ω

γ|v|p dx+
∫

Ω

µv dx. (5.23)

Finally, putting together (5.17) and (5.23), we conclude

lim sup
h

∫
Ω

∇ξL (x, uh,∇vh) · ∇vh exp{ζ(uh)} dx

≤ exp{MR}
∫

Ω

∇ξL∞(x,∇v) · ∇v dx.



90 MARCO SQUASSINA EJDE-2006/MON. 07

In particular, by Fatou’s Lemma, it results

exp{MR}
∫

Ω

∇ξL∞(x,∇v) · ∇v dx

≤ lim inf
h

∫
Ω

∇ξL (x, uh,∇vh) · ∇vh exp{ζ(uh)} dx

≤ exp{MR}
∫

Ω

∇ξL∞(x,∇v) · ∇v dx;

namely,

∇ξL (x, uh,∇vh) · ∇vh exp{ζ(uh)} → exp{MR}∇ξL∞(x,∇v) · ∇v.
L1(Ω). Therefore, since ν|∇vh|p ≤ ∇ξL (x, uh,∇vh)·∇vh exp{ζ(uh)}, again thanks
to Fatou’s Lemma, we conclude that

lim sup
h

∫
Ω

|∇vh|p dx ≤
∫

Ω

|∇v|p dx,

and the proof of (a) is concluded.
Let us now prove assertion (c). Up to a subsequence, exp{−ϑ1(uh)} weakly∗

converges in L∞(Ω) to some η+. Of course, we have

η+(x) =

{
exp

{
−ϑ

}
if v(x) > 0

exp{MR} if v(x) < 0,

exp{−ϑ} ≤ η+(x) ≤ exp{MR} if v(x) = 0.
Then, let us consider in (5.14) as test functions:

ϕ exp{−ϑ1(uh)}, ϕ ∈ C∞c (Ω), ϕ ≥ 0.

Whence, like in the previous argument, we obtain∫
Ω

η+∇ξL∞(x,∇v) · ∇ϕdx ≥
∫

Ω

γη+|v|p−2vϕ dx+
∫

Ω

µη+ϕdx,

for any positive ϕ ∈W 1,p
0 (Ω). Similarly, by means of the test functions

ϕ exp{−ϑ2(uh)}, ϕ ∈ C∞c (Ω), ϕ ≥ 0,

we get for any positive ϕ ∈W 1,p
0 (Ω)∫

Ω

η−∇ξL∞(x,∇v) · ∇ϕdx ≤
∫

Ω

γη−|v|p−2vϕ dx+
∫

Ω

µη−ϕdx,

where η− is the weak∗ limit of some subsequence of exp{−ϑ2(uh)}. �

Consider now

g0(x, s) = g(x, s)− β|s|p−2s+ + α|s|p−2s−, G0(x, s) =
∫ s

0

g0(x, τ) dτ.

Of course, g0 is a Carathéodory function satisfying for a.e. x ∈ Ω and for all s ∈ R

lim
|s|→∞

g0(x, s)
|s|p−2s

= 0, |g0(x, s)| ≤ a(x) + b̃(x)|s|p−1 ,

with b̃ ∈ Ln/p(Ω). Since we are interested in weak solutions u ∈ W 1,p
0 (Ω) of the

equations

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) + tφp−1
1 + ω0 ,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 91

let us define the associated functional ft : W 1,p
0 (Ω) → R, by setting

ft(u) =
∫

Ω

L (x, u,∇u) dx− β

p

∫
Ω

(u+)p dx− α

p

∫
Ω

(u−)p dx

−
∫

Ω

G0(x, u) dx− |t|p−2t

∫
Ω

φp−1
1 u dx− 〈ω0, u〉.

Lemma 5.3. Let (uh) a sequence in W 1,p
0 (Ω) and %h ⊆]0,+∞[ with %h → +∞.

Assume that the sequence (uh

%h
) is bounded in W 1,p

0 (Ω). Then

g0(x, uh)
%p−1
h

→ 0 in L
np′

n+p′ (Ω),
G0(x, uh)

%ph
→ 0 in L1(Ω).

For the proof of the above lemma, argue as in [31, Lemma 3.3]. We now recall
from [132] a compactness property of (CPS)c-sequences.

Theorem 5.4. Let (uh) be a bounded sequence in W 1,p
0 (Ω) and set

〈wh, v〉 =
∫

Ω

∇ξL (x, uh,∇uh) · ∇v dx+
∫

Ω

DsL (x, uh,∇uh)v dx, (5.24)

for all v ∈ C∞c (Ω). If (wh) is strongly convergent to some w in W−1,p′(Ω), then
(uh) admits a strongly convergent subsequence in W 1,p

0 (Ω).

For the proof of the above theorem, see [132, Theorem 3.4].

Lemma 5.5. For each c, t ∈ R the following assertions are equivalent:

(a) ft satisfies the (CPS)c condition ;
(b) every (CPS)c-sequence for ft is bounded in W 1,p

0 (Ω).

Proof. (a) ⇒ (b). It is trivial. (b) ⇒ (a). Let (uh) be a (CPS)c-sequence for ft.
Since (uh) is bounded in W 1,p

0 (Ω), and the map

u 7−→ g(x, u) + tφp−1
1 + ω0 ,

is completely continuous by (5.9), up to a subsequence (g(x, uh) + tφp−1
1 + ω0) is

strongly convergent in L
np′

n+p′ (Ω), hence in W−1,p′(Ω). �

We now come to one of the main tool of this section.

Theorem 5.6. Let c, t ∈ R. Then ft satisfies the (CPS)c condition.

Proof. If (uh) is a (CPS)c-sequence for ft, we have ft(uh) → c and for all v ∈
C∞0 (Ω):∫

Ω

∇ξL (x, uh,∇uh) · ∇v dx+
∫

Ω

DsL (x, uh,∇uh)v dx

− β

∫
Ω

(u+
h )p−1v dx+ α

∫
Ω

(u−h )p−1v dx−
∫

Ω

g0(x, uh)v dx− |t|p−2t

∫
Ω

φ1v dx

= 〈ω0 + σh, v〉,

where σh → 0 in W−1,p′(Ω). Taking into account Theorem 5.4, by Lemma 5.5 it
suffices to show that (uh) is bounded in W 1,p

0 (Ω). Assume by contradiction that,



92 MARCO SQUASSINA EJDE-2006/MON. 07

up to a subsequence, ‖uh‖1,p → +∞ as h → +∞ and set vh = uh‖uh‖−1
1,p. By

Lemma 5.3, we can apply Lemma 5.2 choosing

γh(x) =

{
β if uh(x) ≥ 0
α if uh(x) < 0,

%h = ‖uh‖1,p,

µh =
g0(x, uh)
‖uh‖p−1

1,p

, δh = |t|p−2tφ1 + ω0 + σh.

Then, up to a subsequence, (vh) strongly converges to some v inW 1,p
0 (Ω). Moreover,

putting ϕ = v+ in (c) of Lemma 5.2, we get∫
Ω

η−∇ξL∞(x,∇v+) · ∇v+ dx ≤
∫

Ω

βη−(v+)p dx ,

hence, taking into account (5.12), we have

λ1

∫
Ω

(v+)p dx ≤
∫

Ω

∇ξL∞(x,∇v+) · ∇v+ dx ≤ β

∫
Ω

(v+)p dx.

Since β < λ1, then v+ = 0. By using again the first inequality in (c) of Lemma 5.2,
for each ϕ ≥ 0 we get∫

Ω

η+∇ξL∞(x,∇v) · ∇ϕdx ≥ α

∫
Ω

η+|v|p−2vϕ dx.

namely, since v ≤ 0, we have∫
Ω

∇ξL∞(x,∇v) · ∇ϕdx ≥ α

∫
Ω

|v|p−2vϕ dx.

In a similar way, by the second inequality in (c) of Lemma 5.2 we get∫
Ω

∇ξL∞(x,∇v) · ∇ϕdx ≤ α

∫
Ω

|v|p−2vϕ dx.

Therefore, ∫
Ω

∇ξL∞(x,∇v) · ∇ϕdx = α

∫
Ω

|v|p−2vϕ dx ,

which, in view of [94, Remark 1, pp. 161] is impossible if α differs from λ1. �

5.4. Min-Max estimates. Let us introduce the “asymptotic functional” f∞ :
W 1,p

0 (Ω) → R by setting

f∞(u) =
∫

Ω

L∞(x,∇u) dx− β

p

∫
Ω

(u+)p dx− α

p

∫
Ω

(u−)p dx−
∫

Ω

φp−1
1 u dx.

Then consider the functional f̃t : W 1,p
0 (Ω) → R given by

f̃t(u) =
∫

Ω

L (x, tu,∇u) dx− β

p

∫
Ω

(u+)p dx− α

p

∫
Ω

(u−)p dx

−
∫

Ω

G0(x, tu)
tp

−
∫

Ω

φp−1
1 u dx− 〈ω0, u〉

tp−1
.

Theorem 5.7. The following facts hold:
(a) Assume that (th) ⊂]0,+∞[ with th → +∞ and uh → u in W 1,p

0 (Ω). Then

lim
h
f̃th(uh) = f∞(u).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 93

(b) Assume that (th) ⊂]0,+∞[ with th → +∞ and uh ⇀ u in W 1,p
0 (Ω). Then

f∞(u) ≤ lim inf
h

f̃th(uh).

(c) Assume that (th) ⊂]0,+∞[ with th → +∞, uh ⇀ u in W 1,p
0 (Ω) and

lim sup
h

f̃th(uh) ≤ f∞(u).

Then (uh) strongly converges to u in W 1,p
0 (Ω).

Proof. (a) It is easy to prove. (b) Since uh → u in Lp(Ω), it is sufficient to prove
that ∫

Ω

L∞(x,∇u) dx ≤ lim inf
h

∫
Ω

L (x, thuh,∇uh) dx.

Let us define the Carathéodory function L̃ : Ω× R× Rn → R by setting

L̃ (x, s, ξ) :=

{
L (x, tan(s), ξ) if |s| < π

2

L∞(x, ξ) if |s| ≥ π
2 .

Note that L̃ ≥ 0 and L̃ (x, s, ·) is convex. Up to a subsequence we have

thuh → z a.e. in Ω\{u = 0}, ∇uh ⇀ ∇u in Lp(Ω\{u = 0}),
and

arctan(thuh) → arctan(z) in Lp(Ω\{u = 0}).
Therefore, by [85, Theorem 1] we deduce that∫

Ω\{u=0}
L̃ (x, arctan(z),∇u) dx ≤ lim inf

h

∫
Ω\{u=0}

L̃ (x, arctan(thuh),∇uh) dx ,

that implies ∫
Ω

L∞(x,∇u) dx =
∫

Ω\{u=0}
L∞(x,∇u) dx

≤ lim inf
h

∫
Ω\{u=0}

L (x, thuh,∇uh) dx

= lim inf
h

∫
Ω

L (x, thuh,∇uh) dx.

Let us now prove (c). As above, we obtain

lim inf
h

∫
Ω

L

(
x, thuh,

1
2
∇uh +

1
2
∇u

)
dx ≥

∫
Ω

L∞(x,∇u) dx.

Since we have
lim
h

∫
Ω

L (x, thuh,∇u) dx =
∫

Ω

L∞(x,∇u) dx

and
lim sup

h

∫
Ω

L (x, thuh,∇uh) dx ≤
∫

Ω

L∞(x,∇u) dx , (5.25)

we get

lim sup
h

∫
Ω

(L (x, thuh,∇uh)−L (x, thuh,∇u)) dx ≤ 0.

On the other hand, the strict convexity implies that for each h ∈ N
1
2
L (x, thuh,∇uh) +

1
2
L (x, thuh,∇u)−

1
2
L

(
x, thuh,

1
2
∇uh +

1
2
∇u

)
> 0.



94 MARCO SQUASSINA EJDE-2006/MON. 07

Therefore, the previous limits yield∫
Ω

{
1
2
L (x, thuh,∇uh) +

1
2
L (x, thuh,∇u)−

1
2
L

(
x, thuh,

1
2
∇uh +

1
2
∇u

)}
dx

→ 0.

In particular, up to a subsequence, we have
1
2
L (x, thuh,∇uh) +

1
2
L (x, thuh,∇u)−

1
2
L

(
x, thuh,

1
2
∇uh +

1
2
∇u

)
→ 0,

a.e. in Ω. It easily verified that this can be true only if

∇uh(x) → ∇u(x) for a.e. x ∈ Ω.

Then we have
1
ν

L (x, thuh,∇uh(x)) →
1
ν

L∞(x,∇u(x)) for a.e. x ∈ Ω.

Taking into account (5.25), we deduce

1
ν

∫
Ω

L (x, thuh,∇uh) dx→
1
ν

∫
Ω

L∞(x,∇u) dx,

that by ν|∇uh|p ≤ L (x, thuh,∇uh) yields

lim
h

∫
Ω

|∇uh|p dx =
∫

Ω

|∇u|p dx,

namely the convergence of uh to u in W 1,p
0 (Ω). �

Remark 5.8. Assume that β < λ1 < α. Then the following facts hold:
(a) f ′∞(φ1)(φ1) = 0 ;
(b) lims→−∞ f∞(sφ1) = −∞, where we have set φ1 = φ1

(λ1−β)
1

p−1
.

Proof. (a) It is easy to prove. (b) A direct computation yields that for s < 0

f∞(sφ1) =
λ1 − α

p
|s|p − s.

Since α > λ1, assertion (b) follows. �

Lemma 5.9. For every M > 0 there exists % > 0 such that for each w ∈ W 1,p
0 (Ω)

with ‖w − φ1‖1,p ≤ % we have∫
Ω

L∞(x,−∇w−) dx ≥M

∫
Ω

(w−)p dx.

For the proof of the above lemma, we argue as in [31, Lemma 4.1].

Lemma 5.10. There exists r > 0 such that
(a) for each w ∈W 1,p

0 (Ω), ‖w − φ1‖1,p ≤ r ⇒ f∞(w) ≥ f∞(φ1) ;
(b) for each w ∈W 1,p

0 (Ω), ‖w − φ1‖1,p = r ⇒ f∞(w) > f∞(φ1).

Proof. Let us fix a u ∈ W 1,p
0 (Ω) and define ηu :]0,+∞[→ R by setting ηu(t) =

f∞(tu). It is easy to verify that ηu assumes the minimum value:

M (u) = −
(
1− 1

p

)(1
p

) 1
p−1



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 95

×

[ ∫
Ω
φp−1

1 u dx
] p

p−1

[ ∫
Ω

L∞(x,∇u) dx− β
p

∫
Ω
(u+)p dx− α

p

∫
Ω
(u−)p dx

] 1
p−1

.

Moreover, a direct computation yields for each u 6= φ1

f∞(φ1) < M (u) (5.26)

if and only if

p

∫
Ω

L∞(x,∇u) dx > β

∫
Ω

(u+)p dx+ α

∫
Ω

(u−)p dx+ (λ1 − β)
[ ∫

Ω

φp−1
1 u dx

]p
.

(5.27)
If we now set W =

{
u ∈W 1,p

0 (Ω) :
∫
Ω
φp−1

1 u dx = 0
}
, we obtain

W 1,p
0 (Ω) = span(φ1)⊕W. (5.28)

Let us now prove that (5.27) is fulfilled in a neighborhood of φ1. Since (5.27) is
homogeneous of degree p, we may substitute φ1 with φ1. Let us first consider the
case p ≥ 2 and β > 0. In view of (5.28), by strict convexity, there exists εp > 0
such that for any w ∈W

β

∫
Ω

((φ1 + w)+)p dx+ (λ1 − β)
∫

Ω

φp1 dx

≤ β

∫
Ω

((φ1 + w)+)p dx+ (λ1 − β)
∫

Ω

|φ1 + w|p dx− (λ1 − β)εp
∫

Ω

|w|p dx

≤ β

λ1
p

∫
Ω

L∞(x,∇(φ1 + w)+) dx+
λ1 − β

λ1
p

∫
Ω

L∞(x,∇(φ1 + w)) dx

− (λ1 − β)εp
∫

Ω

|w|p dx.

(5.29)

On the other hand, by Lemma (5.9), for a sufficiently large M we get

α

∫
Ω

((φ1 + w)−)p dx ≤ 1
M

∫
Ω

L∞(x,−∇(φ1 + w)−) dx

≤ β

λ1
p

∫
Ω

L∞(x,−∇(φ1 + w)−) dx ,
(5.30)

for ‖w‖1,p small enough. By combining (5.29) and (5.30) we obtain

β

∫
Ω

((φ1 + w)+)p dx+ α

∫
Ω

((φ1 + w)−)p dx+ (λ1 − β)
∫

Ω

φp1 dx

≤ p

∫
Ω

L∞(x,∇(φ1 + w)) dx− (λ1 − β)εp
∫

Ω

|w|p dx.
(5.31)

Therefore, (5.27) holds in a neighborhood of φ1. In view of (4.4) of [94, Lemma
4.2], the case 1 < p < 2 may be treated in a similar fashion. Let us now note that∫

Ω

|φ1 + w|p dx ≥
∫

Ω

φp1 dx ∀w ∈W.

In the case β ≤ 0, we have

β

∫
Ω

(
(φ1 + w)+

)p
dx+ α

∫
Ω

(
(φ1 + w)−

)p
dx+ (λ1 − β)

∫
Ω

φp1 dx



96 MARCO SQUASSINA EJDE-2006/MON. 07

≤ λ1

2

∫
Ω

|φ1 + w|p dx+ (α− β)
∫

Ω

(
(φ1 + w)−

)p
dx+

(
λ1 −

λ1

2
) ∫

Ω

φp1 dx

so that we reduce to (5.31). �

Proposition 5.11. Let r > 0 as in Lemma 5.10. Then there exist t ∈ R+ and
σ > 0 such that for each t ≥ t and w ∈W 1,p

0 (Ω)

‖w − φ1‖1,p = r ⇒ f̃t(w) ≥ f∞(φ1) + σ.

Proof. By contradiction, let (th) ⊆ R and (wh) ⊆W 1,p
0 (Ω) such that th ≥ h and

‖wh − φ1‖1,p = r, f̃th(wh) < f∞(φ1) +
1
h
. (5.32)

Up to a subsequence we have wh ⇀ w with ‖w − φ1‖1,p ≤ r. Then, by (5.32) and
(a) of the previous Lemma we get

lim sup
h

f̃th(wh) ≤ f∞(φ1) ≤ f∞(w). (5.33)

In view of (c) of Theorem 5.7, wh strongly converges to w and then ‖w−φ1‖1,p = r.
By combining (5.33) with (b) of Lemma 5.10, we get a contradiction. �

Proposition 5.12. Let σ and t be as in the previous proposition. Then there exists
t̃ ≥ t such that for each t ≥ t̃ there exist vt, wt ∈W 1,p

0 (Ω) with

‖vt − φ1‖1,p < r, ft(vt) ≤
σ

2
+ f∞(φ1), (5.34)

‖wt − φ1‖1,p > r, ft(wt) ≤
σ

2
+ f∞(φ1). (5.35)

Moreover, sups∈[0,1] ft(svt + (1− s)wt) < +∞.

Proof. We argue by contradiction. Set t̃ = t + h and suppose that there exists
(th) ⊆ R with th ≥ t̃ such that for every vth and wth in W 1,p

0 (Ω),

‖vth − φ1‖1,p < r, fth(vth) >
σ

2
+ f∞(φ1),

‖wth − φ1‖1,p > r, fth(wth) >
σ

2
+ f∞(φ1).

Take now (zh) going strongly to φ1 in W 1,p
0 (Ω). By (a) of Theorem 5.7 we have

f̃th(zh) → f∞(φ1). On the other hand eventually ‖zh − φ1‖1,p < r and fth(zh) ≤
σ
2 + f∞(φ1), that contradicts our assumptions. Recalling (b) of Remark 5.8, by
arguing as in the previous step, it is easy to prove (5.35). The last statement is
straightforward. �

5.5. Proof of the main result. We now come to the proof of the main result of
the section.

Proof of Theorem 5.1. From Theorem 5.6 we know that ft satisfies the (CPS)c
condition for any c ∈ R. By Proposition 5.11 and Proposition 5.12 we may apply
Theorem 2.9 with u0 = φ1 and obtain existence of at least two weak solutions
u ∈W 1,p

0 (Ω) of problem (5.13) for t > t for a suitable t.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 97

Let us now prove that there exists t such that (5.13) has no solutions for t < t.
If the assertion was false, then we could find a sequence (th) ⊆ R with th → −∞
and a sequence (uh) in W 1,p

0 (Ω) such that for every v ∈ C∞c (Ω),∫
Ω

∇ξL (x, uh,∇uh) · ∇v dx+
∫

Ω

DsL (x, uh,∇uh)v dx

= β

∫
Ω

(u+
h )p−1v dx− α

∫
Ω

(u−h )p−1v dx+
∫

Ω

g0(x, uh)v dx

+ |th|p−2th

∫
Ω

φp−1
1 v dx+ 〈ω0, v〉

Let us first consider the case when, up to a subsequence, th
‖uh‖1,p

→ 0 and set
vh = uh

‖uh‖1,p
. By applying Lemma 5.2 with %h = ‖uh‖1,p, δh = ω0 and

γh(x) =

{
β if uh(x) ≥ 0
α if uh(x) < 0,

µh =
g0(x, uh)
‖uh‖p−1

1,p

+
|th|p−2th

‖uh‖p−1
1,p

φp−1
1 ,

up to a subsequence, (vh) converges strongly to some v in W 1,p
0 (Ω). Then using

the same argument as in the proof of Theorem 5.6 we get a contradiction.
Assume now that there exists M > 0 such that ‖uh‖1,p ≤ −Mth. Then setting

wh = −uht−1
h , wh weakly converges to some w ∈W 1,p

0 (Ω). By applying Lemma 5.2
with %h = −th, δh = ω0 and

γh(x) =

{
β if uh(x) ≥ 0
α if uh(x) < 0,

µh = −g0(x, uh)
|th|p−2th

− φp−1
1 ,

we have that wh strongly converges to w in W 1,p
0 (Ω). The choice of the test function

ϕ = w+ gives, as in the first case, w+ ≡ 0. Arguing as in the end of the proof of
Theorem 5.6 we obtain a contradiction. �

Remark 5.13. Even though we have only considered existence of weak solutions
of (5.13), by [6, Lemma 1.4] the weak solutions u ∈ W 1,p

0 (Ω) of (5.4) belong to
L∞(Ω). Then some nice regularity results can be found in [90].

5.6. Fully nonlinear variational inequalities. Starting from the pioneering pa-
per of Ambrosetti and Prodi [4], jumping problems for semi-linear elliptic equations
of the type

−
n∑

i,j=1

Dj(aij(x)Diu) = g(x, u) in Ω

u = 0 on ∂Ω ,

have been extensively studied; see e.g. [84, 98, 100, 124]. Also the case of semi-
linear variational inequalities with a situation of jumping type has been discussed
in [78, 99]. Very recently, quasi-linear inequalities of the form∫

Ω

{ n∑
i,j=1

aij(x, u)DiuDj(v − u)

+
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju(v − u)
}
dx−

∫
Ω

g(x, u)(v − u) dx



98 MARCO SQUASSINA EJDE-2006/MON. 07

≥ 〈ω, v − u〉 ∀v ∈ K̃ϑ ,

u ∈ Kϑ ,

where Kϑ = {u ∈ H1
0 (Ω) : u ≥ ϑ a.e. in Ω}, K̃ϑ = {v ∈ Kϑ : (v − u) ∈ L∞(Ω)}

and ϑ ∈ H1
0 (Ω), have been considered in [77].

When ϑ ≡ −∞, namely we have no obstacle and the variational inequality
becomes an equation, the problem has been also studied in [31, 34] by A. Canino
and has been extended in [79] to the case of fully nonlinear operators.

The purpose of this section is to study the more general class of nonlinear vari-
ational inequalities of the type∫

Ω

{
∇ξL (x, u,∇u) · ∇(v − u) +DsL (x, u,∇u)(v − u)

}
dx

−
∫
Ω

g(x, u)(v − u) dx

≥ 〈ω, v − u〉 ∀v ∈ K̃ϑ ,

u ∈ Kϑ.

(5.36)

In the main result we shall prove the existence of at least two solutions of (5.36).
The framework is the same of [79], but technical difficulties arise, mainly in the
verification of the Palais-Smale condition. This is due to the fact that such condition
is proved in [79] using in a crucial way test functions of exponential type. Such
test functions are not admissible for the variational inequality, so that a certain
number of modifications is required in particular in the proofs of Theorem 5.18 and
Theorem 5.21.

5.7. The main result. In the following, Ω will denote a bounded domain of Rn,
1 < p < n, ϑ ∈W 1,p

0 (Ω) with ϑ− ∈ L∞(Ω), ω ∈W−1,p′(Ω) and

L : Ω× R× Rn → R
is measurable in x for all (s, ξ) ∈ R × Rn and of class C1 in (s, ξ) a.e. in Ω. We
shall assume that L (x, s, ·) is strictly convex and for each t ∈ R

L (x, s, tξ) = |t|pL (x, s, ξ) (5.37)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn. Furthermore, we assume that:
• there exist ν > 0 and b1 ∈ R such that

ν|ξ|p ≤ L (x, s, ξ) ≤ b1|ξ|p , (5.38)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn ;
• there exist b2, b3 ∈ R such that

|DsL (x, s, ξ)| ≤ b2|ξ|p, (5.39)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn and

|∇ξL (x, s, ξ)| ≤ b3|ξ|p−1, (5.40)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn ;
• there exist R > 0 and a bounded Lipschitzian function ψ : [R,+∞[→

[0,+∞[ such that

s ≥ R⇒ DsL (x, s, ξ) ≥ 0 , (5.41)

s ≥ R⇒ DsL (x, s, ξ) ≤ ψ′(s)∇ξL (x, s, ξ) · ξ , (5.42)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 99

for a.e. x ∈ Ω and for all ξ ∈ Rn. We denote by ψ the limit of ψ(s) as
s→ +∞.

• g(x, s) is a Carathéodory function and G(x, s) =
∫ s
0
g(x, τ) dτ . We assume

that there exist a ∈ L
np

n(p−1)+p (Ω) and b ∈ L
n
p (Ω) such that

|g(x, s)| ≤ a(x) + b(x)|s|p−1 , (5.43)

for a.e. x ∈ Ω and all s ∈ R. Moreover, there exists α ∈ R such that

lim
s→+∞

g(x, s)
sp−1

= α , (5.44)

for a.e. x ∈ Ω.
Set now

lim
s→+∞

L (x, s, ξ) = L∞(x, ξ)

(this limit exists by (5.41)). We also assume that L∞(x, ·) is strictly convex for
a.e. x ∈ Ω. Let us remark that we are not assuming the strict convexity uniformly
in x so that such L∞ is pretty general. Moreover, assume that

sh → +∞, ξh → ξ ⇒ ∇ξL (x, sh, ξh) → ∇ξL∞(x, ξ) , (5.45)

for a.e. x ∈ Ω. Let now

λ1 = min
{
p

∫
Ω

L∞(x,∇u) dx : u ∈W 1,p
0 (Ω),

∫
Ω

|u|p dx = 1
}
, (5.46)

be the first (nonlinear) eigenvalue of

u 7→ −div (∇ξL∞(x,∇u)) .

Observe that by [6, Lemma 1.4] the first eigenfunction φ1 belongs to L∞(Ω) and
by [142, Theorem 1.1] is strictly positive.

Our purpose is to study (5.36) when ω = −tp−1φp−1
1 , namely the family of

problems∫
Ω

{
∇ξL (x, u,∇u) · ∇(v − u) +DsL (x, u,∇u)(v − u)

}
dx

−
∫
Ω

g(x, u)(v − u) dx+ tp−1

∫
Ω

φp−1
1 (v − u) dx ≥ 0 ∀v ∈ K̃ϑ,

u ∈ Kϑ,

(5.47)

where Kϑ =
{
u ∈W 1,p

0 (Ω) : u ≥ ϑ a.e. in Ω
}

and

K̃ϑ =
{
v ∈ Kϑ : (v − u) ∈ L∞(Ω)

}
.

Under the above assumptions, the following is our main result.

Theorem 5.14. Assume that α > λ1. Then there exists t̄ ∈ R such that for all
t ≥ t̄ the problem (5.47) has at least two solutions.

This result extends [77, Theorem 2.1] dealing with Lagrangians of the type

L (x, s, ξ) =
1
2

n∑
i,j=1

aij(x, s)ξiξj −G(x, s)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.



100 MARCO SQUASSINA EJDE-2006/MON. 07

In this particular case, existence of at least three solutions has been proved in
[77] assuming α > µ2 where µ2 is the second eigenvalue of the operator

u 7→ −
n∑

i,j=1

Dj(AijDiu).

In our general setting, since L∞ is not quadratic with respect to ξ, we only
have the existence of the first eigenvalue λ1 and it is not clear how to define higher
order eigenvalues λ2, λ3, . . . . Therefore in our case the comparison of α with higher
eigenvalues has no obvious formulation.

5.8. The bounded Palais-Smale condition. In this section we shall consider
the more general variational inequalities (5.36). To this aim let us now introduce
the functional f : W 1,p

0 (Ω) → R ∪ {+∞}

f(u) =

{∫
Ω

L (x, u,∇u) dx−
∫
Ω
G(x, u) dx− 〈ω, u〉 u ∈ Kϑ

+∞ u 6∈ Kϑ.

Definition 5.15. Let c ∈ R. A sequence (uh) in Kϑ is said to be a concrete
Palais-Smale sequence at level c, ((CPS)c-sequence, for short) for f , if f(uh) → c

and there exists a sequence (ϕh) in W−1,p′(Ω) such that ϕh → 0 and∫
Ω

∇ξL (x, uh,∇uh) · ∇(v − uh) dx+
∫

Ω

DsL (x, uh,∇uh)(v − uh) dx

−
∫

Ω

g(x, uh)(v − uh) dx− 〈ω, v − uh〉

≥ 〈ϕh, v − uh〉 ∀v ∈ K̃ϑ.

We say that f satisfies the concrete Palais-Smale condition at level c, ((CPS)c, for
short), if every (CPS)c-sequence for f admits a strongly convergent subsequence
in W 1,p

0 (Ω).

Theorem 5.16. Let u in Kϑ be such that |df |(u) < +∞. Then there exists ϕ in
W−1,p′(Ω) such that ‖ϕ‖−1,p′ ≤ |df |(u) and∫

Ω

∇ξL (x, u,∇u) · ∇(v − u) dx+
∫

Ω

DsL (x, u,∇u)(v − u) dx

−
∫

Ω

g(x, u)(v − u) dx− 〈ω, v − u〉

≥ 〈ϕ, v − u〉 ∀v ∈ K̃ϑ.

For the proof of the above theorem, we argue as in [77, Theorem 4.6].

Proposition 5.17. Let c ∈ R and assume that f satisfies the (CPS)c condition.
Then f satisfies the (PS)c condition.

The above result is an easy consequence of Theorem 5.16.
Let us note that by combining (5.38) with the convexity of L (x, s, ·), we get

∇ξL (x, s, ξ) · ξ ≥ ν|ξ|p (5.48)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R×Rn. Moreover, there exists M > 0 such that

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ (5.49)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 101

We point out that assumption (5.41) may be strengthened without loss of gen-
erality. Suppose that ϑ(x) > −R for a.e. x ∈ Ω and define

L̃ (x, s, ξ) =

{
L (x, s, ξ) s > −R
L (x,−R, ξ) s ≤ −R.

Such L̃ satisfy our assumptions. On the other hand, if u satisfies∫
Ω

{
∇ξL̃ (x, u,∇u) · ∇(v − u) +DsL̃ (x, u,∇u)(v − u)

}
dx

−
∫
Ω

g(x, u)(v − u) dx+ tp−1

∫
Ω

φp−1
1 (v − u) dx ≥ 0 ∀v ∈ K̃ϑ,

u ∈ Kϑ,

(5.50)

then u satisfies (5.47). Therefore, up to substituting L with L̃ , we can assume
that L satisfies (5.41) for any s ∈ R with |s| > R. (Actually L̃ is only locally
Lipschitz in s but one might always define L̃ (x, s, ξ) = L (x, σ(s), ξ) for a suitable
smooth function σ).

Now, we want to provide in Theorem 5.19 a very useful criterion for the veri-
fication of (CPS)c condition. Let us first prove a local compactness property for
(CPS)c-sequences.

Theorem 5.18. Let (uh) be a sequence in Kϑ and (ϕh) a sequence in W−1,p′(Ω)
such that (uh) is bounded in W 1,p

0 (Ω), ϕh → ϕ and∫
Ω

∇ξL (x, uh,∇uh) · ∇(v − uh) dx+
∫
Ω

DsL (x, uh,∇uh)(v − uh) dx

≥ 〈ϕh, v − uh〉 ∀v ∈ K̃ϑ.

(5.51)

Then it is possible to extract a subsequence (uhk
) strongly convergent in W 1,p

0 (Ω).

Proof. Up to a subsequence, (uh) converges to some u weakly in W 1,p
0 (Ω), strongly

in Lp(Ω) and a.e. in Ω. Moreover, arguing as in step I of [77, Theorem 4.18] it
follows that

∇uh(x) → ∇u(x) for a.e. x ∈ Ω.
We divide the proof into several steps.
I) Let us prove that

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇(−u−h )e−M(uh−R)− dx

≤
∫
Ω

∇ξL (x, u,∇u) · ∇(−u−)e−M(u−R)− dx

(5.52)

where M > 0 is defined in (5.49) and R > 0 has been introduced in hypothesis
(5.41). Consider the test functions

v = uh + ζe−M(uh+R)+

in (5.51) where ζ ∈W 1,p
0 (Ω) ∩ L∞(Ω) and ζ ≥ 0. Then∫

Ω

∇ξL (x, uh,∇uh) · ∇ζe−M(uh+R)+ dx

+
∫
Ω

[DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+]ζe−M(uh+R)+ dx



102 MARCO SQUASSINA EJDE-2006/MON. 07

≥ 〈ϕh, ζe−M(uh+R)+〉.

From (5.41) and (5.49) we deduce that[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh +R)+

]
ζe−M(uh+R)+ ≤ 0,

so that by the Fatou’s Lemma we get∫
Ω

∇ξL (x, u,∇u) · ∇ζe−M(u+R)+ dx

+
∫
Ω

[DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u+R)+]ζe−M(u+R)+ dx

≥ 〈ϕ, ζe−M(u+R)+〉 ∀ζ ∈W 1,p
0 (Ω) ∩ L∞(Ω), ζ ≥ 0.

(5.53)

Now, let us consider the functions

ηk = ηeM(u+R)+ϑk(u),

where η ∈W 1,p
0 (Ω)∩L∞(Ω) with η ≥ 0 and ϑk ∈ C∞(R) is such that 0 ≤ ϑk(s) ≤ 1,

ϑk = 1 on [−k, k], ϑk = 0 outside [−2k, 2k] and |ϑ′k| ≤ c/k for some c > 0.
Putting them in (5.53), for each k ≥ 1 we obtain∫

Ω

∇ξL (x, u,∇u) · ∇(ηϑk(u)) dx+
∫
Ω

DsL (x, u,∇u)ηϑk(u) dx

≥ 〈ϕ, ηϑk(u)〉 ∀η ∈W 1,p
0 (Ω) ∩ L∞(Ω), η ≥ 0.

Passing to the limit as k → +∞ we obtain∫
Ω

∇ξL (x, u,∇u) · ∇η dx+
∫
Ω

DsL (x, u,∇u)η dx ≥ 〈ϕ, η〉 (5.54)

for all η ∈W 1,p
0 (Ω)∩L∞(Ω), η ≥ 0. Taking η = (ϑ−−u−)e−M(u−R)− ∈W 1,p

0 (Ω)∩
L∞(Ω) in (5.54) we get∫

Ω

∇ξL (x, u,∇u) · ∇(ϑ− − u−)e−M(u−R)− dx

≥ −
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u−R)−

]
× (ϑ− − u−)e−M(u−R)−dx+ 〈ϕ, (ϑ− − u−)e−M(u−R)−〉.

(5.55)

On the other hand, taking

v = uh − (ϑ− − u−h )e−M(uh−R)− ≥ uh − (ϑ− − u−h ) = u+
h − ϑ−

in (5.51) we obtain∫
Ω

∇ξL (x, uh,∇uh) · ∇(ϑ− − u−h )e−M(uh−R)− dx

+
∫

Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh −R)−

]
× (ϑ− − u−h )e−M(uh−R)−dx

≤ 〈ϕh, (ϑ− − u−h )e−M(uh−R)−〉.

(5.56)

From (5.41) and (5.49) we deduce that

DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh −R)− ≥ 0.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 103

From (5.56), using Fatou’s Lemma and (5.55) we obtain

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇(ϑ− − u−h )e−M(uh−R)− dx

≤
∫
Ω

∇ξL (x, u,∇u) · ∇(ϑ− − u−)e−M(u−R)− dx.

(5.57)

Since

lim
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇ϑ−e−M(uh−R)− dx

=
∫
Ω

∇ξL (x, u,∇u) · ∇ϑ−e−M(u−R)− dx,

then from (5.57) we deduce (5.52).
II) Let us now prove that

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇u+
h e

−M(uh−R)−dx

≤
∫

Ω

∇ξL (x, u,∇u) · ∇u+e−M(u−R)−dx.

(5.58)

We consider the test functions

v = uh −
[
(u+
h − ϑ+) ∧ k

]
e−M(uh−R)− ≥ ϑ+ (ϑ− − u−h )

in (5.51). By Fatou’s Lemma, we get∫
Ω

∇ξL (x, uh,∇uh) · ∇(u+
h − ϑ+)e−M(uh−R)− dx

+
∫

Ω

[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh −R)−

]
× (u+

h − ϑ+)e−M(uh−R)− dx

≤ 〈ϕh, (u+
h − ϑ+)e−M(uh−R)−〉

(5.59)

from which we deduce that[
DsL (x, uh,∇uh)−M∇ξL (x, uh,∇uh) · ∇(uh −R)−

]
(u+
h − ϑ+)e−M(uh−R)−

belongs to L1(Ω). Using Fatou’s Lemma in (5.59) we obtain

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇(u+
h − ϑ+)e−M(uh−R)− dx

≤ −
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u−R)−

]
× (u+ − ϑ+)e−M(u−R)−dx+ 〈ϕ, (u+ − ϑ+)e−M(u−R)−〉,

(5.60)

from which we also deduce that[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u−R)−

]
(u+ − ϑ+)e−M(u−R)− (5.61)



104 MARCO SQUASSINA EJDE-2006/MON. 07

belongs to L1(Ω). Now, taking ηk = [(u+ − ϑ+) ∧ k] e−M(u−R)− in (5.54), we have∫
Ω

∇ξL (x, u,∇u) · ∇
[
(u+ − ϑ+) ∧ k

]
e−M(u−R)− dx

+
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u−R)−

]
×

[
(u+ − ϑ+) ∧ k

]
e−M(u−R)−dx

≥
〈
ϕ,

[
(u+ − ϑ+) ∧ k

]
e−M(u−R)−

〉
.

(5.62)

Using (5.61) and passing to the limit as k → +∞ in (5.62), it results∫
Ω

∇ξL (x, u,∇u) · ∇(u+ − ϑ+)e−M(u−R)− dx

+
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇(u−R)−

]
(u+ − ϑ+)e−M(u−R)−dx

≥ 〈ϕ, (u+ − ϑ+)e−M(u−R)−〉.
(5.63)

Combining (5.63) with (5.60) we obtain

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇(u+
h − ϑ+)e−M(uh−R)− dx

≤
∫
Ω

∇ξL (x, u,∇u) · ∇(u+ − ϑ+)e−M(u−R)− dx

(5.64)

Since

lim
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇ϑ+e−M(uh−R)− dx

=
∫
Ω

∇ξL (x, u,∇u) · ∇ϑ+e−M(u−R)− dx

from (5.64) we deduce (5.58).
III) Let us prove that uh → u strongly in W 1,p

0 (Ω). We claim that

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uhe−M(uh−R)− dx

≤
∫
Ω

∇ξL (x, u,∇u) · ∇ue−M(u−R)− dx.

In fact using (5.52) and (5.58) we get

lim sup
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uhe−M(uh−R)− dx

≤ lim sup
h

∫
Ω∩{uh≥0}

∇ξL (x, uh,∇uh) · ∇u+
h e

−M(uh−R)− dx

+ lim sup
h

∫
Ω∩{uh≤0}

∇ξL (x, uh,∇uh) · ∇(−u−h )e−M(uh−R)− dx

≤
∫
Ω

∇ξL (x, u,∇u) · ∇ue−M(u−R)− dx

(5.65)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 105

From (5.65) using Fatou’s Lemma we get

lim
h

∫
Ω

∇ξL (x, uh,∇uh) · ∇uhe−M(uh−R)− dx

=
∫
Ω

∇ξL (x, u,∇u) · ∇ue−M(u−R)− dx.

Therefore, since by (5.48) we have

ν exp{−M(R+ ‖ϑ−‖∞)}|∇uh|p ≤ ∇ξL (x, uh,∇uh) · ∇uhe−M(uh−R)− .

It follows that
lim
h

∫
Ω

|∇uh|p dx =
∫
Ω

|∇u|p dx,

namely the strong convergence of (uh) to u in W 1,p
0 (Ω). �

Theorem 5.19. For every c ∈ R the following assertions are equivalent:
(a) f satisfies the (CPS)c condition ;
(b) every (CPS)c-sequence for f is bounded in W 1,p

0 (Ω).

Proof. Since the map {u 7→ g(x, u)} is completely continuous from W 1,p
0 (Ω) to

L
np′

n+p′ (Ω), the proof goes like [77, Theorem 4.37]. �

5.9. The Palais-Smale condition. Let us now set

g0(x, s) = g(x, s)− α(s+)p−1, G0(x, s) =
∫ s

0

g0(x, t)dx.

Of course, g0 is a Carathéodory function satisfying

lim
s→+∞

g0(x, s)
sp−1

= 0, |g0(x, s)| ≤ a(x) + b(x)|s|p−1 ,

for a.e. x ∈ Ω and all s ∈ R where a ∈ L
np

n(p−1)+p (Ω) and b ∈ L
n
p (Ω). Then (5.47)

is equivalent to finding u ∈ Kϑ such that∫
Ω

∇ξL (x, u,∇u) · ∇(v − u) dx+
∫
Ω

DsL (x, u,∇u)(v − u) dx

− α

∫
Ω

(u+)p−1(v − u) dx−
∫
Ω

g0(x, u)(v − u) dx+ tp−1

∫
Ω

φp−1
1 (v − u) dx ≥ 0

for all v ∈ K̃ϑ. Let us define the functional f : W 1,p
0 (Ω) → R ∪ {+∞} by setting

f(u) =

{∫
Ω

L (x, u,∇u)− α
p

∫
Ω
u+p −

∫
Ω
G0(x, u) + tp−1

∫
Ω
φp−1

1 u if u ∈ Kϑ

+∞ if u 6∈ Kϑ.

In view of Theorem 5.16, any critical point of f is a weak solutions of (Pt). Let us
introduce a new functional ft : W 1,p

0 (Ω) → R ∪ {+∞} by setting for each t > 0

ft(u) =

{∫
Ω

L (x, tu,∇u)− α
p

∫
Ω
u+p − 1

tp

∫
Ω
G0(x, tu) +

∫
Ω
φp−1

1 u if u ∈ Kt

+∞ if u 6∈ Kt.

where we have set

Kt =
{
u ∈W 1,p

0 (Ω) : tu ≥ ϑ a.e. in Ω
}
.

From Theorem 5.16 it follows that if u is a critical point of ft then tu satisfies
(5.47).



106 MARCO SQUASSINA EJDE-2006/MON. 07

Lemma 5.20. Let (uh) a sequence in W 1,p
0 (Ω) and %h ⊆]0,+∞[ with %h → +∞.

Assume that the sequence
(
uh

%h

)
is bounded in W 1,p

0 (Ω). Then

g0(x, uh)
%p−1
h

→ 0 in L
np′

n+p′ (Ω),
G0(x, uh)

%ph
→ 0 in L1(Ω).

To prove the above lemma, we argue as in [31, Lemma 3.3]. In view of (5.48)
and (5.39), we can extend ψ to [−N,+∞[ where N is such that ‖ϑ−‖∞ ≤ N , so
that assumption (5.42) becomes

s ≥ −N ⇒ DsL (x, s, ξ) ≤ ψ′(s)∇ξL (x, s, ξ) · ξ. (5.66)

Theorem 5.21. Let α > λ1, c ∈ R and let (uh) in Kϑ be a (CPS)c-sequence for
f . Then (uh) is bounded in W 1,p

0 (Ω).

Proof. By Definition 5.15, there exists a sequence (ϕh) in W−1,p′(Ω) with ϕh → 0
and∫

Ω

∇ξL (x, uh,∇uh) · ∇(v − uh) dx+
∫
Ω

DsL (x, uh,∇uh)(v − uh) dx

− α

∫
Ω

(u+
h )p−1(v − uh) dx−

∫
Ω

g0(x, uh)(v − uh) dx+ tp−1

∫
Ω

φp−1
1 (v − uh) dx

≥ 〈ϕh, v − uh〉 ∀v ∈ Kϑ : (v − uh) ∈ L∞(Ω).
(5.67)

We set now %h = ‖uh‖1,p, and suppose by contradiction that %h → +∞. If we
set zh = %−1

h uh, up to a subsequence, zh converges to some z weakly in W 1,p
0 (Ω),

strongly in Lp(Ω) and a.e. in Ω. Note that z ≥ 0 a.e. in Ω.
We shall divide the proof into several steps.

I) We firstly prove that∫
Ω

∇ξL∞(x,∇z) · ∇z dx ≥ α

∫
Ω

zp dx. (5.68)

Consider the test functions v = uh + (z ∧ k) exp {−ψ(uh)}, where ψ is the function
defined in (5.42). Putting such v in (5.67) and dividing by %p−1

h , we obtain∫
Ω

∇ξL (x, uh,∇zh) · ∇(z ∧ k) exp {−ψ(uh)} dx

+
1

%p−1
h

∫
Ω

[DsL (x, uh,∇uh)− ψ′(uh)∇ξL (x, uh,∇uh) · ∇uh]

× (z ∧ k) exp {−ψ(uh)}dx

≥ α

∫
Ω

(z+
h )p−1(z ∧ k) exp {−ψ(uh)} dx+

∫
Ω

g0(x, uh)
%p−1
h

(z ∧ k) exp {−ψ(uh)} dx

− tp−1

∫
Ω

φp−1
1

%p−1
h

(z ∧ k) exp {−ψ(uh)} dx+
1

%p−1
h

〈ϕh, (z ∧ k) exp {−ψ(uh)}〉.

Observe now that the first term∫
Ω

∇ξL (x, uh,∇zh) · ∇(z ∧ k) exp {−ψ(uh)} dx

passes to the limit, yielding∫
Ω

∇ξL∞(x,∇z) · ∇(z ∧ k) exp {−ψ} dx.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 107

Indeed, by taking into account assumptions (5.45) and (5.40), we may apply [54,
Theorem 5] and deduce that, up to a subsequence,

a.e. in Ω \ {z = 0} : ∇zh(x) → ∇z(x).
Since of course uh(x) → +∞ a.e. in Ω \ {z = 0}, again recalling (5.45), we have

a.e. in Ω \ {z = 0} : ∇ξL (x, uh(x),∇zh(x)) → ∇ξL∞(x,∇z(x)).

Since by (5.40) the sequence (∇ξL (x, uh(x),∇zh(x))) is bounded in Lp
′
(Ω), the

assertion follows. Note also that the term
1

%p−1
h

〈ϕh, (z ∧ k) exp {−ψ(uh)}〉 ,

goes to 0 even if 1 < p < 2. Indeed, in this case, one could use the Cerami-Palais-
Smale condition, which yields %hϕh → 0 in W−1,p′

0 (Ω).
Now, by (5.66) we have

DsL (x, uh,∇uh)− ψ′(uh)∇ξL (x, uh,∇uh) · ∇uh ≤ 0,

then, passing to the limit as h→ +∞, we get∫
Ω

∇ξL∞(x,∇z) · ∇(z ∧ k) exp {−ψ̄} dx ≥ α

∫
Ω

zp−1(z ∧ k) exp {−ψ̄} dx.

Passing to the limit as k → +∞, we obtain (5.68).
II) Let us prove that zh → z strongly in W 1,p

0 (Ω), so that of course ‖z‖1,p = 1.
Consider the function ζ : [−R,+∞[→ R defined by

ζ(s) =

{
MR if s ≥ R

Ms if |s| < R
(5.69)

where M ∈ R is such that for a.e. x ∈ Ω, each s ∈ R and ξ ∈ Rn

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ.
If we choose the test functions

v = uh −
uh − ϑ

exp(MR)
exp(ζ(uh))

in (5.67), we have∫
Ω

∇ξL (x, uh,∇uh) · ∇(uh − ϑ) exp{ζ(uh)} dx

+
∫

Ω

[DsL (x, uh,∇uh) + ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh] (uh − ϑ) exp{ζ(uh)} dx

≤ α

∫
Ω

(u+
h )p−1(uh − ϑ) exp{ζ(uh)} dx+

∫
Ω

g0(x, uh)(uh − ϑ) exp{ζ(uh)} dx

− tp−1

∫
Ω

φp−1
1 (uh − ϑ) exp{ζ(uh)} dx+ 〈ϕh, (uh − ϑ) exp{ζ(uh)}〉 .

Note that

[DsL (x, uh,∇uh) + ζ ′(uh)∇ξL (x, uh,∇uh) · ∇uh] (uh − ϑ) ≥ 0.

Therefore, after division by %ph we get∫
Ω

∇ξL (x, uh,∇zh) · ∇
(
zh −

ϑ

%h

)
exp{ζ(uh)} dx



108 MARCO SQUASSINA EJDE-2006/MON. 07

≤ α

∫
Ω

(z+
h )p−1

(
zh −

ϑ

%h

)
exp{ζ(uh)} dx

+
1

%p−1
h

∫
Ω

g0(x, uh)
(
zh −

ϑ

%h

)
exp{ζ(uh)} dx

− tp−1

%p−1
h

∫
Ω

φp−1
1

(
zh −

ϑ

%h

)
exp{ζ(uh)} dx+

1
%p−1
h

〈
ϕh,

(
zh −

ϑ

%h

)
exp{ζ(uh)}

〉
,

which yields

lim sup
h

∫
Ω

∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} dx ≤ α exp{MR}
∫

Ω

zp dx. (5.70)

By combining (5.70) with (5.68) we get

lim sup
h

∫
Ω

∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} dx

≤ exp{MR}
∫

Ω

∇ξL∞(x,∇z) · ∇z dx

In particular, by Fatou’s Lemma, it results

exp{MR}
∫

Ω

∇ξL∞(x,∇z) · ∇z dx

≤ lim inf
h

∫
Ω

∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} dx

≤ lim sup
h

∫
Ω

∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} dx

≤ exp{MR}
∫

Ω

∇ξL∞(x,∇z) · ∇z dx,

namely, we get∫
Ω

∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} dx→
∫

Ω

exp{MR}∇ξL∞(x,∇z) · ∇z dx.

Therefore, since

ν exp{−MR}|∇zh|p ≤ ∇ξL (x, uh,∇zh) · ∇zh exp{ζ(uh)} ,

thanks to the generalized Lebesgue’s theorem, we conclude that

lim
h

∫
Ω

|∇zh|p dx =
∫

Ω

|∇z|p dx,

and zh converges to z in W 1,p
0 (Ω).

III) Let us consider the test functions v = uh + ϕ exp {−ψ(uh)} such that ϕ in
W 1,p

0 ∩ L∞(Ω) and ϕ ≥ 0. Taking such v in (5.67) and dividing by %p−1
h we obtain∫

Ω

∇ξL (x, uh,∇zh) · ∇ϕ exp {−ψ(uh)} dx

+
1

%p−1
h

∫
Ω

[DsL (x, uh,∇uh)− ψ′(uh)∇ξL (x, uh,∇uh) · ∇uh]ϕ exp {−ψ(uh)} dx

≥ α

∫
Ω

(z+
h )p−1ϕ exp {−ψ(uh)} dx+

∫
Ω

g0(x, uh)
%p−1
h

ϕ exp {−ψ(uh)} dx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 109

− tp−1

∫
Ω

ϕp−1

%p−1
h

exp {−ψ(uh)} dx+
1

%p−1
h

〈ϕh, ϕ exp {−ψ(uh)}〉.

Note that, since by step II we have zh → z in W 1,p
0 (Ω), the term∫

Ω

∇ξL (x, uh,∇zh) · ∇ϕ exp {−ψ(uh)} dx

passes to the limit, yielding∫
Ω

∇ξL∞(x,∇z) · ∇ϕ exp {−ψ} dx.

By means of (5.66), we have

DsL (x, uh,∇uh)− ψ′(uh)∇ξL (x, uh,∇uh) · ∇uh ≤ 0,

then passing to the limit as h→ +∞, we obtain∫
Ω

∇ξL∞(x,∇z) · ∇ϕ exp {−ψ̄} dx− α

∫
Ω

zp−1ϕ exp {−ψ̄} dx ≥ 0 ,

for each ϕ ∈W 1,p
0 ∩ L∞(Ω) with ϕ ≥ 0 which yields∫

Ω

∇ξL∞(x,∇z) · ∇ϕdx ≥ α

∫
Ω

zp−1ϕdx (5.71)

for each ϕ ∈W 1,p
0 (Ω) with ϕ ≥ 0.

In a similar fashion, considering in (5.67) the admissible test functions

v = uh −
(
ϕ ∧ zh − ϑ/%h

exp(ψ)

)
exp(ψ(uh))

with ϕ ∈ W 1,p
0 ∩ L∞(Ω) and ϕ ≥ 0 and dividing by %p−1

h , recalling that zh → z
strongly, we get∫

Ω

∇ξL∞(x,∇z) · ∇
[
ϕ ∧ z

expψ

]
dx ≤ α

∫
Ω

zp−1
[
ϕ ∧ z

expψ

]
dx ,

for each ϕ ∈ W 1,p
0 ∩ L∞(Ω) with ϕ ≥ 0. Actually this holds for any ϕ ∈ W 1,p

0 (Ω)
with ϕ ≥ 0. By substituting ϕ with tϕ with t > 0 we obtain∫

Ω

∇ξL∞(x,∇z) · ∇
[
ϕ ∧ z

t expψ

]
dx ≤ α

∫
Ω

zp−1
[
ϕ ∧ z

t expψ

]
dx.

Letting t→ +∞, and taking into account (5.71), it results∫
Ω

∇ξL∞(x,∇z) · ∇ϕdx = α

∫
Ω

zp−1ϕdx (5.72)

for each ϕ ∈W 1,p
0 (Ω) with ϕ ≥ 0. Clearly (5.72) holds for any ϕ ∈W 1,p

0 (Ω), so that
z is a positive eigenfunction related to α. This is a contradiction by [94, Remark
1, pp. 161]. �

Theorem 5.22. Let c ∈ R, α > λ1 and t > 0. Then ft satisfies the (PS)c-
condition.

Proof. Since ft(u) = f(tu)
tp , it is sufficient to combine Theorem 5.21, Theorem 5.19

and Proposition 5.17. �



110 MARCO SQUASSINA EJDE-2006/MON. 07

5.10. Min-Max estimates. Let us first introduce the “asymptotic functional”
f∞ : W 1,p

0 (Ω) → R ∪ {+∞} by setting

f∞(u) =

{∫
Ω

L∞(x,∇u) dx− α
p

∫
Ω
up dx+

∫
Ω
φp−1

1 u dx if u ∈ K∞

+∞ if u 6∈ K∞

where

K∞ =
{
u ∈W 1,p

0 (Ω) : u ≥ 0 a.e. in Ω
}
.

Proposition 5.23. There exist r > 0, σ > 0 such that

(a) for every u ∈W 1,p
0 (Ω) with 0 < ‖u‖1,p ≤ r then f∞(u) > 0;

(b) for every u ∈W 1,p
0 (Ω) with ‖u‖1,p = r then f∞(u) ≥ σ > 0.

Proof. Let us consider the weakly closed set

K∗ =
{
u ∈ K∞ :

∫
Ω

L∞(x,∇u) dx− α

p

∫
Ω

up dx ≤ 1
2

∫
Ω

L∞(x,∇u) dx
}
.

In K∞ \K∗ the statements are evident. On the other hand, it is easy to see that

inf
{ ∫

Ω

vφp−1
1 dx : v ∈ K∗, ‖v‖1,p = 1

}
= ε > 0

arguing by contradiction. Therefore for each u ∈ K∗ we have

f∞(u) =
∫

Ω

L∞(x,∇u) dx− α

p

∫
Ω

up dx+
∫
Ω

φp−1
1 u dx ≥ c‖u‖p1,p + ε‖u‖1,p

where c ∈ R is a suitable constant. Thus the statements follow. �

Proposition 5.24. Let r > 0 be as in the Proposition 5.23. Then there exist t̄ > 0,
σ′ > 0 such that for every t ≥ t̄ and for every u ∈ W 1,p

0 (Ω) with ‖u‖1,p = r, then
ft(u) ≥ σ′.

Proof. By contradiction, we can find two sequences (th) ⊂ R and (uh) ⊂ W 1,p
0 (Ω)

such that th ≥ h for each h ∈ N, ‖uh‖1,p = r and fth(uh) < 1
h . Up to a subsequence,

(uh) weakly converges in W 1,p
0 (Ω) to some u ∈ K∞. Using (b) of [79, Theorem 5],

it follows that
f∞(u) ≤ lim inf

h
fth(uh) ≤ 0.

By (a) of Proposition 5.23, we have u = 0. On the other hand, since

lim sup
h

fth(uh) ≤ 0 = f∞(u) ,

using (c) of [79, Theorem 5] we deduce that (uh) strongly converges to u inW 1,p
0 (Ω),

namely ‖u‖1,p = r. This is impossible. �

Proposition 5.25. Let σ′, t̄ as in Proposition 5.24. Then there exists t̃ ≥ t̄ such
that for every t ≥ t̃ there exist vt, wt ∈W 1,p

0 (Ω) such that ‖vt‖1,p < r, ‖wt‖1,p > r,
ft(vt) ≤ σ′

2 and ft(wt) ≤ σ′

2 . Moreover we have

sup {ft((1− s)vt + swt) : 0 ≤ s ≤ 1} < +∞.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 111

Proof. We argue by contradiction. We set t̃ = t̄+ h and suppose that there exists
(th) such that th ≥ h+ t̄ and such that for every vth , wth in W 1,p

0 (Ω) with ‖vth‖1,p <
r, ‖wth‖1,p > r it results fth(vth) > σ′

2 and fth(wth) > σ′

2 . It is easy to prove that
there exists a sequence (uh) in Kth which strongly converges to 0 in W 1,p

0 (Ω) and
therefore ‖uh‖1,p < r and fth(uth) ≤ σ′

2 eventually as h → +∞. This contradicts
our assumptions. In a similar way one can prove the statement for wt, while the
last statement is straightforward. �

5.11. Proof of the main result.

Proof of Theorem 5.14. By combining Theorem 5.22, propositions 5.24 and 5.25 we
can apply Theorem 2.9 and deduce the assertion. �

6. Problems with Loss of Compactness

The material in this section comes from [127, 130, 107], to which refer the reader.
Some parts of these publications have been slightly modified to give this collection
a more uniform appearance.

6.1. Positive entire solutions for fully nonlinear problems. In the last few
years there has been a growing interest in the study of positive solutions to vari-
ational quasi-linear equations in unbounded domains of Rn, since these problems
are involved in various branches of mathematical physics (see [20]). Since 1988,
quasi-linear elliptic equations of the form

−div (ϕ(∇u)) = g(x, u) in Rn, (6.1)

have been extensively treated, among the others, in [14, 45, 69, 93, 143] by means
of a combination of topological and variational techniques. Moreover, existence of
a positive solution u ∈ H1(Rn) for the more general equation

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju+ b(x)u = g(x, u) in Rn,

behaving asymptotically (|x| → +∞) like the problem

−∆u+ λu = uq−1 in Rn ,

for some suitable λ > 0 and q > 2, has been firstly studied in 1996 in [48] via
techniques of non-smooth critical point theory. On the other hand, more recently,
in a bounded domain Ω of Rn some existence results for fully nonlinear problems
of the type

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) in Ω
u = 0 on ∂Ω,

(6.2)

have been established in [6, 112, 132]. The goal of this section is to prove existence
of a nontrivial positive solution in W 1,p(Rn) for the nonlinear elliptic equation

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) + b(x)|u|p−2u = g(x, u) in Rn, (6.3)

behaving asymptotically like the p-Laplacian problem

−div
(
|∇u|p−2∇u

)
+ λ|u|p−2u = uq−1 in Rn,



112 MARCO SQUASSINA EJDE-2006/MON. 07

for some suitable λ > 0 and q > p. In other words, equation (6.3) tends to regularize
as |x| → +∞ together with its associated functional f : W 1,p(Rn) → R

f(u) =
∫

Rn

L (x, u,∇u) dx+
1
p

∫
Rn

b(x)|u|p dx−
∫

Rn

G(x, u) dx. (6.4)

Since in general f is continuous but not even locally Lipschitzian, unless L does
not depend on u or the growth conditions on L are very restrictive, we shall refer
to the non-smooth critical point theory developed in [36, 50, 58, 86, 87] and we
shall follow the approach of [48].

We assume that 1 < p < n, the function L : Rn × R × Rn → R is measurable
in x for all (s, ξ) ∈ R × Rn, of class C1 in (s, ξ) for a.e. x ∈ Rn and L (x, s, ·) is
strictly convex and homogeneous of degree p. Take b ∈ L∞(Rn) with b ≤ b(x) ≤ b
for a.e. x ∈ Rn for some b, b > 0. We shall assume the following:

• There exists ν > 0 such that

ν|ξ|p ≤ L (x, s, ξ) ≤ 1
p
|ξ|p , (6.5)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R× Rn ;
• there exists c1 > 0 such that:

|DsL (x, s, ξ)| ≤ c1|ξ|p , (6.6)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R× Rn.
Moreover, there exist c2 > 0 and a ∈ Lp′(Rn) such that

|∇ξL (x, s, ξ)| ≤ a(x) + c2|s|
p∗
p′ + c2|ξ|p−1, (6.7)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R× Rn ;
• there exists R > 0 such that

s ≥ R⇒ DsL (x, s, ξ)s ≥ 0 , (6.8)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R× Rn.
• uniformly in s ∈ R and ξ, η ∈ Rn with |ξ| ≤ 1 and |η| ≤ 1

lim
|x|→+∞

∇ξL (x, s, ξ) · η = |ξ|p−2ξ · η , (6.9)

lim
|x|→+∞

DsL (x, s, ξ)s = 0 , (6.10)

lim
|x|→+∞

b(x) = λ , (6.11)

for some λ > 0 and with b(x) ≤ λ for a.e. x ∈ Rn.
• G : Rn × R → R is a Carathéodory function, G(x, s) =

∫ s
0
g(x, t) dt and

there exist β > 0 and q > p such that

s > 0 ⇒ 0 < qG(x, s) ≤ g(x, s)s , (6.12)

(q − p)L (x, s, ξ)−DsL (x, s, ξ)s ≥ β|ξ|p, (6.13)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R×Rn. Moreover there exist σ ∈ ]p, p∗[
and c > 0 such that:

|g(x, s)| ≤ d(x) + c|s|σ−1 , (6.14)

for a.e. x ∈ Rn and all s > 0, where d ∈ Lr(Rn) with r ∈
[
np′

n+p′ , p
′
[
.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 113

• Also

lim
|x|→+∞

g(x, s)
sq−1

= 1, (6.15)

uniformly in s > 0 and

lim
|s|→0

G(x, s)
|s|p

= 0, (6.16)

uniformly in x ∈ Rn and g(x, s) ≥ sq−1 for each s > 0.
Under the above assumptions, the following is our main result.

Theorem 6.1. The Euler’s equation of f

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) + b|u|p−2u = g(x, u) in Rn (6.17)

admits at least one nontrivial positive solution u ∈W 1,p(Rn).

This result extends to a more general setting [48, Theorem 2] dealing with the
case

L (x, s, ξ) =
1
2

n∑
i,j=1

aij(x, s)ξiξj ,

and Theorem 2.1 of [45] involving integrands of the type:

L (x, ξ) =
1
p
a(x)|ξ|p ,

where a ∈ L∞(R) and 1 < p < n. Let us remark that we assume (6.8) for large
values of s, while in [48] it was supposed that for a.e. x ∈ Rn and all ξ ∈ Rn

∀s ∈ R :
n∑

i,j=1

sDsaij(x, s)ξiξj ≥ 0.

This assumption has been widely considered in literature, not only in studying
existence but also to ensure local boundedness of weak solutions (see e.g. [6]).

Condition (6.13) has been already used in [6, 112, 132] and seems to be a natural
extension of what happens in the quasi-linear case [36].

We point out that in a bounded domain, conditions (6.12) and (6.13) may be
assumed for large values of s (see e.g. [132]). Finally (6.9), (6.10), (6.11) and (6.15)
fix the asymptotic behavior of (6.3). By (6.9) and (6.10) there exist two maps
ε1 : Rn × R× Rn × Rn → R and ε2 : Rn × R× Rn → R such that

∇ξL (x, s, ξ) · η = |ξ|p−2ξ · η + ε1(x, s, ξ, η)|ξ|p−1|η| (6.18)

DsL (x, s, ξ)s = ε2(x, s, ξ)|ξ|p (6.19)

where ε1(x, s, ξ, η) → 0 and ε2(x, s, ξ) → 0 as |x| → +∞ uniformly in s ∈ R and
ξ, η ∈ Rn.

6.2. The concrete Palais-Smale condition. Let us now set for a.e. x ∈ Rn and
for all (s, ξ) ∈ R× Rn:

L̃ (x, s, ξ) =

{
L (x, s, ξ) if s ≥ 0
L (x, 0, ξ) if s < 0

g̃(x, s) =

{
g(x, s) if s ≥ 0
0 if s < 0.

(6.20)

We define a modified functional f̃ : W 1,p(Rn) → R by setting

f̃(u) =
∫

Rn

L̃ (x, u,∇u) dx+
1
p

∫
Rn

b(x)|u|p dx−
∫

Rn

G̃(x, u) dx. (6.21)



114 MARCO SQUASSINA EJDE-2006/MON. 07

Then the Euler’s equation of f̃ is given by :

−div
(
∇ξL̃ (x, u,∇u)

)
+DsL̃ (x, u,∇u) + b(x)|u|p−2u = g̃(x, u) in Rn. (6.22)

Lemma 6.2. If u ∈W 1,p(Rn) is a solution of (6.22), then u is a positive solution
of (6.17).

Proof. Let Q : R → R the Lipschitz map defined by:

Q(s) =


0 if s ≥ 0
s if − 1 ≤ s ≤ 0
−1 if s ≤ −1.

Testing f̃ ′(u) with Q(u) ∈W 1,p ∩L∞(Rn) and taking into account (6.20) we have:

0 = f̃ ′(u)(Q(u))

=
∫

Rn

∇ξL̃ (x, u,∇u) · ∇Q(u) dx

+
∫

Rn

DsL̃ (x, u,∇u)Q(u) dx+
∫

Rn

b(x)|u|p−2uQ(u) dx−
∫

Rn

g̃(x, u)Q(u) dx

=
∫
{−1<u<0}

∇ξL (x, 0,∇u) · ∇u dx+
∫
{u<0}

DsL̃ (x, u,∇u)Q(u) dx

+
∫

Rn

b(x)|u|p−2uQ(u) dx−
∫
{u<0}

g̃(x, u)Q(u) dx

=
∫
{−1<u<0}

pL (x, 0,∇u) dx+
∫
{u<0}

b(x)|u|p−2uQ(u) dx

≥ b
∫

Rn

|u|p−2uQ(u) dx ≥ 0.

In particular, it results Q(u) = 0, namely u ≥ 0. �

Therefore, without loss of generality we shall suppose that

∀s ≤ 0 : g(x, s) = 0, L (x, s, ξ) = L (x, 0, ξ)

for a.e. x ∈ Rn and all ξ ∈ Rn.
Lemma 6.3. Let c ∈ R. Then each (CPS)c-sequence for f is bounded in W 1,p(Rn).
Proof. If (uh) is a (CPS)c-sequence for f , arguing as in [48, Lemma 2], since

f(uh)−
1
q
f ′(uh)(uh) = c+ o(1)

as h→ +∞, by (6.12) and (6.13) we get:

β

∫
Rn

|∇uh|p dx+
q − p

p
b

∫
Rn

|uh|p dx ≤ C

for some C > 0, hence the assertion. �

Let us note that there exists M > 0 such that:

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ (6.23)

for a.e. x ∈ Rn and for all (s, ξ) ∈ R× Rn.
We now prove a local compactness property for (CPS)c-sequences. In the fol-

lowing, Ω b Rn will always denote an open and bounded subset of Rn.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 115

Theorem 6.4. Let (uh) be a bounded sequence in W 1,p(Rn) and for each v ∈
C∞c (Rn) set

〈wh, v〉 =
∫

Rn

∇ξL (x, uh,∇uh) · ∇v dx+
∫

Rn

DsL (x, uh,∇uh)v dx. (6.24)

If (wh) is strongly convergent to some w in W−1,p′(Ω) for each Ω b Rn, then (uh)
admits a strongly convergent subsequence in W 1,p(Ω) for each Ω b Rn.

Proof. Since (uh) is bounded in W 1,p(Rn), we find a u in W 1,p(Rn) such that, up
to a subsequence, uh ⇀ u in W 1,p(Rn). Moreover, for each Ω b Rn we have:

uh → u in Lp(Ω), uh(x) → u(x) for a.e. x ∈ Rn.

By a natural extension of [22, Theorem 2.1] to unbounded domains, we have
∇uh(x) → ∇u(x) for a.e. x ∈ Rn. Then, following the blueprint of [132, The-
orem 3.2] we obtain for each v ∈ C∞c (Rn)

〈w, v〉 =
∫

Rn

∇ξL (x, u,∇u) · ∇v dx+
∫

Rn

DsL (x, u,∇u)v dx. (6.25)

Choose now Ω b Rn and fix a positive smooth cut-off function η on Rn with η = 1
on Ω. Moreover, let ϑ : R → R be the function defined by

ϑ(s) =


Ms if 0 < s < R

MR if s ≥ R

−Ms if −R < s < 0
MR if s ≤ −R ,

(6.26)

where M is as in (6.23). Since by [132, Proposition 3.1] vh = ηuh exp{ϑ(uh)} are
admissible test functions for (6.24), we get∫

Rn

∇ξL (x, uh,∇uh) · ∇uhη exp{ϑ(uh)} dx− 〈wh, ηuh exp{ϑ(uh)}〉

+
∫

Rn

∇ξL (x, uh,∇uh) · ∇ηuh exp{ϑ(uh)} dx

+
∫

Rn

[DsL (x, uh,∇uh) + ϑ′(uh)∇ξL (x, uh,∇uh) · ∇uh] ηuh exp{ϑ(uh)} dx

= 0.

Let us observe that

∇ξL (x, uh,∇uh) · ∇uh → ∇ξL (x, u,∇u) · ∇u for a.e. x ∈ Rn.

Since for each h ∈ N we have

[−DsL (x, uh,∇uh)− ϑ′(uh)∇ξL (x, uh,∇uh) · ∇uh] ηuh exp{ϑ(uh)} ≤ 0,

Fatou’s Lemma yields

lim sup
h

∫
Rn

[−DsL (x, uh,∇uh)− ϑ′(uh)∇ξL (x, uh,∇uh) · ∇uh]

× ηuh exp{ϑ(uh)} dx

≤
∫

Rn

[−DsL (x, u,∇u)− ϑ′(u)∇ξL (x, u,∇u) · ∇u] ηu exp{ϑ(u)} dx.



116 MARCO SQUASSINA EJDE-2006/MON. 07

Therefore, we conclude that

lim sup
h

∫
Rn

∇ξL (x, uh,∇uh) · ∇uhη exp{ϑ(uh)} dx

≤ lim sup
h

{∫
Rn

[−DsL (x, uh,∇uh)− ϑ′(uh)∇ξL (x, uh,∇uh) · ∇uh]

× ηuh exp{ϑ(uh)} dx+ 〈wh, ηuh exp{ϑ(uh)}〉

−
∫

Rn

∇ξL (x, uh,∇uh) · ∇ηuh exp{ϑ(uh)} dx
}

≤
{∫

Rn

[−DsL (x, u,∇u)− ϑ′(u)∇ξL (x, u,∇u) · ∇u] ηu exp{ϑ(u)} dx

+ 〈w, ηu exp{ϑ(u)}〉 −
∫

Rn

∇ξL (x, u,∇u) · ∇ηu exp{ϑ(u)} dx
}

=
∫

Rn

∇ξL (x, u,∇u) · ∇uη exp{ϑ(u)} dx,

where we used (6.25) with v = ηu exp{ϑ(u)}. In particular, we have∫
Rn

∇ξL (x, u,∇u) · ∇u η exp{ϑ(u)} dx

≤ lim inf
h

∫
Rn

∇ξL (x, uh,∇uh) · ∇uhη exp{ϑ(uh)} dx

≤ lim sup
h

∫
Rn

∇ξL (x, uh,∇uh) · ∇uhη exp{ϑ(uh)} dx

≤
∫

Rn

∇ξL (x, u,∇u) · ∇u η exp{ϑ(u)} dx ,

namely

lim
h

∫
Rn

∇ξL (x, uh,∇uh) · ∇uhη exp{ϑ(uh)} dx

=
∫

Rn

∇ξL (x, u,∇u) · ∇uη exp{ϑ(u)} dx.

Since L (x, s, ·) is p-homogeneous, by (6.5) for each h ∈ N we have

νηp|∇uh|p ≤ η exp{ϑ(uh)}∇ξL (x, uh,∇uh) · ∇uh ,
by the generalized Lebesgue’s theorem we deduce that:

lim
h

∫
Rn

η|∇uh|p dx =
∫

Rn

η|∇u|p dx.

Up to substituting η with ηp, we get:

lim
h

∫
Rn

|η∇uh|p dx =
∫

Rn

|η∇u|p dx ,

which implies that
η∇uh → η∇u in Lp(Rn) ,

namely ∇uh → ∇u in Lp(Ω). �

Let us remark that, in general, since the imbedding W 1,p(Rn) ↪→ Lp(Rn) is not
compact, we cannot have strong convergence of (CPS)c sequences on unbounded
domains of Rn. Nevertheless, we have the following result.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 117

Lemma 6.5. Let (uh) be a (CPS)c-sequence for f . Then there exists u in W 1,p(Rn)
such that, up to a subsequence, the following facts hold:

(a) (uh) converges to u weakly in W 1,p(Rn) ;
(b) (uh) converges to u strongly in W 1,p(Ω) for each Ω b Rn ;
(c) u is a positive weak solution to (6.3).

Proof. Since by Lemma 6.3 the sequence (uh) is bounded in W 1,p(Rn), of course
(a) holds. Now, fixed Ω b Rn, if we set

wh = γh + g(x, uh)− b|uh|p−2uh ∈W−1,p′(Ω), γh → 0 in W−1,p′(Ω),

(b) follows by Theorem 6.4 with w = g(x, u)− b|u|p−2u. Finally by Lemma 6.2, (c)
is a consequence of equation (6.25). �

Let us now prove a technical Lemma that we shall use later.

Lemma 6.6. Let c ∈ R and (uh) be a bounded (CPS)c-sequence for f . Then for
each ε > 0 there exists % > 0 such that∫

{|uh|≤%}
|∇uh|p dx ≤ ε

for each h ∈ N.

Proof. Let ε, % > 0 and define for δ ∈]0, 1[ the function ϑδ : R → R by setting

ϑδ(s) =


s if |s| ≤ %

%+ δ%− δs if % < s < %+ %
δ

−%− δ%− δs if − %− %
δ < s < −%

0 if |s| ≥ %+ %
δ .

(6.27)

Since ϑδ(uh) ∈W 1,p(Rn) ∩ L∞(Rn), we get

〈wh, ϑδ(uh)〉 =
∫

Rn

∇ξL (x, uh,∇uh) · ∇ϑδ(uh) dx

+
∫

Rn

DsL (x, uh,∇uh)ϑδ(uh) dx+
∫

Rn

b|uh|p−2uhϑδ(uh)

−
∫

Rn

g(x, uh)ϑδ(uh) dx.

Then condition (6.8), b(x) > 0 and |ϑδ(uh)| ≤ % yield∫
Rn

∇ξL (x, uh,∇uh) · ∇ϑδ(uh) dx

≤
∫

Rn

g(x, uh)ϑδ(uh) dx+ %‖uh‖p1,p +
1

p′p
p′
p δ

p′
p

‖wh‖p
′

−1,p′ + δ‖uh‖p1,p.

Since (uh) is bounded in W 1,p(Rn), there exists δ > 0 such that δ‖uh‖p1,p ≤ εν/8
and

δ

∫
Rn

∇ξL (x, uh,∇uh) · ∇uh dx ≤
εν

2
, (6.28)

uniformly with h ∈ N so large that 1

p′p
p′
p δ

p′
p

‖wh‖p
′

−1,p′ ≤
εν
8 . Now, since∫

Rn

g(x, uh)ϑδ(uh) dx ≤
∫
{|uh|≤%+ %

δ}
g(x, uh)uh dx



118 MARCO SQUASSINA EJDE-2006/MON. 07

≤ ‖d‖r
( ∫
{|uh|≤%+ %

δ}
|uh|r

′
dx

)1/r′

+ c

∫
{|uh|≤%+ %

δ}
|uh|σ dx,

we can find % > 0 such that∫
Rn

g(x, uh)ϑδ(uh) dx ≤
εν

8

and %‖uh‖p1,p ≤ εν
8 . Therefore we obtain∫
{|uh|≤%+ %

δ}
∇ξL (x, uh,∇uh) · ∇ϑδ(uh) dx ≤

εν

2
,

namely, taking into account (6.28)∫
{|uh|≤%}

∇ξL (x, uh,∇uh) · ∇uh dx ≤ εν.

By (6.5) the proof is complete. �

Let us now introduce the “asymptotic functional” f∞ : W 1,p(Rn) → R by setting

f∞(u) =
1
p

∫
Rn

|∇u|p dx+
λ

p

∫
Rn

|u|p dx− 1
q

∫
Rn

|u+|q dx

and consider the associated p-Laplacian problem

−div
(
|∇u|p−2∇u

)
+ λ|u|p−2u = uq−1 in Rn.

(See [45] for the case p > 2 and [19] for the case p = 2).
We now investigate the behavior of the functional f over its (CPS)c-sequences.

Lemma 6.7. Let (uh) be a (CPS)c-sequence for f and u its weak limit. Then

f(uh) ≈ f(u) + f∞(uh − u), (6.29)

f ′(uh)(uh) ≈ f ′(u)(u) + f ′∞(uh − u)(uh − u) (6.30)
as h→ +∞, where the notation Ah ≈ Bh means Ah −Bh → 0.

Proof. By [37, Lemma 2.2] we have the splitting:∫
Rn

G(x, uh) dx−
∫

Rn

G(x, u) dx− 1
q

∫
Rn

|(uh − u)+|q dx = o(1),

as h→ +∞. Moreover, we easily get:∫
Rn

b|uh|p dx−
∫

Rn

b|u|p dx− λ

∫
Rn

|uh − u|p dx = o(1),

as h→ +∞. Observe now that thanks to (6.18) we have∫
{|x|>%}

∇ξL (x, uh,∇uh) · ∇uh dx−
∫
{|x|>%}

|∇uh|p dx→ 0, as %→ +∞,

uniformly in h ∈ N and∫
{|x|>%}

∇ξL (x, u,∇u) · ∇u dx−
∫
{|x|>%}

|∇u|p dx→ 0, as %→ +∞.

Therefore, taking into account that for each σ > 0 there exists cσ > 0 with

|∇uh|p ≤ cσ|∇u|p + (1 + σ)|∇uh −∇u|p ,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 119

we deduce that for each ε > 0 there exists % > 0 such that for each h ∈ N∫
{|x|>%}

∇ξL (x, uh,∇uh) · ∇uh dx−
∫
{|x|>%}

∇ξL (x, u,∇u) · ∇u dx

−
∫
{|x|>%}

|∇(uh − u)|p dx < c̃ε ,

for some c̃ > 0. On the other hand, since by Lemma 6.5 we have

∇uh → ∇u in Lp(B(0, %),Rn),
we deduce∫

{|x|≤%}
∇ξL (x, uh,∇uh) · ∇uh dx =

∫
{|x|≤%}

∇ξL (x, u,∇u) · ∇u dx+ o(1),

as h→ +∞. Then, for each ε > 0 there exists h ∈ N such that∫
{|x|≤%}

∇ξL (x, uh,∇uh) · ∇uh dx−
∫
{|x|≤%}

∇ξL (x, u,∇u) · ∇u dx

−
∫
{|x|≤%}

|∇(uh − u)|p dx < ĉε ,

for each h ≥ h, for some ĉ > 0. Putting the previous inequalities together, we have∫
Rn

∇ξL (x, uh,∇uh) · ∇uh dx

=
∫

Rn

∇ξL (x, u,∇u) · ∇u dx+
∫

Rn

|∇(uh − u)|p dx+ o(1)

as h→ +∞. Taking into account that L (x, s, ·) is homogeneous of degree p, (6.29)
is proved. To prove (6.30), by the previous step and condition (6.15), it suffices to
show that∫

Rn

DsL (x, uh,∇uh)uh dx =
∫

Rn

DsL (x, u,∇u)u dx+ o(1), (6.31)

as h → +∞. By (6.19), we find b1, b2 > 0 such that for each ε > 0 there exists
% > 0 with∫

{|x|>%}
DsL (x, uh,∇uh)uh dx ≤ b1ε,

∫
{|x|>%}

DsL (x, u,∇u)u dx ≤ b2ε,

uniformly in h ∈ N. On the other hand, combining (b) of Lemma 6.5 with (6.13),
the generalized Lebesgue’s Theorem yields∫

{|x|≤%}
DsL (x, uh,∇uh)uh dx =

∫
{|x|≤%}

DsL (x, u,∇u)u dx+ o(1),

as h→ +∞. Then, (6.30) follows by the arbitrariness of ε. �

Let us recall from [97, Lemma I.1] the following result.

Lemma 6.8. Let 1 < p ≤ ∞ and 1 ≤ q < ∞ with q 6= p∗. Assume that (uh) is a
bounded sequence in Lq(Rn) with (∇uh) bounded in Lp(Rn) and there exists R > 0
such that:

sup
y∈Rn

∫
y+BR

|uh|q dx = o(1) ,

as h→ +∞. Then uh → 0 in Lα(Rn) for each α ∈]q, p∗[.



120 MARCO SQUASSINA EJDE-2006/MON. 07

Let (uh) denote a concrete Palais-Smale sequence for f and let us assume that
its weak limit u is 0. If np′

n+p′ < r < p′, recalling that by (6.31) it results∫
Rn

DsL (x, uh,∇uh)uh dx = o(1),

as h→ +∞, we get

pc = pf(uh)− f ′(uh)(uh) + o(1) ≤
∫

Rn

g(x, uh)uh dx+

+ o(1) ≤ ‖d‖r‖uh‖r′ + c‖uh‖σσ + o(1).

Hence, either ‖uh‖r′ or ‖uh‖σ does not converge strongly to 0. If we now apply
Lemma 6.8 with p = q (note also that p < r′, σ < p∗), taking into account that (uh)
is bounded in W 1,p(Rn) we find C > 0 and a sequence (yh) ⊂ Rn with |yh| → +∞
such that ∫

yh+BR

|uh|p dx ≥ C ,

for some R > 0. In particular, if τhuh(x) = uh(x− yh), we have∫
BR

|τhuh|p dx ≥ C

and there exists u 6≡ 0 such that:

τhuh ⇀ u in W 1,p(Rn). (6.32)

If r = np′

n+p′ , the same can be obtained in a similar fashion since for each ε > 0 there
exist

d1,ε ∈ L`(Rn) ` ∈
] np′

n+ p′
, p′

[
, d2,ε ∈ L

np′
n+p′ (Rn)

such that
d = d1,ε + d2,ε, ‖d2,ε‖ np′

n+p′
≤ ε.

We now show that u is a weak solution of

−div
(
|∇u|p−2∇u

)
+ λ|u|p−2u = uq−1 in Rn. (6.33)

Lemma 6.9. Let (uh) a (CPS)c-sequence for f with uh ⇀ 0. Then u is a weak
solution of (6.33). Moreover u > 0.

Proof. For all ϕ ∈ C∞c (Rn) and h ∈ N we set

∀x ∈ Rn : (τhϕ)(x) := ϕ(x+ yh).

Since (uh) is a (CPS)c-sequence for f , we have that

∀ϕ ∈ C∞c (Rn) : f ′(uh)(τhϕ) = o(1),

namely, as h→ +∞∫
Rn

∇ξL (x, uh,∇uh) · ∇τhϕdx+
∫

Rn

DsL (x, uh,∇uh)τhϕdx

+
∫

Rn

b(x)|uh|p−2uhτ
hϕdx−

∫
Rn

g(x, uh)τhϕdx = o(1).

Of course, as h→ +∞ we have∫
Rn

b(x)|uh|p−2uhτ
hϕdx =

∫
suppϕ

b(x− yh)|τhuh|p−2τhuhϕdx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 121

→ λ

∫
Rn

|u|p−2uϕdx,∫
Rn

g(x, uh)τhϕdx =
∫

suppϕ

g(x− yh, τhuh)ϕdx→
∫

Rn

|u+|q−1ϕdx.

Next, we have ∫
Rn

∇ξL (x, uh,∇uh) · ∇τhϕdx

=
∫

suppϕ

∇ξL (x− yh, τhuh,∇τhuh) · ∇ϕdx

→
∫

Rn

|∇u|p−2∇u · ∇ϕdx.

Now, for each ε > 0, Lemma 6.6 gives a % > 0 such that∫
Rn

DsL (x, uh,∇uh)τhϕdx ≤ c̃ε+
∫
{|uh|>%}

DsL (x, uh,∇uh)τhϕdx.

On the other hand, by (6.10) we have∫
{|uh|>%}

DsL (x, uh,∇uh)τhϕdx

=
∫

suppϕ∩{|τhuh|>%}
DsL (x− yh, τhuh,∇τhuh)ϕdx = o(1),

as h→ +∞. By arbitrariness of ε we conclude the proof. Finally u ≥ 0 follows by
Lemma 6.2 and u > 0 follows by [142, Theorem 1.1]. �

Lemma 6.10. Let (uh) be a (CPS)c-sequence for f with uh ⇀ 0. Then

f∞(u) ≤ lim inf
h

f∞(τhuh).

Proof. Since (uh) weakly goes to 0, Lemma 6.7 gives f ′∞(uh)(uh) → 0 as h→ +∞,
so that

f ′∞(τhuh)(τhuh) → 0 as h→ +∞,

namely ∫
Rn

|∇τhuh|p dx+ λ

∫
Rn

|τhuh|p dx−
∫

Rn

(τhu+
h )q dx→ 0

as h→ +∞. Therefore

f∞(τhuh)−
(1
p
− 1
q

) ∫
Rn

(τhu+
h )q dx→ 0.

Similarly, Lemma 6.9 yields

f∞(u) =
(1
p
− 1
q

) ∫
Rn

|u|q dx,

and the assertion follows by Fatou’s Lemma. �

Lemma 6.11. If (uh) is a (CPS)c-sequence for f with uh ⇀ 0, then f∞(u) ≤ c.

Proof. Since Lemma 6.7 yields

f(uh) ≈ f∞(τhuh), as h→ +∞,

by the previous Lemma we conclude the proof. �



122 MARCO SQUASSINA EJDE-2006/MON. 07

We finally come to the proof of the main result of this section.

Proof of Theorem 6.1. Since G is super-linear at +∞, (6.12), for all u in the space
W 1,p(Rn)\{0},

u ≥ 0 ⇒ lim
t→+∞

f(tu) = −∞.

Let v ∈ C∞c (Rn) positive be such that for all t > 1 : f(tv) < 0, and define the
min-max class

Γ =
{
γ ∈ C([0, 1],W 1,p(Rn)) : γ(0) = 0, γ(1) = v

}
,

and the min-max value
c = inf

γ∈Γ
max
t∈[0,1]

f(γ(t)).

Let us remark that for each u ∈W 1,p(Rn)

f(u) ≥ ν‖∇u‖pp +
b
p
‖u‖pp −

∫
Rn

G(x, u) dx.

Then, since by (6.16) it results

lim
h

∫
Rn G(x,wh)
‖wh‖p1,p

= 0

for each (wh) that goes to 0 in W 1,p(Rn), f has a mountain pass geometry, and by
the deformation Lemma of [36] there exists a (CPS)c-sequence (uh) ⊂ W 1,p(Rn)
for f . By Lemma 6.5 it results that (uh) converges weakly to a positive weak
solution u of (6.3). Therefore, if u 6= 0, we are done. On the other hand, if u = 0
let us consider u. We now prove that u is a weak solution to our problem. Since
we have for each u ∈W 1,p(Rn)\{0}

u ≥ 0 ⇒ lim
t→+∞

f∞(tu) = −∞ ,

we find R > 0 so large that

∀a, b ≥ 0 : a+ b = R⇒ f∞(au+ bv) < 0.

Define the path γ : [0, 1] →W 1,p(Rn) by

γ(t) =


3Rtu if t ∈

[
0, 1

3

]
(3t− 1)Rv + (2− 3t)Ru if t ∈

[
1
3 ,

2
3

]
(3R+ 3t− 3Rt− 2)v if t ∈

[
2
3 , 1

]
.

Of course we have γ ∈ Γ, f∞(γ(t)) < 0 for each t ∈] 13 , 1] and by [45, Lemma 2.4]

max
t∈[0, 13 ]

f∞(γ(t)) = f∞(u).

Hence, by Lemma (6.11) and the assumptions on L and g, we have

c ≤ max
t∈[0,1]

f(γ(t)) ≤ max
t∈[0,1]

f∞(γ(t)) = f∞(u) ≤ c.

Therefore, since γ is an optimal path in Γ, by the non-smooth deformation Lemma
of [36], there exists t ∈]0, 1[ such that γ(t) is a critical point of f at level c. Moreover
γ(t) = u, otherwise

f(γ(t)) ≤ f∞(γ(t)) < f∞(u) = c,

in contradiction with f(γ(t)) = c. Then u is a positive solution to (6.3). �



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 123

Remark 6.12. Let 1 < p < n, q > p and λ > 0. As a by-product of Theorem 6.1,
taking

L (x, s, ξ) =
1
p
|ξ|p +

λ

p
|s|p − 1

q
|s|q,

we deduce that the problem

−div
(
|∇u|p−2∇u

)
+ λ|u|p−2u = |u|q−2u in Rn , (6.34)

has at least one nontrivial positive solution u ∈W 1,p(Rn). (see also [45, 143]).
In some sense, Theorem 6.1 implies that the ε-perturbed problem

−div
(
(1 + ε(x, u,∇u))|∇u|p−2∇u

)
+ λ|u|p−2u = |u|q−2u in Rn , (6.35)

has at least one nontrivial positive solution u ∈W 1,p(Rn).

Remark 6.13. By [6, Lemma 1.4] we have a local boundedness property for solu-
tions of problem (6.3), namely for each Ω b Rn each weak solution u ∈W 1,p(Ω) of
(6.3) belongs to L∞(Ω) provided that in (6.14) is d ∈ Ls(Ω) for a sufficiently large
s. (see [6, 36]).

6.3. Fully nonlinear problems at critical growth. Let Ω ⊂ Rn be a bounded
domain, 1 < p < n and p < q < p∗ = np

n−p . In this section we are concerned with
the existence of two nontrivial solutions in W 1,p

0 (Ω) of the problem (6.36),

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = |u|p
∗−2u+ λ|u|q−2u+ εh in Ω

u = 0 on ∂Ω
(6.36)

with h ∈ Lp′(Ω), h 6≡ 0, provided that ε > 0 is small and λ > 0 is large.
Motivations for investigating problems as (6.36) come from various situations in

geometry and physics which present lack of compactness (see e.g. [28]). A typical
example is Yamabe’s problem, i.e. find u > 0 such that

−4
n− 1
n− 2

∆Mu = R′u(n+2)/(n−2) −R(x)u on M,

for some constant R′, where M is an n-dimensional Riemannian manifold, R(x)
its scalar curvature and −∆M is the Laplace-Beltrami operator on M . Since p∗

is the critical Sobolev exponent for which the embedding W 1,p
0 (Ω) ↪→ Lp

∗
(Ω) fails

to be compact, as known, one encounters serious difficulties in applying variational
methods to (6.36). As known, in general, if h ≡ 0 and λ = 0, to obtain a solution
of

−∆pu = |u|p
∗−2u in Ω

u = 0 on ∂Ω,

one has to consider in detail the geometry of Ω (see e.g. [16]) or has to replace
the critical term up

∗−1 with up
∗−1−ε and then investigate the limits of uε as ε→ 0

(nearly critical growth, see [72] and references therein). Let us now assume that
h ≡ 0 and λ 6= 0. As we showed in Corollary 9.21 by the general Pohozǎev identity
of Pucci and Serrin [116], if

p∗∇xL (x, s, ξ) · x− nDsL (x, s, ξ)s ≥ 0,

a.e. in Ω and for all (s, ξ) ∈ R × Rn, then (6.36) admits no nontrivial smooth
solution for each λ ≤ 0 when the domain Ω is star-shaped and L is sufficiently
smooth. Therefore, in this case we are reduced to consider positive λ. Let us briefly



124 MARCO SQUASSINA EJDE-2006/MON. 07

recall the historical background of existence results for problems at critical growth
with lower-order perturbations. In 1983, in a pioneering paper [28], Brézis and
Nirenberg proved that the problem

−∆u = u(n+2)/(n−2) + λu in Ω
u > 0 in Ω
u = 0 on ∂Ω

has at least one solution u ∈ H1
0 (Ω) provided that

λ ∈

{
(0, λ1) if n ≥ 4,
(λ1/4, λ1) if n = 3 and Ω = B(0, R),

where λ1 is the first eigenvalue of −∆ in Ω. The extension to the p-Laplacian was
achieved by Garcia Azorero and Peral Alonso in [70, 71] (see also [11]). Namely,
they proved the existence of a nontrivial solution of:

−∆pu = |u|p
∗−2u+ λ|u|q−2u in Ω

u = 0 on ∂Ω

provided that

λ ∈


(0, λ1) if 1 < p = q < p∗ and p2 ≤ n ;
(λ0,∞) if 1 < p < q < p∗ and p2 > n ;
(0,∞) if 1 < p < q < p∗ and p2 ≤ n ;
(0,∞) if max{p, p∗ − p

p−1} < q < p∗ ,

where λ1 is the first eigenvalue of −∆p and λ0 is a suitable positive real number.
Finally, for bifurcation and multiplicity results in the semi-linear case (p = 2), we
refer to the paper of Cerami, Fortunato and Struwe [38].

Let us now assume h 6≡ 0. Then, a natural question is whether inhomogeneous
problems like (6.36) have more than one solution. For bounded domains one of the
first answers was given in 1992 by Tarantello in [140], where it is shown that the
problem

−∆u = |u|2
∗−2u+ h(x) in Ω

u = 0 on ∂Ω

admits two distinct solutions u1, u2 ∈ H1
0 (Ω) if ‖h‖2 is small. The existence of two

nontrivial solutions for the p-Laplacian problem

−∆pu = |u|p
∗−2u+ λ|u|q−2u+ h(x) in Ω
u = 0 on ∂Ω

for 1 < p < q < p∗, λ large and ‖h‖p′ small enough, has been proven in 1995 by
Chabrowski in [41]. This achievement has been recently extended by Zhou in [148]
to the equation:

−∆pu+ c|u|p−2u = |u|p
∗−2u+ f(x, u) + h(x)

on the entire Rn, where f(x, u) is a lower-order perturbation of |u|p∗−2u. This
case involves a double loss of compactness, one due to the unboundedness of the



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 125

domain and the other due to the critical Sobolev exponent. Now, more recently,
some results for the more general problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = g(x, u) in Ω
u = 0 on ∂Ω

with g subcritical and super-linear have been considered in [6, 112] and [132]. It is
therefore natural to see what happens when g has a critical growth.

A first answer was given in 1998 by Arioli and Gazzola in [10], where they proved
the existence of a nontrivial solution u ∈ H1

0 (Ω) for a class of quasi-linear equations
of the type

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju = |u|2
∗−2u+ λu, (6.37)

where the coefficients (aij(x, s)) satisfy some suitable assumptions, including a semi-
linear asymptotic behavior as s→ +∞ (see remark 6.15).

Now, in view of the above mentioned results for −∆, −∆p, we expect that
problems (Pε,λ) admits at least two nontrivial solutions for ε small and λ large.
To prove this, we shall argue on the functional fε,λ : W 1,p

0 (Ω) → R given by

fε,λ(u) =
∫

Ω

L (x, u,∇u) dx− 1
p∗

∫
Ω

|u|p
∗
dx− λ

q

∫
Ω

|u|q dx− ε

∫
Ω

hu dx, (6.38)

where W 1,p
0 (Ω) will be endowed with the norm ‖u‖1,p =

(∫
Ω
|∇u|p dx

)1/p.
The first solution is obtained via a local minimization argument while the second

solution will follow by the mountain pass theorem without Palais-Smale condition
in its non-smooth version (see [36]).

In general, under reasonable assumptions on L , fε,λ is continuous but not even
locally Lipschitzian unless L does not depend on u or is subjected to some very
restrictive growth conditions. Then, we shall refer to the non-smooth critical point
theory developed in [36, 50, 58].

We assume that L (x, s, ξ) : Ω× R× Rn → R is measurable in x for all (s, ξ) ∈
R × Rn, of class C1 in s and of class C2 in ξ and that L (x, s, ·) is strictly convex
and p-homogeneous with L (x, s, 0) = 0. Moreover, we shall assume that:

• There exists ν > 0 such that

L (x, s, ξ) ≥ ν

p
|ξ|p

a.e. in Ω and for all (s, ξ) ∈ R× Rn ;
• there exists c1, c2 ∈ R such that

|DsL (x, s, ξ)| ≤ c1|ξ|p

a.e. in Ω and for all (s, ξ) ∈ R× Rn and∣∣∇2
ξξL (x, s, ξ)

∣∣ ≤ c2|ξ|p−2 (6.39)

a.e. in Ω and for all (s, ξ) ∈ R× Rn ;
• there exist R > 0 and γ ∈]0, q − p[ such that

|s| ≥ R⇒ DsL (x, s, ξ)s ≥ 0 (6.40)

a.e. in Ω and for all (s, ξ) ∈ R× Rn and

DsL (x, s, ξ)s ≤ γL (x, s, ξ) (6.41)



126 MARCO SQUASSINA EJDE-2006/MON. 07

a.e. in Ω and for all (s, ξ) ∈ R× Rn.
Assumptions (6.40) and (6.41) have already been considered in literature (see [6,
112, 132]). Under the previous assumptions, the following is our main result:

Theorem 6.14. There exists λ0 > 0 such that for all λ > λ0 there exists ε0 > 0
such that (6.36) has at least two nontrivial solutions in W 1,p

0 (Ω) for each 0 < ε < ε0.

This result extends the achievements of [41, 140] to a more general class of
elliptic boundary value problems. We stress that, unlike in [41], we proved our
result without any use of concentration-compactness techniques. Indeed, to prove
the existence of the first solution as a local minimum of fε,λ, we showed that our
functional is weakly lower semi-continuous on small balls of W 1,p

0 (Ω). From this
point of view, our approach seems to be simpler and more direct. Furthermore,
we gave in Theorem 6.25 a precise range of compactness for fε,λ. This, to our
knowledge, has not been previously stated for fully nonlinear elliptic problems and
not even for the quasi-linear elliptic equation (6.37). In fact, in [10] it was only
found a “nontrivial energy range” for the functional, inside which weak limits of
Palais-Smale sequences are nontrivial and are solutions of (6.37).

Remark 6.15. Note that no asymptotic behavior has been assumed on L (x, s, ξ)
and DsL (x, s, ξ)s when s goes to +∞, while in [10], to prove that problem (6.37)
has a solution, it was assumed that

lim
s→+∞

aij(x, s) = δij , lim
s→+∞

sDsaij(x, s) = 0, (i, j = 1, . . . , n)

uniformly with respect to x ∈ Ω, namely problem (6.37) converges “in some sense”
to the semi-linear equation −∆u = |u|2∗−2u+ λu.

Remark 6.16. We point out that we assumed (6.40) just for |s| ≥ R, while in [10],
for problem (6.37), it was assumed that:

∀s ∈ R :
n∑

i,j=1

sDsaij(x, s)ξiξj ≥ 0

for a.e. x ∈ Ω and each ξ ∈ Rn.

6.4. The first solution. Let us note that by combining L (x, s, 0) = 0 and (6.39),
one finds b1, b2 > 0 such that:

L (x, s, ξ) ≤ b1|ξ|p, (6.42)

for a.e. x ∈ Ω and each (s, ξ) ∈ R× Rn and

|∇ξL (x, s, ξ)| ≤ b2|ξ|p−1 (6.43)

for a.e. x ∈ Ω and each (s, ξ) ∈ R × Rn. We now prove a weakly lower semi-
continuity property for fε,λ.

Theorem 6.17. There exists % > 0 such that the functional fε,λ is weakly lower
semi-continuous on

{
u ∈W 1,p

0 (Ω) : ‖u‖1,p ≤ %
}
, for each λ ∈ R and ε > 0.

Proof. Let (uh) ⊂W 1,p
0 (Ω) and u with uh ⇀ u in W 1,p

0 (Ω) and ‖uh‖1,p ≤ %. Taking
into account that up to a subsequence we have

uh → u in Lp(Ω) , ∇uh ⇀ ∇u in Lp(Ω) , (6.44)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 127

and uh(x) → u(x) for a.e. x ∈ Ω, by the growth condition (6.42), it results:∫
Ω

L (x, uh,∇u) dx =
∫

Ω

L (x, u,∇u) dx+ o(1),

as h→ +∞. Also note that∫
Ω

|uh|q dx =
∫

Ω

|u|q dx+ o(1),∫
Ω

huh dx =
∫

Ω

hu dx+ o(1),

as h→ +∞. In particular, it suffices to show that for % small:

lim inf
h

{∫
Ω

L (x, uh,∇uh) dx−
∫

Ω

L (x, uh,∇u) dx

− 1
p∗

∫
Ω

|uh|p
∗
dx+

1
p∗

∫
Ω

|u|p
∗
dx

}
≥ 0

(6.45)

Let us now consider for each k ≥ 1 the function Tk : R → R given by

Tk(s) =


−k if s ≤ −k
s if − k ≤ s ≤ k

k if s ≥ k

and let Rk : R → R be the map defined by Rk = Id− Tk, namely

Rk(s) =


s+ k if s ≤ −k
0 if − k ≤ s ≤ k

s− k if s ≥ k.

It is easily seen that∫
Ω

L (x, uh,∇uh) dx =
∫

Ω

L (x, uh,∇Tk(uh)) dx+
∫

Ω

L (x, uh,∇Rk(uh)) dx,

(6.46)
for each k ∈ N. Of course, we also have∫

Ω

L (x, uh,∇u) dx =
∫

Ω

L (x, uh,∇Tk(u)) dx+
∫

Ω

L (x, uh,∇Rk(u)) dx, (6.47)

for each k ∈ N. Now, taking into account that∫
Ω

|u|p
∗−1|uh − u| dx = o(1)

as h→ +∞, and that for any k ∈ N∫
Ω

|Tk(uh)− Tk(u)|p
∗
dx = o(1)



128 MARCO SQUASSINA EJDE-2006/MON. 07

as h→ +∞, there exist c1, c2, c3 > 0 such that

1
p∗

∫
Ω

|uh|p
∗
dx− 1

p∗

∫
Ω

|u|p
∗
dx

≤ c1

∫
Ω

(
|uh|p

∗−1 + |u|p
∗−1

)
|uh − u| dx

≤ c2

∫
Ω

|uh − u|p
∗
dx+ o(1)

≤ c3

∫
Ω

|Tk(uh)− Tk(u)|p
∗
dx+ c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx+ o(1)

= c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx+ o(1)

(6.48)

for any k fixed, as h→ +∞. For each h, k ∈ N we have∫
Ω

L (x, uh,∇Rk(uh)) dx ≥
ν

p

∫
Ω

|∇Rk(uh)|p dx.

On the other hand, by the definition of Rk we have∫
Ω

L (x, uh,∇Rk(u)) dx ≤ c1

∫
Ω

|∇Rk(u)|p dx ≤ ≤ ν

p

∫
Ω

|∇Rk(u)|p dx+ o(1) ,

as k → +∞, uniformly in h ∈ N. In particular, since for each k ∈ N it holds

lim inf
h

{∫
Ω

L (x, uh, Tk(∇uh)) dx−
∫

Ω

L (x, uh, Tk(∇u)) dx
}
≥ 0,

by (6.46), (6.47) and (6.48) there exists cp > 0 such that:

lim inf
h

{∫
Ω

L (x, uh,∇uh) dx−
∫

Ω

L (x, uh,∇u) dx

− 1
p∗

∫
Ω

|uh|p
∗
dx+

1
p∗

∫
Ω

|u|p
∗
dx

}
≥ lim inf

h

{∫
Ω

L (x, uh,∇Rk(uh)) dx−
∫

Ω

L (x, uh,∇Rk(u)) dx

− c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx

}
≥ lim inf

h

{ν
p

∫
Ω

|∇Rk(uh)|p dx−
ν

p

∫
Ω

|∇Rk(u)|p dx

− c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx

}
− o(1) ≥

≥ lim inf
h

{
cp

∫
Ω

|∇Rk(uh)−∇Rk(u)|p dx

− c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx

}
− o(1)

(6.49)

as k → +∞. Now, by Sobolev inequality, we find b1, b2 > 0 with

lim inf
h

{
cp

∫
Ω

|∇Rk(uh)−∇Rk(u)|p dx− c3

∫
Ω

|Rk(uh)−Rk(u)|p
∗
dx

}
≥ lim inf

h
‖Rk(uh)−Rk(u)‖pp∗

{
b1 − b2‖Rk(uh)−Rk(u)‖p

∗−p
p∗

}
≥ 0



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 129

provided that ‖uh‖1,p ≤ % with % sufficiently small and independent of ε and λ. In
particular, (6.45) follows by (6.49) by the arbitrariness of k. �

Lemma 6.18. For each λ ∈ R there exist ε > 0 and %, η > 0 such that

∀u ∈W 1,p
0 (Ω) : ‖u‖1,p = %⇒ fε,λ(u) > η.

Proof. Since

fε,λ(u) ≥
ν

p

∫
Ω

|∇u|p dx− 1
p∗

∫
Ω

|u|p
∗
dx− λ

q

∫
Ω

|u|q dx− ε

∫
Ω

hu dx,

arguing as in [41, Lemma 2], one gets

fε,λ(u) ≥ ‖u‖1,p
{
‖u‖p−1

1,p ϕλ(‖u‖1,p)− ε‖h‖p′cLn(Ω)
p∗−p
pp∗

}
(6.50)

where ϕλ : [0,+∞[→ R is given by

ϕλ(τ) =
ν

p
− S−p

∗

p∗
τp

∗−p − λ

q
cqLn(Ω)

p∗−q
p∗ τ q−p

for some c > 0. The assertion now follows. �

Proposition 6.19. For each λ ∈ R there exists ε0 > 0 such that (6.36) admits at
least one nontrivial solution u1 ∈W 1,p

0 (Ω) for each ε < ε0. Moreover fε,λ(u1) < 0.

Proof. Let us choose φ ∈ W 1,p
0 (Ω) in such a way that:

∫
Ω
hφ dx > 0. Therefore,

since for each t > 0 it results

fε,λ(tφ) = tp
∫

Ω

L (x, tφ,∇φ) dx− tp
∗

p∗

∫
Ω

|φ|p
∗
dx− λtq

q

∫
Ω

|φ|q dx− εt

∫
Ω

hφ dx,

there exists tε,λ > 0 such that fε,λ(tφ) < 0 for each t ∈]0, tε,λ[. In particular,

inf
‖u‖1,p≤%

fε,λ(u) < 0 ,

for each % > 0 sufficiently small. Now, by Theorem 6.17 there exist % > 0 and
u1 ∈W 1,p

0 (Ω) with ‖u1‖1,p ≤ % such that:

fε,λ(u1) = min
‖u‖1,p≤%

fε,λ(u) < 0.

Moreover, up to reducing %, it has to be ‖u1‖1,p < % if ε > 0 is small enough,
otherwise by Lemma 6.18 we would get fε,λ(u1) ≥ 0. In particular, u1 is a solution
of (6.36). �

Remark 6.20. Note that by (6.50), one can get a weak solution of (6.36) for each
ε > 0 on domains Ω with Ln(Ω) sufficiently small.

Remark 6.21. Following Lemmas 3 and 4 in [41], one obtains existence of a weak
solution also in the case p ≥ q. On the other hand we remark that if p ≥ q and
λ > 0 one has to require that Ln(Ω) is sufficiently small.



130 MARCO SQUASSINA EJDE-2006/MON. 07

6.5. The concrete Palais-Smale condition. In this section we prove that fε,λ
satisfies the concrete Palais-Smale condition at levels c within a suitable range of
values.

Lemma 6.22. Let c ∈ R. Then each (CPS)c-sequence for fε,λ is bounded.

Proof. Let c ∈ R and let (uh) be a (CPS)c-sequence for fε,λ. Set:

〈wh, ϕ〉 =
∫

Ω

∇ξL (x, uh,∇uh) · ∇ϕdx+
∫

Ω

DsL (x, uh,∇uh)ϕdx

−
∫

Ω

gε,λ(x, uh)ϕdx−
∫

Ω

|uh|p
∗−2uhϕdx

for all ϕ ∈ C∞c (Ω) where ‖wh‖−1,p′ → 0 as h→ +∞ and

gε,λ(x, s) = λ|s|q−2s+ εh(x).

for a.e. x ∈ Ω and all s ∈ R. It is easily verified that for each α ∈ [p, p∗[ there
exists bα ∈ L1(Ω) such that:

gε,λ(x, s)s+ |s|p
∗
≥ α

{λ
q
|s|q +

1
p∗
|s|p

∗
+ εh(x)s

}
− bα(x)

a.e. in Ω and for each s ∈ R. Now, from f ′ε,λ(uh)(uh)

‖uh‖1,p
= o(1) as h → +∞, one

deduces that∫
Ω

pL (x, uh,∇uh) dx+
∫

Ω

DsL (x, uh,∇uh)uh dx

=
∫

Ω

gε,λ(x, uh)uh dx+
∫

Ω

|uh|p
∗
dx+ 〈wh, uh〉

≥ α
{λ
q

∫
Ω

|uh|q dx+
1
p∗

∫
Ω

|uh|p
∗
dx+ ε

∫
Ω

huh dx
}

−
∫

Ω

bα(x) dx+ 〈wh, uh〉

≥ α

∫
Ω

L (x, uh,∇uh) dx− αfε,λ(uh)−
∫

Ω

bα(x) dx+ 〈wh, uh〉.

On the other hand, by (6.41) one obtains

ν

p
(α− γ − p)

∫
Ω

|∇uh|p dx ≤ (α− γ − p)
∫

Ω

L (x, uh,∇uh) dx

≤ αfε,λ(uh) +
∫

Ω

bα(x) dx+ ‖wh‖−1,p′‖uh‖1,p.

Choosing now α > p in such a way that α−γ−p > 0, one obtains the assertion. �

Remark 6.23. By exploiting the proof of Lemma 6.22 one notes that

sup
{∣∣ ∫

Ω

hu dx
∣∣ : u is a critical point of fε,λ at level c ∈ R

}
≤ σ

for some σ > 0 independent on ε > 0 and λ > 0.

Remark 6.24. Let 1 ≤ p < ∞. It is readily seen that the following proposition
holds: assume that uh → u strongly in Lp(Ω) and vh → v weakly in Lp

′
(Ω) and

a.e. in Ω. Then uhvh → uv strongly in L1(Ω).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 131

Let now S denote the best Sobolev constant (cf. [138])

S = inf
{
‖∇u‖pp : u ∈W 1,p

0 (Ω), ‖u‖p∗ = 1
}
.

The next result is the main technical tool of this section.

Theorem 6.25. There exist K > 0 and ε0 > 0 such that fε,λ satisfies (CPS)c
with

0 < c <
p∗ − γ − p

p∗(γ + p)
(νS)n/p −Kε (6.51)

for each ε < ε0 and λ > 0.

Proof. Let (uh) be a concrete Palais-Smale sequence for fε,λ at level c. Since (uh)
is bounded in W 1,p

0 (Ω) by Lemma 6.22, up to a subsequence we have

uh → u in Lp(Ω) , ∇uh ⇀ ∇u in Lp(Ω).

Moreover, as shown in [22], we also have:

for a.e. x ∈ Ω : ∇uh(x) → ∇u(x).

Arguing as in [132, Theorem 3.2] we get

〈wε,λ, u〉+ ‖u‖p
∗

p∗ =
∫

Ω

∇ξL (x, u,∇u) · ∇u dx+
∫

Ω

DsL (x, u,∇u)u dx ,

where wε,λ ∈W−1,p′(Ω) is defined by

〈wε,λ, v〉 = λ

∫
Ω

|u|q−2uv dx+ ε

∫
Ω

hv dx.

This, following again [132, Theorem 3.2], yields the existence of d ∈ R with

lim sup
h

{∫
Ω

∇ξL (x, uh,∇uh) · ∇uh −
∫

Ω

|uh|p
∗
dx

}
≤ d ≤

{ ∫
Ω

∇ξL (x, u,∇u) · ∇u−
∫

Ω

|u|p
∗
dx

}
.

(6.52)

Of course, we have{
∇ξL (x, uh,∇uh)−∇ξL (x, uh,∇(uh − u))

}
⇀ ∇ξL (x, u,∇u)

in Lp
′
(Ω). Let us note that it actually holds the strong limit{

∇ξL (x, uh,∇uh)−∇ξL (x, uh,∇(uh − u))
}
→ ∇ξL (x, u,∇u)

in Lp
′
(Ω), since by (6.39) there exist τ ∈]0, 1[ and c > 0 with

|∇ξL (x, uh,∇uh)−∇ξL (x, uh,∇(uh − u))|
≤

∣∣∇2
ξξL (x, uh,∇uh + (τ − 1)∇u)

∣∣ |∇u|
≤ c|∇uh|p−2|∇u|+ c|∇u|p−1.

Therefore, by Remark 6.24, we have

∇ξL (x, uh,∇uh) · ∇uh
= ∇ξL (x, uh,∇(uh − u)) · ∇uh +∇ξL (x, u,∇u) · ∇uh + o(1)

= ∇ξL (x, uh,∇(uh − u)) · ∇(uh − u) +∇ξL (x, u,∇u) · ∇u+ o(1) in L1(Ω) ,



132 MARCO SQUASSINA EJDE-2006/MON. 07

as h→ +∞, namely

∇ξL (x, uh,∇uh) · ∇uh −∇ξL (x, u,∇u) · ∇u (6.53)

= ∇ξL (x, uh,∇(uh − u)) · ∇(uh − u) + o(1) in L1(Ω) , (6.54)

as h→ +∞. In a similar way, since there exists c̃ > 0 with∣∣∣|uh|p∗ − |uh|p∗−p|uh − u|p
∣∣∣ ≤ c̃

[
|uh|p

∗−p(|uh|p−1 + |u|p−1)
]
|u|,

one obtains {
|uh|p

∗
− |uh|p

∗−p|uh − u|p
}
→ |u|p

∗
in L1(Ω). (6.55)

In particular, by combining (6.52), (6.53) and (6.55), it results:

lim sup
h

∫
Ω

[
∇ξL (x, uh,∇(uh−u)) · ∇(uh−u)− |uh|p

∗−p|uh−u|p
]
dx ≤ 0. (6.56)

On the other hand, by Hölder and Sobolev inequalities, we get∫
Ω

[
∇ξL (x, uh,∇(uh − u)) · ∇(uh − u)− |uh|p

∗−p|uh − u|p
]
dx (6.57)

≥ ν‖∇(uh − u)‖pp −
1
S
‖uh‖p

∗−p
p∗ ‖∇(uh − u)‖pp (6.58)

= {ν − 1
S
‖uh‖p

∗−p
p∗ }‖∇(uh − u)‖pp , (6.59)

which turns out to be coercive if

lim sup
h

‖uh‖p
∗

p∗ < (νS)n/p. (6.60)

Now, from fε,λ(uh) → c we deduce∫
Ω

L (x, uh,∇uh) dx−
1
p∗
‖uh‖p

∗

p∗ =
λ

q
‖u‖qq + ε

∫
Ω

hu dx+ c+ o(1) , (6.61)

as h→ +∞. On the other hand, by using (6.41), from f ′ε,λ(uh)(uh) → 0 we obtain

γ + p

p

∫
Ω

L (x, uh,∇uh) dx−
1
p
‖uh‖p

∗

p∗ ≥
λ

p
‖u‖qq +

ε

p

∫
Ω

hu dx+ o(1) , (6.62)

as h→ +∞. Multiplying (6.61) by γ+p
p , we obtain

γ + p

p

∫
Ω

L (x, uh,∇uh) dx−
γ + p

pp∗
‖uh‖p

∗

p∗ (6.63)

=
γ + p

pq
λ‖u‖qq +

γ + p

p
ε

∫
Ω

hu+
γ + p

p
c+ o(1) , (6.64)

as h→ +∞. Therefore, by combining (6.63) with (6.62), one gets

p∗ − γ − p

pp∗
‖uh‖p

∗

p∗ ≤ −q − γ − p

pq
λ‖u‖qq + c′ε

∫
Ω

hu dx+
γ + p

p
c+ o(1) (6.65)

≤ c′ε

∫
Ω

hu dx+
γ + p

p
c+ o(1) , (6.66)

as h→ +∞. Now, taking into account Remark 6.23, we deduce

‖uh‖p
∗

p∗ ≤
p∗(γ + p)
p∗ − γ − p

c+ K̃ε+ o(1),



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 133

as h→ +∞ for some K̃ > 0. In particular, condition (6.60) is fulfilled if

p∗(γ + p)
p∗ − γ − p

c+ K̃ε < (νS)n/p

which yields range (6.51) for ε small and a suitable K > 0. By combining (6.56)
and (6.57) we conclude that uh goes to u strongly in W 1,p

0 (Ω). �

Remark 6.26. We observe that for the equation

−∆pu = |u|p
∗−2u+ λ|u|q−2u+ εh in Ω,

being γ = 0 and ν = 1, our range of compactness (6.51) reduces to:

0 < c <
Sn/p

n
−Kε.

See also the results of [41].

6.6. The second solution. Let us finally come to the proof of Theorem 6.14.

Proof. Let us choose φ ∈W 1,p
0 ∩ L∞(Ω) such that

‖φ‖p∗ = 1 and
∫

Ω

hφ dx < 0.

It is easily seen that
lim

t→+∞
fε,λ(tφ) = −∞ ,

so that there exists tλ,ε > 0 with

fε,λ(tλ,εφ) = sup
t≥0

fε,λ(tφ) > 0. (6.67)

Taking into account (6.41), the value tλ,ε must satisfy

ε

∫
Ω

hφ = tq−1
λ,ε

{
tp−qλ,ε

[ ∫
Ω

pL (x, tλ,εφ,∇φ) dx

+
∫

Ω

DsL (x, tλ,εφ,∇φ)tλ,εφdx
]
− tp

∗−q
λ,ε − λ

∫
Ω

|φ|q dx
}

≤ tq−1
λ,ε

{
tp−qλ,ε M

∫
Ω

|∇φ|p dx− tp
∗−q
λ,ε − λ

∫
Ω

|φ|q dx
}
,

for some M > 0. Now, being

lim
λ→+∞

{
tp
∗−q
λ,ε + λ

∫
Ω

|φ|q dx
}

= +∞

it has to be tλ,ε → 0 as λ→ +∞. In particular, by (6.67) we obtain

lim
λ→+∞

sup
t≥0

fε,λ(tφ) = 0 ,

so that there exists λ0 > 0 such that:

0 < sup
t≥0

fε,λ(tφ) <
p∗ − γ − p

p∗(γ + p)
(νS)n/p −Kε (6.68)

for each λ ≥ λ0 and ε < ε0. Let w = tφ with t so large that fε,λ(w) < 0 and set

Φ =
{
γ ∈ C([0, 1],W 1,p

0 (Ω)) : γ(0) = 0, γ(1) = w
}

and
βε,λ = inf

γ∈Φ
max
t∈[0,1]

fε,λ(γ(t))



134 MARCO SQUASSINA EJDE-2006/MON. 07

Taking into account Lemma 6.18, by Theorem 2.10 one finds (uh) ⊂W 1,p
0 (Ω) with:

fε,λ(uh) → βε,λ, |dfε,λ|(uh) → 0 ,

0 < η ≤ βε,λ = inf
γ∈Φ

max
t∈[0,1]

fε,λ(γ(t)) ≤ sup
t≥0

fε,λ(tφ). (6.69)

By Theorem 6.25 fε,λ satisfies (CPS)βε,λ
, since by (6.68) and (6.69)

λ ≥ λ0 ⇒ 0 < βε,λ <
p∗ − γ − p

p∗(γ + p)
(νS)n/p −Kε

for each ε < ε0. Therefore there exist a subsequence of (uh) ⊂ W 1,p(Ω) strongly
convergent to some u2 which solves (6.36). Since fε,λ(u1) < 0 and fε,λ(u2) > 0, of
course u1 6≡ u2. �

Remark 6.27. In the case 1 < q ≤ p < p∗, in general, our method is inconclusive
since it may happen that

lim
λ→+∞

sup
t≥0

fε,λ(tφ) 6= 0.

See section 4 of [41] where this is discussed for the p-Laplacian.

6.7. One solution for a more general nonlinearity. Assume that L (x, s, ξ) :
Ω×R×Rn → R is measurable in x for all (s, ξ) ∈ R×Rn, of class C1 in (s, ξ) and
that L (x, s, ·) is strictly convex and p-homogeneous with L (x, s, 0) = 0. Moreover:

• there exist ν > 0 and c1, c2 > 0 such that:

L (x, s, ξ) ≥ ν

p
|ξ|p, |DsL (x, s, ξ)| ≤ c1|ξ|p, (6.70)

a.e. in Ω and for all (s, ξ) ∈ R× Rn and

|∇ξL (x, s, ξ)| ≤ c2|ξ|p−1, (6.71)

a.e. in Ω and for all (s, ξ) ∈ R× Rn ;
• there exist R,R′ > 0 and γ ∈ (0, p∗ − p) such that:

|s| ≥ R⇒ DsL (x, s, ξ)s ≥ 0, (6.72)

|s| ≥ R′ ⇒ DsL (x, s, ξ)s ≤ γL (x, s, ξ), (6.73)
a.e. in Ω and for all (s, ξ) ∈ R× Rn ;

• Let λ1 be the first eigenvalue of −∆p with homogeneous boundary condi-
tions.

Let g : Ω× R → R be a Carathéodory function such that

∀ε > 0 ∃aε ∈ L
np

n(p−1)+p (Ω) : |g(x, s)| ≤ aε(x) + ε|s|p
∗−1, (6.74)

lim sup
s→0

G(x, s)
|s|p

<
νλ1

p
, G(x, s) ≥ 0, (6.75)

uniformly for a.e. x ∈ Ω and each s ∈ R. Moreover, we assume that there exists a
nonempty open set Ω0 ⊂ Ω such that

• if n < p2 (critical dimensions),

lim
s→+∞

G(x, s)

sp(np+p−2n)
/

(p−1)(n−p)
= +∞ (6.76)

uniformly for a.e. x ∈ Ω0.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 135

• if n = p2 : ∃µ > 0, there exist µ, a > 0 such that

∀s ∈ [0, a] : G(x, s) ≥ µ|s|p or ∀s ≥ a : G(x, s) ≥ µ(|s|p − ap), (6.77)

for a.e. x ∈ Ω0.
• if n > p2, there exists µ > 0 and b > a such that

∀s ∈ [a, b] : G(x, s) ≥ µ (6.78)

for a.e. x ∈ Ω0.
Conditions (6.70), (6.71), (6.72) and (6.73) have already been considered in [6, 132],
while assumptions (6.74), (6.75), (6.76), (6.77) and (6.78) can be found in [11]. Note
that g(x, u) is neither assumed to be positive nor homogeneous in u.

Under additional assumptions (6.81) and (6.82), that will be stated in the next
sections, we have the following result.

Theorem 6.28. Cg admits at least one nontrivial solution.

This result extends the achievements of [10, 11] to a more general class of elliptic
boundary value problems. We remark that we assume (6.72) and (6.73) for |s| ≥ R,
while in [10] these assumptions are requested for each s ∈ R.

6.8. Existence of one nontrivial solution. Let us first prove that the concrete
Palais-Smale sequences of

f(u) =
∫

Ω

L (x, u,∇u) dx− 1
p∗

∫
Ω

|u|p
∗
dx−

∫
Ω

G(x, u) dx (6.79)

are bounded. We will make a new choice of test function, which also removes some
of the technicalities involved in [132].

Lemma 6.29. Let c ∈ R. Then each (CPS)c-sequence for f is bounded.

Proof. Let c ∈ R and let (uh) be a (CPS)c-sequence for f . In the usual notations,
one has ‖wh‖−1,p′ → 0 as h → +∞. It is easily verified that for each α ∈ [p, p∗[
there exists bα ∈ L1(Ω) with

g(x, s)s+ |s|p
∗
≥ α

{
G(x, s) +

1
p∗
|s|p

∗}
− bα(x)

a.e. in Ω and for each s ∈ R. Let now M > 0, k ≥ 1 and ϑk : R → R,

ϑk(s) =



s if s ≥ kM
M
M−1s−

M
M−1k if k ≤ s ≤ kM

0 if − k ≤ s ≤ k
M
M−1s+ M

M−1k if − kM ≤ s ≤ −k
s if s ≤ −kM

Since for each k ∈ N we have f ′(uh)(ϑk(uh)) = o(1) as h → +∞, there exists
Ck,M > 0 such that∫

{|uh|≥kM}
pL (x, uh,∇uh) +

M

M − 1

∫
{k≤|uh|≤kM}

pL (x, uh,∇uh)

+
∫
{|uh|≥kM}

DsL (x, uh,∇uh)uh

+
M

M − 1

∫
{k≤|uh|≤kM}

DsL (x, uh,∇uh)(uh ± k)



136 MARCO SQUASSINA EJDE-2006/MON. 07

=
∫
{|uh|≥kM}

g(x, uh)uh dx+
M

M − 1

∫
{k≤|uh|≤kM}

g(x, uh)(uh ± k)

+
∫
{|uh|≥kM}

|uh|p
∗

+
M

M − 1

∫
{k≤|uh|≤kM}

|uh|p
∗−2uh(uh ± k) + 〈wh, ϑk(uh)〉

≥
∫

Ω

g(x, uh)uh − kM

∫
{|uh|≤kM}

|g(x, uh)|

+
M

M − 1

∫
{k≤|uh|≤kM}

g(x, uh)(uh ± k) +
∫

Ω

|uh|p
∗
dx

− kM

∫
{|uh|≤kM}

|uh|p
∗−1 +

M

M − 1

∫
{k≤|uh|≤kM}

|uh|p
∗−2uh(uh ± k) dx

+ 〈wh, ϑk(uh)〉

≥ α

[∫
Ω

G(x, uh) +
1
p∗

∫
Ω

|uh|p
∗
dx

] ∫
Ω

bα(x)

− kM

∫
{|uh|≤kM}

|g(x, uh)| dx+
M

M − 1

∫
{k≤|uh|≤kM}

g(x, uh)(uh ± k)

− kM

∫
{|uh|≤kM}

|uh|p
∗−1 +

M

M − 1

∫
{k≤|uh|≤kM}

|uh|p
∗−2uh(uh ± k)

+ 〈wh, ϑk(uh)〉

≥ α

∫
Ω

L (x, uh,∇uh)− αf(uh)−
∫

Ω

bα(x)− Ck,M + 〈wh, ϑk(uh)〉.

On the other hand, by (6.73) and (6.72) one obtains∫
{|uh|≥k}

DsL (x, uh,∇uh)uh dx ≤ γ

∫
{|uh|≥k}

L (x, uh,∇uh) dx, (6.80)

and

−k
∫
{k≤uh≤kM}

DsL (x, uh,∇uh) dx ≤ 0,

k

∫
{−kM≤uh≤−k}

DsL (x, uh,∇uh) dx ≤ 0,

for some k ≥ 1 so that k ≥ max{R,R′}. Therefore, we find C̃k,M > 0 with

ν

p

(
α− M

M − 1
γ − M

M − 1
p

) ∫
Ω

|∇uh|p dx

≤
(
α− M

M − 1
γ − M

M − 1
p

) ∫
Ω

L (x, uh,∇uh) dx

≤ αf(uh) +
∫

Ω

bα(x) dx+ C̃k,M + ‖wh‖−1,p′‖ϑk(uh)‖1,p.

To conclude, choose α ∈]p, p∗[ and M > 0 so that α− M
M−1γ −

M
M−1p > 0. �

Remark 6.30. It has to be pointed out that with the choice of test function ϑk
there is no need of using [132, Lemma 3.3], which involves lots of very technical
computations.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 137

Lemma 6.31. Let c ∈ R and let (uh) be a (CPS)c-sequence for f such that uh ⇀ 0.
Then for each ε > 0 and % > 0 we have∫

{|uh|≤%}
L (x, uh,∇uh) dx ≤ ε

∫
{|uh|>%}

L (x, uh,∇uh) dx+ o(1),

uniformly as h→ +∞.

Proof. It is a consequence of [132, Lemma 3.3], taking into account that∫
Ω

(g(x, uh) + |uh|p
∗−1)ϑδ(uh) dx→ 0

as h→ +∞ (where ϑδ is the bounded test function defined in the proof). �

Assume now furthermore that (asymptotic behavior):

lim
s→+∞

L (x, s, ξ) =
1
p
|ξ|p, (6.81)

lim
s→+∞

DsL (x, s, ξ)s = 0, (6.82)

uniformly with respect to x ∈ Ω and to ξ ∈ Rn with |ξ| ≤ 1. This means that there
exist ε1 : Ω× R× Rn → R and ε2 : Ω× R× Rn → R such that

L (x, s, ξ) =
1
p
|ξ|p + ε1(x, s, ξ)|ξ|p

DsL (x, s, ξ)s = ε2(x, s, ξ)|ξ|p

where ε1,2(x, s, ξ) → 0 as s → +∞ uniformly in x ∈ Ω and ξ ∈ Rn. Let S denote
the best Sobolev constant

S = inf
{
‖∇u‖pp : u ∈W 1,p

0 (Ω), ‖u‖p∗ = 1
}
.

Lemma 6.32. Let (uh) ⊂ W 1,p
0 (Ω) be a concrete Palais-Smale sequence for f at

level c with
0 < c <

1
n
Sn/p.

Assume that uh ⇀ u. Then u 6≡ 0.

Proof. Assume by contradiction that u ≡ 0. In particular, u→ 0 in Ls(Ω) for each
1 ≤ s < p∗. Therefore, taking into account (6.74) and the p-homogeneity of L
with respect to ξ, from f ′(uh)(uh) → 0 we obtain∫

Ω

pL (x, uh,∇uh) dx+
∫

Ω

DsL (x, uh,∇uh)uh dx−
∫

Ω

|uh|p
∗
dx = o(1) , (6.83)

as h→ +∞. Let us now prove that for each % > 0

lim
h

∣∣∣ ∫
{|uh|≤%}

DsL (x, uh,∇uh)uh dx
∣∣∣ ≤ C ′′

%
, (6.84)

for some C ′′ > 0. Indeed, since uh ⇀ 0, by Lemma 6.31 and (6.70), one has∣∣∣ ∫
{|uh|≤%}

DsL (x, uh,∇uh)uh dx
∣∣∣ ≤ C%

∫
{|uh|≤%}

L (x, uh,∇uh) dx

≤ C%ε

∫
{|uh|>%}

L (x, uh,∇uh) dx+ o(1)



138 MARCO SQUASSINA EJDE-2006/MON. 07

≤ C ′%ε

∫
Ω

|∇uh|p dx+ o(1) ≤ C ′′%ε+ o(1),

for each % > 0 and ε > 0 uniformly as h → +∞. Then (6.84) follows by choosing
ε = 1/%2. In particular, since condition (6.82) yields

lim
%→+∞

∫
{|uh|>%}

DsL (x, uh,∇uh)uh dx = 0, (6.85)

uniformly in h ∈ N, by combining (6.84) with (6.85), one gets

lim
h

∫
Ω

DsL (x, uh,∇uh)uh dx = 0. (6.86)

In a similar way, by (6.81), one shows that, as h→ +∞,∫
Ω

L (x, uh,∇uh) dx =
1
p

∫
Ω

|∇uh|p dx+ o(1). (6.87)

Therefore, by (6.83) one gets

‖uh‖p1,p − ‖uh‖
p∗

p∗ = o(1),

as h→ +∞. In particular, from the definition of S, it holds

‖uh‖p1,p
(
1− S−p

∗/p‖uh‖p
∗−p

1,p

)
≤ o(1) ,

as h→ +∞. Since c > 0 it has to be

‖uh‖p1,p ≥ Sn/p + o(1), ‖uh‖p
∗

p∗ ≥ Sn/p + o(1),

as h→ +∞. Hence, by (6.86) and (6.87) one deduces that

f(uh) =
1
n
‖uh‖p1,p +

1
p∗

(‖uh‖p1,p + ‖uh‖p
∗

p∗) + o(1) ≥ 1
n
Sn/p ,

contradicting the assumption. �

Proof of Theorem 6.28. Let us consider the min-max class

Γ =
{
γ ∈ C([0, 1],W 1,p

0 (Ω)) : γ(0) = 0, γ(1) = w
}

with f(tw) < 0 for t large and

β = inf
γ∈Φ

max
t∈[0,1]

f(γ(t)).

Then, by the mountain pass theorem in its non-smooth version (see [36]), one finds
a Palais-Smale sequence for f at level β. We have to prove that

0 < β <
1
n
Sn/p.

Consider the family of maps on Rn

Tδ,x0(x) =
cnδ

n−p
p(p−1)(

δ
p

p−1 + |x− x0|
p

p−1

)n−p
p

with δ > 0 and x0 ∈ Rn. Tδ,x0 is a solution of −∆pu = up
∗−1 on Rn. Taking a

function φ ∈ C∞c (Ω) with 0 ≤ φ ≤ 1 and φ = 1 in a neighborhood of x0 and setting
vδ = φTδ,x0 , it results

‖vδ‖p1,p = Sn/p + o
(
ε(n−p)/(p−1)

)
, ‖vδ‖p

∗

p∗ = Sn/p + o
(
εn/(p−1)

)
(6.88)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 139

as δ → 0, so that, as δ → 0,

tpδ
p
‖vδ‖p1,p −

tp
∗

δ

p∗
‖vδ‖p

∗

p∗ ≤
1
n
Sn/p + o

(
ε(n−p)/(p−1)

)
. (6.89)

Assume by contradiction that for each δ > 0 there exists tδ > 0 with

f(tδvδ) =
tpδ
p
‖vδ‖p1,p + tpδ

∫
Ω

{
L (x, tδvδ,∇vδ)−

1
p
|∇vδ|p

}
dx (6.90)

−
∫

Ω

G(x, tδvδ) dx−
tp
∗

δ

p∗
‖vδ‖p

∗

p∗ ≥
1
n
Sn/p (6.91)

In particular, there exist M1,M2 > 0 with M1 ≤ tδ ≤ M2. Moreover, as proved in
[11, Lemma 5], there exists τ : [0, 1] → R with τ(ε) → +∞ and∫

Ω

G(x, tδvδ) dx ≥ τ(ε)ε(n−p)/(p−1). (6.92)

as ε→ 0. By (6.72) and (6.81) one also has∫
Ω

{
L (x, tδvδ,∇vδ)−

1
p
|∇vδ|p

}
dx ≤ 0 (6.93)

for each δ > 0. By putting together (6.89), (6.90), (6.92), (6.93), one concludes

f(tδvδ) ≤
1
n
Sn/p + (C − τ(ε))ε(n−p)/(p−1)

which contradict (6.90) for ε sufficiently small. �

6.9. Problems with nearly critical growth. Let Ω be a bounded domain of
Rn, 1 < p < n and p∗ = np

n−p . In 1989 Guedda and Veron [81] proved that the
p-Laplacian problem at critical growth

−∆pu = up
∗−1 in Ω

u > 0 in Ω
u = 0 on ∂Ω,

(6.94)

has no non-trivial solution u ∈W 1,p
0 (Ω) if the domain Ω is star-shaped. As known,

this non-existence result is due to the failure of compactness for the critical Sobolev
embedding W 1,p

0 (Ω) ↪→ Lp
∗
(Ω), which causes a loss of global Palais-Smale condition

for the functional associated with (6.94). On the other hand, if for instance one
considers annular domains

Ωr1,r2 = {x ∈ Rn : 0 < r1 < |x| < r2} ,
then the radial embedding

W 1,p
0,rad(Ωr1,r2) ↪→ Lq(Ωr1,r2)

is compact for each q < +∞ and one can find a non-trivial radial solution of (6.94)
(see [88]). Therefore, we see how the existence of non-trivial solutions of (6.94) is
related to the shape of the domain and not just to the topology. In the case p = 2,
the problem

−∆u = u(n+2)/(n−2) in Ω
u > 0 in Ω
u = 0 on ∂Ω,

(6.95)



140 MARCO SQUASSINA EJDE-2006/MON. 07

has been deeply studied and existence results have been obtained provided that Ω
satisfies suitable assumptions. In a striking paper [16], Bahri and Coron have proved
that if Ω has a non-trivial topology, i.e. if Ω has a non-trivial homology in some
positive dimension, then (6.95) always admits a non-trivial solution. Moreover,
Dancer [56] constructed for each n ≥ 3 a contractible domain Ωn, homeomorphic
to a ball, for which (6.95) has a non-trivial solution. See also [111] and references
therein for more recent existence and multiplicity results.

We remark that, to our knowledge, this type of achievements are not known
when p 6= 2. In our opinion, one of the main difficulties is the fact, that differently
from the case p = 2, it is not proven that all positive solutions of −∆pu = up

∗−1

in Rn are Talenti’s radial functions, which attain the best Sobolev constant (see
Proposition 6.37).

Now, there is a second approach in the study of problem (6.94), which in gen-
eral does not require any geometrical or topological assumption on Ω, namely to
investigate the asymptotic behavior of solutions uε of problems with nearly critical
growth

−∆pu = |u|p
∗−2−εu in Ω

u = 0 on ∂Ω,
(6.96)

as ε goes to 0. If Ω is a ball and p = 2, Atkinson and Peletier [12] showed in 1987
the blow-up of a sequence of radial solutions. The extension to the case p 6= 2
was achieved by Knaap and Peletier [89] in 1989. On a general bounded domain,
instead, the study of limits of solutions of (6.96) was performed by Garcia Azorero
and Peral Alonso [72] around 1992.

Let now ε > 0 and consider the following general class of Euler-Lagrange equa-
tions with nearly critical growth

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = |u|p
∗−2−εu in Ω

u = 0 on ∂Ω,
(6.97)

associated with the functional fε : W 1,p
0 (Ω) → R given by

fε(u) =
∫

Ω

L (x, u,∇u) dx− 1
p∗ − ε

∫
Ω

|u|p
∗−ε dx. (6.98)

As noted in [132], in general these functionals are not even locally Lipschitzian
under natural growth assumptions. Nevertheless, via techniques of non-smooth
critical point theory (see [132] and references therein) it can be shown that (6.97)
admits a non-trivial solution uε ∈W 1,p

0 (Ω).
Let (uε)ε>0 denote a sequence of solutions of (6.97). The main goal of this

section is to prove that if the weak limit of (|∇uε|p)ε>0 has no blow-up points in
Ω, then the limit problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = |u|p
∗−2u in Ω

u = 0 on ∂Ω.
(6.99)

has a non-trivial solution (the weak limit of (uε)ε>0), provided that fε(uε) → c
with

p∗ − p− γ

pp∗
(νS)n/p < c < 2

p∗ − p− γ

pp∗
(νS)n/p, (6.100)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 141

where ν > 0 and γ ∈ (0, p∗ − p) will be defined later. In our framework (6.100)
plays the role of a generalized second critical energy range (if γ = 0 and ν = 1, one
finds the usual range Sn/p

n < c < 2S
n/p

n for problem (6.96)).
The plan is as follows: in Section 6.10 we shall state our main results ; in Sec-

tion 6.11 we shall collect the main tools, namely the lower bounds on the non-
vanishing Dirac masses and on the non-trivial weak limits ; in Section 6.12 we shall
prove our main results ; finally, in Section 6.13 we shall see that at the mountain
pass levels the sequence (uε)ε>0 blows up. Moreover, we shall state a non-existence
results obtained via the Pucci-Serrin variational identity.

In the following, we shall always consider the space W 1,p
0 (Ω) endowed with the

standard norm ‖u‖p1,p =
∫
Ω
|∇u|p dx and we shall denote by ‖ · ‖p the usual norm

of Lp(Ω).

6.10. The main results. Let Ω be any bounded domain of Rn and assume that
L : Ω×R×Rn → R is measurable in x for all (s, ξ) ∈ R×Rn, of class C1 in (s, ξ)
a.e. in Ω, that L (x, s, ·) is strictly convex and p-homogeneous with L (x, s, 0) = 0.
Moreover:

• There exist b0 > 0 and ν > 0 such that
ν

p
|ξ|p ≤ L (x, s, ξ) ≤ b0|s|p + b0|ξ|p (6.101)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn ;
• there exists b1 > 0 such that for each δ > 0 there exists aδ ∈ L1(Ω) with

|DsL (x, s, ξ)| ≤ aδ(x) + δ|s|p
∗

+ b1|ξ|p (6.102)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn, and

|∇ξL (x, s, ξ)| ≤ a1(x) + b1|s|
p∗
p′ + b1|ξ|p−1 (6.103)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn, where a1 ∈ Lp
′
(Ω) ;

• for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn,

DsL (x, s, ξ)s ≥ 0 (6.104)

and there exists γ ∈ (0, p∗ − p) such that:

DsL (x, s, ξ)s ≤ γL (x, s, ξ) (6.105)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn.
The previous assumptions are natural in the quasi-linear setting and were con-

sidered in [132] and in a stronger form in [6].
We stress that although as noted in the introduction fε fails to be differentiable

on W 1,p
0 (Ω), one may compute the derivatives along the L∞-directions; namely

∀u ∈W 1,p
0 (Ω), ∀ϕ ∈W 1,p

0 ∩ L∞(Ω):

f ′ε(u)(ϕ)

=
∫

Ω

∇ξL (x, u,∇u) · ∇ϕdx+
∫

Ω

DsL (x, u,∇u)ϕdx−
∫

Ω

|u|p
∗−2−εuϕdx.

By combining the following proposition with (3.25), one can also compute f ′ε(u)(u).



142 MARCO SQUASSINA EJDE-2006/MON. 07

Proposition 6.33. Let u, v ∈W 1,p
0 (Ω) be such that DsL (x, u,∇u)v ≥ 0 and

〈w,ϕ〉 =
∫

Ω

∇ξL (x, u,∇u) · ∇ϕdx+
∫

Ω

DsL (x, u,∇u)ϕdx. (6.106)

for all ϕ ∈ C∞c (Ω) and with w ∈ W−1,p′(Ω). Then DsL (x, u,∇u)v ∈ L1(Ω) and
one can take ϕ = v in (6.106).

For the proof of the above proposition, see [132, Proposition 3.1].
Under the preceding assumptions, by [132, Theorem 1.1], for each ε > 0 one

deduces that (6.97) admits at least one non-trivial solution uε ∈ W 1,p
0 (Ω) (by

solution we shall always mean weak solution, namely f ′ε(uε) = 0 in the sense of
distributions). We point out that the technical aspects in the verification of the
Palais-Smale condition are, in our opinion, interesting and not trivial. As a starting
point, let us show that (uε) is bounded in W 1,p

0 (Ω).

Lemma 6.34. Let (uε)ε>0 ⊂ W 1,p
0 (Ω) be a sequence of solutions of (6.97) such

that
lim
ε→0

fε(uε) < +∞.

Then (uε)ε>0 is bounded in W 1,p
0 (Ω).

Proof. If uε is a solution of (6.97), we have f ′ε(uε)(ϕ) = 0 for each ϕ ∈ C∞c (Ω). On
the other hand, taking into account (6.104), by Proposition 6.33 one can choose
ϕ = uε. Therefore, in view of (6.105) and the p-homogeneity of L (x, s, ·), one
obtains

lim
ε→0

fε(uε) = lim
ε→0

(
fε(uε)−

1
p∗ − ε

f ′ε(uε)(uε)
)

= lim
ε→0

( ∫
Ω

L (x, uε,∇uε) dx−
p

p∗ − ε

∫
Ω

L (x, uε,∇uε) dx

− 1
p∗ − ε

∫
Ω

DsL (x, uε,∇uε)uε dx
)

≥ lim
ε→0

p∗ − p− ε− γ

p∗ − ε

∫
Ω

L (x, uε,∇uε) dx

≥ p∗ − p− γ

pp∗
ν lim
ε→0

∫
Ω

|∇uε|p dx.

In particular, (uε)ε>0 is bounded in W 1,p
0 (Ω). �

As a consequence, one may apply P.L. Lions’ concentration-compactness prin-
ciple (see [95, 96]) and obtain a subsequence of (uε)ε>0, u ∈ W 1,p

0 (Ω) and two
bounded positive measures µ and σ such that:

uε ⇀ u in W 1,p
0 (Ω), uε → u in Lq(Ω), 1 < q < p∗, (6.107)

|∇uε|p ⇀ µ, |uε|p
∗
⇀ σ (in the sense of measures) , (6.108)

µ ≥ |∇u|p +
∞∑
j=1

µjδxj
, µj ≥ 0 , (6.109)

σ = |u|p
∗

+
∞∑
j=1

σjδxj
, σj ≥ 0 , (6.110)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 143

µj ≥ Sσ
p

p∗

j , (6.111)

where δxj
denotes the Dirac measure at xj ∈ Ω and S denotes the best Sobolev

constant for the embedding W 1,p
0 (Ω) ↪→ Lp

∗
(Ω) (see e.g. [138]).

The following is our main result.

Theorem 6.35. Let (uε)ε>0 be any sequence of solutions of (6.97) with fε(uε) → c
and

p∗ − p− γ

pp∗
(νS)n/p < c < 2

p∗ − p− γ

pp∗
(νS)n/p.

Then µj = 0 for j ≥ 2 and the following alternative holds:
(a) µ1 = 0 and u is a non-trivial solution of (6.99) ;
(b) µ1 6= 0 and u ≡ 0.

This result extends [72, Theorem 9] to fully nonlinear elliptic problems.

Theorem 6.36. Let (uε)ε>0 be any sequence of solutions of (6.97) with

lim
ε→0

fε(uε) =
p∗ − p− γ

pp∗
(νS)n/p.

Then u ≡ 0.

As we shall see in section (6.13), this is also the behavior when one considers
critical levels of mountain-pass type.

6.11. The weak limit. Let us briefly summarize the main properties of the best
Sobolev constant.

Proposition 6.37. Let 1 < p < n and S be the best Sobolev constant, i.e.

S = inf
{∫

Ω

|∇u|p dx : u ∈W 1,p
0 (Ω),

∫
Ω

|u|p
∗
dx = 1

}
. (6.112)

Then, the following facts hold:
(a) S is independent on Ω ⊂ Rn; it depends only on the dimension n ;
(b) the infimum (6.112) is never achieved on bounded domains Ω ⊂ Rn ;
(c) the infimum (6.112) is achieved if Ω = Rn by the family of functions on Rn

Tδ,x0(x) =
(
nδ

(n− p

p− 1
)p−1

)n−p

p2
(δ + |x− x0|

p
p−1 )−

n−p
p (6.113)

with δ > 0 and x0 ∈ Rn. Moreover for each δ > 0 and x0 ∈ Rn, Tδ,x0 is a
solution of the equation −∆pu = up

∗−1 on Rn.

For the proof of the above proposition, see [138]. The next result establishes
uniform lower bounds for the Dirac masses.

Lemma 6.38. If σj 6= 0, then σj ≥ ν
n
p S

n
p and µj ≥ ν

n
p∗ S

n
p .

Proof. Let xj ∈ Ω the point which supports the Dirac measure of coefficient σj 6= 0.
Denoting with B(xj , δ) the open ball of center xj and radius δ > 0, we can consider
a function ψδ ∈ C∞c (Rn) such that 0 ≤ ψδ ≤ 1, |∇ψδ| ≤ 2

δ , ψδ(x) = 1 if x ∈ B(xj , δ)
and ψδ(x) = 0 if x 6∈ B(xj , 2δ).



144 MARCO SQUASSINA EJDE-2006/MON. 07

By Proposition 6.33 and the p-homogeneity of L (x, s, ·), we have

0 = f ′ε(uε)(ψδuε)

=
∫

Ω

uε∇ξL (x, uε,∇uε) · ∇ψδ dx+ p

∫
Ω

ψδL (x, uε,∇uε) dx

+
∫

Ω

ψδDsL (x, uε,∇uε)uε dx−
∫

Ω

|uε|p
∗−εψδ dx

(6.114)

Applying Hölder inequality and (6.103) to the first term of the decomposition and
keeping into account that (uε)ε>0 is bounded in W 1,p

0 (Ω) and uε → u in Lq(Ω) for
every q < p∗, one find c1 > 0 and c2 > 0 with

lim
ε→0

∣∣ ∫
Ω

uε∇ξL (x, uε,∇uε) · ∇ψδ dx
∣∣

≤
( ∫

B(xj ,2δ)

|a1|
p

p−1 dx
) p−1

p
( ∫

B(xj ,2δ)

|u|p
∗
dx

) 1
p∗

( ∫
B(xj ,2δ)

|∇ψδ|n dx
) 1

n

+ b1

( ∫
B(xj ,2δ)

|u|p
∗
dx

)n−1
n

( ∫
B(xj ,2δ)

|∇ψδ|n dx
) 1

n

+ b̃1

( ∫
B(xj ,2δ)

|u|p
∗
dx

) 1
p∗

( ∫
B(xj ,2δ)

|∇ψδ|n dx
) 1

n

≤ c1

( ∫
B(xj ,2δ)

|u|p
∗
dx

) 1
p∗

+ c2

( ∫
B(xj ,2δ)

|u|p
∗
dx

)n−1
n

= βδ

(6.115)
with βδ → 0 as δ → 0. Then, taking into account (6.104) and (6.101) one has

0 ≥ −βδ + lim
ε→0

ν

∫
Ω

|∇uε|pψδ dx− lim
ε→0

Ln(Ω)
ε

p∗

(∫
Ω

|uε|p
∗
ψδ dx

) p∗−ε
p∗

≥ −βδ + ν

∫
Ω

ψδ dµ−
∫

Ω

ψδ dσ.

Letting δ → 0, it results νµj ≤ σj . By means of (6.111) one concludes the proof. �

Next result establishes uniform lower bounds for the non-zero weak limits.

Lemma 6.39. If u 6= 0, then
∫
Ω
|∇u|p dx > ν

n
p∗ Sn/p and

∫
Ω
|u|p∗ dx > νn/pSn/p.

Proof. By Lemma 6.38, we may assume that µ has at most r Dirac masses µ1, . . . , µr
at x1, . . . , xr. Let now 0 < δ < 1

4 mini 6=j |xi − xj | and ψδ ∈ C∞c (Rn) be such that
0 ≤ ψδ ≤ 1, |∇ψδ| ≤ 2

δ , ψδ(x) = 1 if x ∈ B(xj , δ) and ψδ(x) = 0 if x 6∈ B(xj , 2δ).
Taking into account (6.104), for each ε, δ > 0 we have

∫
Ω

DsL (x, uε,∇uε)uε(1− ψδ) dx ≥ 0.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 145

Then, since one can choose (1− ψδ)uε as test, one obtains

0 = f ′ε(uε)((1− ψδ)uε)

=
∫

Ω

pL (x, uε,∇uε)(1− ψδ) dx−
∫

Ω

∇ξL (x, uε,∇uε) · ∇ψδuε dx

+
∫

Ω

DsL (x, uε,∇uε)uε(1− ψδ) dx−
∫

Ω

|uε|p
∗−ε(1− ψδ) dx

≥ ν

∫
Ω

|∇uε|p(1− ψδ) dx−
∫

Ω

∇ξL (x, uε,∇uε) · ∇ψδuε dx

− Ln(Ω)
ε

p∗

(∫
Ω

|uε|p
∗
(1− ψδ) dx

) p∗−ε
p∗

.

(6.116)

On the other hand, arguing as for (6.115), one gets

lim
ε→0

∣∣ ∫
Ω

uε∇ξL (x, uε,∇uε) · ∇ψδ dx
∣∣ ≤ βδ (6.117)

for each δ > 0. Now, it results

lim
ε→0

∫
Ω

|∇uε|p(1− ψδ) dx =
∫

Ω

(1− ψδ) dµ

≥
∫

Ω

|∇u|p(1− ψδ) dx+
r∑
j=1

µj(1− ψδ(xj))

=
∫

Ω

|∇u|p dx+ o(1)

(6.118)

as δ → 0 and

lim
ε→0

∫
Ω

|uε|p
∗
(1− ψδ) dx =

∫
Ω

(1− ψδ) dσ

=
∫

Ω

|u|p
∗
(1− ψδ) dx+

r∑
j=1

σj(1− ψδ(xj))

=
∫

Ω

|u|p
∗
dx+ o(1)

(6.119)

as δ → 0. Therefore, in view of (6.117), (6.118) and (6.119), by letting δ → 0 and
ε→ 0 in (6.116), one concludes that

ν

∫
Ω

|∇u|p dx ≤
∫

Ω

|u|p
∗
dx. (6.120)

As Ω is bounded, by (b) of Proposition 6.37 one has∫
Ω

|∇u|p dx > S
( ∫

Ω

|u|p
∗
dx

)p/p∗
,

which combined with (6.120) yields the assertion. �

In the next result we show that weak limits of (uε)ε>0 are indeed solutions of
(6.99).

Lemma 6.40. Let (uε)ε>0 ⊂W 1,p
0 (Ω) be a sequence of solutions of (6.97) and let

u be its weak limit. Then u is a solution of (6.99).



146 MARCO SQUASSINA EJDE-2006/MON. 07

Proof. For each ε > 0 one has for all ϕ ∈ C∞c (Ω):∫
Ω

∇ξL (x, uε,∇uε) · ∇ϕdx+
∫

Ω

DsL (x, uε,∇uε)ϕdx

=
∫

Ω

|uε|p
∗−2−εuεϕdx.

(6.121)

Since (uε)ε>0 is bounded in W 1,p
0 (Ω), up to a subsequence, as ε→ 0, u satisfies

∇uε ⇀ ∇u in Lp(Ω), uε → u in Lp(Ω), uε(x) → u(x) for a.e. x ∈ Ω.

Moreover, by [22, Theorem 1], up to a subsequence, we have ∇uε(x) → ∇u(x) for
a.e. x ∈ Ω. Therefore, in view of (6.103) one deduces that

∇ξL (x, uε,∇uε) ⇀ ∇ξL (x, u,∇u) in Lp
′
(Ω,Rn). (6.122)

By (6.101) and (6.102) one finds M > 0 such that for each δ > 0

|DsL (x, s, ξ)| ≤M∇ξL (x, s, ξ) · ξ + aδ(x) + δ|s|p
∗

(6.123)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R × Rn. If we test equation (6.121) with the
functions

ϕε = ϕ exp{−Mu+
ε } , ϕ ∈W 1,p

0 ∩ L∞(Ω) , ϕ ≥ 0

for each ε > 0 we obtain∫
Ω

∇ξL (x, uε,∇uε) · ∇ϕ exp{−Mu+
ε } dx−

∫
Ω

|uε|p
∗−2−εuεϕ exp{−Mu+

ε } dx

+
∫

Ω

[
DsL (x, uε,∇uε)−M∇ξL (x, uε,∇uε) · ∇u+

ε

]
ϕ exp{−Mu+

ε } dx = 0.

Since by inequalities (6.104) and (6.123) for each ε > 0 and δ > 0 we have[
DsL (x, uε,∇uε)−M∇ξL (x, uε,∇uε) · ∇u+

ε

]
ϕ exp{−Mu+

ε } ≤ aδ(x) + δ|uε|p
∗
,

arguing as in [132, Theorem 3.4], one obtains

lim sup
ε→0

∫
Ω

[
DsL (x, uε,∇uε)−M∇ξL (x, uε,∇uε) · ∇u+

ε

]
ϕ exp{−Mu+

ε } dx

≤
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇u+

]
ϕ exp{−Mu+} dx.

Therefore, taking into account (6.122) and since as ε→ 0,∫
Ω

|uε|p
∗−2−εuεϕdx→

∫
Ω

|u|p
∗−2uϕdx

for each ϕ ∈W 1,p
0 ∩ L∞(Ω) positive, one may conclude that∫

Ω

∇ξL (x, u,∇u) · ∇ϕ exp{−Mu+} dx−
∫

Ω

|u|p
∗−2uϕ exp{−Mu+} dx

+
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇u+

]
ϕ exp{−Mu+} dx ≥ 0.

for each ϕ ∈W 1,p
0 ∩ L∞(Ω) positive. Testing now (6.121) with

ϕk = ϕϑ(
u

k
) exp{Mu+} , ϕ ∈ C∞c (Ω) , ϕ ≥ 0 ,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 147

where ϑ is smooth, ϑ = 1 in
[
− 1

2 ,
1
2

]
and ϑ = 0 in ] −∞,−1] ∪ [1,+∞[, it follows

that ∫
Ω

∇ξL (x, u,∇u) · ∇ϕk exp{−Mu+} dx−
∫

Ω

|u|p
∗−2uϕϑ(

u

k
) dx

+
∫

Ω

[
DsL (x, u,∇u)−M∇ξL (x, u,∇u) · ∇u+

]
ϕϑ(

u

k
) dx ≥ 0.

which, arguing again as [132, Theorem 3.4], yields as k → +∞∫
Ω

∇ξL (x, u,∇u) · ∇ϕdx+
∫

Ω

DsL (x, u,∇u)ϕdx ≥
∫

Ω

|u|p
∗−2uϕdx.

for each ϕ ∈ C∞c (Ω) positive. Working analogously with ϕε = ϕ exp{−Mu−ε }, one
obtains the opposite inequality, i.e. u is a solution of (6.99). �

6.12. Proof of the main results. Let us now consider a sequence (uε)ε>0 of
solutions of (6.97) with fε(uε) → c and

p∗ − p− γ

pp∗
(νS)n/p < c < 2

p∗ − p− γ

pp∗
(νS)n/p. (6.124)

Then, there exist a subsequence of (uε)ε>0 and two bounded positive measures µ
and σ verifying (6.107), (6.108), (6.109), (6.110) and (6.111).

Proof of Theorem 6.35. Let us first show that there exists at most one j such that
µj 6= 0. Suppose that µj 6= 0 for every j = 1, . . . r ; in view of Lemma 6.38 one has
that µj ≥ ν

n
p∗ S

n
p . Following the proof of Lemma 6.34, we obtain

c = lim
ε→0

fε(uε) ≥
p∗ − p− γ

pp∗
ν lim
ε→0

∫
Ω

|∇uε|p dx

≥ p∗ − p− γ

pp∗
ν

∫
Ω

dµ

≥ p∗ − p− γ

pp∗
ν

r∑
j=1

µj

≥ r
p∗ − p− γ

pp∗
(νS)

n
p .

Taking into account (6.124) one has

2
p∗ − p− γ

pp∗
(νS)n/p > c ≥ r

p∗ − p− γ

pp∗
(νS)

n
p ,

hence r ≤ 1. Now, arguing as in Lemma 6.34 one obtains

2
p∗ − p− γ

pp∗
(νS)n/p > c = lim

ε→0
fε(uε)

= lim
ε→0

[
fε(uε)−

1
p∗ − ε

f ′ε(uε)(uε)
]

≥ p∗ − p− γ

pp∗
ν lim
ε→0

∫
Ω

|∇uε|p dx

≥ p∗ − p− γ

pp∗

(
ν

∫
Ω

|∇u|p dx+ νµ1

)
.



148 MARCO SQUASSINA EJDE-2006/MON. 07

If both summands were non-zero, by Lemma 6.38 and Lemma 6.39 we would obtain

ν

∫
Ω

|∇u|p dx > (νS)n/p, νµ1 ≥ (νS)
n
p

and thus a contradiction. Viceversa, let us assume that u ≡ 0 and µ1 = 0. Let
ψ ∈ C1

c (Ω) with ψ ≥ 0. By testing our equation with ψuε and using Hölder
inequality, one gets∫

Ω

uε∇ξL (x, uε,∇uε) · ∇ψ dx+ p

∫
Ω

ψL (x, uε,∇uε) dx

+
∫

Ω

DsL (x, uε,∇uε)ψuε dx

=
∫

Ω

|uε|p
∗−εψ dx

≤
( ∫

Ω

|uε|p
∗
ψ dx

) p∗−ε
p∗ Ln(Ω)

ε
p∗ .

(6.125)

Since (uε)ε>0 is bounded in W 1,p
0 (Ω), by (6.103) there exists C > 0 such that∣∣∣∣∫

Ω

uε∇ξL (x, uε,∇uε) · ∇ψ dx
∣∣∣∣ ≤ C ‖uε‖p ,

which, by uε → 0 in Lp(Ω), yields

lim
ε→0

∫
Ω

uε∇ξL (x, uε,∇uε) · ∇ψ dx = 0.

Moreover, since by (6.104) we get∫
Ω

DsL (x, uε,∇uε)ψuε dx ≥ 0,

taking into account (6.101) and passing to the limit in (6.125), we get

∀ψ ∈ Cc(Ω) : ψ ≥ 0 ⇒ ν

∫
Ω

ψ dµ ≤
∫

Ω

ψ dσ. (6.126)

On the other hand µ1 = 0 and u = 0 imply σ = 0. Then, since µ ≥ 0, by (6.126),
we get µ = 0. In particular, one gets

c = lim
ε→0

fε(uε)

= lim
ε→0

[p∗ − p− ε

p∗ − ε

∫
Ω

L (x, uε,∇uε) dx − 1
p∗ − ε

∫
Ω

DsL (x, uε,∇uε)uε dx
]

≤ pb0
n

lim
ε→0

(∫
Ω

|uε|p dx+
∫

Ω

|∇uε|p dx
)

=
pb0
n

∫
Ω

dµ = 0

which is not possible. Therefore, either µ1 = 0 and u 6≡ 0, or µ1 6= 0 and u ≡ 0. �

Remark 6.41. If (6.124) is replaced by the (k + 1)-th critical energy range

k
p∗ − p− γ

pp∗
(νS)n/p < c < (k + 1)

p∗ − p− γ

pp∗
(νS)n/p

for k ∈ N, k ≥ 1, in a similar way one can prove that µj = 0 for any j ≥ k + 1 and
(a) if µj = 0 for every j ≥ 1, then u is a non-trivial solution of (6.99) ;



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 149

(b) if µj 6= 0 for every 1 ≤ j ≤ k, then u ≡ 0.

Remark 6.42. Let f0 : W 1,p
0 (Ω) → R be the functional associated with (6.99) and

u ∈W 1,p
0 (Ω), u 6≡ 0, a solution of (6.99) (obtained as weak limit of (uε)ε>0). Then

f0(u) >
p∗ − p− γ

pp∗
(νS)n/p. (6.127)

Indeed,

f0(u) = f0(u)−
1
p∗
f ′0(u)(u)

≥ p∗ − p− γ

p∗

∫
Ω

L (x, u,∇u) dx

≥ p∗ − p− γ

pp∗
ν

∫
Ω

|∇u|p dx ,

which yields (6.127) in view of Lemma 6.39. This, in some sense, explains why one
chooses c greater than p∗−p−γ

pp∗ (νS)n/p in Theorem 6.35.

Let now (uε)ε>0 be a sequence of solutions of (6.97) with fε(uε) → c and

lim
ε→0

fε(uε) =
p∗ − p− γ

pp∗
(νS)n/p.

Proof of Theorem 6.36. Let us first note that

f0(u) ≤ lim
ε→0

fε(uε) +
1
p∗

∞∑
j=1

σj . (6.128)

Indeed, taking into account that by [53, Theorem 3.4]∫
Ω

L (x, u,∇u) dx ≤ lim
ε→0

∫
Ω

L (x, uε,∇uε) dx,

(6.128) follows by combining Hölder inequality with (6.110).
Now assume by contradiction that u 6≡ 0. Then, there exists j0 ∈ N such that

µj0 6= 0 and σj0 6= 0 otherwise, by Remark 6.42 and (6.128) we would get

p∗ − p− γ

pp∗
(νS)n/p < f0(u) ≤ lim

ε→0
fε(uε) =

p∗ − p− γ

pp∗
(νS)n/p.

Then, arguing as in Lemma 6.34 and applying Lemma 6.38, we obtain

p∗ − p− γ

pp∗
(νS)n/p = lim

ε→0
fε(uε)

≥ p∗ − p− γ

pp∗

(
ν

∫
Ω

|∇u|p dx+ νµj0

)
≥ p∗ − p− γ

pp∗
ν

∫
Ω

|∇u|p dx+
p∗ − p− γ

pp∗
(νS)n/p ,

which implies u ≡ 0, a contradiction. �



150 MARCO SQUASSINA EJDE-2006/MON. 07

6.13. Mountain-pass critical values. In this section, we shall investigate the
asymptotics of (uε) in the case of critical levels of min-max type. We assume that
L satisfies a stronger assumption, i.e.

L (x, s, ξ) ≤ 1
p
|ξ|p (6.129)

for a.e. x ∈ Ω and for all (s, ξ) ∈ R× Rn. In particular, it results that ν ≤ 1. Let
uε be a critical point of fε associated with the mountain pass level

cε = inf
η∈Cε

max
t∈[0,1]

fε(η(t)), (6.130)

where
Cε = {η ∈ C([0, 1],W 1,p

0 (Ω)) : η(0) = 0, η(1) = wε}
and wε ∈W 1,p

0 (Ω) is chosen in such a way that fε(wε) < 0. If u is the weak limit of
(uε)ε>0, as before one can apply P.L. Lions’ concentration-compactness principle.

Lemma 6.43. lim
ε→0

fε(uε) ≤ 1
nS

n/p.

Proof. Let x0 ∈ Ω and δ > 0 and consider the functions Tδ,x0 as in (6.113). By (c)
of Proposition 6.37, one has:

‖∇Tδ,x0‖
p
p,Rn = ‖Tδ,x0‖

p∗

p∗,Rn = S
n
p .

Moreover, taking a function φ ∈ C∞c (Ω) with 0 ≤ φ ≤ 1 and φ = 1 in a neighbor-
hood of x0 and setting vδ = φTδ,x0 , it results

‖∇vδ‖pp = S
n
p + o(1), ‖vδ‖p

∗

p∗ = S
n
p + o(1) , (6.131)

as δ → 0 (see [81, Lemma 3.2]). We want to prove that, for any t ≥ 0,

lim
ε→0

fε(tvδ) ≤
1
n
S

n
p + o(1)

as δ → 0. By (6.129) one has

lim
ε→0

fε(tvδ) = tp
∫

Ω

L (x, tvδ,∇vδ) dx− lim
ε→0

tp
∗−ε

p∗ − ε

∫
Ω

|vδ|p
∗−ε dx

≤ tp

p

∫
Ω

|∇vδ|p dx−
tp
∗

p∗

∫
Ω

|vδ|p
∗
dx.

Keeping into account (6.131) and the fact that tp

p −
tp
∗

p∗ ≤ 1
n for every t ≥ 0 , one

gets

lim
ε→0

fε(tvδ) ≤
tp

p
S

n
p − tp

∗

p∗
S

n
p + o(1) ≤ 1

n
S

n
p + o(1)

as δ → 0. Now choose t0 > 0 such that fε(t0vδ) < 0; by (6.130) we have that

lim
ε→0

fε(uε) ≤ lim
ε→0

max
s∈[0,1]

fε(st0vδ) ≤
1
n
S

n
p + o(1)

and conclude the proof letting δ → 0. �

Theorem 6.44. Suppose that the number of non-zero Dirac masses is[ pp∗

(p∗ − p− γ)nν
n
p

]
where [x] denotes the integer part of x. Then u ≡ 0.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 151

Proof. Taking into account the previous lemma and arguing as in Lemma 6.34, one
obtains

1
n
S

n
p ≥ lim

ε→0
fε(uε)

≥ p∗ − p− γ

pp∗
ν
( ∫

Ω

|∇u|p dx+
r∑
j=1

µj

)
≥ p∗ − p− γ

pp∗
ν

∫
Ω

|∇u|p dx+ r
p∗ − p− γ

pp∗
νn/pS

n
p ,

where r denotes the number of non-vanishing masses. Hence it must be

0 ≤ r ≤
[ pp∗

(p∗ − p− γ)nν
n
p

]
.

In particular, if r is maximum and u 6≡ 0, by virtue of Lemma 6.39 one obtains
p∗ − p− γ

pp∗
νn/pS

n
p >

p∗ − p− γ

pp∗
ν

∫
Ω

|∇u|p dx > p∗ − p− γ

pp∗
νn/pS

n
p ,

which is a contradiction. �

7. The Singularly Perturbed Case, I

Let Ω be a possibly unbounded smooth domain of RN with N ≥ 3. Since the
pioneering work of Floer and Einstein [68] in the one space dimension, much interest
has been directed in the last decade to singularly perturbed elliptic problems of the
form

−ε2∆u+ V (x)u = f(u) in Ω
u > 0 in Ω
u = 0 on ∂Ω

(7.1)

for a super-linear and subcritical nonlinearity f with f(s)/s nondecreasing.
Typically, there exists a family of solutions (uε)ε>0 which exhibits a spike shape

around the local minima (possibly degenerate) of the function V (x) and decade
elsewhere as ε goes to zero (see e.g. [3, 62, 63, 64, 83, 108, 109, 119, 125, 126, 145]
and references therein). A natural question is now whether these concentration
phenomena are a special feature of the semi-linear case or we can expect a similar
behavior to hold for more general elliptic equations which possess a variational
structure.

In this section we will give a positive answer to this question for the following
class of singularly perturbed quasi-linear elliptic problems

−ε2
N∑

i,j=1

Dj(aij(x, u)Diu) +
ε2

2

N∑
i,j=1

Dsaij(x, u)DiuDju+ V (x)u = f(u) in Ω

u > 0 in Ω
u = 0 on ∂Ω

(7.2)
under suitable assumptions on the functions aij , V and f . Notice that if aij(x, s) =
δij then equation (7.2) reduces to (7.1), in which case the problem originates from
different physical and biological models and, in particular, in the study of the so
called standing waves for the nonlinear Schrödinger equation.



152 MARCO SQUASSINA EJDE-2006/MON. 07

Existence and multiplicity results for equations like (7.2) have been object of a
very careful analysis since 1994 (see e.g. [6, 7, 33, 36, 132] for the case where Ω
is bounded and [48, 130] for Ω unbounded). On the other hand, to the author’s
knowledge, no result on the asymptotic behavior of the solutions (as ε vanishes)
of (7.2) can be found in literature. In particular no achievement is known so far
concerning the concentration phenomena for the solutions uε of (7.2) around the
local minima, not necessarily non-degenerate, of V .

We stress that various difficulties arise in comparison with the study of the semi-
linear equation (7.1) (see Section 7.4 for a list of properties which are not known
to hold in our framework).

A crucial step in proving our main result is to show that the Mountain-Pass
energy level of the functional J associated with the autonomous limiting equation

−
N∑

i,j=1

Dj(aij(x̂, u)Diu) +
1
2

N∑
i,j=1

Dsaij(x̂, u)DiuDju+ V (x̂)u = f(u) in RN

(7.3)
with x̂ ∈ RN , is the least among other nontrivial critical values (Lemma 7.10).
Notice that, no uniqueness result is available, to our knowledge, for this general
equation (on the contrary in the semi-linear case some uniqueness theorems for
ground state solutions have been obtained by performing an ODE analysis in radial
coordinates, see e.g. [44]). The least energy problem for (7.3) is also related to the
fact:

u ∈ H1(RN ), u ≥ 0 and u solution of (7.3) implies that J(u) = max
t≥0

J(tu) (7.4)

Unfortunately, as remarked in [48, section 3], if one assumes that condition
(7.10) holds, then property (7.4) cannot hold true even if the map s 7→ f(s)/s is
nondecreasing.

To show the minimality property for the Mountain-Pass level and to study the
uniform limit of uε on ∂Λ, inspired by the recent work of Jeanjean and Tanaka [82],
we make a repeated use of the Pucci-Serrin identity [116], which has turned out to
be a very powerful tool (Lemmas 7.10 and 7.11).

Notice that the functional associated with (7.2) (see (7.16)) is not even locally
Lipschitz and tools of non-smooth critical point theory will be employed (see [50, 58]
and references therein). Also the proof of a suitable Palais-Smale type condition
for a modification of the functional Iε becomes more involved.

We assume that f ∈ C1(R+) and there exist 1 < p < N+2
N−2 and 2 < ϑ ≤ p + 1

with

lim
s→+∞

f(s)
sp

= 0, lim
s→0+

f(s)
s

= 0, (7.5)

0 < ϑF (s) ≤ f(s)s for every s ∈ R+, (7.6)

where F (s) =
∫ s
0
f(t) dt for every s ∈ R+.

Furthermore, let V : RN → R be a locally Hölder continuous function bounded
below away from zero, that is, there exists α > 0 with

V (x) ≥ α for every x ∈ RN . (7.7)

The functions aij(x, s) : Ω × R+ → R are continuous in x and of class C1 with
respect to s, aij(x, s) = aji(x, s) for every i, j = 1, . . . , N and there exists a positive



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 153

constant C with
|aij(x, s)| ≤ C, |Dsaij(x, s)| ≤ C

for every x ∈ Ω and s ∈ R+. Finally, let R, ν > 0 and 0 < γ < ϑ− 2 be such that

N∑
i,j=1

aij(x, s)ξiξj ≥ ν|ξ|2, (7.8)

N∑
i,j=1

sDsaij(x, s)ξiξj ≤ γ
N∑

i,j=1

aij(x, s)ξiξj , (7.9)

s ≥ R ⇒
N∑

i,j=1

Dsaij(x, s)ξiξj ≥ 0 (7.10)

for every x ∈ Ω, s ∈ R+ and ξ ∈ RN .
Hypothesis (7.5), (7.6) and (7.7) on f and V are standard. Observe that neither

monotonicity assumptions on the function f(s)/s nor uniqueness conditions on the
limiting equation (7.3) are considered. Finally, (7.9) and (7.10) have already been
used, for instance in [6, 7, 33, 36, 48], in order to tackle these general equations.

Let HV (Ω) be the weighted Hilbert space defined by

HV (Ω) =
{
u ∈ H1

0 (Ω) :
∫

Ω

V (x)u2 < +∞
}
,

endowed with the scalar product (u, v)V =
∫
Ω
DuDv + V (x)uv and denote by

‖ · ‖HV (Ω) the corresponding norm.
Let Λ be a compact subset of Ω such that there exists x0 ∈ Λ with

V (x0) = min
Λ
V < min

∂Λ
V, (7.11)

N∑
i,j=1

aij(x0, s)ξiξj = min
x∈Λ

N∑
i,j=1

aij(x, s)ξiξj (7.12)

for every s ∈ R+ and ξ ∈ RN . Let us set

σ := sup
{
s > 0 : f(t) ≤ tV (x0) for every t ∈ [0, s]

}
, (7.13)

M :=
{
x ∈ Λ : V (x) = V (x0)

}
. (7.14)

The following is the main result of the section.

Theorem 7.1. Assume that conditions (7.5), (7.6), (7.7), (7.8), (7.9), (7.10),
(7.11), (7.12) hold. Then there exists ε0 > 0 such that, for every ε ∈ (0, ε0),
there exist uε ∈ HV (Ω) ∩ C(Ω) and xε ∈ Λ satisfying the following properties:

(a) uε is a weak solution of the problem

−ε2
N∑

i,j=1

Dj(aij(x, u)Diu) +
ε2

2

N∑
i,j=1

Dsaij(x, u)DiuDju+ V (x)u = f(u) in Ω

u > 0 in Ω
u = 0 on ∂Ω ;

(7.15)



154 MARCO SQUASSINA EJDE-2006/MON. 07

(b) there exists σ′ > 0 such that

uε(xε) = sup
Ω
uε, σ < uε(xε) < σ′, lim

ε→0
d(xε,M ) = 0

where σ is as in (7.13) and M is as in (7.14) ;
(c) for every % > 0 we have

lim
ε→0

‖uε‖L∞(Ω\B%(xε)) = 0 ;

(d) we have
lim
ε→0

‖uε‖HV (Ω) = 0

and, as a consequence, lim
ε→0

‖uε‖Lq(Ω) = 0 for every 2 ≤ q < +∞.

The proof of the theorem is variational and in the spirit of a well-known paper
by del Pino and Felmer [62], where it was successfully developed into a local setting
the global approach initiated by Rabinowitz [119].

We will consider the functional Iε : HV (Ω) → R associated with the problem
(7.15),

Iε(u) :=
ε2

2

N∑
i,j=1

∫
Ω

aij(x, u)DiuDju+
1
2

∫
Ω

V (x)u2 −
∫

Ω

F (u) (7.16)

and construct a new functional Jε which satisfies the Palais-Smale condition (in
a suitable sense) at every level (Iε does not, in general) and to which the (non-
smooth) Mountain-Pass Theorem can be directly applied to get a critical point uε
with precise energy estimates.

Then we will prove that uε goes to zero uniformly on ∂Λ as ε goes to zero (this
is the hardest step, here we repeatedly use the Pucci-Serrin identity in a suitable
form) and show that uε is actually a solution of the original problem with all of the
stated properties.

Remark 7.2. We do not know whether the solutions of problem (7.15) obey to
the following exponential decay

uε(x) ≤ α exp
{
− β

ε
|x− xε|

}
for every x ∈ Ω, for some α, β ∈ R+, (7.17)

which is a typical feature in the semi-linear case. This fact would follow if we had a
suitable Gidas-Ni-Nirenberg [76] type result for the equation (7.3) to be combined
with some results by Rabier and Stuart [117] on the exponential decay of second
order elliptic equations.

As pointed out in [64], the concentration around the minima of the potential is,
in some sense, a model situation for other phenomena such as concentration around
the maxima of d(x, ∂Ω). Furthermore it seems to be the technically simplest case,
thus suitable for a first investigation in the quasi-linear case.

7.1. The del Pino-Felmer penalization scheme. We now define a suitable
modification of the functional Iε in order to regain the (concrete) Palais-Smale
condition at any level and apply the Mountain Pass Theorem. Let us consider the
positive constant

` := sup
{
s > 0 :

f(t)
t

≤ α

k
for every 0 ≤ t ≤ s

}



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 155

for some k > ϑ/(ϑ− 2). We define the function f̃ : R+ → R by setting

f̃(s) :=

{
α
k s if s > `

f(s) if 0 ≤ s ≤ `

and the maps g,G : Ω× R+ → R,

g(x, s) := χΛ(x)f(s) + (1− χΛ(x))f̃(s), G(x, s) =
∫ s

0

g(x, τ) dτ

for every x ∈ Ω. Then the function g(x, s) is measurable in x, of class C1 in s and
it satisfies the following assumptions:

lim
s→+∞

g(x, s)
sp

= 0, lim
s→0+

g(x, s)
s

= 0 uniformly in x ∈ Ω, (7.18)

0 < ϑG(x, s) ≤ g(x, s)s for every x ∈ Λ and s ∈ R+, (7.19)

0 ≤ 2G(x, s) ≤ g(x, s)s ≤ 1
k
V (x)s2 for every x ∈ Ω \ Λ and s ∈ R+. (7.20)

Without loss of generality, we may assume that

g(x, s) = 0 for every x ∈ Ω and s < 0,

aij(x, s) = aij(x, 0) for every x ∈ Ω, s < 0 and i, j = 1, . . . , N .

Let Jε : HV (Ω) → R be the functional

Jε(u) :=
ε2

2

N∑
i,j=1

∫
Ω

aij(x, u)DiuDju+
1
2

∫
Ω

V (x)u2 −
∫

Ω

G(x, u).

The next result provides the link between the critical points of the modified
functional Jε and the solutions of the original problem.

Proposition 7.3. Assume that uε ∈ HV (Ω) is a critical point of Jε and that there
exists a positive number ε0 such that

uε(x) ≤ ` for every ε ∈ (0, ε0) and x ∈ Ω \ Λ.

Then uε is a solution of (7.15).

Proof. By assertion (a) of Corollary 2.25, it results that uε is a solution of the
penalized problem. Since uε ≤ ` on Ω \ Λ, we have

G(x, uε(x)) = F (uε(x)) for every x ∈ Ω.

Moreover, by arguing as in the proof of [130, Lemma 1], one gets uε > 0 in Ω. Then
uε is a solution of (7.15). �

The next Lemma - which is nontrivial - provides a local compactness property
for bounded concrete Palais-Smale sequences of Jε. For the proof, we refer the
reader to [130, Theorem 2 and Lemma 3].

Lemma 7.4. Assume that conditions (7.5), (7.6), (7.7), (7.8), (7.9), (7.10) hold.
Let ε > 0. Assume that (uh) ⊂ H1(RN ) is a bounded sequence and

〈wh, ϕ〉 = ε2
N∑

i,j=1

∫
RN

aij(x, uh)DiuhDjϕ+
ε2

2

N∑
i,j=1

∫
RN

Dsaij(x, uh)DiuhDjuhϕ



156 MARCO SQUASSINA EJDE-2006/MON. 07

for every ϕ ∈ C∞c (RN ), where (wh) is strongly convergent in H−1(Ω̃) for a given
bounded domain Ω̃ of RN .

Then (uh) admits a strongly convergent subsequence in H1(Ω̃). In particular,
if (uh) is a bounded concrete Palais-Smale condition for Jε at level c and u is its
weak limit, then, up to a subsequence, Duh → Du in L2(Ω̃,RN ) for every bounded
subset Ω̃ of Ω.

Since Ω may be unbounded, in general, the original functional Iε does not satisfy
the concrete Palais-Smale condition. In the following Lemma we prove that, instead,
the functional Jε satisfies it for every ε > 0 at every level c ∈ R.

Lemma 7.5. Assume that conditions (7.5), (7.6), (7.7), (7.8), (7.9), (7.10) hold.
Let ε > 0. Then Jε satisfies the concrete Palais-Smale condition at every level
c ∈ R.

Proof. Let (uh) ⊂ HV (Ω) be a concrete Palais-Smale sequence for Jε at level c. We
divide the proof into two steps:
Step I. Let us prove that (uh) is bounded in HV (Ω). Since Jε(uh) → c, from
inequalities (7.19) and (7.20), we get

ϑε2

2

N∑
i,j=1

∫
Ω

aij(x, uh)DiuhDjuh +
ϑ

2

∫
Ω

V (x)u2
h

≤
∫

Λ

g(x, uh)uh +
ϑ

2k

∫
Ω\Λ

V (x)u2
h + ϑc+ o(1)

(7.21)

as h → +∞. Moreover, we have J ′ε(uh)(uh) = o(‖uh‖HV (Ω)) as h → +∞. Then,
again by virtue of (7.20), we deduce

ε2
N∑

i,j=1

∫
Ω

aij(x, uh)DiuhDjuh +
ε2

2

N∑
i,j=1

∫
Ω

Dsaij(x, uh)uhDiuhDjuh +
∫

Ω

V (x)u2
h

≥
∫

Λ

g(x, uh)uh + o(‖uh‖HV (Ω)),

as h→ +∞, which, by (7.9), yields

(γ
2

+ 1
)
ε2

N∑
i,j=1

∫
Ω

aij(x, uh)DiuhDjuh +
∫

Ω

V (x)u2
h

≥
∫

Λ

g(x, uh)uh + o(‖uh‖HV (Ω))

(7.22)

as h → +∞. Then, in view of (7.8), by combining inequalities (7.21) and (7.22)
one gets

min
{(ϑ

2
− γ

2
− 1

)
νε2,

ϑ

2
− ϑ

2k
− 1

} ∫
Ω

(
|Duh|2 + V (x)u2

h

)
≤ ϑc+ o(‖uh‖HV (Ω)) + o(1)

(7.23)

as h→ +∞, which implies the boundedness of (uh) in HV (Ω).
Step II. By virtue of Step I, there exists u ∈ HV (Ω) such that, up to a subsequence,
(uh) weakly converges to u in HV (Ω).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 157

Let us now prove that actually (uh) converges strongly to u in HV (Ω). By taking
into account Lemma 7.4 (applied with Ω̃ = B%(0) for every % > 0), it suffices to
prove that for every δ > 0 there exists % > 0 such that

lim sup
h

∫
Ω\B%(0)

(
|Duh|2 + V (x)u2

h

)
< δ. (7.24)

We may assume that Λ ⊂ B%/2(0). Consider a cut-off function ψ% ∈ C∞(Ω) with
ψ% = 0 on B%/2(0), ψ% = 1 on Ω \ B%(0), |Dψ%| ≤ c/% on Ω for some positive
constant c. Let M be a positive number such that∣∣∣1

2

N∑
i,j=1

Dsaij(x, s)ξiξj
∣∣∣ ≤M

N∑
i,j=1

aij(x, s)ξiξj (7.25)

for every x ∈ Ω, s ∈ R+, ξ ∈ RN and let ζ : R → R be the map defined by

ζ(s) :=


0 if s < 0
Ms if 0 ≤ s < R

MR if s ≥ R,

(7.26)

being R > 0 the constant defined in (7.10). Notice that
N∑

i,j=1

[1
2
Dsaij(x, s) + ζ ′(s)aij(x, s)

]
ξiξj ≥ 0, (7.27)

for every x ∈ Ω, s ∈ R, ξ ∈ RN . Of course J ′ε(uh)(ψ%uh exp{ζ(uh)}) can be
computed. Since (uh) is bounded in HV (Ω) and (7.27) holds, we get

o(1) = J ′ε(uh)(ψ%uh exp{ζ(uh)})

= ε2
N∑

i,j=1

∫
Ω

aij(x, uh)DiuhDjuhψ% exp{ζ(uh)}

+ ε2
N∑

i,j=1

∫
Ω

[1
2
Dsaij(x, uh) + ζ ′(uh)aij(x, uh)

]
DiuhDjuhuhψ% exp{ζ(uh)}

+ ε2
N∑

i,j=1

∫
Ω

aij(x, uh)uhDiuhDjψ% exp{ζ(uh)}+
∫

Ω

V (x)u2
hψ% exp{ζ(uh)}

−
∫

Ω

g(x, uh)uhψ% exp{ζ(uh)}

≥
∫

Ω

(
ε2ν|Duh|2 + V (x)u2

h

)
ψ% exp{ζ(uh)}

+ ε2
N∑

i,j=1

∫
Ω

aij(x, uh)uhDiuhDjψ% exp{ζ(uh)}

−
∫

Ω

g(x, uh)uhψ% exp{ζ(uh)}.

Therefore, in view of (7.20), it results

o(1) ≥
∫

Ω

(
ε2ν|Duh|2 + V (x)u2

h

)
ψ% exp{ζ(uh)}



158 MARCO SQUASSINA EJDE-2006/MON. 07

+ ε2
N∑

i,j=1

∫
Ω

aij(x, uh)uhDiuhDjψ% exp{ζ(uh)}

− 1
k

∫
Ω

V (x)u2
hψ% exp{ζ(uh)}

as %→ +∞. Taking into account that∣∣∣ N∑
i,j=1

∫
Ω

aij(x, uh)uhDiuhDjψ% exp{ζ(uh)}
∣∣∣ ≤ exp{MR}C̃

%
‖Duh‖2‖uh‖2,

there exists C ′ > 0 (which depends only on ε, ν and k) such that, as %→ +∞,

lim sup
h

∫
Ω\B%(0)

(
|Duh|2 + V (x)u2

h

)
≤ C ′

%
,

which yields (7.24). Therefore uh → u strongly in HV (Ω) and the proof is complete.
�

7.2. Energy estimates and concentration. Let us now introduce the functional
J0 : H1(RN ) → R defined by

J0(u) :=
1
2

N∑
i,j=1

∫
RN

aij(x0, u)DiuDju+
1
2

∫
RN

V (x0)u2 −
∫

RN

F (u)

where x0 is as in (7.11). Let us set

c̄ := inf
γ∈P0

sup
t∈[0,1]

J0(γ(t)),

where P0 is the family defined by

P0 :=
{
γ ∈ C([0, 1],HV (RN )) : γ(0) = 0, J0(γ(1)) < 0

}
. (7.28)

Let us also set

Pε :=
{
γ ∈ C([0, 1],HV (Ω)) : γ(0) = 0, Jε(γ(1)) < 0

}
. (7.29)

In the following, if necessary, we will assume that, for every γ ∈ Pε, for every
t ∈ [0, 1] the map γ(t) is extended to zero outside Ω.

In the next Lemma we get a critical point uε of Jε with a precise energy upper
bound.

Lemma 7.6. For ε > 0 sufficiently small Jε admits a critical point uε ∈ HV (Ω)
such that

Jε(uε) ≤ εN c̄+ o(εN ). (7.30)

Proof. Let ε > 0. By Lemma 7.5 the functional Jε satisfies the concrete Palais-
Smale condition at every level c ∈ R. Moreover, since g(x, s) = o(s) as s → 0
uniformly in x, it is readily seen that Jε verifies the Mountain-Pass geometry.
Finally, if z is a positive function in HV (Ω) \ {0} such that supt(z) ⊂ Λ, by (7.6)
it results Jε(tz) → −∞ as t → +∞. Therefore, by minimaxing over the family
(7.29), the functional Jε admits a nontrivial critical point uε ∈ HV (Ω) such that

Jε(uε) = inf
γ∈Pε

sup
t∈[0,1]

Jε(γ(t)).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 159

Since c̄ is the Mountain-Pass value of the limiting functional J0, for every δ > 0
there exists a continuous path γ : [0, 1] → HV (RN ) such that

c̄ ≤ sup
t∈[0,1]

J0(γ(t)) ≤ c̄+ δ, γ(0) = 0, J0(γ(1)) < 0. (7.31)

Let ζ ∈ C∞c (RN ) be a cut-off function with ζ = 1 in a neighborhood U of x0 in
Λ. We define the continuous path Γε : [0, 1] → HV (Ω) by setting Γε(τ)(x) :=
ζ(x)γ(τ)

(
x−x0
ε

)
for every τ ∈ [0, 1] and x ∈ Ω. Then, for every τ ∈ [0, 1], after

extension to zero outside Ω, we have

Jε(Γε(τ))

=
ε2

2

N∑
i,j=1

∫
RN

aij

(
x, ζ(x)γ(τ)

(x− x0

ε

))
DiζDjζ γ

2(τ)
(x− x0

ε

)
+

1
2

N∑
i,j=1

∫
RN

aij

(
x, ζ(x)γ(τ)

(x− x0

ε

))
(Diγ(τ))

(x− x0

ε

)
(Djγ(τ))

(x− x0

ε

)
ζ2

+ ε
N∑

i,j=1

∫
RN

aij

(
x, ζ(x)γ(τ)

(x− x0

ε

))
Diζ(Djγ(τ))

(x− x0

ε

)
ζγ(τ)

(x− x0

ε

)
+

1
2

∫
RN

V (x)ζ2(x)γ2(τ)
(x− x0

ε

)
−

∫
RN

G

(
x, ζ(x)γ(τ)

(x− x0

ε

))
.

Then, after the change of coordinates, for every τ ∈ [0, 1], we get

Jε(Γε(τ)) =
εN+2

2

N∑
i,j=1

∫
RN

aij (εy + x0, ζ(εy + x0)γ(τ)(y))Diζ(εy + x0)

×Djζ(εy + x0) γ2(τ)(y)

+ εN+1
N∑

i,j=1

∫
RN

aij (εy + x0, ζ(εy + x0)γ(τ)(y))Diζ(εy + x0)

×Djγ(τ)(y)ζ(εy + x0)γ(τ)(y)

+
εN

2

N∑
i,j=1

∫
RN

aij (εy + x0, ζ(εy + x0)γ(τ)(y))Diγ(τ)(y)

×Djγ(τ)(y)ζ2(εy + x0) +
εN

2

∫
RN

V (εy + x0)ζ2(εy + x0)γ2(τ)(y)

− εN
∫

RN

G(εy + x0, ζ(εy + x0)γ(τ)(y)).

Taking into account that for every τ ∈ [0, 1],

lim
ε→0

∫
RN

V (εy + x0)ζ2(εy + x0)γ2(τ)(y) =
∫

RN

V (x0)γ2(τ)(y),

lim
ε→0

∫
RN

G(εy + x0, ζ(εy + x0)γ(τ)(y)) =
∫

RN

F (γ(τ)(y)),



160 MARCO SQUASSINA EJDE-2006/MON. 07

and

lim
ε→0

N∑
i,j=1

∫
RN

aij(εy + x0, ζ(εy + x0)γ(τ)(y))Diγ(τ)(y)Djγ(τ)(y)ζ2(εy + x0)

=
N∑

i,j=1

∫
RN

aij(x0, γ(τ)(y))Diγ(τ)(y)Djγ(τ)(y),

we obtain

Jε(Γε(τ)) = εN
{1

2

N∑
i,j=1

∫
RN

aij(x0, γ(τ)(y))Diγ(τ)(y)Djγ(τ)(y)

+
1
2

∫
RN

V (x0)γ2(τ)(y)−
∫

RN

F (γ(τ)(y))
}

+ o(εN )

as ε→ 0, namely

Jε(Γε(τ)) = εNJ0(γ(τ)) + o(εN ) (7.32)

as ε→ 0, where o(εN ) is independent of τ (by a compactness argument). Then, by
(7.31) and (7.32), it follows that Γε ∈ Pε for every ε > 0 sufficiently small and,

Jε(uε) = inf
γ∈Pε

sup
t∈[0,1]

Jε(γ(t)) ≤ sup
t∈[0,1]

Jε(Γε(t))

= εN sup
t∈[0,1]

J0(γ(t)) + o(εN )

≤ εN c̄+ o(εN ) + δεN for every δ > 0.

By the arbitrariness of δ one concludes the proof. �

In the following result we get some priori estimates for the rescalings of uε.

Corollary 7.7. Let (εh) ⊂ R+, (xh) ⊂ Λ and assume that (uεh
) ⊂ HV (Ω) is as in

Lemma 7.6. Let us set

vh ∈ HV (Ωh), Ωh := ε−1
h (Ω− xh), vh(x) := uεh

(xh + εhx)

and put vh = 0 outside Ωh. Then there exists a positive constant C such that for
every h ∈ N,

‖vh‖H1(RN ) ≤ C . (7.33)

Proof. We consider the functional Jh : HV (Ωh) → R given by

Jh(v) :=
1
2

N∑
i,j=1

∫
Ωh

aij(xh + εhx, v)DivDjv (7.34)

+
1
2

∫
Ωh

V (xh + εhx)v2 −
∫

Ωh

G(xh + εhx, v). (7.35)

Since Jh(vh) = εh
−NJεh

(uεh
), by virtue of Lemma 7.6 we have Jh(vh) ≤ c̄ + o(1)

as h → +∞. Therefore, if we set Λh = ε−1
h (Λ − xh), from inequalities (7.19) and



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 161

(7.20), we get

ϑ

2

N∑
i,j=1

∫
RN

aij(xh + εhx, vh)DivhDjvh +
ϑ

2

∫
RN

V (xh + εhx)v2
h

≤
∫

Λh

g(xh + εhx, vh)vh +
ϑ

2k

∫
RN\Λh

V (xh + εhx)v2
h + ϑc̄+ o(1)

(7.36)

as h → +∞. Moreover, since it results J ′h(vh)(vh) = 0 for every h ∈ N, again by
(7.20), we get

N∑
i,j=1

∫
RN

aij(xh + εhx, vh)DivhDjvh

+
1
2

N∑
i,j=1

∫
RN

Dsaij(xh + εhx, vh)vhDivhDjvh +
∫

RN

V (xh + εhx)v2
h

≥
∫

Λh

g(xh + εhx, vh)vh,

which, in view of (7.9), yields(γ
2

+ 1
) N∑
i,j=1

∫
RN

aij(xh + εhx, vh)DivhDjvh +
∫

RN

V (xh + εhx)v2
h

≥
∫

Λh

g(xh + εhx, vh)vh.

(7.37)

Then, recalling (7.7) and (7.8), by combining inequality (7.36) and (7.37) one gets

min
{(ϑ

2
− γ

2
− 1

)
ν,

(ϑ
2
− ϑ

2k
− 1

)
α
}∫

RN

(
|Dvh|2 + v2

h

)
≤ ϑc̄+ o(1) (7.38)

as h→ +∞, which yields the assertion. �

Corollary 7.8. Assume that (uε)ε>0 ⊂ HV (Ω) is as in Lemma 7.6. Then

lim
ε→0

‖uε‖HV (Ω) = 0.

Proof. We may argue as in Step I of Lemma 7.5 with uh replaced by uε and c
replaced by Jε(uε). Thus, from inequality (7.23), for every ε > 0 we get∫

Ω

(
|Duε|2 + V (x)u2

ε

)
≤ ϑ

min
{(

ϑ
2 −

γ
2 − 1

)
νε2, ϑ2 −

ϑ
2k − 1

} Jε(uε).
By virtue of Lemma 7.6, this yields∫

Ω

(
|Duε|2 + V (x)u2

ε

)
≤ 2ϑc̄

(ϑ− γ − 2)ν
εN−2 + o(εN−2)

for every ε sufficiently small, which implies the assertion. �

Let L : RN×R×RN → R be a function of class C1 such that the function ∇ξL
is of class C1 and let ϕ ∈ L∞loc(RN ). We now recall the Pucci-Serrin variational
identity [116].



162 MARCO SQUASSINA EJDE-2006/MON. 07

Lemma 7.9. Let u : RN → R be a C2 solution of

−div (DξL (x, u,Du)) +DsL (x, u,Du) = ϕ in D ′(RN ).

Then for every h ∈ C1
c (RN ,RN ),

N∑
i,j=1

∫
RN

Dih
jDξi

L (x, u,Du)Dju

−
∫

RN

[
(div h)L (x, u,Du) + h ·DxL (x, u,Du)

]
=

∫
RN

(h ·Du)ϕ .

(7.39)

We refer the reader to [59], where the above variational relation is proved for C1

solutions. We now derive an important consequence of the previous identity which
will play an important role in the proof of Lemma 7.11.

Lemma 7.10. Let µ > 0 and h,H : R+ → R be the continuous functions defined
by

h(s) = −µs+ f(s), H(s) =
∫ s

0

h(t) dt,

where f satisfies (7.5) and (7.6). Moreover, let bij ∈ C1(R+) ∩ L∞(R+) with
b′ij ∈ L∞(R+) and assume that there exist ν′ > 0 and R′ > 0 with

N∑
i,j=1

bij(s)ξiξj ≥ ν′|ξ|2, s ≥ R′ ⇒
N∑

i,j=1

b′ij(s)ξiξj ≥ 0 (7.40)

for every s ∈ R+ and ξ ∈ RN . Let u ∈ H1(RN ) be any nontrivial positive solution
of the equation

−
N∑

i,j=1

Dj(bij(u)Diu) +
1
2

N∑
i,j=1

b′ij(u)DiuDju = h(u) in RN . (7.41)

We denote by Ĵ the associated functional

Ĵ(u) :=
1
2

N∑
i,j=1

∫
RN

bij(u)DiuDju−
∫

RN

H(u). (7.42)

Then it results Ĵ(u) ≥ b, where

b := inf
γ∈ cP sup

t∈[0,1]

Ĵ(γ(t)),

P̂ :=
{
γ ∈ C([0, 1],H1(RN )) : γ(0) = 0, Ĵ(γ(1)) < 0

}
.

Proof. By condition (7.40),

Ĵ(v) ≥ 1
2

min {ν′, µ} ‖v‖2H1(RN ) −
∫

RN

F (v) for every v ∈ H1(RN ).

Then, since for every ε > 0 there exists Cε > 0 with

0 ≤ F (s) ≤ εs2 + Cε|s|
2N

N−2 for every s ∈ R+,

it is readily seen that there exist %0 > 0 and δ0 > 0 such that Ĵ(v) ≥ δ0 for every
v with ‖v‖1,2 = %0. In particular Ĵ has a Mountain-Pass geometry. As we will see,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 163

P̂ 6= ∅, so that b is well defined. Let u be a nontrivial positive solution of (7.41)
and consider the dilation path

γ(t)(x) :=

{
u(xt ) if t > 0
0 if t = 0.

Notice that ‖γ(t)‖2H1 = tN−2‖Du‖22 + tN‖u‖22 for every t ∈ R+, which implies that
the curve γ belongs to C([0,+∞[,H1(RN )). For every t ∈ R+ it results that

Ĵ(γ(t)) =
1
2

N∑
i,j=1

∫
RN

bij(γ(t))Diγ(t)Djγ(t)−
∫

RN

H(γ(t))

=
tN−2

2

N∑
i,j=1

∫
RN

bij(u)DiuDju− tN
∫

RN

H(u)

which yields, for every t ∈ R+

d

dt
Ĵ(γ(t)) =

N − 2
2

tN−3
N∑

i,j=1

∫
RN

bij(u)DiuDju−NtN−1

∫
RN

H(u). (7.43)

By (7.40), arguing like at the end of Step I of Lemma 7.11 (namely using the local
Serrin estimates) it results that u ∈ L∞loc(RN ). Hence by the regularity results of
[90], it follows that u is of class C2. Then we can use Lemma 7.9 by choosing ϕ = 0,

L (s, ξ) :=
1
2

N∑
i,j=1

bij(s)ξiξj −H(s) for every s ∈ R+ and ξ ∈ RN , (7.44)

h(x) := hk(x) = T (
x

k
)x for every x ∈ RN and k ≥ 1, (7.45)

being T ∈ C1
c (RN ) such that T (x) = 1 if |x| ≤ 1 and T (x) = 0 if |x| ≥ 2. In

particular, it results that hk ∈ C1
c (RN ,RN ) for every k ≥ 1 and

Dih
j
k(x) = DiT (

x

k
)
xj
k

+ T (
x

k
)δij for every x ∈ RN and i, j = 1, . . . , N

(div hk)(x) = DT (
x

k
) · x
k

+NT (
x

k
) for every x ∈ RN .

Then, since DxL (u,Du) = 0, it follows by (7.39) that for every k ≥ 1
n∑

i,j=1

∫
RN

DiT (
x

k
)
xj
k
DjuDξiL (u,Du) +

∫
RN

T (
x

k
)DξL (u,Du) ·Du

−
∫

RN

DT (
x

k
) · x
k

L (u,Du)−
∫

RN

NT (
x

k
)L (u,Du) = 0.

Since there exists C > 0 with

DiT (
x

k
)
xj
k
≤ C for every x ∈ RN , k ≥ 1 and i, j = 1, . . . , N ,

by the Dominated Convergence Theorem, letting k → +∞, we obtain∫
RN

[
NL (u,Du)−DξL (u,Du) ·Du

]
= 0,



164 MARCO SQUASSINA EJDE-2006/MON. 07

namely, by (7.44),

N − 2
2

N∑
i,j=1

∫
RN

bij(u)DiuDju = N

∫
RN

H(u). (7.46)

By plugging this formula into (7.43), we obtain
d

dt
Ĵ(γ(t)) = N(1− t2)tN−3

∫
RN

H(u)

which yields d
dt Ĵ(γ(t)) > 0 for t < 1 and d

dt Ĵ(γ(t)) < 0 for t > 1, i.e.

sup
t∈[0,L]

Ĵ(γ(t)) = Ĵ(γ(1)) = Ĵ(u).

Moreover, observe that

γ(0) = 0 and Ĵ(γ(T )) < 0 for T > 0 sufficiently large.

Then, after a suitable scale change in t, γ ∈ P̂ and the assertion follows. �

The following is one of the main tools of the section.

Lemma 7.11. Assume that (uε)ε>0 ⊂ HV (Ω) is as in Lemma 7.6. Then

lim
ε→0

max
∂Λ

uε = 0. (7.47)

Proof. The following auxiliary fact is sufficient to prove assertion (7.47): if εh → 0
and (xh) ⊂ Λ are such that uεh

(xh) ≥ c for some c > 0, then

lim
h
V (xh) = min

Λ
V. (7.48)

Indeed, assume by contradiction that there exist (εh) ⊂ R+ with εh → 0 and
(xh) ⊂ ∂Λ such that uεh

(xh) ≥ c for some c > 0. Up to a subsequence, we have
xh → x̄ ∈ ∂Λ. Then by (7.48) it results

min
∂Λ

V ≤ V (x̄) = lim
h
V (xh) = min

Λ
V

which contradicts assumption (7.11).
We divide the proof of (7.48) into four steps:

Step I. Up to a subsequence, xh → x̂ for some x̂ ∈ Λ. By contradiction, we assume
that

V (x̂) > min
Λ
V = V (x0).

Since for every h ∈ N the function uεh
solves (Pεh

), the sequence

vh ∈ HV (Ωh), Ωh = ε−1
h (Ω− xh), vh(x) = uεh

(xh + εhx)

satisfies

−
N∑

i,j=1

Dj(aij(xh + εhx, vh)Divh) +
1
2

N∑
i,j=1

Dsaij(xh + εhx, vh)DivhDjvh = wh

in Ωh, vh > 0 in Ωh and vh = 0 on ∂Ωh, where we have set

wh := g(xh + εhx, vh)− V (xh + εhx)vh for every h ∈ N.
Setting vh = 0 outside RN , by Corollary 7.7, up to a subsequence, vh → v weakly
in H1(RN ). Notice that the sequence (χΛ(xh + εhx)) converges weak* in L∞

to a measurable function 0 ≤ χ ≤ 1. In particular, taking into account that



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 165

|wh| ≤ c1|vh| + c2|vh|p, (wh) is strongly convergent in H−1(Ω̃) for every bounded
subset Ω̃ of RN . Therefore, by a simple variant of Lemma 7.4, we conclude that (vh)
is strongly convergent to v in H1(Ω̃) for every bounded subset Ω̃ ⊂ RN (actually,
as we will see, vh → v uniformly over compacts). Then it follows that the limit v
is a solution of the equation

−
N∑

i,j=1

Dj(aij(x̂, v)Div) +
1
2

N∑
i,j=1

Dsaij(x̂, v)DivDjv + V (x̂)v = g0(x, v) in RN

(7.49)
where g0(x, s) := χ(x)f(s) + (1− χ(x))f̃(s) for every x ∈ RN and s ∈ R+.

We now prove that v 6= 0. Let us set

dh(x) :=

{
V (xh + εhx)− g(x,vh(x))

vh(x) if vh(x) 6= 0

0 if vh(x) = 0,

Aj(x, s, ξ) :=
N∑
i=1

aij(xh + εhx, s)ξi for j = 1, . . . , N,

B(x, s, ξ) := dh(x)s,

C(x, s) :=
1
2

N∑
i,j=1

Dsaij(xh + εhx, s)Divh(x)Djvh(x)

for every x ∈ RN , s ∈ R+ and ξ ∈ RN . Taking into account the assumptions on
the coefficients aij(x, s), it results that

A(x, s, ξ) · ξ ≥ ν|ξ|2, |A(x, s, ξ)| ≤ c|ξ|, |B(x, s, ξ)| ≤ dh(x)|s|.

Moreover, by (7.10) we have

s ≥ R ⇒ C(x, s)s ≥ 0

for every x ∈ RN and s ∈ R+. By the growth condition on g, dh ∈ L
N

2−δ (B2%(0))
for every % > 0 and

S = sup
h
‖dh‖

L
N

2−δ (B2%(0))
≤ D% sup

h∈N
‖vh‖L2∗ (B2%(0)) < +∞

for some δ > 0 sufficiently small. Since div(A(x, vh, Dvh)) = B(x, vh, Dvh) +
C(x, vh) for every h ∈ N, by virtue of [122, Theorem 1 and Remark at p.261] there
exists a positive constant M(δ,N, c, %δS) and a radius % > 0, sufficiently small, such
that

sup
h∈N

max
x∈B%(0)

|vh(x)| ≤M(δ,N, c, %δS)(2%)−N/2 sup
h∈N

‖vh‖L2(B2%(0)) < +∞

so that (vh) is uniformly bounded in B%(0). Then, by [122, Theorem 8], (vh)
is bounded in some C1,α(B%/2(0)). Up to a subsequence this implies that (vh)
converges uniformly to v in B%/2(0). This yields v(0) = limh vh(0) = limh uεh

(xh) ≥
c > 0.

In a similar fashion one shows that vh → v uniformly over compacts.



166 MARCO SQUASSINA EJDE-2006/MON. 07

Step II. We prove that v actually solves the following equation

−
N∑

i,j=1

Dj(aij(x̂, v)Div) +
1
2

N∑
i,j=1

Dsaij(x̂, v)DivDjv + V (x̂)v = f(v) in RN .

(7.50)
In general the function χ of Step I is given by χ = χTΛ(bx), where TΛ(x̂) is the
tangent cone of Λ at x̂. On the other hand, since we may assume without loss of
generality that Λ is smooth, it results (up to a rotation) that χ(x) = χ{x1<0}(x)
for every x ∈ RN . In particular, v is a solution of the problem

−
N∑

i,j=1

Dj(aij(x̂, v)Div) +
1
2

N∑
i,j=1

Dsaij(x̂, v)DivDjv + V (x̂)v

= χ{x1<0}(x)f(v) + χ{x1>0}(x)f̃(v) in RN .

(7.51)

Let us first show that v(x) ≤ ` on {x1 = 0}. To this aim, let us use again Lemma 7.9,
by choosing this time

ϕ(x) := χ{x1<0}(x)f(v(x)) + χ{x1>0}(x)f̃(v(x)) for every x ∈ RN

L (s, ξ) :=
1
2

N∑
i,j=1

aij(x̂, s)ξiξj +
V (x̂)

2
s2 for every s ∈ R and ξ ∈ RN ,

h(x) := hk(x) =
(
T (
x

k
), 0, . . . , 0

)
for every x ∈ RN and k ≥ 1.

Then hk ∈ C1
c (RN ,RN ) and, since DxL (v,Dv) = 0, for every k ≥ 1, it results∫
RN

[1
k

N∑
i=1

DiT (
x

k
)D1vDξi

L (v,Dv)−D1T (
x

k
)
1
k

L (v,Dv)
]

=
∫

RN

T (
x

k
)ϕ(x, v)D1v.

Again by the Dominated Convergence Theorem, letting k → +∞, it results∫
RN

ϕ(x, v)D1v = 0,

that is, after integration by parts,∫
RN−1

[
F (v(0, x′))− F̃ (v(0, x′))

]
dx′ = 0.

Taking into account that F (s) ≥ F̃ (s) with equality only if s ≤ `, we get v(0, x′) ≤ `
for every x′ ∈ RN−1. To prove that actually v(x1, x

′) ≤ ` for every x1 > 0 and
x′ ∈ RN−1, we test (7.51) with

η(x) :=

{
0 if x1 < 0
(v(x1, x

′)− `)+ exp{ζ(v(x1, x
′))} if x1 > 0

where ζ(s) is as in (7.26) and then we argue as in Section 7.3 (see the computations
in formula (7.58)). In particular,

ϕ(x, v(x)) = f(v(x)) for every x ∈ RN , (7.52)

so that v is a nontrivial solution of (7.50).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 167

Step III If Jh : HV (Ωh) → R is as in (7.34), the function vh is a critical point
of Jh and Jh(vh) = εh

−NJεh
(uεh

) for every h ∈ N. Let us consider the functional
Jbx : H1(RN ) → R defined as

Jbx(u) :=
1
2

N∑
i,j=1

∫
RN

aij(x̂, u)DiuDju+
1
2

∫
RN

V (x̂)u2 −
∫

RN

F (u).

We now want to prove that

Jbx(v) ≤ lim inf
h

Jh(vh). (7.53)

Let us set for every h ∈ N and x ∈ Ωh

ξh(x) :=
1
2

N∑
i,j=1

aij(xh + εhx, vh)DivhDjvh +
1
2
V (xh + εhx)v2

h −G(xh + εhx, vh).

(7.54)
Since vh → v in H1 over compact sets, in view of (7.52), for every % > 0 one gets

lim
h

∫
B%(0)

ξh(x) =
1
2

∫
B%(0)

( N∑
i,j=1

aij(x̂, v)DivDjv + V (x̂)v2
)
−

∫
B%(0)

F (v).

Moreover, as v belongs to H1(RN ),

1
2

∫
B%(0)

( N∑
i,j=1

aij(x̂, v)DivDjv + V (x̂)v2
)
−

∫
B%(0)

F (v) = Jbx(v)− o(1)

as % → +∞. Therefore, it suffices to show that for every δ > 0 there exists % > 0
with

lim inf
h

∫
Ωh\B%(0)

ξh(x) ≥ −δ. (7.55)

Consider a function η% ∈ C∞(RN ) such that 0 ≤ η% ≤ 1, η% = 0 on B%−1(0), η% = 1
on RN \B%(0) and |Dη%| ≤ c. Let us set for every h ∈ N,

βh(%) :=
N∑

i,j=1

∫
B%(0)\B%−1(0)

aij(xh + εhx, vh)DivhDj(η%vh)

+
1
2

N∑
i,j=1

∫
B%(0)\B%−1(0)

Dsaij(xh + εhx, vh)η%vhDivhDjvh

+
∫
B%(0)\B%−1(0)

V (xh + εhx)v2
hη% −

∫
B%(0)\B%−1(0)

g(xh + εhx, vh)η%vh.

After some computations, in view of (7.9) and (7.54), one gets

− βh(%) + J ′h(vh)(η%vh)

≤ (γ + 2)
∫

Ωh\B%(0)

ξh(x)−
γ

2

∫
Ωh\B%(0)

V (xh + εhx)v2
h

+ (γ + 2)
∫

Ωh\B%(0)

G(xh + εhx, vh)−
∫

Ωh\B%(0)

g(xh + εhx, vh)vh



168 MARCO SQUASSINA EJDE-2006/MON. 07

Notice that, by virtue of (7.19), for % large enough, setting Λh = ε−1
h (Λ− xh), we

get

− γ

2

∫
Λh\B%(0)

V (xh + εhx)v2
h + (γ + 2)

∫
Λh\B%(0)

G(xh + εhx, vh)

−
∫

Λh\B%(0)

g(xh + εhx, vh)vh

≤ −(ϑ− 2− γ)
∫

Λh\B%(0)

G(xh + εhx, vh) ≤ 0.

Analogously, in view of (7.20), we obtain

− γ

2

∫
Ωh\(B%(0)∪Λh)

V (xh + εhx)v2
h + (γ + 2)

∫
Ωh\(B%(0)∪Λh)

G(xh + εhx, vh)

−
∫

Ωh\(B%(0)∪Λh)

g(xh + εhx, vh)vh

≤ −γ
2

∫
Ωh\(B%(0)∪Λh)

V (xh + εhx)v2
h +

γ

2k

∫
Ωh\(B%(0)∪Λh)

V (xh + εhx)v2
h ≤ 0.

Therefore, since J ′h(vh)(η%vh) = 0 for every h ∈ N and

lim sup
h

βh(%) = o(1) as %→ +∞,

inequality (7.55) follows and thus (7.53) holds true.
Step IV. In this step we get the desired contradiction. By combining Lemma 7.6
with the inequality (7.53), one immediately gets

Jbx(v) ≤ c̄ = inf
γ∈P0

sup
t∈[0,1]

J0(γ(t)). (7.56)

Since v is a nontrivial solution of (7.50), by applying Lemma 7.10 with

µ = V (x̂), ν′ = ν, R′ = R, bij(s) = aij(x̂, s),

being P̂ ⊂ P0, V (x̂) > V (x0) and, by (7.12),
N∑

i,j=1

aij(x̂, s)ξiξj ≥
N∑

i,j=1

aij(x0, s)ξiξj for every s ∈ R+ and ξ ∈ RN ,

it follows that

Jbx(v) ≥ inf
γ∈ cP sup

t∈[0,1]

Jbx(γ(t)) > inf
γ∈P0

sup
t∈[0,1]

J0(γ(t)) = c̄, (7.57)

which contradicts (7.56). �

7.3. Proof of the main result. We are now ready to prove Theorem 7.1.
Step I. We prove that (a) holds. By Lemma 7.11 there exists ε0 > 0 such that

uε(x) < ` for every ε ∈ (0, ε0) and x ∈ ∂Λ.

Then, since uε ∈ HV (Ω), for every ε ∈ (0, ε0), if ζ is defined as in (7.26), the
function

vε(x) :=

{
0 if x ∈ Λ
(uε(x)− `)+ exp{ζ(uε(x))} if x ∈ Ω \ Λ



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 169

belongs to H1
0 (Ω) and it is an admissible test for the equation

−ε2
N∑

i,j=1

Dj(aij(x, uε)Diuε) +
ε2

2

N∑
i,j=1

Dsaij(x, uε)DiuεDjuε + V (x)uε = g(x, uε).

After some computations, one obtains

ε2
N∑

i,j=1

∫
Ω\Λ

aij(x, uε)Di

[
(uε − `)+

]
Dj

[
(uε − `)+

]
exp{ζ(uε)}

+ ε2
N∑

i,j=1

∫
Ω\Λ

[
1
2
Dsaij(x, uε) + ζ ′(uε)aij(x, uε)

]
DiuεDjuε(uε − `)+ exp{ζ(uε)}

+
∫

Ω\Λ
Φε(x)

[
(uε − `)+

]2 exp{ζ(uε)}+
∫

Ω\Λ
Φε(x)`(uε − `)+ exp{ζ(uε)} = 0,

(7.58)
where Φε : Ω → R is the function given by

Φε(x) := V (x)− g(x, uε(x))
uε(x)

.

Notice that, by virtue of condition (7.20), one has

Φε(x) > 0 for every x ∈ Ω \ Λ.

Therefore, taking into account (7.27), all the terms in (7.58) must be equal to zero.
We conclude that (uε − `)+ = 0 on Ω \ Λ, namely,

uε(x) ≤ ` for every ε ∈ (0, ε0) and x ∈ Ω \ Λ. (7.59)

Hence, by Proposition 7.3, uε is a positive solution of the original problem (7.15).
Moreover, by virtue of (7.10), using again the argument at the end of Step I of
Lemma 7.11 it results that uε ∈ L∞loc(Ω), which, by the regularity results of [90],
yields uε ∈ C(Ω). Notice that by arguing in a similar fashion testing with

vε(x) :=

{
0 if x ∈ Λ
(uε(x)− sup∂Λ uε)+ exp{ζ(uε(x))} if x ∈ Ω \ Λ

it results uε → 0 uniformly outside Λ.
Step II. We prove that (b) holds. If xε denotes the maximum of uε in Λ, since
uε → 0 uniformly outside Λ, it results that uε(xε) = supΩ uε. By arguing as at the
end of Step I of Lemma 7.11, setting vε(x) = uε(xε+εx) it results that the sequence
(vε(0)) is bounded in R. Then there exists σ′ > 0 such that uε(xε) = vε(0) ≤ σ′.
Assume now by contradiction that uε(xε) ≤ σ for some ε ∈ (0, ε0). Then, taking
into account the definition of σ and that uε → 0 uniformly outside Λ, it holds (with
strict inequality in some subset of Ω)

V (x)− f(uε(x))
uε(x)

≥ 0 for every x ∈ Ω. (7.60)

Let ζ : R+ → R be the map defined in (7.26). Then the function uε exp{ζ(uε)} can
be chosen as an admissible test in the equation

−ε2
N∑

i,j=1

Dj(aij(x, uε)Diuε) +
ε2

2

N∑
i,j=1

Dsaij(x, uε)DiuεDjuε + V (x)uε = f(uε).



170 MARCO SQUASSINA EJDE-2006/MON. 07

After some computations, one obtains

ε2
N∑

i,j=1

∫
Ω

aij(x, uε)DiuεDjuε exp{ζ(uε)}

+ ε2
N∑

i,j=1

∫
Ω

[
1
2
Dsaij(x, uε) + ζ ′(uε)aij(x, uε)

]
DiuεDjuεuε exp{ζ(uε)}

+
∫

Ω

(
V (x)− f(uε)

uε

)
u2
ε exp{ζ(uε)} = 0.

(7.61)

Then, by (7.8), (7.27) and (7.60) all the terms in equation (7.61) must be equal to
zero, namely uε ≡ 0, which is not possible. Then uε(xε) ≥ σ for every ε ∈ (0, ε0)
and by (7.48) we also get d(xε,M ) → 0 as ε→ 0.

Step III. We prove that (c) holds. Assume by contradiction that there exists % > 0,
δ > 0, εh → 0 and yh ∈ Λ \B%(xεh

) such that

lim sup
h

uεh
(yh) ≥ δ. (7.62)

Then, arguing as in Lemma 7.11, we can assume that yh → y, xεh
→ ỹ and

vh(y) := uεh
(yh + εhy) → v, ṽh(y) := uεh

(xεh
+ εhy) → ṽ strongly in H1

loc(RN ),
where v is a solution of

−
N∑

i,j=1

Dj(aij(y, v)Div) +
1
2

N∑
i,j=1

Dsaij(y, v)DivDjv + V (y)v = f(v) in RN

and ṽ is a solution of

−
N∑

i,j=1

Dj(aij(ỹ, v)Div) +
1
2

N∑
i,j=1

Dsaij(ỹ, v)DivDjv + V (ỹ)v = f(v) in RN .

Observe that v 6= 0 and ṽ 6= 0. Indeed, arguing as in Step I of Lemma 7.11 it results
that (vh) and (ṽh) converge uniformly in a neighborhood of zero, so that from (7.62)
and uεh

(xεh
) ≥ σ we get v(0) ≥ δ and ṽ(0) ≥ σ. Now, setting zh := xεh

−yh

εh
and

ξh(y) :=
1
2

N∑
i,j=1

aij(yh + εhy, vh)DivhDjvh +
1
2
V (yh + εhy)v2

h −G(yh + εhy, vh),

if ψ ∈ C∞(R), 0 ≤ ψ ≤ 1, ψ(s) = 0 for s ≤ 1 and ψ(s) = 1 for s ≥ 2, arguing as in
Lemma 7.11 by testing the equation satisfied by vh with

ϕh,R(y) = vh(y)
[
ψ

( |y|
R

)
+ ψ

( |y − zh|
R

)
− 1

]
,

taking into account that

lim
h

∣∣∣ ∫
B2R(0)∪B2R(zh)\(BR(0)∪BR(zh))

ξh(y)
∣∣∣ = o(1)

as R→ +∞, it turns out that for every δ > 0 there exists R > 0 with

lim inf
h

∫
Ωh\(BR(0)∪BR(zh))

ξh(y) ≥ −δ.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 171

Moreover, for every R > 0, we have

lim inf
h

∫
BR(0)∪BR(zh)

ξh(y) = lim inf
h

∫
BR(0)

1
2

N∑
i,j=1

aij(yh + εhy, vh)DivhDjvh

+
1
2
V (yh + εhy)v2

h −G(yh + εhy, vh)

+ lim inf
h

∫
BR(0)

1
2

N∑
i,j=1

aij(xεh
+ εhy, ṽh)DiṽhDj ṽh

+
1
2
V (xεh

+ εhy)ṽ2
h −G(xεh

+ εhy, ṽh)

=
∫
BR(0)

1
2

N∑
i,j=1

aij(y, v)DivDjv +
1
2
V (y)v2 − F (v)

+
∫
BR(0)

1
2

N∑
i,j=1

aij(ỹ, ṽ)DiṽDj ṽ +
1
2
V (ỹ)ṽ2 − F (ṽ).

Therefore, we deduce that

lim inf
h

εh
−NJεh

(uεh
) = lim inf

h

∫
Ωh

ξh(y) ≥ Jy(v) + Jey(ṽ).
Let by and bey be the Mountain-Pass values of Jy and Jey. By Lemma 7.10, (7.11)
and (7.12) we have Jy(v) ≥ by ≥ c̄ and Jey(ṽ) ≥ bey ≥ c̄. Therefore we conclude that

lim inf
h

εh
−NJεh

(uεh
) ≥ 2c̄,

which contradicts Lemma 7.6.
Step IV. We prove that (d) holds. By Corollary 7.8, we have ‖uε‖HV (Ω) → 0. In
particular, uε → 0 in Lq(Ω) for every 2 ≤ q ≤ 2∗. As a consequence uε → 0 in
Lq(Ω) also for every q > 2∗. Indeed, if q > 2∗, we have∫

Ω

|uε|q =
∫

Ω

|uε|q−2∗ |uε|2
∗
≤ σ′

q−2∗
∫

Ω

|uε|2
∗
→ 0

as ε→ 0. The proof is now complete. �

7.4. A few related open problems. We quote here a few (open) problems related
to the main result.

Problem 7.12. Under suitable assumptions, does a Gidas-Ni-Nirenberg [76] type
result (radial symmetry) hold for the solutions of autonomous equations of the type

−
N∑

i,j=1

Dj(bij(u)Diu) +
1
2

N∑
i,j=1

b′ij(u)DiuDju = h(u) in RN? (7.63)

Problem 7.13. Under suitable assumptions on bij and h, is it possible to prove,
as in the semi-linear case, a uniqueness result for the solutions of equation (7.63)?

Problem 7.14. Is it true that for each ε > 0 the solution uε of problem (7.15)
admits a unique maximum point inside Λ?

Problem 7.15. Is it true that the solutions uε of problem (7.15) decay exponen-
tially as for the semi-linear case (see formula (7.17))?



172 MARCO SQUASSINA EJDE-2006/MON. 07

8. The Singularly Perturbed Case, II

In this section we turn to a more delicate situation, namely the study of the
multi-peak case, also for possibly degenerate operators. We refer the reader to [74].
Some parts of this publication has been slightly modified to give this collection a
more uniform appearance.

Assume that V : RN → R is a C1 function and there exists a positive constant
α such that

V (x) ≥ α for every x ∈ RN . (8.1)
Moreover let Λ1, . . . ,Λk be k disjoint compact subsets of Ω and xi ∈ Λi with

V (xi) = min
Λi

V < min
∂Λi

V, i = 1, . . . , k. (8.2)

Let us set for all i = 1, . . . , k

Mi :=
{
x ∈ Λi : V (x) = V (xi)

}
. (8.3)

Let 1 < p < N , p∗ = Np
N−p and let WV (Ω) be the weighted Banach space

WV (Ω) :=
{
u ∈W 1,p

0 (Ω) :
∫

Ω

V (x)|u|p < +∞
}

endowed with the natural norm ‖u‖pWV
:=

∫
Ω
|Du|p +

∫
Ω
V (x)|u|p. For all A,B ⊂

RN , let us denote their distance by dist(A,B).
The following is the first of our main results.

Theorem 8.1. Assume that (8.1) and (8.2) hold and let 1 < p ≤ 2, p < q < p∗.
Then there exists ε0 > 0 such that, for every ε ∈ (0, ε0), there exist uε in

WV (Ω) ∩ C1,β
loc (Ω) and k points xε,i ∈ Λi satisfying the following properties:

(a) uε is a weak solution of the problem

−εp∆pu+ V (x)up−1 = uq−1 in Ω
u > 0 in Ω
u = 0 on ∂Ω;

(8.4)

(b) there exist σ, σ′ ∈]0,+∞[ such that for every i = 1, . . . , k we have

uε(xε,i) = sup
Λi

uε, σ < uε(xε,i) < σ′, lim
ε→0

dist(xε,i,Mi) = 0

where Mi is as in (8.3);
(c) for every r < min{dist(Mi,Mj) : i 6= j} we have

lim
ε→0

‖uε‖L∞(Ω\
Sk

i=1 Br(xε,i))
= 0 ;

(d) it results
lim
ε→0

‖uε‖WV
= 0.

Moreover, if k = 1 the assertions hold for every 1 < p < N .

Actually, this result will follow by a more general achievement involving a larger
class of quasi-linear operators. Before stating it, we make a few assumptions. As-
sume that 1 < p < N , f ∈ C1(R+) and there exist p < q < p∗ and p < ϑ ≤ q
with

lim
s→0+

f(s)
sp−1

= 0, lim
s→+∞

f(s)
sq−1

= 0, (8.5)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 173

0 < ϑF (s) ≤ f(s)s for every s ∈ R+, (8.6)

where F (s) =
∫ s
0
f(t) dt for every s ∈ R+.

The function j(x, s, ξ) : Ω × R+ × RN → R is continuous in x and of class
C1 with respect to s and ξ, the function {ξ 7→ j(x, s, ξ)} is strictly convex and
p-homogeneous and there exist two positive constants c1, c2 with

|js(x, s, ξ)| ≤ c1|ξ|p, |jξ(x, s, ξ)| ≤ c2|ξ|p−1 (8.7)

for a.e. x ∈ Ω and every s ∈ R+, ξ ∈ RN (js and jξ denote the derivatives of j with
respect of s and ξ respectively). Let R, ν > 0 and 0 < γ < ϑ− p with

j(x, s, ξ) ≥ ν|ξ|p, (8.8)

js(x, s, ξ)s ≤ γj(x, s, ξ) (8.9)

a.e. in Ω, for every s ∈ R+ and ξ ∈ RN , and

js(x, s, ξ) ≥ 0 for every s ≥ R (8.10)

a.e. in Ω and for every ξ ∈ RN . For every fixed x̄ ∈ Ω, the limiting equation

−div(jξ(x̄, u,Du)) + js(x̄, u,Du) + V (x̄)up−1 = f(u) in RN (8.11)

admits a unique positive solution (up to translations). Finally, we assume that

j(xi, s, ξ) = min
x∈Λi

j(x, s, ξ), i = 1, . . . , k (8.12)

for every s ∈ R+ and ξ ∈ RN , where the xis are as in (8.2).
We point out that assumptions (8.1), (8.2), (8.5) and (8.6) are the same as in

[62, 63]. Conditions (8.7)-(8.10) are natural assumption, already used, throughout
this monograph.

The following result is an extension of Theorem 8.1.

Theorem 8.2. Assume that (8.1), (8.2), (8.5), (8.6), (8.7), (8.8), (8.9), (8.10),
(8.11), (8.12) hold. Then there exists ε0 > 0 such that, for every ε ∈ (0, ε0),
there exist uε in WV (Ω) ∩ C1,β

loc (Ω) and k points xε,i ∈ Λi satisfying the following
properties:

(a) uε is a weak solution of the problem

−εp div(jξ(x, u,Du)) + εpjs(x, u,Du) + V (x)up−1 = f(u) in Ω
u > 0 in Ω
u = 0 on ∂Ω ;

(8.13)

(b) there exist σ, σ′ ∈]0,+∞[ such that for every i = 1, . . . , k we have

uε(xε,i) = sup
Λi

uε, σ < uε(xε,i) < σ′, lim
ε→0

dist(xε,i,Mi) = 0

where Mi is as in (8.3);
(c) for every r < min{dist(Mi,Mj) : i 6= j} we have

lim
ε→0

‖uε‖L∞(Ω\
Sk

i=1 Br(xε,i))
= 0 ;

(d) it results
lim
ε→0

‖uε‖WV
= 0.



174 MARCO SQUASSINA EJDE-2006/MON. 07

Notice that if k = 1 assumption (8.11) can be dropped: in fact following the
arguments of [131] it is possible to prove that the previous result holds without
any uniqueness assumption, which instead, as in the semi-linear case, seems to be
necessary for the case k > 1. This holds true for the p-Laplacian problem (8.4) and
for more general situation we refer the reader to [123].

Various difficulties arise in comparison with the semi-linear framework (see also
Section 5 of [131]). To study the concentration properties of uε inside the Λis
(see Section 8.3), inspired by the recent work of Jeanjean and Tanaka [82], we
make a repeated use of a Pucci-Serrin type identity [59] which has turned out to
be a very powerful tool (see Section 8.2). It has to be pointed out that, in our
possibly degenerate setting, we cannot hope to have C2 solutions, but at most C1,β

solutions (see [65, 141]). Therefore, the classical Pucci-Serrin identity [116] is not
applicable in our framework. On the other hand, it has been recently shown in [59]
that, under minimal regularity assumptions, the identity holds for locally Lipschitz
solutions, provided that the operator is strictly convex in the gradient, which, from
our viewpoint, is a very natural requirement (see Theorem 8.6). Under uniqueness
assumptions this identity has also turned out to be useful in characterizing the exact
energy level of the solution of (8.11). More precisely, we prove that (8.11) admits
a least energy solution having the Mountain-Pass energy level (see Theorem 8.7).

8.1. Penalization and compactness. In this section, following the approach of
del Pino and Felmer [63], we define a suitable penalization of the functional Iε :
WV (Ω) → R associated with the problem (8.13),

Iε(u) := εp
∫

Ω

j(x, u,Du) +
1
p

∫
Ω

V (x)|u|p −
∫

Ω

F (u).

By the growth condition on j, it is easily seen that Iε is a continuous functional.
Let α > 0 be as in (8.1) and consider the positive constant

` := sup
{
s > 0 :

f(t)
tp−1

≤ α

κ
for every 0 ≤ t ≤ s

}
(8.14)

for some fixed κ > ϑ/(ϑ− p). We define the function f̃ : R+ → R by setting

f̃(s) :=

{
α
κ s

p−1 if s > `

f(s) if 0 ≤ s ≤ `

and the map g : Ω× R+ → R as

g(x, s) := 1Λ(x)f(s) + (1− 1Λ(x))f̃(s), Λ =
k⋃
i=1

Λi

for a.e. x ∈ Ω and every s ∈ R+. The function g(x, s) is measurable in x, of class
C1 in s and it satisfies the following properties:

lim
s→+∞

g(x, s)
sq−1

= 0, lim
s→0+

g(x, s)
sp−1

= 0 uniformly in x, (8.15)

0 < ϑG(x, s) ≤ g(x, s)s for x ∈ Λ and s ∈ R+, (8.16)

0 ≤ pG(x, s) ≤ g(x, s)s ≤ 1
κ
V (x)sp for x ∈ Ω \ Λ and s ∈ R+, (8.17)

where we have set G(x, s) :=
∫ s
0
g(x, τ) dτ .



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 175

Without loss of generality, we may assume that

g(x, s) = 0 for a.e. x ∈ Ω and every s < 0, (8.18)

j(x, s, ξ) = j(x, 0, ξ) for every x ∈ Ω, s < 0 and ξ ∈ RN . (8.19)

Let now Jε : WV (Ω) → R be the functional defined as

Jε(u) := εp
∫

Ω

j(x, u,Du) +
1
p

∫
Ω

V (x)|u|p −
∫

Ω

G(x, u).

If x̄ is in one of the Λis, we also consider the “limit” functionals on W 1,p(RN ),

Ix̄(u) :=
∫

RN

j(x̄, u,Du) +
1
p

∫
RN

V (x̄)|u|p −
∫

RN

F (u) (8.20)

whose positive critical points solve equation (8.11). We denote by cx̄ the Mountain-
Pass value of Ix̄, namely

cx̄ := inf
γ∈Px̄

sup
t∈[0,1]

Ix̄(γ(t)), (8.21)

Px̄ :=
{
γ ∈ C([0, 1],W 1,p(RN )) : γ(0) = 0, Ix̄(γ(1)) < 0

}
. (8.22)

We set ci := cxi for every i = 1, . . . , k. Considering σi > 0 such that
k∑
i=1

σi <
1
2

min
{
ci : i = 1, . . . , k

}
,

we claim that, up to making Λis smaller, we may assume that

ci ≤ cx̄ ≤ ci + σi for all x̄ ∈ Λi. (8.23)

In fact ci ≤ cx̄ follows because xi is a minimum of V in Λi and (8.12) holds. On
the other hand, let us consider x̄h → xi such that limh cx̄h

= lim supx̄→xi
cx̄. Let

γ ∈ Px̄ be such that maxτ∈[0,1] Ixi(γ(τ)) ≤ ci + σi. Since Ix̄h
→ Ixi uniformly on

γ, we have that for h large enough, γ ∈ Px̄h
and there exists τh ∈ [0, 1] such that

cx̄h
≤ Ix̄h

(γ(τh)) ≤ Ixi(γ(τh)) + o(1) ≤ ci + σi + o(1).

We deduce that lim supx̄→xi
cx̄ ≤ ci + σi so that the claim is proved.

If Λ̂i denote mutually disjoint open sets compactly containing Λi, we introduce
the functionals Jε,i : W 1,p(Λ̂i) → R as

Jε,i(u) := εp
∫

Λ̂i

j(x, u,Du) +
1
p

∫
Λ̂i

V (x)|u|p −
∫

Λ̂i

G(x, u) (8.24)

for every i = 1, . . . , k.
Finally, let us define the penalized functional Eε : WV (Ω) → R by setting

Eε(u) := Jε(u) + Pε(u), (8.25)

Pε(u) := M
k∑
i=1

(
(Jε,i(u)+)1/2 − εN/2(ci + σi)1/2

)2

+
, (8.26)

where M > 0 is chosen so that

M >
c1 + · · ·+ ck

mini=1,...,k

{
(2ci)1/2 − (ci + σi)1/2

} .
The functionals Jε, Jε,i and Eε are merely continuous.



176 MARCO SQUASSINA EJDE-2006/MON. 07

The next result provides the link between the critical points of Eε and the weak
solutions of the original problem.

Proposition 8.3. Let uε ∈ WV (Ω) be any critical point of Eε and assume that
there exists a positive number ε0 such that the following conditions hold

uε(x) < ` for every ε ∈ (0, ε0) and x ∈ Ω \ Λ, (8.27)

ε−NJε,i(uε) < ci + σi for every ε ∈ (0, ε0) and i = 1, . . . , k. (8.28)

Then, for every ε ∈ (0, ε0), uε is a solution of (8.13).

Proof. Let ε ∈ (0, ε0). By condition (8.28) and the definition of P (uε), uε is actually
a critical point of Jε. In view of (a) of Proposition 2.25, uε is a weak solution of

−εp div(jξ(x, u,Du)) + εpjs(x, u,Du) + V (x)|u|p−2u = G(x, u).

Moreover, by (8.27) and the definition of f̃ , it results G(x, uε(x)) = F (uε(x)) for
a.e. x ∈ Ω. By (8.18) and (8.19) and arguing as in the proof of [130, Lemma 1],
one gets uε > 0 in Ω. Thus uε is a solution of (8.13). �

The next Lemma is a variant of a local compactness property for bounded con-
crete Palais-Smale sequences (cf. [130, Theorem 2 and Lemma 3]; see also [48]).

Lemma 8.4. Assume that (8.7), (8.8), (8.10) hold and let (ψh) ⊂ L∞(RN ) bounded
with ψh(x) ≥ λ > 0. Let ε > 0 and assume that (uh) ⊂ W 1,p(RN ) is a bounded
sequence such that

〈wh, ϕ〉 = εp
∫

RN

ψh(x)jξ(x, uh, Duh) ·Dϕ+ εp
∫

RN

ψh(x)js(x, uh, Duh)ϕ

for every ϕ ∈ C∞c (RN ), where (wh) is strongly convergent in W−1,p′(Ω̃) for a given
bounded domain Ω̃ of RN . Then (uh) admits a strongly convergent subsequence in
W 1,p(Ω̃).

Since Ω may be unbounded, in general the original functional Iε does not satisfy
the concrete Palais-Smale condition. In the following Lemma we prove that, instead,
for every ε > 0 the functional Eε satisfies it at every level c ∈ R.

Lemma 8.5. Assume that conditions (8.1), (8.5), (8.6), (8.7), (8.8), (8.9), (8.10)
hold. Let ε > 0.

Then Eε satisfies the concrete Palais-Smale condition at every level c ∈ R.

Proof. Let (uh) ⊂ WV (Ω) be a concrete Palais-Smale sequence for Eε at level c.
We divide the proof into two steps:
Step I. We prove that (uh) is bounded in WV (Ω). From (8.16) and (8.17), we get

ϑεp
∫

Ω

j(x, uh, Duh) +
ϑ

p

∫
Ω

V (x)|uh|p

≤
∫

Λ

g(x, uh)uh +
ϑ

pκ

∫
Ω\Λ

V (x)|uh|p + ϑJε(uh)
(8.29)

for every h ∈ N. Moreover, for every h ∈ N we can compute J ′ε(uh)(uh); in view of
(8.17) we obtain∫

Λ

g(x, uh)uh + J ′ε(uh)[uh]



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 177

≤ εp
∫

Ω

jξ(x, uh, Duh) ·Duh + εp
∫

Ω

js(x, uh, Duh)uh +
∫

Ω

V (x)|uh|p

for every h ∈ N. Notice that by (8.9) and the p-homogeneity of the map {ξ 7→
j(x, s, ξ)}, it results

js(x, uh, Duh)uh ≤ γj(x, uh, Duh),

jξ(x, uh, Duh) ·Duh = pj(x, uh, Duh)

for every h ∈ N. Therefore,∫
Λ

g(x, uh)uh + J ′ε(uh)[uh] ≤ (γ + p)εp
∫

Ω

j(x, uh, Duh) +
∫

Ω

V (x)|uh|p (8.30)

for every h ∈ N. In view of (8.8), by combining inequalities (8.29) and (8.30) one
gets

min
{

(ϑ− γ − p) νεp,
ϑ

p
− ϑ

pκ
− 1

} ∫
Ω

(
|Duh|p + V (x)|uh|p

)
≤ ϑJε(uh)− J ′ε(uh)[uh]

(8.31)

for every h ∈ N. In a similar fashion, arguing on the functionals Jε,i, it results

min
{

(ϑ− γ − p) νεp,
ϑ

p
− ϑ

pκ
− 1

} ∫
Λ̂i

(
|Duh|p + V (x)|uh|p

)
≤ ϑJε,i(uh)− J ′ε,i(uh)[uh] for every h ∈ N and i = 1, . . . , k.

(8.32)

In particular, notice that one obtains

ϑ̄Jε,i(uh)− J ′ε,i(uh)[uh] ≥ 0 for every h ∈ N and i = 1, . . . , k

and every γ + p < ϑ̄ < ϑ. Then, after some computations, one gets

ϑ̄Pε(uh)− P ′ε(uh)[uh]

≥ −ϑ̄MεN/2
k∑
i=1

(ci + σi)1/2
(
(Jε,i(uh)+)1/2 − εN/2(ci + σi)1/2

)
+

≥ −CεN/2Pε(uh)1/2

which implies, by Young’s inequality, the existence of a constant d > 0 such that

ϑPε(uh)− P ′ε(uh)[uh] ≥ −dεN (8.33)

for every h ∈ N. By combining (8.31) with (8.33), since

Eε(uh) = c+ o(1), E′ε(uh)[uh] = o(‖uh‖WV
)

as h→ +∞, one obtains∫
Ω

(
|Duh|p + V (x)|uh|p

)
≤ ϑc+ dεN

min
{

(ϑ− γ − p) νεp, ϑp −
ϑ
pκ − 1

} + o(‖uh‖WV
) + o(1)

(8.34)

as h→ +∞, which yields the boundedness of (uh) in WV (Ω).



178 MARCO SQUASSINA EJDE-2006/MON. 07

Step II. By virtue of Step I, there exists u ∈WV (Ω) such that, up to a subsequence,
(uh) weakly converges to u inWV (Ω). Let us now prove that actually (uh) converges
strongly to u in WV (Ω). If we define for every h ∈ N the weights

θh,i = M
[
(Jε,i(uh)+)1/2 − εN/2(ci + σi)1/2

]
+
(Jε,i(uh)+)−1/2, i = 1, . . . , k

and put θh(x) =
∑k
i=1 θh,i1Λ̂i

(x) with 0 ≤ θh,i ≤ M . After a few computations,
one gets

〈wh, ϕ〉 = εp
∫

Ω

(1 + θh)jξ(x, uh, Duh) ·Dϕ+ εp
∫

Ω

(1 + θh)js(x, uh, Duh)ϕ

for every ϕ ∈ C∞c (Ω), where

wh = (1 + θh)g(x, uh)− (1 + θh)V (x)|uh|p−2uh + ξh,

with ξh → 0 strongly in W−1,p′(Ω). Since, up to a subsequence, (wh) strongly
converges to w := (1 + θ̄)g(x, u) − (1 + θ̄)V (x)|u|p−2u in W−1,p′(B%) for every
% > 0, by applying Lemma 8.4 with Ω̃ = B% ∩ Ω and ψh(x) = 1 + θh(x), it suffices
to show that, for every δ > 0, there exists % > 0 such that

lim sup
h

∫
Ω\B%

(
|Duh|p + V (x)|uh|p

)
< δ. (8.35)

Consider a cut-off function χ% ∈ C∞(RN ) with 0 ≤ χ% ≤ 1, χ% = 0 on B%/2, χ% = 1
on RN \B% and |Dχ%| ≤ a/% for some a > 0. By taking % large enough, we have

k⋃
i=1

Λ̂i ∩ supt(χ%) = ∅. (8.36)

Let now ζ : R → R be the map defined by

ζ(s) :=


0 if s < 0
M̄s if 0 ≤ s < R

M̄R if s ≥ R,

(8.37)

being R > 0 the constant defined in (8.10) and M̄ a positive number (which exists
by the growths (8.7) and (8.8)) such that

|js(x, s, ξ)| ≤ pM̄j(x, s, ξ) (8.38)

for every x ∈ Ω, s ∈ R and ξ ∈ RN . Notice that, by combining (8.10) and (8.38),
we obtain

js(x, s, ξ) + pζ ′(s)j(x, s, ξ) ≥ 0 for every x ∈ Ω, s ∈ R and ξ ∈ RN . (8.39)

By (8.36) it is easily proved that P ′ε(uh)(χ%uhe
ζ(uh)) = 0 for every h. Therefore,

since the sequence (χ%uheζ(uh)) is bounded in WV (Ω), taking into account (8.39)
and (8.19) we obtain

o(1) = J ′ε(uh)(χ%uhe
ζ(uh))

= εp
∫

Ω

jξ(x, uh, Duh) ·Duhχ%eζ(uh)

+ εp
∫

Ω

jξ(x, uh, Duh) ·Dχ%uheζ(uh)

+ εp
∫

Ω

[
js(x, uh, Duh) + pζ ′(uh)j(x, uh, Duh)

]
uhχ%e

ζ(uh)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 179

+
∫

Ω

V (x)|uh|pχ%eζ(uh) −
∫

Ω

g(x, uh)uhχ%eζ(uh)

≥
∫

Ω

(
pεpj(x, uh, Duh) + V (x)|uh|p

)
χ%e

ζ(uh)

+ εp
∫

Ω

jξ(x, uh, Duh) ·Dχ%uheζ(uh) −
∫

Ω

g(x, uh)uhχ%eζ(uh)

as h→ +∞. Therefore, in view of (8.17) and (8.36), it results

o(1) ≥
∫

Ω

(
pεpν|Duh|p + V (x)|uh|p

)
χ%e

ζ(uh)

+ εp
∫

Ω

jξ(x, uh, Duh) ·Dχ%uheζ(uh) − 1
κ

∫
Ω

V (x)|uh|pχ%eζ(uh)

as h→ +∞ for % large enough. Since by (8.7) we have∣∣∣ ∫
Ω

jξ(x, uh, Duh) ·Dχ%uheζ(uh)
∣∣∣ ≤ C

%
‖Duh‖p−1

p ‖uh‖p ≤
C̃

%
,

there exists a positive constant C ′ such that

lim sup
h

∫
Ω\B%

(
|Duh|p + V (x)|uh|p

)
≤ C ′

%

which yields (8.35). The proof is now complete. �

8.2. Two consequences of the Pucci-Serrin identity. Let L : RN×R×RN →
R be a function of class C1 such that the function {ξ 7→ L (x, s, ξ)} is strictly convex
for every (x, s) ∈ RN × R, and let ϕ ∈ L∞loc(RN ).

We now recall a Pucci-Serrin variational identity for locally Lipschitz continuous
solutions of a general class of Euler equations, recently obtained in [59]. Notice that
the classical identity [116] is not applicable here, since it requires the C2 regularity
of the solutions while in our degenerate setting the maximal regularity is C1,β

loc (see
[65, 141]).

Theorem 8.6. Let u : RN → R be a locally Lipschitz solution of

−div (DξL (x, u,Du)) +DsL (x, u,Du) = ϕ in D ′(RN ).

Then for every h ∈ C1
c (RN ,RN ),

N∑
i,j=1

∫
RN

Dih
jDξi

L (x, u,Du)Dju

−
∫

RN

[
(div h)L (x, u,Du) + h ·DxL (x, u,Du)

]
=

∫
RN

(h ·Du)ϕ .

(8.40)

We want to derive two important consequences of the previous variational iden-
tity. In the first we show that the Mountain-Pass value associated with a large class
of elliptic autonomous equations is the minimal among other nontrivial critical val-
ues.

Theorem 8.7. Let x̄ ∈ RN and assume that conditions (8.1), (8.5), (8.6), (8.7),
(8.8), (8.9), (8.10) hold. Then the equation

−div(jξ(x̄, u,Du)) + js(x̄, u,Du) + V (x̄)up−1 = f(u) in RN (8.41)



180 MARCO SQUASSINA EJDE-2006/MON. 07

admits a least energy solution u ∈W 1,p(RN ), that is

Ix̄(u) = inf
{
Ix̄(w) : w ∈W 1,p(RN ) \ {0} is a solution of (8.41)

}
,

where Ix̄ is as in (8.20). Moreover, Ix̄(u) = cx̄, that is u is at the Mountain-Pass
level.

Proof. We divide the proof into two steps. Step I. Let u be any nontrivial solution
of (8.41), and let us prove that Ix̄(u) ≥ cx̄. By the assumptions on V and f , it
is readily seen that there exist %0 > 0 and δ0 > 0 such that Ix̄(v) ≥ δ0 for every
v ∈ W 1,p(RN ) with ‖v‖1,p = %0. In particular Ix̄ has a Mountain-Pass geometry.
As we will see, Px̄ 6= ∅, so that cx̄ is well defined. Let now u be a positive solution
of (8.41) and consider the dilation path

γ(t)(x) :=

{
u(x/t) if t > 0
0 if t = 0.

Notice that ‖γ(t)‖p1,p = tN−p‖Du‖pp + tN‖u‖pp for every t ∈ R+, which implies that
the curve γ belongs to C(R+,W 1,p(RN )). For the sake of simplicity, we consider
the continuous function H : R+ → R defined by

H(s) =
∫ s

0

h(t) dt, where h(s) = −V (x̄)sp−1 + f(s).

For every t ∈ R+ it results that

Ix̄(γ(t)) =
∫

RN

j(x̄, γ(t), Dγ(t))−
∫

RN

H(γ(t))

= tN−p
∫

RN

j(x̄, u,Du)− tN
∫

RN

H(u)

which yields, for every t ∈ R+

d

dt
Ix̄(γ(t)) = (N − p)tN−p−1

∫
RN

j(x̄, u,Du)−NtN−1

∫
RN

H(u). (8.42)

By virtue of (8.8) and (8.10), a standard argument yields u ∈ L∞loc(RN ) (see [122,
Theorem 1]); by the regularity results of [65, 141], it follows that u ∈ C1,β

loc (RN )
for some 0 < β < 1. Then, since {ξ 7→ j(x, s, ξ)} is strictly convex, we can use
Theorem 8.6 by choosing in (8.40) ϕ = 0 and

L (s, ξ) := j(x̄, s, ξ)−H(s) for every s ∈ R+ and ξ ∈ RN ,

h(x) = hk(x) := T (
x

k
)x for every x ∈ RN and k ≥ 1,

(8.43)

being T ∈ C1
c (RN ) such that T (x) = 1 if |x| ≤ 1 and T (x) = 0 if |x| ≥ 2. In

particular, for every k we have that hk ∈ C1
c (RN ,RN ) and

Dih
j
k(x) = DiT (

x

k
)
xj
k

+ T (
x

k
)δij for every x ∈ RN , i, j = 1, . . . , N ,

(div hk)(x) = DT (
x

k
) · x
k

+NT (
x

k
) for every x ∈ RN .

Then, since DxL (u,Du) = 0, it follows by (8.40) that
n∑

i,j=1

∫
RN

DiT (
x

k
)
xj
k
DjuDξi

L (u,Du) +
∫

RN

T (
x

k
)DξL (u,Du) ·Du



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 181

−
∫

RN

DT (
x

k
) · x
k

L (u,Du)−
∫

RN

NT (
x

k
)L (u,Du) = 0

for every k ≥ 1. Since there exists C > 0 with

DiT (
x

k
)
xj
k
≤ C for every x ∈ RN , k ≥ 1 and i, j = 1, . . . , N ,

by the Dominated Convergence Theorem, letting k → +∞, we obtain∫
RN

[
NL (u,Du)−DξL (u,Du) ·Du

]
= 0,

namely, by (8.43) and the p-homogeneity of {ξ 7→ j(x, s, ξ)},

(N − p)
∫

RN

j(x̄, u,Du) = N

∫
RN

H(u). (8.44)

In particular notice that
∫

RN H(u) > 0. By plugging this formula into (8.42), we
obtain

d

dt
Ix̄(γ(t)) = N(1− tp)tN−p−1

∫
RN

H(u)

which yields d
dtIx̄(γ(t)) > 0 for 0 < t < 1 and d

dtIx̄(γ(t)) < 0 for t > 1, namely

sup
t∈[0,+∞[

Ix̄(γ(t)) = Ix̄(γ(1)) = Ix̄(u).

Moreover, observe that γ(0) = 0 and Ix̄(γ(T )) < 0 for T > 0 sufficiently large.
Then, after a suitable scale change in t, γ ∈ Px̄ and the assertion follows.

Step II Let us now prove that (8.41) has a nontrivial solution u ∈W 1,p(RN ) such
that cx̄ ≥ Ix̄(u). Let (uh) be a Palais-Smale sequence for Ix̄ at the level cx̄. Since
(uh) is bounded in W 1,p(RN ), considering the test uheζ(uh) with ζ as in (8.37), and
recalling (8.39), we have

pcx̄ + o(1) = pIx̄(uh)− I ′x̄(uh)[uhe
ζ(uh)]

=
∫

RN

p(1− eζ(uh))j(x̄, uh, Duh) +
∫

RN

(1− eζ(uh))V (x̄)|uh|p

−
∫

RN

[pζ ′(uh)j(x̄, uh, Duh) + js(x̄, uh, Duh)]uheζ(uh)

−
∫

RN

pF (uh) +
∫

RN

f(uh)uheζ(uh)

≤ −
∫

RN

pF (uh) +
∫

RN

f(uh)uheζ(uh)

≤ C

∫
RN

|uh|p + |uh|q

for some C > 0. By [97, Lemma I.1], we conclude that (uh) may not vanish in Lp,
that is there exists xh ∈ RN , R > 0 and λ > 0 such that for h large∫

xh+BR

|uh|p ≥ λ. (8.45)

Let vh(x) := uh(xh + x) and let u ∈ W 1,p(RN ) be such that vh ⇀ u weakly in
W 1,p(RN ). Since vh is a Palais-Smale sequence for Ix̄ at level cx̄, by Lemma 8.4,



182 MARCO SQUASSINA EJDE-2006/MON. 07

we have that vh → u strongly in W 1,p
loc (RN ). By (8.45), we deduce that u is a

nontrivial solution of (8.41). Let δ > 0; we claim that there exists % > 0 such that

lim inf
h

∫
RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
≥ −δ. (8.46)

In fact, let % > 0, and let η% be a smooth function such that 0 ≤ η% ≤ 1, η% = 0 on
B%−1, η% = 1 on RN \B% and ‖Dη%‖∞ ≤ 2. Testing with η%vh, we get

〈wh, η%vh〉 −
∫
B%\B%−1

[
jξ(x̄, vh, Dvh) ·D(η%vh)

+ js(x̄, vh, Dvh)η%vh + V (x̄)|vh|pη% − f(vh)vhη%
]

=
∫

RN\B%

[
jξ(x̄, vh, Dvh) ·D(η%vh) + js(x̄, vh, Dvh)η%vh

+ V (x̄)|vh|pη% − f(vh)vhη%
]

where wh → 0 strongly in W−1,p′(RN ). For the right hand side we have∫
RN\B%

[
jξ(x̄, vh, Dvh) ·D(η%vh) + js(x̄, vh, Dvh)η%vh

+ V (x̄)|vh|pη% − f(vh)vhη%
]

=
∫

RN\B%

[
pj(x̄, vh, Dvh) + js(x̄, vh, Dvh)vh + V (x̄)|vh|p − f(vh)vh

]
,

and by (8.9) we have∫
RN\B%

[
pj(x̄, vh, Dvh) + js(x̄, vh, Dvh)vh + V (x̄)|vh|p − f(vh)vh

]
≤ (p+ γ)

∫
RN\B%

j(x̄, vh, Dvh) +
∫

RN\B%

V (x̄)|vh|p − f(vh)vh

= (p+ γ)
∫

RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
− p+ γ

p

∫
RN\B%

V (x̄)|vh|p +
∫

RN\B%

V (x̄)|vh|p

+
∫

RN\B%

[
(p+ γ)F (vh)− f(vh)vh

]
≤ (p+ γ)

∫
RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
+

∫
RN\B%

[
(p+ γ)F (vh)− ϑF (vh)

]
≤ (p+ γ)

∫
RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
.

We conclude that

(p+ γ)
∫

RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
≥ 〈wh, η%vh〉 −

∫
B%\B%−1

[
jξ(x̄, vh, Dvh) ·D(η%vh) + js(x̄, vh, Dvh)η%vh



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 183

+ V (x̄)|vh|pη% − f(vh)vhη%
]
.

Since by Lemma 8.4 we have vh → u strongly in W 1,p(B%), we get

lim
h

∫
B%\B%−1

[
jξ(x̄, vh, Dvh) ·D(η%vh) + js(x̄, vh, Dvh)η%vh

+ V (x̄)|vh|pη% − f(vh)vhη%
]

=
∫
B%\B%−1

[
jξ(x̄, u,Du) ·D(η%u) + js(x̄, u,Du)η%u+ V (x̄)|u|pη% − f(u)uη%

]
,

and so we deduce that for every δ > 0 there exists %̄ > 0 such that for all % > %̄ we
have

lim inf
h

∫
RN\B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
≥ −δ.

Furthermore we have

lim
h

∫
B%

[
j(x̄, vh, Dvh) +

1
p
V (x̄)|vh|p − F (vh)

]
= Ix̄(u,B%),

where

Ix̄(u,B%) :=
∫
B%

[
j(x̄, u,Du) +

1
p
V (x̄)|u|p − F (u)

]
,

and so we conclude that for all % > %̄

cx̄ ≥ Ix̄(u,B%)− δ.

Letting % → +∞ and since δ is arbitrary, we get cx̄ ≥ Ix̄(u), and the proof is
complete. �

The second result can be considered as an extension (also with a different proof)
of [63, Lemma 2.3] to a general class of elliptic equations. Again we stress that, in
this degenerate setting, Theorem 8.6 plays an important role.

Lemma 8.8. Let u ∈W 1,p(RN ) be a positive solution of the equation

− div(jξ(x̄, u,Du)) + js(x̄, u,Du) + V (x̄)up−1

= 1{x1<0}(x)f(u) + 1{x1>0}(x)f̃(u) in RN .
(8.47)

Then u is actually a solution of the equation

−div(jξ(x̄, u,Du)) + js(x̄, u,Du) + V (x̄)up−1 = f(u) in RN . (8.48)

Proof. Let us first show that u(x) ≤ ` on the set {x1 = 0}. As in the proof of
Theorem 8.7 it follows that u ∈ C1,β

loc (RN ) for some 0 < β < 1. Then we can apply
again Theorem 8.6 by choosing this time in (8.40):

L (s, ξ) := j(x̄, s, ξ) +
V (x̄)
p

sp for every s ∈ R+ and ξ ∈ RN ,

ϕ(x) := 1{x1<0}(x)f(u(x)) + 1{x1>0}(x)f̃(u(x)) for every x ∈ RN ,

h(x) = hk(x) :=
(
T (
x

k
), 0, . . . , 0

)
for every x ∈ RN and k ≥ 1



184 MARCO SQUASSINA EJDE-2006/MON. 07

being T ∈ C1
c (RN ) such that T (x) = 1 if |x| ≤ 1 and T (x) = 0 if |x| ≥ 2. Then

hk ∈ C1
c (RN ,RN ) and, taking into account that DxL (u,Du) = 0, we have∫

RN

[1
k

N∑
i=1

DiT (
x

k
)D1uDξi

L (u,Du)−D1T (
x

k
)
1
k

L (u,Du)
]

=
∫

RN

T (
x

k
)ϕ(x)D1u

for every k ≥ 1. Again by the Dominated Convergence Theorem, letting k → +∞,
it follows

∫
RN ϕ(x)Dx1u = 0, that is, after integration by parts,∫

RN−1

[
F (u(0, x′))− F̃ (u(0, x′))

]
dx′ = 0.

Taking into account that F (s) ≥ F̃ (s) with equality only if s ≤ `, we get

u(0, x′) ≤ ` for every x′ ∈ RN−1. (8.49)

To prove that actually

u(x1, x
′) ≤ ` for every x1 > 0 and x′ ∈ RN−1, (8.50)

let us test equation (8.47) with the function

η(x) =

{
0 if x1 < 0
(u(x1, x

′)− `)+eζ(u(x1,x
′)) if x1 > 0,

where ζ : R+ → R is the map defined in (8.37). Notice that, in view of (8.49), the
function η belongs to W 1,p(RN ). After some computations, one obtains∫

{x1>0}
pj(x̄, u,D(u− `)+)eζ(u)

+
∫
{x1>0}

[js(x̄, u,Du) + pζ ′(u)j(x̄, u,Du)] (u− `)+eζ(u)

+
∫
{x1>0}

[
V (x̄)− α

κ

]
up−1(u− `)+eζ(u) = 0.

(8.51)

By (8.1) and (8.39) all the terms in (8.51) must be equal to zero. We conclude that
(u − `)+ = 0 on {x1 > 0}, namely (8.50) holds. In particular ϕ(x) = f(u(x)) for
every x ∈ RN , so that u is a solution of (8.48). �

8.3. Energy estimates. Let dε,i be the Mountain-Pass critical value which corre-
sponds to the functional Jε,i defined in (8.24). More precisely,

dε,i := inf
γi∈Γi

sup
t∈[0,1]

Jε,i(γi(t)) (8.52)

where
Γi :=

{
γi ∈ C([0, 1],W 1,p(Λ̂i)) : γi(0) = 0, Jε,i(γi(1)) < 0

}
.

Then the following result holds.

Lemma 8.9. For i = 1, . . . , k, we have

lim
ε→0+

ε−Ndε,i = ci .



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 185

Proof. The inequality
dε,i ≤ εNci + o(εN ) (8.53)

can be easily derived (see the first part of the proof of Lemma 8.10). Let us prove
the opposite inequality, which is harder. To this aim, we divide the proof into two
steps.
Step I. Let wε be a Mountain-Pass critical point for Jε,i. We have wε ≥ 0, and by
regularity results wε ∈ L∞(Λ̂i) ∩ C1,α

loc (Λ̂i). Let us define

Mε := sup
x∈Λ̂i

wε(x) < +∞,

and for all δ > 0 define the set

Uδ :=
{
x ∈ Λ̂i : wε(x) > Mε − δ

}
.

We may use the following nontrivial test for the equation satisfied by wε

ϕδ := [wε − (Mε − δ)]+eζ(wε),

where the map ζ : R+ → R is defined as in (8.37). We have

Dϕδ = eζ(wε)Dwε1Uδ
+ ϕδζ

′(wε)Dwε,

and so we obtain

εp
∫
Uδ

pj(x,wε, Dwε)eζ(wε) + εp
∫
Uδ

[
pζ ′(wε)j(x,wε, Dwε) + js(x,wε, Dwε)

]
ϕδ

=
∫
Uδ

[
−V (x)wp−1

ε + g(x,wε)
]
ϕδ.

Then, by (8.39), it results∫
Uδ

[
−V (x)wp−1

ε + g(x,wε)
]
ϕδ ≥ εp

∫
Uδ

pj(x,wε, Dwε)eζ(wε) > 0. (8.54)

Suppose that Uδ ∩ Λi = ∅ for some δ > 0; we have that g(x,wε) = f̃(wε) on Uδ, so
that ∫

Uδ

[
−V (x)wp−1

ε + f̃(wε)
]
ϕδ > 0. (8.55)

On the other hand, we note that by construction f̃(wε) ≤ 1
kV (x)wp−1

ε with strict
inequality on an open subset of Uδ. We deduce that (8.55) cannot hold, and so
Uδ ∩Λi 6= ∅ for all δ. Since Λi is compact, we conclude that wε admits a maximum
point xε in Λi. Moreover, we have wε(xε) ≥ `, where ` is as in (8.14), since otherwise
(8.54) cannot hold.

Let us now consider the functions vε(y) := wε(xε + εy) and let εj → 0. We have
that, up to a subsequence, xεj → x̄ ∈ Λi. Since wε is a Mountain-Pass critical
point of Jε,i, arguing as in Step I of Lemma 8.5 there exists C > 0 such that∫

RN

(
εp|Dwε|p + V (x)|wε|p

)
≤ Cdε,i,

which, by (8.53) implies, up to subsequences, vεj ⇀ v weakly in W 1,p(RN ). We
now prove that v 6= 0. Let us set

dj(y) :=

V (xεj + εjy)−
g(xεj

+εjy,vεj
(y))

vp−1
εj

(y)
if vεj (y) 6= 0

0 if vεj (y) = 0,



186 MARCO SQUASSINA EJDE-2006/MON. 07

A(y, s, ξ) := jξ(xεj
+ εjy, s, ξ),

B(y, s, ξ) := dj(y)sp−1,

C(y, s) := js(xεj + εjy, s,Dvεj (y))

for every y ∈ RN , s ∈ R+ and ξ ∈ RN . Taking into account the growth of condition
on jξ, the strict convexity of j in ξ and condition (8.8), we get

A(y, s, ξ) · ξ ≥ ν|ξ|p, |A(y, s, ξ)| ≤ c2|ξ|p−1, |B(y, s, ξ)| ≤ |dj(y)||s|p−1.

Moreover, by condition (8.10) we have

s ≥ R ⇒ C(y, s) ≥ 0

for every y ∈ RN and s ∈ R+. By the growth of conditions on g, we have that for
δ sufficiently small dj ∈ L

N
p−δ (B2%) for every % > 0 and

S = sup
j
‖dj‖

L
N

p−δ (B2%)
≤ D

(
1 + sup

j∈N
‖vεj‖Lp∗ (B2%)

)
< +∞

for some D = D% > 0. Since we have div(A(y, vεj , Dvεj )) = B(y, vεj , Dvεj ) +
C(y, vεj

) for every j ∈ N, by virtue of [122, Theorem 1 and Remark at p.261] there
exists a radius % > 0 and a positive constant M = M(ν, c2, S%δ) such that

sup
j∈N

max
y∈B%

|vεj
(y)| ≤M(2%)−N/p sup

j∈N
‖vεj

‖Lp(B2%) < +∞

so that (vεj
) is uniformly bounded in B%. Then, by [122, Theorem 8], up to a

subsequence (vεj
) converges uniformly to v in a small neighborhood of zero. This

yields v(0) = limj vεj (0) = limj wεj (xεj ) ≥ `.
Up to a rotations and translation, it is easily seen that v is a positive solution of

−div(jξ(x̄, v,Dv)) + js(x̄, v,Dv) + V (x̄)vp−1 = 1{x1<0}f(v) + 1{x1>0}f̃(v).

By Lemma 8.8 it follows that v is actually a nontrivial solution of

−div(jξ(x̄, v,Dv)) + js(x̄, v,Dv) + V (x̄)vp−1 = f(v).

Then, by Theorem 8.7 and (8.23), we have Ix̄(v) = cx̄ ≥ ci. To conclude the proof,
it is sufficient to prove that

lim inf
j

ε−Nj dεj ,i = lim inf
j

ε−Nj Jεj ,i(wεj
) ≥ Ix̄(v). (8.56)

Step II. We prove (8.56). It results

ε−Nj Jεj ,i(wεj ) =
∫

Λ̂εj
,i

j(xεj + εjy, vεj , Dvεj )

+
1
p

∫
Λ̂εj,i

V (xεj
+ εjy)vpεj

−
∫

Λ̂εj,i

G(xεj
+ εjy, vεj

)

where Λ̂εj ,i = {y ∈ RN : xεj
+εjy ∈ Λ̂i}. By Lemma 8.4, we have vεj

→ v strongly
in W 1,p

loc (RN ). Following the same computations of Theorem 8.7, Step II, we deduce
that for all δ > 0 there exists %̄ > 0 such that for all % > %̄ we have

lim inf
j

∫
Λ̂εj,i\B%

[
j(xεj

+εjy, vεj
, Dvεj

)+
1
p
V (xεj

+εjy)vpεj
−G(xεj

+εjy, vεj
)
]
≥ −δ.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 187

Furthermore,

lim
j

∫
B%

[
j(xεj + εjy, vεj , Dvεj ) +

1
p
V (xεj + εjy)vpεj

−G(xεj + εjy, vεj )
]

= Ix̄(v,B%),

where
Ix̄(v,B%) :=

∫
B%

[
j(x̄, v,Dv) +

1
p
V (x̄)vp − F (v)

]
.

We conclude that for all % > %̄

lim inf
j

ε−Nj Jεj ,i(wεj
) ≥ Ix̄(v,B%)− δ,

and (8.56) follows letting %→ +∞ and δ → 0. �

Let us now consider the class

Γε :=
{
γ ∈ C([0, 1]k,WV (Ω)) : γ satisfies conditions (a), (b), (c), (d)

}
,

where:
(a) γ(t) =

∑k
i=1 γi(ti) for every t ∈ ∂[0, 1]k, with γi ∈ C([0, 1],WV (Ω)) ;

(b) supt(γi(ti)) ⊂ Λi for every ti ∈ [0, 1] and i = 1, . . . , k ;
(c) γi(0) = 0 and Jε(γi(1)) < 0 for every i = 1, . . . , k ;
(d) ε−NEε(γ(t)) ≤

∑k
i=1 ci + σ for every t ∈ ∂[0, 1]k,

where 0 < σ < 1
2 min{ci : i = 1, . . . , k}. We set

cε := inf
γ∈Γε

sup
t∈[0,1]k

Eε(γ(t)). (8.57)

Lemma 8.10. For ε small enough Γε 6= ∅ and

lim
ε→0+

ε−Ncε =
k∑
i=1

ci. (8.58)

Proof. Firstly, let us prove that for ε small Γε 6= ∅ and

cε ≤ εN
k∑
i=1

ci + o(εN ). (8.59)

By definition of ci, for all δ > 0 there exists γi ∈ Pi with

ci ≤ max
τ∈[0,1]

Ixi
(γi(τ)) ≤ ci +

δ

2k
(8.60)

where the xis are as in (8.2) and

Pi :=
{
γi ∈ C([0, 1],W 1,p(RN )) : γi(0) = 0, Ixi

(γi(1)) < 0
}
.

We choose δ so that δ < min{σ, kσi}. Let us set

γ̂i(τ)(x) = ηi(x)γi(τ)
(x− xi

ε

)
for every τ ∈ [0, 1] and x ∈ Ω,

where ηi ∈ C∞c (RN ), 0 ≤ ηi ≤ 1, supp ηi ⊂ Λi, and xi ∈ int({ηi = 1}). We have

Jε(γ̂i(τ)) =
∫

Ω

εpj(x, γ̂i(τ), Dγ̂i(τ)) +
1
p

∫
Ω

V (x)|γ̂i(τ)|p −
∫

Ω

G(x, γ̂i(τ)). (8.61)

Since it results

Dγ̂i(τ) = Dηi(x)γi(τ)
(x− xi

ε

)
+

1
ε
ηi(x)Dγi(τ)

(x− xi
ε

)
,



188 MARCO SQUASSINA EJDE-2006/MON. 07

and for all ξ1, ξ2 ∈ RN there exists t ∈ [0, 1] with

j(x, s, ξ1 + ξ2) = j(x, s, ξ2) + jξ(x, s, tξ1 + ξ2) · ξ1,
taking into account the p-homogeneity of j, the term

εp
∫

Ω

j(x, γ̂i(τ), Dγ̂i(τ))

has the same behavior of∫
Ω

j

(
x, ηi(x)γi(τ)

(x− xi
ε

)
, ηi(x)Dγi

(x− xi
ε

))
(8.62)

up to an error given by

εp
∫

Ω

jξ(x, s(x), t(x)ξ1(x) + ξ2(x)) · ξ1(x), (8.63)

where we have set

s(x) := γ̂i(τ)(x),

ξ1(x) := Dηi(x)γi(τ)
(x− xi

ε

)
,

ξ2(x) :=
1
ε
ηi(x)Dγi(τ)

(x− xi
ε

)
,

and t(x) is a function with 0 ≤ t(x) ≤ 1 for every x ∈ Ω. We proceed in the
estimation of (8.63). We obtain

εp
∣∣∣ ∫

Ω

jξ(x, s(x), t(x)ξ1(x) + ξ2(x)) · ξ1(x)
∣∣∣

≤ c̃2ε
p

∫
Ω

|ξ1(x)|p + c̃2ε
p

∫
Ω

|ξ2(x)|p−1|ξ1(x)|.

Making the change of variable y = x−xi

ε , we obtain

εp
∣∣∣ ∫

Ω

jξ(x, s(x), t(x)ξ1(x) + ξ2(x)) · ξ1(x)
∣∣∣

≤ c̃2ε
p+N

∫
RN

|Dηi(xi + εy)|p|γi(τ)(y)|p

+ c̃2ε
N+1

∫
RN

|ηi(xi + εy)|p−1|Dγi(τ)(y)|p−1|Dηi(xi + εy)||γi(τ)(y)|

= o(εN )

where o(εN ) is independent of τ , since γi has compact values in W 1,p(RN ). Chang-
ing the variable also in (8.62) yields∫

Ω

j

(
x, ηi(x)γi(τ)

(x− xi
ε

)
, ηi(x)Dγi(τ)

(x− xi
ε

))
= εN

∫
RN

j(xi + εy, ηi(xi + εy)γi(τ)(y), ηi(xi + εy)Dγi(τ)(y)).

By the Dominated Convergence Theorem we get

lim
ε→0

∫
RN

j(xi + εy, ηi(xi + εy)γi(τ)(y), ηi(xi + εy)Dγi(τ)(y))

=
∫

RN

j(xi, γi(τ)(y), Dγi(τ)(y))



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 189

uniformly with respect to τ . Reasoning in a similar fashion for the other terms in
(8.61), we conclude that for ε small enough

Jε(γ̂i(τ)) = εNIxi
(γi(τ)) + o(εN ) (8.64)

for every τ ∈ [0, 1] with o(εN ) independent of τ . Let us now set

γ0(τ1, . . . , τk) :=
k∑
i=1

γ̂i(τi).

Since supp γ̂i(τ) ⊂ Λi for every τ , we have that Jε,i(γ̂i(τ)) = Jε(γ̂i(τ)); then, by
the choice of δ, we get for ε small

[Jε,i(γ̂i(τ))+]
1
2 − ε

N
2 (ci + σi)

1
2 = [Jε(γ̂i(τ))+]

1
2 − ε

N
2 (ci + σi)

1
2

= ε
N
2 [Ixi

(γi(τ)) + o(1)]
1
2 − ε

N
2 (ci + σi)

1
2

≤ ε
N
2
[
ci +

δ

2k
+ o(1)

] 1
2 − ε

N
2 (ci + σi)

1
2 ≤ 0,

and

Eε(γ0(τ1, . . . , τk)) = Jε(γ0(τ1, . . . , τk)) =
k∑
i=1

Jε(γ̂i(τi)).

By (8.60) and (8.64) we obtain that for ε small enough

Eε(γ0(τ)) ≤ εN
k∑
i=1

(
ci +

δ

2k
)
≤ εN

( k∑
i=1

ci + σ
)

so that the class Γε is not empty. Moreover, we have

lim sup
ε→0+

cε
εN

≤
k∑
i=1

ci + δ

and, by the arbitrariness of δ, we have conclude that (8.59) holds. Let us now prove
that

cε ≥ εN
k∑
i=1

ci + o(εN ). (8.65)

Given γ ∈ Γε, by a variant of [51, Proposition 3.4] there exists t̄ ∈ [0, 1]k such that

Jε,i(γ(t̄)) ≥ dε,i

for all i = 1, . . . , k, where the dε,is are as in (8.52). Then we have by Lemma 8.9

sup
t∈[0,1]k

Jε(γ(t)) ≥ sup
t∈[0,1]k

k∑
i=1

Jε,i(γ(t)) ≥
k∑
i=1

dε,i = εN
k∑
i=1

ci + o(εN ),

which implies the assertion. �

Corollary 8.11. For every ε > 0 there exists a critical point uε ∈ WV (Ω) of the
functional Eε such that cε = Eε(uε). Moreover ‖uε‖WV

→ 0 as ε→ 0.

Proof. By Lemma 8.5 it results that Eε satisfies the Palais-Smale condition for
every c ∈ R (see Definition 2.15). Then, by Lemma 8.10, for every ε > 0 the (non-
smooth) Mountain-Pass Theorem (see [50]) for the class Γε provides the desired
critical point uε of Eε. To prove the second assertion we may argue as in Step I of



190 MARCO SQUASSINA EJDE-2006/MON. 07

Lemma 8.5 with uh replaced by uε and c replaced by Eε(uε). Thus, from inequality
(8.34), for every ε > 0 we get∫

Ω

(
|Duε|p + V (x)|uε|p

)
≤ ϑEε(uε) + dεN

min
{

(ϑ− γ − p) νεp, ϑp −
ϑ
2κ − 1

} . (8.66)

By virtue of Lemma 8.10, this yields∫
Ω

(
|Duε|p + V (x)|uε|p

)
≤

{
ϑ(c1 + · · ·+ ck) + d

(ϑ− γ − p)ν

}
εN−p + o(εN−p),

as ε→ 0, which implies the assertion. �

Let us now set

Ωε :=
{
y ∈ RN : εy ∈ Ω

}
, vε(y) := uε(εy) ∈W 1,p(Ωε),

Λ̂ε,i :=
{
y ∈ RN : εy ∈ Λ̂i

}
, Λε :=

{
y ∈ RN : εy ∈ Λ

}
.

Lemma 8.12. The function vε is a solution of the equation

− div
(
(1 + θε(εy))jξ(εy, v,Dv)

)
+ (1 + θε(εy))js(εy, v,Dv)

+ (1 + θε(εy))V (εy)vp−1

= (1 + θε(εy))g(εy, v) in Ωε,

(8.67)

where for every ε > 0

θε(x) :=
k∑
i=1

θε,i1Λ̂i
(x), θε,i ∈ [0,M ],

θε,i := M
[
(Jε,i(uε)+)1/2 − εN/2(ci + σi)1/2

]
+
(Jε,i(uε)+)−1/2.

(8.68)

Proof. It suffices to expand E′ε(uε)(ϕ) = 0 for every ϕ ∈ C∞c (Ω). �

Corollary 8.13. The sequence (vε) is bounded in W 1,p(RN ).

Proof. It suffices to combine Lemma 8.10 with the inequality∫
RN

(
|Dvε|p + V (x)|vε|p

)
≤ ϑε−Ncε + d

min
{

(ϑ− γ − p) ν, ϑp −
ϑ
2κ − 1

}
which follows by (8.66). �

The following lemma “kills” the second penalization term of Eε.

Lemma 8.14. For i = 1, . . . , k, we have

lim
ε→0

ε−NJε,i(uε) = ci . (8.69)

Proof. Let us first prove that, as %→ +∞,

lim sup
ε→0+

∫
Ωε\N%(Λε)

(
|Dvε|p + |vε|p

)
= o(1), (8.70)

where N%(Λε) := {y ∈ RN : dist(y,Λε) < %}. We can test equation (8.67) with
ψε,%vεe

ζ(vε), where ψε,% = 1−
∑k
i=1 ψ

i
ε,%, ψ

i
ε,% ∈ C∞(RN ),

ψiε,% =

{
1 if dist(y,Λε,i) < %/2,
0 if dist(y,Λε,i) > %



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 191

and the function ζ is defined as in (8.37). By virtue of (8.1), (8.7), the boundedness
of (vε) in W 1,p(RN ) and (8.39) there exist C,C ′ > 0 such that

C

∫
Ωε\N%(Λε)

(
|Dvε|p + |vε|p

)
≤

∫
Ωε\Λε

(1 + θε(εy))
[
pj(εy, vε, Dvε) +

{
V (εy)− f̃(vε)

vp−1
ε

}
vpε

]
ψε,%e

ζ(vε)

= −
∫

Ωε\Λε

(1 + θε(εy))
[
js(εy, vε, Dvε) + pζ ′(vε)j(εy, vε, Dvε)

]
vεψε,%e

ζ(vε)

−
∫

Ωε\Λε

(1 + θε(εy))jξ(εy, vε, Dvε) ·Dψε,%vεeζ(vε)

≤ 2eM̄R

∫
Ωε\Λε

|Dψε,%||jξ(εy, vε, Dvε)|vε ≤
C̃

%
‖Dvε‖p−1

p ‖vε‖p ≤
C ′

%
,

which implies (8.70). Now, to prove (8.69), we adapt the argument of [63, Lemma
2.1] to our context. It is sufficient to prove that

lim
ε→0

ε−NJε,i(uε) ≤ ci + σi (8.71)

for every i = 1, . . . , k. Then (8.69) follows by arguing exactly as in [63, Lemma
2.4]. By contradiction, let us suppose that for some εj → 0 we have

lim sup
j

ε−Nj Jεj ,i(uεj
) > ci + σi. (8.72)

Then there exists λ > 0 with∫
Λ̂εj,i

(
|Dvεj |p + |vεj |p

)
≥ λ,

and so by (8.70) there exists % > 0 such that for j large enough∫
N%(Λεj,i)

(
|Dvεj

|p + |vεj
|p

)
≥ λ

2
.

Following [63, Lemma 2.1], P.L. Lions’ concentration compactness argument [97]
yields the existence of S > 0, ρ > 0 and a sequence yj ∈ Λεj ,i such that for j large
enough ∫

BS(yj)

vpεj
≥ ρ. (8.73)

Let us set vj(y) := vεj (yj + y), and let εjyj → x̄ ∈ Λi. By Corollary 8.13, we
may assume that vj weakly converges to some v in W 1,p(RN ). By Lemma 8.4, we
have that vj → v strongly in W 1,p

loc (RN ); note that v 6= 0 by (8.73). In the case
dist(yj , ∂Λεj ,i) → +∞, since vj satisfies in −yj + Λεj ,i the equation

−div(jξ(εjyj + εjy, vj , Dvj))+ js(εjyj + εjy, vj , Dvj)+V (εjyj + εjy)vp−1 = f(vj),

v satisfies on RN the equation

−div(jξ(x̄, v,Dv)) + js(x̄, v,Dv) + V (x̄)vp−1 = f(v). (8.74)

If dist(yj , ∂Λεj ,i) ≤ C < +∞, we deduce that v satisfies an equation of the form
(8.47), and by Lemma 8.8, we conclude that v satisfies equation (8.74). Since
v is a nontrivial critical point for Ix̄, by (8.11) and Theorem 8.7, recalling that



192 MARCO SQUASSINA EJDE-2006/MON. 07

ci ≤ cx̄ ≤ ci + σi, we get ci ≤ Ix̄(v) ≤ ci + σi. Then we can find a sequence
Rj → +∞ such that

lim
j

∫
BRj

(yj)

j(εjy, vεj
, Dvεj

) +
1
p
V (εjy)|vεj

|p −G(εjy, vεj
) = Ix̄(v) ≤ ci + σi.

Then by (8.72) we deduce that for j large enough∫
Λ̂εj,i\BRj

(yj)

(
|Dvεj |p + |vεj |p

)
≥ λ > 0.

Reasoning as above, there exist S̃, ρ̃ > 0 and a sequence ỹj ∈ Λεj ,i \ BRj
(yj) such

that ∫
B eS(eyj)

vpεj
≥ ρ̃ > 0. (8.75)

Let εj ỹj → x̃ ∈ Λi; then we have ṽj(y) := vεj
(ỹj + y) → ṽ weakly in W 1,p(RN ),

where ṽ is a nontrivial solution of the equation

−div(jξ(x̃, v,Dv)) + js(x̃, v,Dv) + V (x̃)vp−1 = f(v).

As before we get Iex(ṽ) ≥ ci. We are now in a position to deduce that

lim inf
j

ε−Nj Jεj ,i(uε) > Ix̄(v) + Iex(ṽ) ≥ 2ci.

In fact, vεj
satisfies in Λ̂εj ,i the equation

−div(jξ(εjy, vεj , Dvεj )) + js(εjy, vεj , Dvεj ) + V (εjy)vp−1
εj

= g(εjy, vεj ). (8.76)

Since yj , ỹj ∈ Λεj ,i, for j large enough Bj,R := B(yj , R) ∪B(ỹj , R) ⊂ Λ̂εj ,i, and so
we can test (8.76) with

ϕ(y) =
[
ψ

( |y − yj |
R

)
+ ψ

( |y − ỹj |
R

)
− 1

]
vεj

(y)

where ψ ∈ C∞(R) with 0 ≤ ψ ≤ 1, ψ(s) = 0 for s ≤ 1 and ψ(s) = 1 for s ≥ 2.
Reasoning as in Lemma 8.9, we have that for all δ > 0 there exists R̄ such that for
all R > R̄ we have∫

Λ̂εj,i\Bj,R

[
j(εjy, vεj , Dvεj ) +

1
p
V (εjy)|vεj |p −G(εjy, vεj )

]
≥ −δ

so that
lim inf

j
ε−Nj Jεj ,i(uεj

) ≥ Ix̄(v,BR) + Iex(ṽ, BR)− δ.

Letting R→ +∞ and δ → 0, we get

lim inf
j

ε−Nj Jεj ,i(uεj
) > 2ci. (8.77)

The same arguments apply to the functional Jε: we have that

lim inf
j

ε−Nj Jεj
(uεj

) ≥ 2ci. (8.78)

Then by combining (8.77) and (8.78) we obtain

lim inf
j

ε−Nj Eεj (uεj ) ≥ 2ci +M
[
(2ci)1/2 − (ci + σi)1/2

]2

+
.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 193

By Lemma 8.10, we have

M
[
(2ci)1/2 − (ci + σi)1/2

]2

+
≤

k∑
i=1

ci,

against the choice of M . �

8.4. Proofs of the main results. We are now ready to prove the main results of
the section.

Proof of Theorem 8.2. Let us consider the sequence (uε) of critical points of Eε
given by Corollary 8.11. We have that ‖uε‖WV

→ 0. Since uε satisfies

− div
(
(1 + θε(x))jξ(x, v,Dv)

)
+ (1 + θε(x))js(x, v,Dv) + (1 + θε(x))V (x)vp−1

= (1 + θε(x))g(x, v) in Ω,

with θε defined as in (8.68), by the regularity results of [90] uε is locally Hölder
continuous in Ω. We claim that there exists σ > 0 such that

uε(xε,i) = sup
Λi

uε > σ > 0 (8.79)

for every ε sufficiently small and i = 1, . . . , k: moreover

lim
ε→0

dist(xε,i,Mi) = 0 (8.80)

for i = 1, . . . , k, where the Mis are the sets of minima of V in Λi. In fact, let
us assume that there exists i0 ∈ {1, . . . , k} such that uε(xε,i0) → 0 as ε → 0.
Therefore, uε → 0 uniformly on Λi0 as ε→ 0, which implies that

sup
y∈Λε,i0

vε(y) → 0 as ε→ 0, (8.81)

where vε(y) := uε(εy). On the other hand, since by (8.69) we have

lim
ε→0

ε−NJε,i0(uε) = ci0 > 0,

considering Λ̃i0 relatively compact in Λi0 , following the proof of Lemma 8.14, we
find S > 0 and % > 0 such that

sup
y∈eΛε,i0

∫
BS(y)

vpε ≥ %

for every ε ∈ (0, ε0), which contradicts (8.81). We conclude that (8.79) holds. To
prove (8.80), it is sufficient to prove that

lim
ε→0

V (xε,i) = min
Λi

V

for every i = 1, . . . , k. Assume by contradiction that for some i0

lim
ε→0

V (xε,i0) > min
Λi0

V = bi0 .

Then, up to a subsequence, xεj ,i0 → xi0 ∈ Λi0 and V (xi0) > bi0 . Then, arguing as
in the proof of Lemma 8.14 and using Theorem 8.7, we would get

lim inf
j

εj
−NJεj ,i0(uεj

) ≥ Ixi0
(v) = cxi0

> ci0

which is impossible, in view of (8.69).



194 MARCO SQUASSINA EJDE-2006/MON. 07

We now prove that

lim
ε→0

uε = 0 uniformly on Ω \
k⋃
i=1

int(Λi). (8.82)

Let us first prove that

lim
ε→0

sup
∂Λi

uε = 0 for i = 1, . . . , k.

By contradiction, let i0 ∈ {1, . . . , k} and σ > 0 with uεj
(xj) ≥ σ for (xj) ⊂ ∂Λi0 .

Up to a subsequence, xj → x0 ∈ ∂Λi0 . Therefore, taking into account Lemma 8.8
and the local regularity estimates of [122] (see also the end of Step I of Lemma 8.9),
the sequence vj(y) := uεj (xj + εjy) converges weakly to a nontrivial solution v ∈
W 1,p(RN ) of

−div(jξ(x0, v,Dv)) + js(x0, v,Dv) + V (x0)vp−1 = f(v) in RN .

As V (x0) > V (xi0), we have

lim inf
j

ε−Nj Jεj ,i0(uεj
) ≥ Ix0(v) > ci0 ,

which violates (8.69). Testing the equation with

(uε −max
i

sup
∂Λi

uε)+1Ω\Λe
ζ(uε),

as in Lemma 8.8, this yields that uε(x) ≤ maxi sup∂Λi
uε for every x ∈ Ω \ Λ, so

that (8.82) holds.
By Proposition 8.3, uε is actually a solution of the original problem (8.13) be-

cause the penalization terms are neutralized by the facts Jε,i(uε) < ci + σi and
uε < ` on Ω \ Λ for ε small. By regularity results, it follows uε ∈ C1,β

loc (Ω),
and so point (a) is proved. Taking into account (8.79) and (8.82), we get that
uε has a maximum x̄ε ∈ Ω which coincides with one of the xε,is. Considering
v̄ε(y) := uε(xε,i + εy), since v̄ε is uniformly bounded in W 1,p

loc (RN ), by the local
regularity estimates [122], there exists σ′ with

uε(xε,i) ≤ σ′

for all i = 1, . . . , k. In view of (8.79), (8.80) and Corollary 8.11, we conclude that
points (b) and (d) are proved. Let us now come to point (c). Let us assume by
contradiction that there exists r̄, δ, i0 and εj → 0 such that there exists yj ∈
Λi0 \Br̄(xεj ,i0) with

lim sup
j

uεj (yj) ≥ δ.

We may assume that yj → ȳ, xεj ,i0 → x̄, and v̄j(y) := uεj (yj + εjy) → v̄, vj(y) :=
uεj

(xεj ,i0 + εjy) → v strongly in W 1,p
loc (RN ): then, arguing as in Lemma 8.14, it

turns out that
lim inf

j
ε−Nj Jεj ,i0(uεj ) ≥ Ix̄(v) + Iȳ(v̄) ≥ 2ci0

which is against (8.69). We conclude that point (c) holds, and the proof is con-
cluded. �

Proof of Theorem 8.1. If 1 < p ≤ 2 and p < q < p∗, the equation

−∆pu+ V (x̄)up−1 = uq−1 in RN (8.83)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 195

admits a unique positive C1 solution (up to translations). Indeed, a solution u ∈
C1(RN ) of (8.83) exists by Theorem 8.7. By [93, Theorem 1] we have u(x) → 0
as |x| → ∞. Moreover, by [55, Theorem 1.1], the solution u is radially symmetric
about some point x0 ∈ RN and radially decreasing. Then u is a radial ground state
solution of (8.83). By [123, Theorem 1], u is unique (up to translations). Then
(8.11) is satisfied and the assertions follow by Theorem 8.2 applied to the functions
j(x, s, ξ) = 1

p |ξ|
p and f(s) = sq−1. �

9. Nonexistence Results

Some parts in this section have been slightly modified to give this collection a
more uniform appearance. We refer the reader to source in [59].

9.1. A general Pucci-Serrin type identity. In 1965 Pohozǎev discovered a very
important identity for solutions of the problem

∆u+ g(u) = 0 in Ω
u = 0 on ∂Ω.

This variational identity enabled him to show that the above problem has no non-
trivial solution provided that Ω is a bounded star-shaped domain of Rn and g
satisfies

∀s ∈ R : s 6= 0 ⇒ (n− 2)sg(s)− 2nG(s) > 0
where G is the primitive of g with G(0) = 0.

Let Ω be a bounded open subset of Rn with smooth boundary and outer normal
ν. Assume that L is a function of class C1 on Ω×R×Rn with L (x, 0, 0) = 0 and
that the vector valued function

∇ξL (x, s, ξ) =
(
∂L

∂ξ1
(x, s, ξ), · · · , ∂L

∂ξn
(x, s, ξ)

)
is of class C1 in Ω×R×Rn. Moreover, let G be a continuous function in Ω×R×Rn.
Consider the problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = G (x, u,∇u) in Ω
u = 0 on ∂Ω.

(9.1)

Let us recall the celebrated identity proved by Pucci and Serrin [116].

Theorem 9.1. Assume that u ∈ C2(Ω) ∩ C1(Ω) is a solution of (9.1). Then∫
∂Ω

[
L (x, 0,∇u)−∇ξL (x, 0,∇u) · ∇u

]
h · ν dH n−1

=
∫

Ω

[
L (x, u,∇u) div h+ h · ∇xL (x, u,∇u)

]
dx

−
n∑

i,j=1

∫
Ω

[
DjuDihj + uDia

]
Dξi

L (x, u,∇u) dx

−
∫

Ω

a
[
∇ξL (x, u,∇u) · ∇u+ uDsL (x, u,∇u)

]
dx

+
∫

Ω

[
h · ∇u+ au

]
G (x, u,∇u) dx

(9.2)

for each a ∈ C1(Ω) and h ∈ C1(Ω,Rn).



196 MARCO SQUASSINA EJDE-2006/MON. 07

Remark 9.2. Identity (9.2) follows by testing the equation with h ·∇u+au. More
generally, it is satisfied by solutions u ∈ C1(Ω) ∩W 2,2

loc (Ω).

Theorem 9.1 generalizes a well-known identity of Pohožaev [115] which has
turned out to be a powerful tool in proving non-existence of solutions of prob-
lem (9.1). On the other hand, in some cases the requirement that u is of class
C2(Ω) seems too restrictive, while C1(Ω) is not (cf. [141]). See e.g. problems in
which the p-Laplacian operator is involved [81].

The aim of this section is to remove the C2(Ω) assumption on u, by imposing
the strict convexity of L (x, s, ·). The main result is the following:

Theorem 9.3. Assume that u ∈ C1(Ω) is a solution of (9.1) and that the map

ξ 7→ L (x, s, ξ)

is strictly convex for each (x, s) ∈ Ω × R. Then (9.2) holds for all a ∈ C1(Ω),
h ∈ C1(Ω,Rn).

Let us observe that the strict convexity of L (x, s, ·) is indeed usually assumed
in the applications and it is also natural if one expects the solution u to be of class
C1(Ω). In some particular situations (see Section 9.3), we are also able to assume
only the convexity of L (x, s, ·). This is the case, for instance, if one takes

L (x, s, ξ) = α(x, s)β(ξ) + γ(x, s).

Note that if the test functions a and h have compact support in Ω, we obtain the
variational identity also when u is only locally Lipschitz in Ω. This seems to be
useful in particular when L (x, s, ·) is merely convex, as a C1 regularity of u cannot
be expected.

Finally, we refer the reader to [116] for various applications of the previous result
to non-existence theorems.

9.2. The approximation argument. Let Ω be an open subset of Rn, not nec-
essarily bounded. Assume that ξ 7→ L (x, s, ξ) is strictly convex for each (x, s) ∈
Ω× R.

Lemma 9.4. Let u : Ω → R be a locally Lipschitz solution of

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = G (x, u,∇u) in Ω. (9.3)

Then ∫
Ω

(
L (x, u,∇u) div h+ h · ∇xL (x, u,∇u)

)
dx

=
n∑

i,j=1

∫
Ω

DihjDξi
L (x, u,∇u)Dju dx−

∫
Ω

G (x, u,∇u)h · ∇u dx
(9.4)

for every h ∈ C1
c (Ω,Rn).

Proof. Since h has compact support, without loss of generality we may assume that
Ω is bounded. Let R > 0 with |∇u(x)| ≤ R for every x ∈ supth and let ϑ ∈ C∞(R)
be such that ϑ = 1 on [−R,R] and ϑ = 0 outside [−R − 1, R + 1]. Define now
L (x, s, ξ) by

L (x, s, ξ) = L (x, s, ϑ(|ξ|)ξ)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 197

for each (x, s, ξ) ∈ Ω× R× Rn. Then there exists ω > 0 such that
n∑

i,j=1

∇2
ξiξj

L (x, s, ξ)ηiηj > −ω|η|2

for each (x, s, ξ) ∈ Ω×R×Rn and η ∈ Rn. Let us now introduce Λ ∈ C1([0,+∞[)
by

Λ(τ) =

{
0 if 0 ≤ τ ≤ R

ω′(τ −R)2 if τ ≥ R,

where ω′ > ω. Moreover, let L̃ : Ω× Rn → R be given by

L̃ (x, ξ) = L (x, u(x), ξ) + Λ(|ξ|) (9.5)

for each (x, ξ) ∈ Ω×Rn. Then L̃ (x, ·) is strictly convex and there are ν, c > 0 with

L̃ (x, ξ) ≥ ν|ξ|2 − c

for each (x, ξ) ∈ Ω × Rn. In particular, since u solves (9.3), then it is the unique
minimum of the functional f : H1

0 (Ω) → R given by

f(w) =
∫

Ω

L̃ (x,∇w) dx+
∫

Ω

[
DsL (x, u,∇u)− G (x, u,∇u)

]
w dx.

On the other hand, if uk ∈ H1
0 (Ω) denotes the minimum of the modified functional

fk(w) = f(w) +
1
k

∫
Ω

|∇w|2 dx,

by standard regularity arguments, uk ∈ C1(Ω) ∩W 2,2
loc (Ω). Since f(uk) → f(u) as

k → +∞, we get uk ⇀ u in H1
0 (Ω) and∫

Ω

L̃ (x,∇uk) dx→
∫

Ω

L̃ (x,∇u) dx,

which by [144, Theorem 3] implies uk → u in H1
0 (Ω). In particular, ∇uk(x) →

∇u(x) a.e. in Ω, up to a subsequence. Put now

L̂ (x, ξ) = L̃ (x, ξ) +
1
k
|ξ|2.

Since uk satisfies the Euler’s equation of fk

div(∇ξL̂ (x,∇uk)) = DsL (x, u,∇u)− G (x, u,∇u),

by (9.2) it results∫
Ω

(
L̂ (x,∇uk) div h+ h · ∇xL̂ (x,∇uk)

)
dx

=
n∑

i,j=1

∫
Ω

DihjDξi
L̂ (x,∇uk)Djuk dx

+
∫

Ω

[
DsL (x, u,∇u)− G (x, u,∇u)

]
h · ∇uk dx,

namely∫
Ω

L (x, u(x), ϑ(|∇uk|)∇uk) div h dx+
∫

Ω

Λ(|∇uk|) div h dx



198 MARCO SQUASSINA EJDE-2006/MON. 07

+
1
k

∫
Ω

|∇uk|2 div h dx+
∫

Ω

h · ∇xL̃ (x,∇uk) dx

−
n∑

i,j=1

∫
Ω

DihjDξiL (x, u(x),∇uk)Djuk dx−
n∑

i,j=1

∫
Ω

DihjDξiΛ(|∇uk|)Djuk dx

− 2
k

n∑
i,j=1

∫
Ω

DihjDiukDjuk dx−
∫

Ω

[
DsL (x, u,∇u)− G (x, u,∇u)

]
h · ∇uk dx

= 0 .

Notice that
1
k

∫
Ω

|∇uk|2 div h dx→ 0,
2
k

n∑
i,j=1

∫
Ω

DihjDiukDjuk dx→ 0

as k → +∞. Moreover, since Djuk → Dju and DξiΛ(|∇uk|) ⇀ DξiΛ(|∇u|) in
L2(Ω),

n∑
i,j=1

∫
Ω

DihjDξi
Λ(|∇uk|)Djuk dx→

n∑
i,j=1

∫
Ω

DihjDξi
Λ(|∇u|)Dju dx = 0

as k → +∞ and ∫
Ω

Λ(|∇uk|) div h dx→
∫

Ω

Λ(|∇u|) div h dx = 0

as k → +∞. Since∫
Ω

h · ∇xL̃ (x,∇uk) dx =
∫

Ω

h · ∇xL (x, u(x), ϑ(|∇uk|)∇uk) dx

+
∫

Ω

DsL (x, u(x), ϑ(|∇uk|)∇uk)h · ∇u dx

and being

|∇xL (x, u(x), ϑ(|∇uk|)∇uk)| ≤ c1, |DsL (x, u(x), ϑ(|∇uk|)∇uk)| ≤ c2

for some c1, c2 > 0, one obtains∫
Ω

h · ∇xL̃ (x,∇uk) dx→
∫

Ω

h · ∇xL (x, u,∇u) dx+
∫

Ω

DsL (x, u,∇u)h · ∇u dx.

Furthermore, since there exists c3 > 0 with |L (x, u(x), ϑ(|∇uk|)∇uk)| ≤ c3, one
gets ∫

Ω

L (x, u, ϑ(|∇uk|)∇uk) div h dx→
∫

Ω

L (x, u,∇u) div h dx.

Taking into account that there exists c4 > 0 with∣∣∣∣Dξi
L (x, u(x), ϑ(|∇uk|)∇uk)

[
ϑ′(|∇uk|)

Diuk∇uk
|∇uk|

+ ϑ(|∇uk|)ei
]∣∣∣∣ ≤ c4

and that Djuk → Dju in L2(Ω), one deduces
n∑

i,j=1

∫
Ω

DihjDξiL (x, u,∇uk)Djuk dx→
n∑

i,j=1

∫
Ω

DihjDξiL (x, u,∇u)Dju dx

as k → +∞. Noting that, of course∫
Ω

[
DsL (x, u,∇u)− G (x, u,∇u)

]
h · ∇ukdx



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 199

converges to ∫
Ω

[
DsL (x, u,∇u)− G (x, u,∇u)

]
h · ∇udx

as k →∞, the proof is complete. �

Remark 9.5. Let us observe that a (different) approximation technique was also
used by Guedda and Véron [81] to deal with the particular case L (x, s, ξ) = 1

p |ξ|
p.

Let us now assume that Ω is bounded with Lipschitz boundary and let ν(x)
denote the outer normal to ∂Ω at x (which exists for H n−1-a.e. x ∈ ∂Ω).

Lemma 9.6. Let u ∈ C1(Ω) be a weak solution of (9.1). Then∫
Ω

(L (x, u,∇u) div h+ h · ∇xL (x, u,∇u)) dx

=
n∑

i,j=1

∫
Ω

DihjDξi
L (x, u,∇u)Dju dx−

∫
Ω

G (x, u,∇u)h · ∇u dx

+
∫
∂Ω

[L (x, 0,∇u)−∇ξL (x, 0,∇u) · ∇u] (h · ν) dH n−1

for every h ∈ C1(Ω,Rn).

Proof. Let k ≥ 1 and ϕk : R → [0, 1] be given by

ϕk(s) =


0 if s ≤ 1

k

ks− 1 if 1
k ≤ s ≤ 2

k

1 if s ≥ 2
k ,

(9.6)

and define the Lipschitz map ψk : Rn → [0, 1] by setting

ψk(x) = ϕk(d(x,Rn \ Ω)).

Applying Lemma 9.4 on Rn with ψkh in place of h, one deduces∫
Rn

ψkL (x, u,∇u) div h dx+
∫

Rn

L (x, u,∇u)∇ψk · h dx

+
∫

Rn

ψkh · ∇xL (x, u,∇u) dx

=
n∑

i,j=1

∫
Rn

hjDiψkDξi
L (x, u,∇u)Dju dx

+
n∑

i,j=1

∫
Rn

ψkDihjDξi
L (x, u,∇u)Dju dx−

∫
Rn

G (x, u,∇u)ψkh · ∇u dx.

Taking into account that (ψk) is bounded in BV (Rn) and

∀η ∈ C(Rn,Rn) :
∫

Rn

∇ψk · η dx→ −
∫
∂Ω

η · ν dH n−1,

one has ∫
Rn

L (x, u,∇u)∇ψk · h dx→ −
∫
∂Ω

L (x, 0,∇u)(h · ν) dH n−1



200 MARCO SQUASSINA EJDE-2006/MON. 07

as k → +∞ and
n∑

i,j=1

∫
Rn

hjDiψkDξiL (x, u,∇u)Dju dx→ −
n∑

i,j=1

∫
∂Ω

νihjDξiL (x, 0,∇u)Dju dx.

As observed in [116], clearly one has
n∑

i,j=1

νihjDξi
L (x, 0,∇u)Dju = ∇ξL (x, 0,∇u) · ∇u (h · ν) on ∂Ω.

Since of course ψk(x) → χΩ(x) for each x ∈ Rn, the proof is complete. �

Proof of Theorem 9.3. Clearly, if u ∈ C1(Ω) is a solution of (9.1) one has∫
Ω

a
[
∇ξL (x, u,∇u) · ∇u+ uDsL (x, u,∇u)− uG (x, u,∇u)

]
dx

+
∫

Ω

u∇a · ∇ξL (x, u,∇u) dx = 0
(9.7)

for each a ∈ C1(Ω). The assertion follows by combining (9.7) with Lemma 9.6. �

Remark 9.7. Let N ≥ 2. It is easily seen that Theorem 9.3 has a vectorial
counterpart for solutions u ∈ C1(Ω,RN ) of the system

−div (∇ξk
L (x, u,∇u)) +Dsk

L (x, u,∇u) = Gk(x, u,∇u) in Ω
u = 0 on ∂Ω
k = 1, . . . , N.

See also [116, Proposition 3].

9.3. Non-strict convexity in some particular cases. In this section we will
see that, in some particular cases, the assumption of strict convexity of L (x, s, ·)
can be relaxed to the weaker assumption of convexity. Let Ω be a bounded open
subset of Rn with Lipschitz boundary.

Lemma 9.8. Let F : Ω× Rn → R be a function with F (x, ·) convex and C1 and
F (·, ξ) measurable. Assume that there exist a0 ∈ L1(Ω), a1 ∈ Lp

′
(Ω), 1 < p < +∞,

and b, d > 0 with

|∇ξF (x, ξ)| ≤ a1(x) + b|ξ|p−1, (9.8)

F (x, ξ) ≥ d|ξ|p − a0(x) (9.9)

for a.e. x ∈ Ω and all ξ ∈ Rn. Let (wk) ⊂ Lp(Ω,Rn) and w be such that

wk ⇀ w in Lp(Ω,Rn),
∫

Ω

F (x,wk) dx→
∫

Ω

F (x,w) dx

as k → +∞. Then

F (x,wk) ⇀ F (x,w) in L1(Ω), (9.10)

∇ξF (x,wk) → ∇ξF (x,w) in Lp
′
(Ω) (9.11)

as k → +∞. Moreover, up to a subsequence, |wk|p ≤ ψ for some ψ ∈ L1(Ω).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 201

Proof. Let us define F̃ : Ω× Rn → R by setting

F̃ (x, ξ) = F (x,w(x) + ξ)−F (x,w(x))−∇ξF (x,w(x)) · ξ.

Note that F̃ ≥ 0, F̃ (x, 0) = 0, ∇ξF̃ (x, 0) = 0 and∫
Ω

F̃ (x,wk − w) dx→ 0 as k → +∞. (9.12)

Therefore, since for each ϕ ∈ L∞(Ω)∫
Ω

ϕ∇ξF (x,w) · (wk − w) dx→ 0 as k → +∞,

one has ∫
Ω

ϕ
[
F (x,wk)−F (x,w)

]
dx→ 0 as k → +∞,

which proves (9.10).
Note that, in view of (9.12), up to a subsequence one has F̃ (x,wk(x)−w(x)) → 0

for a.e. x ∈ Ω. Fix now such an x; then by (9.9) up to a subsequence wk(x) → y

for some y ∈ Rn, which yields F̃ (x, y−w(x)) = 0. In particular, y−w(x) is a local
minimum for F̃ (x, ·), so that ∇ξF̃ (x, y − w(x)) = 0. Hence we conclude

∇ξF (x,wk(x)) → ∇ξF (x,w(x)). (9.13)

Now, since by (9.12) there exists ψ̃ ∈ L1(Ω) such that

F (x,wk)−F (x,w)−∇ξF (x,w) · (wk − w) ≤ ψ̃,

by (9.9) and Young’s inequality one finds c1, c2 > 0 such that

c1|wk|p ≤ a0 + F (x,w)−∇ξF (x,w) · w + ψ̃ + c2|∇ξF (x,w)|p
′

In particular, in view of (9.8) one deduces |∇ξF (x,wk)| ≤ g for some g ∈ Lp′(Ω),
which combined with (9.13) yields the second assertion. �

9.4. The splitting case. In this subsection we shall deal with the case when
L (x, s, ξ) is of the form α(x, s)β(ξ) + γ(x, s).

Lemma 9.9. Let α, γ ∈W 1,∞(Ω) with α ≥ 0 and β ∈ C1(Rn) convex such that

F (x, ξ) = α(x)β(ξ) + γ(x)

with d|ξ|p − b ≤ β(ξ) ≤ b(1 + |ξ|p), 1 < p < +∞, for some b, d > 0. Let (wk) and
w with

wk ⇀ w in Lp(Ω,Rn),
∫

Ω

F (x,wk) dx→
∫

Ω

F (x,w) dx

as k → +∞. Then

β(wk)∇α(x) ⇀ β(w)∇α(x) in L1(Ω), (9.14)

as k → +∞.

Proof. If Ω0 denotes the set where α = 0, one may argue on

Ω \ Ω0 =
+∞⋃
h=1

Ωh, Ωh =
{
x ∈ Ω : α(x) >

1
h

}
.

By Lemma 9.8 there exists ψ ∈ L1(Ω) such that

χΩ\Ωh
(x)β(wk(x))∇α(x) ≤ ψ(x)



202 MARCO SQUASSINA EJDE-2006/MON. 07

up to a subsequence; hence for each ε > 0 one finds h0 ≥ 1 such that∫
Ω\Ωh0

β(wk(x))∇α(x) dx < ε

uniformly with respect to k. On the other hand, again by Lemma 9.8 one knows
that

F (x,wk) ⇀ F (x,w) in L1(Ωh0)
as k → +∞, which implies

α(x)β(wk) ⇀ α(x)β(w) in L1(Ωh0).

Then since 1/α ∈ L∞(Ωh0) one gets β(wk) ⇀ β(w) in L1(Ωh0), which yields
(9.14). �

Theorem 9.10. Let u : Ω → R be a locally Lipschitz solution of (9.3). Assume
that there exist α, γ ∈ C1(Ω×R) and β ∈ C2(Rn) convex such that α ≥ 0, β(0) = 0
and

L (x, s, ξ) = α(x, s)β(ξ) + γ(x, s).
Then

n∑
i,j=1

∫
Ω

[
DjuDihj + uDia

]
DξiL (x, u,∇u) dx

+
∫

Ω

a
[
∇ξL (x, u,∇u) · ∇u+ uDsL (x, u,∇u)

]
dx

=
∫

Ω

[
L (x, u,∇u) div h+ h · ∇xL (x, u,∇u)

]
dx

=
∫

Ω

[
h · ∇u+ au

]
G (x, u,∇u) dx

(9.15)

holds for each a ∈ C1
c (Ω) and h ∈ C1

c (Ω,Rn).

Proof. Let θ, Λ, L̃ and (uk) ⊂ H1
0 (Ω) be as in Lemma 9.4. We apply Lemma 9.8

choosing

wk = ∇uk, F (x, ξ) = Λ(|ξ|) or F (x, ξ) = β(θ(|ξ|)ξ).

By (9.11) one has
n∑

i,j=1

∫
Ω

DihjDξi
Λ(|∇uk|)Djuk dx→

n∑
i,j=1

∫
Ω

DihjDξi
Λ(|∇u|)Dju dx = 0,

and the term
n∑

i,j=1

∫
Ω

Dihjα(x, u)Dξi
β(ϑ(|∇uk|)∇uk)Djuk dx

goes to
n∑

i,j=1

∫
Ω

Dihjα(x, u)Dξi
β(∇u)Dju dx

as k → +∞. Moreover, by (9.10) one obtains∫
Ω

Λ(|∇uk|) div h dx→
∫

Ω

Λ(|∇u|) div h dx = 0,



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 203∫
Ω

α(x, u)β(ϑ(|∇uk|)∇uk) div h dx→
∫

Ω

α(x, u)β(∇u) div h dx

as k → +∞. Finally, by (9.14) of Lemma 9.9 one gets∫
Ω

h · ∇xL̃ (x,∇uk) dx→
∫

Ω

[
h · ∇xα(x, u) +Dsα(x, u)h · ∇u

]
β(∇u) dx

+
∫

Ω

[∇xγ(x, u) +Dsγ(x, u)]h · ∇u dx

as k → +∞. Then (9.15) follows by exploiting the proof of Lemma 9.4. �

At this point, arguing as in Lemma 9.6 and taking into account (9.7), we obtain
the following result.

Theorem 9.11. Let u ∈ C1(Ω) be a weak solution of (9.1). Let α, γ ∈ C1(Ω×R)
and β ∈ C2(Rn) convex such that α ≥ 0, β(0) = 0 and

L (x, s, ξ) = α(x, s)β(ξ) + γ(x, s).

Then (9.2) holds for each a ∈ C1(Ω) and h ∈ C1(Ω,Rn).

9.5. The one-dimensional case. In this subsection we assume that Ω is an in-
terval in R and L : Ω×R×R → R is of class C1 with L (x, s, ·) convex and DξL
of class C1.

Theorem 9.12. Let u : Ω → R be a locally Lipschitz solution of (9.3). Then
(9.15) holds for each a ∈ C1

c (Ω) and h ∈ C1
c (Ω).

Theorem 9.13. Let u ∈ C1(Ω) be a weak solution of (9.1). Then (9.2) holds for
each a ∈ C1(Ω) and h ∈ C1(Ω).

Taking into account next result, the above theorems follow arguing as in the
proof of Lemmas 9.4 and 9.6.

Lemma 9.14. Let F : Ω × R → R be a C1 function with F (x, ·) convex. Let
(wk) ⊂ Lp(Ω) and w be such that

wk ⇀ w in Lp(Ω),
∫

Ω

F (x,wk) dx→
∫

Ω

F (x,w) dx

as k → +∞ and assume that (DxF (x,wk)) is bounded in Lq for some q > 1. Then

DxF (x,wk) ⇀ DxF (x,w) in L1(Ω) (9.16)
as k → +∞.

Proof. Let us set, for each x ∈ Ω,

y[(x) = lim inf
k

wk(x), y](x) = lim sup
k

wk(x).

Notice that one has
y[(x) ≤ w(x) ≤ y](x) (9.17)

for a.e. x ∈ Ω. Without loss of generality, one can replace wk(x) by its projection
onto [y[(x), y](x)] ; in particular

y[(x) ≤ wk(x) ≤ y](x) (9.18)

for a.e. x ∈ Ω. Arguing as in the proof of Lemma 9.8 one obtains

F̃ (x, y[(x)− w(x)) = 0, F̃ (x, y](x)− w(x)) = 0



204 MARCO SQUASSINA EJDE-2006/MON. 07

for a.e. x ∈ Ω. Then, by F̃ ≥ 0 and the convexity of F̃ (x, · ) one has

F̃ (x, (1− θ)y[(x) + θy](x)− w(x)) = 0

for every θ ∈ [0, 1] and a.e. x ∈ Ω. This yields

F
(
x, (1− ϑ)y[(x) + ϑy](x)

)
= (1− ϑ)F (x, y[(x)) + ϑF (x, y](x)) (9.19)

for a.e. x ∈ Ω. For each m ≥ 1 let us set

Ωm =
{
x ∈ Ω : y](x)− y[(x) ≥

1
m

}
.

By Lusin’s theorem, for each ε > 0 there exists a closed subset Cm,ε ⊂ Ωm such
that

y[
∣∣
Cm,ε

, y]
∣∣
Cm,ε

are continuous, L1(Ωm \ Cm,ε) < ε,

where L1 denotes the one-dimensional Lebesgue measure. We also cut off from Cm,ε
the negligible set of isolated points. Let us now take x ∈ Cm,ε and (xk) ⊂ Cm,ε
with xk → x. If δ > 0 is sufficiently small, by continuity one has

y[(xk) ≤ y[(x) + δ < y](x)− δ ≤ y](xk) (9.20)

for each k ∈ N large enough. By (9.19), for each ϑ ∈ [0, 1] one obtains

F (x, (1− ϑ)(y[(x) + δ) + ϑ(y](x)− δ))

= (1− ϑ)F (x, y[(x) + δ) + ϑF (x, y](x)− δ).

Moreover, (9.20) implies

F (xk, (1− ϑ)(y[(x) + δ)+ ϑ(y](x)− δ))

= (1− ϑ)F (xk, y[(x) + δ) + ϑF (xk, y](x)− δ).

Therefore, combining the previous identities yields

DxF (x, (1− ϑ)(y[(x) + δ)+ ϑ(y](x)− δ))

= (1− ϑ)DxF (x, y[(x) + δ) + ϑDxF (x, y](x)− δ)

for each ϑ ∈ [0, 1]. Letting δ → 0 one obtains

DxF (x, (1− ϑ)y[(x)+ϑy](x)) = (1− ϑ)DxF (x, y[(x)) + ϑDxF (x, y](x))

for each ϑ ∈ [0, 1]. By (9.17) and (9.18) we can choose

ϑ =
w(x)− y[(x)
y](x)− y[(x)

, ϑk =
wk(x)− y[(x)
y](x)− y[(x)

.

Then one gets

DxF (x,w(x)) =
y](x)− w(x)
y](x)− y[(x)

DxF (x, y[(x)) +
w(x)− y[(x)
y](x)− y[(x)

DxF (x, y](x))

and

DxF (x,wk(x)) =
y](x)− wk(x)
y](x)− y[(x)

DxF (x, y[(x)) +
wk(x)− y[(x)
y](x)− y[(x)

DxF (x, y](x)).

In particular, one concludes

DxF (x,wk(x))

= DxF (x,w(x)) + (wk(x)− w(x))
DxF (x, y](x))−DxF (x, y[(x))

y](x)− y[(x)



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 205

for all x ∈ Cm,ε, which implies that

∀ϕ ∈ L∞(Cm,ε) :
∫
Cm,ε

DxF (x,wk)ϕdx→
∫
Cm,ε

DxF (x,w)ϕdx

as k → +∞. On the other hand, since (DxF (x,wk)) is bounded in Lq(Ω), for any
ϕ ∈ L∞(Cm,ε) there exists c > 0 such that∣∣∣ ∫

Ωm\Cm,ε

DxF (x,wk)ϕdx
∣∣∣ ≤ cL1(Ωm \ Cm,ε) < cε.

Letting ε→ 0, one gets

∀ϕ ∈ L∞(Ωm) :
∫

Ωm

DxF (x,wk)ϕdx→
∫

Ωm

DxF (x,w)ϕdx

for each m ≥ 1. Moreover, since on the set

Ω∞ =
{
x ∈ Ω : y](x) = y[(x)

}
one has wk → w pointwise, then

∀ϕ ∈ L∞(Ω∞) :
∫

Ω∞

DxF (x,wk)ϕdx→
∫

Ω∞

DxF (x,w)ϕdx

which concludes the proof. �

9.6. Non-existence results. In the following we want to recall from [116] a gen-
eral variational identity that holds both for scalar-valued and vector-valued ex-
tremals of multiple integrals of calculus of variations that will allow us to get non-
existence results for various classes of problems.

Let Ω be a bounded open subset of Rn, n ≥ 3 and k ≥ 1. For each α ∈ Nn we
set

ξα := ξi · · · ξk, Cα :=
|α|!

α1! · · ·αn!
, Lξα := CαDξi...k

L .

Let now f : W k,p
0 (Ω) → R be the k-th order functional of calculus of variations

f(u) =
∫

Ω

L (x, u,∇u, . . . ,∇ku) dx.

By direct calculation, the Euler-Lagrange’s equation of f is given by
k∑

|α|=0

(−1)|α|DαLξα(x, u, . . . ,∇ku) = 0 in Ω. (9.21)

If u ∈W k,p
0 (Ω) is a weak solution to (9.21) and λ ∈ Ck(Ω), v ∈ Ck(Ω,Rn), we set

ϑ := v · ∇u+ λu, ξγ :=
n∑
i=1

ξα+βξi, ∂αλ,v := [Dα(v ·D + λ)− (v ·D)Dα] .

We now recall the following Pohozǎev-type identity for general lagrangians.

Proposition 9.15. Assume that u ∈ Ck(Ω) is a weak solution to (9.21). Then

div
{
vL (x, u, . . . ,∇ku)−

k−1∑
|α+β|=0

(−1)|β|
CαCβ
Cγ

DαϑDβLξγ (x, u, . . . ,∇ku)
}

= div(v)L (x, u, . . . ,∇ku) + v · ∇xL (x, u, . . . ,∇ku)



206 MARCO SQUASSINA EJDE-2006/MON. 07

−
k∑

|α|=0

∂αλ,vu ·Lξα(x, u, . . . ,∇ku)

for a.e. x ∈ Ω and for each v ∈ Ck(Ω,Rn).

The proof of the above identity follows by direct computation. See [116, section
5].

We now come to the main non-existence result for first order scalar-valued ex-
tremals.

Theorem 9.16. Assume that Ω is star-shaped with respect to 0. Suppose also that

ξ · ∇ξL (x, 0, ξ)−L (x, 0, ξ) ≥ 0 (9.22)

for a.e. x ∈ Ω and all ξ ∈ Rn and that there exists λ ∈ R such that

nL (x, s, ξ)+x ·∇xL (x, s, ξ)−λsDsL (x, s, ξ)− (λ+1)ξ ·∇ξL (x, s, ξ) ≥ 0 (9.23)

for a.e. x ∈ Ω and each (s, ξ) ∈ R × Rn. Let s = 0 or ξ = 0 whenever equality
holds in (8.65). Then the elliptic boundary value problem

−div (∇ξL (x, u,∇u)) +DsL (x, u,∇u) = 0 in Ω (9.24)

has no weak solution u ∈ C1(Ω).

Proof. Let u ∈ C1(Ω) be a weak solution of (9.21). By applying the divergence
Theorem to identity of Proposition 9.15 choosing v(x) = x and k = 1, since u = 0
on ∂Ω, we get∫

∂Ω

[L (x, 0,∇u)−∇ξL (x, 0,∇u) · ∇u] (x · ν) dHn−1∫
Ω

{
nL (x, u,∇u) + x · ∇xL (x, u,∇u)

− λuDsL (x, u,∇u)− (λ+ 1)∇u · ∇ξL (x, u,∇u)
}
dx.

Taking into account that on ∂Ω it is (x · ν) > 0, conditions (9.22) and (9.23) yield
a contradiction. �

Corollary 9.17. Assume that there exists λ ∈ R such that
n∑

i,j=1

(
(n− 2λ− 2)aij(x, s) + x · ∇xaij(x, s)− λsDsaij(x, s)

)
ξiξj ≥ 0

for a.e. x ∈ Ω and each (s, ξ) ∈ R× Rn and

λsg(x, s)− nG(x, s)− x · ∇xG(x, s) > 0

and for a.e. x ∈ Ω and each s ∈ R\{0}. Then the quasi-linear problem

−
n∑

i,j=1

Dj(aij(x, u)Diu) +
1
2

n∑
i,j=1

Dsaij(x, u)DiuDju = g(x, u) in Ω (9.25)

has no weak solution u ∈ C1(Ω).



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 207

Proof. It comes straightforward from the previous result taking

L (x, s, ξ) =
1
2

n∑
i,j=1

aij(x, s)ξiξj

for a.e. x ∈ Ω and each (s, ξ) ∈ R× Rn. �

We now come to the main non-existence result for first order vector-valued ex-
tremals.

Theorem 9.18. Assume that Ω is star-shaped with respect to 0. Suppose also that

ξ · ∇ξL (x, 0, ξ)−L (x, 0, ξ) ≥ 0, (9.26)
for a.e. x ∈ Ω and each ξ ∈ RnN and that there exists λ ∈ R such that

nL (x, s, ξ)+x·∇xL (x, s, ξ)−λu·∇sL (x, s, ξ)−(λ+1)ξ ·∇ξL (x, s, ξ) ≥ 0 (9.27)

for a.e. x ∈ Ω and each (s, ξ) ∈ RN × RnN . Assume further that equality holds
only when either s = 0 or ξ = 0. Then the nonlinear elliptic system

div(∇ξL (x, u,∇u)) +∇sL (x, u,∇u) = 0 (9.28)

has no weak solution u ∈ C2(Ω,RN ) ∩ C1(Ω,RN ).

Proof. Arguing as in the scalar case we obtain the following variational identity

Di

{
viL (x, u,∇u)−

(
vjDju

k + λuk
)
Dξk

i
L (x, u,∇u)

}
= DiviL (x, u,∇u) + viDxi

L (x, u,∇u)−
(
Dju

kDivj + ukDiλ
)
Dξk

i
L (x, u,∇u)

− λ
(
Diu

kDξk
i
L (x, u,∇u) + ukDsk

L (x, u,∇u)
)

where i, j are understood to be summed from 1 to n and k from 1 to N . Therefore,
it suffices to argue as in Theorem 9.16. �

Corollary 9.19. Assume that there exists λ ∈ R such that
n∑

i,j=1

N∑
h,k=1

(
(n− 2λ− 2)ahkij (x, s) + x · ∇xa

hk
ij (x, s)− λs ·Dsa

hk
ij (x, s)

)
ξhi ξ

k
j ≥ 0

for a.e. x ∈ Ω and each (s, ξ) ∈ RN × RnN and

λs · g(x, s)− nG(x, s)− x · ∇xG(x, s) > 0.

and for a.e. x ∈ Ω and each s ∈ RN\{0}. Then the quasi-linear system (` =
1, . . . , N)

−
n∑

i,j=1

N∑
h=1

Dj(ah`ij (x, u)Diuh) +
1
2

n∑
i,j=1

N∑
h,k=1

Ds`
ahkij (x, u)DiuhDjuk = g`(x, u)

(9.29)
has no weak solution u ∈ C2(Ω,RN ) ∩ C1(Ω,RN ).

Proof. It follows by Theorem (9.18) choosing

L (x, s, ξ) =
1
2

n∑
i,j=1

N∑
h,k=1

ahkij (x, s)ξhi ξ
k
j −G(x, s)

for a.e. x ∈ Ω and each (s, ξ) ∈ Rn × RnN . �



208 MARCO SQUASSINA EJDE-2006/MON. 07

Theorem 9.20. Let Ω be star-shaped with respect to the origin and

∇xL (x, s, ξ) · x− n

p∗
DsL (x, s, ξ)s+

{ n
p∗
− n

q

}
λ |s|q ≥ 0 , (9.30)

for a.e. x ∈ Ω and all (s, ξ) ∈ R × Rn. Then (P0,λ) has no nontrivial solution
u ∈ C1(Ω).

Proof. If we define F : Ω× R× Rn → R by setting

∀(x, s, ξ) ∈ Ω× R× Rn : F (x, s, ξ) = L (x, s, ξ)− λ

q
|s|q − 1

p∗
|s|p

∗
,

the first assertion follows, after some computations, by the inequality

nF +∇xF · x− aDsFs− (a+ 1)∇ξF · ξ ≥ 0

where we have chosen a = (n− p)/p (see [116, Theorem 1]). �

Corollary 9.21. Let Ω be star-shaped with respect to the origin, λ ≤ 0 and

p∗∇xL (x, s, ξ) · x− nDsL (x, s, ξ)s ≥ 0, (9.31)

for a.e. x ∈ Ω and all (s, ξ) ∈ R× Rn. Then (P0,λ) admits no nontrivial solution
u ∈ C1(Ω).

Proof. Since q < p∗ and λ ≤ 0, condition (9.31) implies condition (9.30). �

Assume that λ ≤ 0 and L does not depend on x. Then, by the previous result,
the non-existence condition becomes DsL (s, ξ)s ≤ 0. Note that this is precisely
the contrary of our assumption (6.40). Then, from this point of view (6.40) seems
to be natural.

References

[1] Adimurthi, Critical points for multiple integrals of the calculus of variations, Ann. Sc.
Norm. Sup. Pisa 17 (1990), 393–413.

[2] A. Ambrosetti, A perturbation theorem for superlinear boundary value problems, Math.

Res. Center, Univ. Wisconsin-Madison, Tech. Sum. Report 1446 (1974).
[3] A. Ambrosetti, M. Badiale, S. Cingolani, Semiclassical states of nonlinear Schrödinger

equations, Arch. Ration. Mech. Anal. 140 (1997), 285–300.
[4] A. Ambrosetti, G. Prodi, On the inversion of some differentiable mappings with singular-

ities between Banach spaces, Ann. Mat. Pura Appl. 93 (1972), 231–247.

[5] A. Ambrosetti, P.H. Rabinowitz, Dual variational methods in critical point theory and
applications, J. Funct. Anal. 14 (1973), 349–381.

[6] D. Arcoya, L. Boccardo, Critical points for multiple integrals of the calculus of variations,

Arch. Rat. Mech. Anal. 134 (1996), 249–274.
[7] , Some remarks on critical point theory for nondifferentiable functionals, NoDEA

Nonlinear Differential Equations Appl. 6 (1999), 79–100.

[8] G. Arioli, F. Gazzola, Weak solutions of quasilinear elliptic PDE’s at resonance, Ann.
Fac. Sci. Toulouse 6 (1997), 573–589.

[9] , Existence and multiplicity results for quasilinear elliptic differential systems, Comm.

Partial Differential Equations 25 (2000), 125–153
[10] , Quasilinear elliptic equations at critical growth, NoDEA, Nonlinear Differential

Equations Appl. 5 (1998), 83–97.
[11] , Some results on p-Laplace equations with a critical growth term, Differential Inte-

gral Equations 11 (1998), 311–326.

[12] F.V. Atkinson, L.A. Peletier, Elliptic equations with nearly critical growth, J. Differen-
tial Equations 70 (1987), 349–365.

[13] J.P. Aubin, I. Ekeland, Applied Nonlinear Analysis, Wiley-Interscience, New York, 1984.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 209

[14] M. Badiale, G. Citti, Concentration compactness principle and quasilinear elliptic equa-

tions in Rn, Comm. Partial Differential Equations 16 (1991), 1795–1818.

[15] A. Bahri, H. Berestyki, A perturbation method in critical point theory and applications,
Trans. Amer. Math. Soc. 267 (1981), 1–32.

[16] A. Bahri, J.M. Coron, On a nonlinear elliptic equation involving the critical Sobolev

exponent: the effect of the topology of the domain, Comm. Pure Appl. Math. 41 (1988),
253–294.

[17] A. Bahri, P.L. Lions, Morse index of some min-max critical points. I. Applications to

multiplicity results, Comm. Pure Appl. Math. 41 (1988), 1027–1037.
[18] , Solutions of superlinear elliptic equations and their Morse indices, Comm. Pure

Appl. Math. 45 (1992), 1205–1215.

[19] V. Benci, G. Cerami, Positive solutions of some nonlinear elliptic problems in exterior
domains, Arch. Rational Mech. Anal. 99 (1987), 283–300.

[20] H. Berestycki, P.L. Lions, Nonlinear scalar field equation I, existence of a ground state,
Arch. Rat. Mech. Anal. 82 (1983), 313–346.

[21] L. Boccardo, The Bensoussan & Co. technique for the study of some critical points prob-

lems, Optimal Control and PDE, J.L. Menaldi et al. ed. Book in honour of Professor Alain
Bensoussan’s 60th birthday, IOS Press (2000).

[22] L. Boccardo, F. Murat, Almost everywhere convergence of the gradients of solutions to

elliptic and parabolic equations, Nonlinear Anal. 19 (1992), 581–597.
[23] L. Boccardo, B. Pellacci, Bounded positive critical points of some multiple integrals of

the calculus of variations, Progr. Nonlinear Differential Equations Appl. 54, Birkhäuser,

Basel, 2003, 33–51.
[24] P. Bolle, On the Bolza Problem, J. Differential Equations 152 (1999), 274–288.

[25] P. Bolle, N. Ghoussoub, H. Tehrani, The multiplicity of solutions in nonhomogeneous

boundary value problems, Manuscripta Math. 101 (2000), 325–350.
[26] H. Brezis, Problèmes unilatéraux, J. Math. Pures Appl. 51 (1972), 1–68.

[27] H. Brézis, F.E. Browder, Sur une propriété des espaces de Sobolev, C. R. Acad. Sc. Paris
287 (1978), 113–115.

[28] H. Brézis, L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical

Sobolev exponent, Comm. Pure Appl. Math. 36 (1983), 437–477.
[29] A.M. Candela, A. Salvatore, Multiplicity results of an elliptic equation with non homo-

geneous boundary conditions, Topol. Meth. Nonlinear Anal. 11 (1998), 1–18.

[30] A.M. Candela, A. Salvatore, M. Squassina, Multiple solutions for semilinear elliptic
systems with nonhomogeneous boundary conditions, Nonlinear Anal. 51 (2002), 249–270.

[31] A. Canino, On a jumping problem for quasilinear elliptic equations, Math. Z. 226 (1997),

193–210.
[32] , On a variational approach to some quasilinear problems, Serdica Math. J. 22 (1996),

399–426.
[33] , Multiplicity of solutions for quasilinear elliptic equations, Topol. Meth. Nonlinear

Anal. 6 (1995), 357–370.

[34] , On the existence of three solutions for jumping problems involving quasilinear
operators, Topol. Meth. Nonlinear Anal. 18 (2001), 1–16.

[35] , Variational bifurcation for quasilinear elliptic equations, Calc. Var. Partial Differ-

ential Equations 18 (2003), 269–286.
[36] A. Canino, M. Degiovanni, Nonsmooth critical point theory and quasilinear elliptic equa-

tions Topological Methods in Differential Equations and Inclusions, 1–50 - A. Granas, M.

Frigon, G. Sabidussi Eds. - Montreal (1994), NATO ASI Series - Kluwer A.P. (1995).
[37] D. Cao, X. Zhu, The concentration–compactness principle in nonlinear elliptic equations,

Acta Math. Sci. 9 (1989), 307–328.

[38] G. Cerami, D. Fortunato, M. Struwe, Bifurcation and multiplicity results for nonlinear
elliptic problems involving critical Sobolev exponents, Ann. Inst. Henri Poincaré Anal.

Nonlin. 1 (1984), 341–350.
[39] J. Chabrowski, Variational methods for potential operator equations, Walter de Gruyter

Studies in Mathematics 24, Walter de Gruyter & Co., Berlin - New York, 1997.

[40] J. Chabrowski, Weak Convergence Methods for Semilinear Elliptic Equations, World Sci-
entific, Singapore, New Jersey, London, Hong Kong, 1999.



210 MARCO SQUASSINA EJDE-2006/MON. 07

[41] , On multiple solutions for the nonhomogeneous p-Laplacian with a critical Sobolev

exponent, Differential Integral Equations, 8 (1995), 705–716.

[42] K.C. Chang, Infinite dimensional Morse theory and multiple solution problems, Progress
in Nonlinear Differential Equations and their Applications, 6. Birkhäuser, (1993).

[43] , Variational methods for nondifferentiable functionals and their applications to par-

tial differential equations, J. Math. Anal. Appl. 80 (1981), 102–129.
[44] C.C. Chen, C.S. Lin, Uniqueness of the ground state solutions of ∆u + f(u) = 0 in RN ,

N ≥ 3, Comm. Partial Differential Equations 16 (1991), 1549–1572.

[45] G. Citti, Existence of positive solutions of quasilinear degenerate elliptic equations on
unbounded domains, Ann. Mat. Pura Appl. (4) 158 (1991), 315–330.

[46] M. Clapp, Critical point theory for perturbations of symmetric functionals, Comment.

Math. Helv. 71 (1996), 570–593.
[47] F.H. Clarke, I. Ekeland, Nonlinear oscillations and boundary value problems for Hamil-

tonian systems, Arch. Rat. Mech. Anal. 78 (1982), 315–333.
[48] M. Conti, F. Gazzola, Positive entire solutions of quasilinear elliptic problems via non-

smooth critical point theory, Topol. Meth. Nonlinear Anal. 8 (1996), 275–294.

[49] J.N. Corvellec, M. Degiovanni, Nontrivial solutions of quasilinear equations via non-
smooth Morse theory, J. Differential Equations 136 (1997), 268–293.

[50] J.N. Corvellec, M. Degiovanni, M. Marzocchi, Deformation properties for continuous

functionals and critical point theory, Topol. Meth. Nonlinear Anal. 1 (1993), 151–171.
[51] V. Coti Zelati, P. Rabinowitz, Homoclinic orbits for second order Hamiltonian systems

possessing superquadratic potentials, J. Amer. Math. Soc. 4 (1991), 693–727.

[52] R. Courant, D. Hilbert, Methods of Mathematical Physics, Vols. I and II, Interscience,
New York, (1953) and (1962).

[53] B. Dacorogna, Direct methods in the calculus of variations, Springer Verlag, (1988).

[54] G. Dal Maso, F. Murat, Almost everywhere convergence of gradients of solutions to
nonlinear elliptic systems, Nonlinear Anal. 31 (1998), 405–412.

[55] L. Damascelli, M. Ramaswamy, Symmetry of C1 solutions of p-Laplace equations in RN ,
Adv. Nonlinear Stud. 1 (2001), 40–64.

[56] E.N. Dancer, A note on an equation with critical exponent, Bull. London Math. Soc. 20

(1988), 600–602.
[57] E. De Giorgi, Un esempio di estremali discontinui per un problema variazionale di tipo

ellittico, Boll. UMI 4 (1968), 135–137.

[58] M. Degiovanni, M. Marzocchi, A critical point theory for nonsmooth functionals, Ann.
Mat. Pura Appl. (6) 167 (1994), 73–100.

[59] M. Degiovanni, A. Musesti, M. Squassina, On the regularity of solutions in the Pucci-

Serrin identity, Calc. Var. Partial Differential Equations 18 (2003), 317–334.
[60] M. Degiovanni, M. Marzocchi, V. Rădulescu, Multiple solutions of hemivariational in-

equalities with area-type term, Calc. Var. Partial Differential Equations 10 (2000), 355–387.

[61] M. Degiovanni, S. Zani, Euler equations involving nonlinearities without growth condi-
tions, Potential Anal. 5 (1996), 505–512.

[62] M. Del Pino, P. Felmer, Local mountain passes for semilinear elliptic problems in un-
bounded domains, Calc. Var. Partial Differential Equations 4 (1996), 121–137.

[63] , Multi-peak bound states for nonlinear Schrödinger equations, Ann. Inst. H.

Poincaré Anal. Non Linéaire 15 (1998), 127–149.
[64] , Semi-classical states for nonlinear Schrödinger equations, J. Functional Anal. 149

(1997), 245–265.
[65] E. DiBenedetto, C1,α local regularity of weak solutions of degenerate elliptic equations,

Nonlinear Anal. 7 (1983), 827–850.

[66] G-C. Dong, S. Li, On the existence of infinitely many solutions of the dirichlet problem for
some nonlinear elliptic equation, Univ. of Wisconsin Math. Research Center Teach. Rep.

N02161, Dec. (1980).

[67] I. Ekeland, N. Ghoussoub, H. Tehrani, Multiple solutions for a classical problem in the
calculus of variations, J. Differential Equations 131 (1996), 229–243.

[68] A. Floer, A. Weinstein, Nonspreading wave packets for the cubic Schrödinger equation

with a bounded potential, J. Funct. Anal. 69 (1986), 397–408.
[69] B. Franchi, E. Lanconelli, J. Serrin, Existence and uniqueness of ground state solutions

of quasilinear elliptic equations, Springer-Verlag, New York 1 (1988), 293–300.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 211

[70] J. Garcia Azorero, I. Peral Alonso, Existence and nonuniqueness for the p-Laplacian,

Comm. Partial Differential Equations 12 (1987), 1389–1430.

[71] , Multiplicity of solutions for elliptic problems with critical exponent or with a non-
symmetric term, Trans. Amer. Math. Soc. 323 (1991), 887–895.

[72] , On limits of solutions of elliptic problems with nearly critical growth, Comm. Partial

Differential Equations 17 (1992), 2113–2126.
[73] N. Ghoussoub, Duality and perturbation methods in critical point theory, Cambridge Tracts

in Math. 107, Cambridge University Press, Cambridge 1993.

[74] A. Giacomini, M. Squassina, Multi-peak solutions for a class of degenerate elliptic equa-
tions, Asymptotic Anal. 36 (2003), 115–147.

[75] M. Giaquinta, Multiple integrals in the calculus of variations, Princeton University Press,

Princeton, (1983).
[76] B. Gidas, W.M. Ni, L. Nirenberg, Symmetry of positive solutions of nonlinear elliptic

equations in Rn, Mathematical analysis and applications, Part A, pp. 369–402, Adv. in
Math. Suppl. Stud., 7a, Academic Press, New York-London, 1981.

[77] A. Groli, Jumping problems for quasilinear elliptic variational inequalities, NoDEA, Non-

linear Differential Equations Appl. 9 (2002), 117–147.
[78] A. Groli, A. Marino, C. Saccon, Variational theorems of mixed type and asymptotically

linear variational inequalities, Topol. Meth. Nonlinear Anal. 12 (1998), 109–136.

[79] A. Groli, M. Squassina, On the existence of two solutions for a general class of jumping
problems, Topol. Meth. Nonlin. Anal. 21 (2003), 325–344.

[80] , Jumping problems for fully nonlinear elliptic variational inequalities, J. Convex

Anal. 8 (2001), 471–488.
[81] M. Guedda, L. Veron, Quasilinear elliptic problems involving critical Sobolev exponents,

Nonlinear Anal. 13 (1989), 879–902.
[82] L. Jeanjean, K. Tanaka, A remark on least energy solutions in RN , Proc. Amer. Math.

Soc. 131 (2003), 2399–2408.

[83] , Singularly perturbed elliptic problems with superlinear or asymptotically linear
nonlinearities, preprint, (2002).

[84] H. Hofer, Variational and topological methods in partially ordered Hilbert spaces, Math.

Ann. 261 (1982), 493–514.
[85] A. Ioffe, On lower semicontinuity of integral functionals I, Siam J. Control Optim. 15

(1977), 521–538.

[86] A. Ioffe, E. Schwartzman, Metric critical point theory 1. Morse regularity and homotopic
stability of a minimum, J. Math. Pures Appl. 75 (1996), 125–153.

[87] G. Katriel, Mountain pass theorems and global homeomorphism theorems, Ann. Inst. H.

Poincaré Anal. Non Linéaire 11 (1994), 189–209.
[88] J.L. Kazdan, J. Warner, Remarks on some quasilinear elliptic equations, Comm. Pure

Appl. Math. 28 (1975), 567–597.
[89] M.C. Knaap, L.A. Peletier, Quasilinear elliptic equations with nearly critical growth,

Comm. Partial Differential Equations 14 (1989), 1351–1383.

[90] O.A. Ladyzhenskaya, N.N. Ural’tseva, Linear and Quasilinear Elliptic equations Nauka
Press, Academic Press, New York, (1968).

[91] S. Lancelotti, Morse index estimates for continuous functionals associated with quasilinear

elliptic equations, Adv. Differential Equations 7 (2002), 99–128.
[92] S. Lancelotti, A. Musesti, M. Squassina, Infinitely many solutions for poly-harmonic

problems with broken symmetries, Math. Nachr. 253 (2003), 35–44.

[93] G. Li, S. Yan, Eigenvalue problems for quasilinear elliptic equations in Rn, Comm. Partial
Differential Equations. 14 (1989), 1291–1314.

[94] P. Lindqvist, On the equation div(|∇u|p−2∇u) + λ|u|p−2u = 0, Proc. Amer. Math. Soc.

109 (1990), 157–162.
[95] P.L. Lions, The concentration–compactness principle in the calculus of variations. The limit

case, part I, Rev. Matemática Iberoamericana 1 (1985), 145–201.

[96] , The concentration–compactness principle in the calculus of variations. The limit

case, part II, Rev. Matemática Iberoamericana 1 (1985), 45–121.

[97] , The concentration compactness principle in the calculus of variations. The locally
compact case, Ann. Inst. H. Poincaré Anal. Non Linéaire 1 (1984), Part I, 109–145.



212 MARCO SQUASSINA EJDE-2006/MON. 07

[98] A. Marino, A. Micheletti, A. Pistoia, Some variationals results on semilinear ellip-

tic problems with asymptotically nonsymmetric behaviour, Quaderni Sc. Norm. Sup. Pisa

(1991), 243–256.
[99] A. Marino, D. Passaseo, A jumping behavior induced by an obstacle, In M. Gilardi, M.

Matzeu and F. Pacella, editors, Progress in Variational Methods in Hamiltonian Systems

and Elliptic Equations, Pitman (1992), 127–143.
[100] A. Marino, C. Saccon, Some variational theorems of mixed type and elliptic problems

with jumping nonlinearities, Ann. Sc. Norm. Sup. di Pisa, 25 (1997), 631–665.

[101] A. Marino, D. Scolozzi, Geodesics with obstacles, Boll. Un. Mat. Ital. B 2 (1983), 1–31.
[102] M. Marzocchi, Multiple solutions of quasilinear equations involving an area-type term, J.

Math. Anal. Appl. 196 (1995), 1093–1104.

[103] J. Mawhin, M. Willem, Critical point theory and Hamiltonian systems. Applied Mathe-
matical Sciences, 74, Springer-Verlag, New York, (1989).

[104] C. Miranda, Alcune osservazioni sulla maggiorazione in Lν(Ω) delle soluzioni deboli delle
equazioni ellittiche del secondo ordine, Ann. Mat. Pura Appl. 61 (1963), 151–168.

[105] C.B. Morrey, Multiple integrals in the calculus of variations, Springer, (1966).

[106] J. Moser, A sharp form of an inequality by N. Trudinger, Indiana Math. J. 20 (1971),
1077–1092.

[107] A. Musesti, M. Squassina, Asymptotics of solutions for fully nonlinear problems with

nearly critical growth, Z. Anal. Anwendungen 21 (2002), 185–201.
[108] Y.-G. Oh, Existence of semiclassical bound states of nonlinear Schrödinger equations with

potentials of the class (V )a, Comm. Partial Differential Equations 13 (1988), 1499–1519.

[109] , Stability of semiclassical bound states of nonlinear Schrödinger equations with
potentials, Comm. Math. Phys. 121 (1989), 11–33.

[110] S. Paleari, M. Squassina, Multiplicity results for perturbed symmetric quasilinear elliptic
systems, Differential Integral Equation 14 (2001), 785–800.

[111] D. Passaseo, The effect of the domain shape on the existence of positive solutions of the

equation ∆u + u2∗−1 = 0, Topol. Meth. Nonlinear Anal. 3 (1994), 27–54.
[112] B. Pellacci, Critical points for non differentiable functionals, Boll. Un. Mat. Ital. B 11

(1997), 733–749.

[113] , Critical points for some functionals of the calculus of variations, Topol. Meth.
Nonlinear Anal. 17 (2001), 285–305.

[114] B. Pellacci, M. Squassina, Existence of unbounded critical points for a class of lower

semicontinuous functionals, J. Differential Equations 201 (2004), 25–62.
[115] S.I. Pohozaev, Eigenfunctions of the equations ∆u + λf(u) = 0, Soviet Math. Dokl. 6

(1965), 1408–1411.

[116] P. Pucci, J. Serrin, A General Variational Identity, Indiana Math. J. 35 (1986), 681–703.
[117] P. Rabier, C. Stuart, Exponential decay of the solutions of quasilinear second-order equa-

tions and Pohozaev identities, J. Differential Equations 165 (2000), 199–234.
[118] P.H. Rabinowitz, Critical points of perturbed symmetric functionals, Trans. Amer. Math.

Soc. 272, (1982).

[119] , On a class of nonlinear Schrödinger equations, Z. Angew. Math. Phys. 43 (1992),
270–291.

[120] , Minimax methods in critical point theory with applications to differential equations,

CBMS Reg. Conf. Series Math. 65, Amer. Math. Soc., Providence, R.I. (1986).
[121] , Variational methods for nonlinear eigenvalue problems. Eigenvalues of non-linear

problems (Centro Internaz. Mat. Estivo C.I.M.E.), III Ciclo, Varenna, 1974), pp. 139–195.

Edizioni Cremonese, Rome, 1974.
[122] J. Serrin, Local behavior of solutions of quasi-linear equations, Acta Math. 111 (1964),

247–302.

[123] J. Serrin, M. Tang, Uniqueness of ground states for quasilinear elliptic equations, Indiana
Univ. Math. J. 49 (2000), 897–923.

[124] S. Solimini, Some remarks on the number of solutions of some nonlinear elliptic problems,
Ann. Inst. H. Poincaré Anal. Non Linéaire 2 (1985), 143–156.

[125] G. Spradlin, A singularly perturbed elliptic partial differential equation with an almost
periodic term, Calc. Var. Partial Differential Equations 9 (1999), 207–232.

[126] , An elliptic equation with spike solutions concentrating at local minima of the Lapla-
cian of the potential, Electron. J. Differential Equations (2000), No. 32, 14 pp.



EJDE-2006/MON. 07 ON A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS 213

[127] M. Squassina, Two solutions for inhomogeneous fully nonlinear elliptic equations at critical

growth, NoDEA, Nonlinear Differential Equations Appl. 11 (2004), 53–71.

[128] , Multiple critical points for perturbed symmetric functionals associated with quasi-
linear elliptic problems, Nonlinear Anal. 47 (2001), 1605–1616.

[129] , Perturbed S1-symmetric Hamiltonian systems, Applied Math. Letters 14 (2001),

375–380.
[130] , Existence of positive entire solutions for nonlinear elliptic problems, Topol. Meth.

Nonlinear Anal. 17 (2001), 23–39.

[131] , Spike solutions for a class of singularly perturbed quasilinear elliptic equations,
Nonlinear Anal. 54 (2003), 1307-1336.

[132] , Existence of weak solutions to general Euler’s equations via nonsmooth critical

point theory, Ann. Fac. Sci. Toulouse (6) 9 (2000), 113–131.
[133] , Existence of multiple solutions for quasilinear diagonal elliptic systems, Electron.

J. Differential Equations 14 (1999), 1–12.
[134] , On a class of nonlinear elliptic BVP at critical growth, Dyn. Contin. Discrete

Impuls. Syst. Ser. A Math. Anal. 10 (2003), 193–208.

[135] G. Stampacchia, Èquations elliptiques du second ordre à coefficients discontinus, Séminaire

de Mathématiques Supérieures, Les Presses de l’Université de Montréal, Montreal, Que. 16

(1966), 326pp.
[136] M. Struwe, Variational methods. Applications to nonlinear partial differential equations

and Hamiltonian systems, Springer, Berlin (1996).

[137] , Quasilinear elliptic eigenvalue problems, Comment. Math. Helvetici 58, (1983),
509–527.

[138] G. Talenti, Best constant in Sobolev inequality, Ann. Mat. Pura Appl. 110 (1976), 353–
372.

[139] K. Tanaka, Morse indices at critical points related to the symmetric mountain pass theorem

and applications, Comm. Partial Differential Equations 14 (1989), 99–128.
[140] G. Tarantello, On nonhomogeneous elliptic equations involving critical Sobolev exponent,

Ann. Inst. H. Poincaré Anal. Non Linéaire, 9 (1992), 281–304.

[141] P. Tolksdorf, Regularity for a more general class of quasilinear elliptic equations, J. Dif-
ferential Equations 51 (1984), 126–150.

[142] N. Trudinger, On Harnack type inequalities and their application to quasilinear elliptic

equations, Comm. Pure Appl. Math. 20 (1967), 721–747.
[143] L.S. Yu, Nonlinear p-Laplacian problems on unbounded domains, Proc. Amer. Math. Soc.

115 (1992), 1037–1045.

[144] A. Visintin, Strong convergence results related to strict convexity, Comm. Partial Differ-
ential Equations 9 (1984), 439–466.

[145] X. Wang, On concentration of positive bound states of nonlinear Schrödinger equations,
Comm. Math. Phys. 153 (1993), 229–244.

[146] M. Willem, Minimax theorems, Birkhauser Verlag, Basel, 1996.

[147] E. Zeidler, Nonlinear Functional Analysis and its Applications II/A, Springer-Verlag, New
York, 1989.

[148] H.S. Zhou, Solutions for a quasilinear elliptic equation with critical sobolev exponent and

perturbations on Rn, Differential Integral Equations 13 (2000), 595–612.

Marco Squassina
Dipartimento di Matematica e Applicazioni

Università degli Studi di Milano-Bicocca

Via R. Cozzi 53, I-20125 Milano, Italy
E-mail address: marco.squassina@unimib.it


	Preface
	Notation
	1. Introduction
	2. Review of Critical Point Theory
	2.1. Notions of non-smooth analysis
	2.2. The case of lower semi-continuous functionals
	2.3. Functionals of the calculus of variations

	3. Super-linear Elliptic Problems
	3.1. Quasi-linear elliptic systems
	3.2. The concrete Palais-Smale condition
	3.3. Existence of multiple solutions for elliptic systems
	3.4. Regularity of weak solutions for elliptic systems
	3.5. Fully nonlinear scalar problems
	3.6. The concrete Palais-Smale condition
	3.7. Existence of a weak solution
	3.8. Super-linear problems with unbounded coefficients
	3.9. General setting and main results
	3.10. Verification of the key condition
	3.11. The variational setting
	3.12. A compactness result for J
	3.13. Proofs of the main Theorems
	3.14. Summability results

	4. Perturbation from Symmetry
	4.1. Quasi-linear elliptic systems
	4.2. Symmetry perturbed functionals
	4.3. Boundedness of concrete Palais-Smale sequences
	4.4. Compactness of concrete Palais-Smale sequences
	4.5. Existence of multiple solutions
	4.6. Semi-linear systems with nonhomogeneous data
	4.7. Reduction to homogeneous boundary conditions
	4.8. The Palais-Smale condition
	4.9. Comparison of growths for min-max values
	4.10. Bolle's method for non-symmetric problems
	4.11. Application to semi-linear elliptic systems
	4.12. The diagonal case

	5. Problems of Jumping Type
	5.1. Fully nonlinear elliptic equation
	5.2. The main result
	5.3. The concrete Palais-Smale condition
	5.4. Min-Max estimates
	5.5. Proof of the main result
	5.6. Fully nonlinear variational inequalities
	5.7. The main result
	5.8. The bounded Palais-Smale condition
	5.9. The Palais-Smale condition
	5.10. Min-Max estimates
	5.11. Proof of the main result

	6. Problems with Loss of Compactness
	6.1. Positive entire solutions for fully nonlinear problems
	6.2. The concrete Palais-Smale condition
	6.3. Fully nonlinear problems at critical growth
	6.4. The first solution
	6.5. The concrete Palais-Smale condition
	6.6. The second solution
	6.7. One solution for a more general nonlinearity
	6.8. Existence of one nontrivial solution
	6.9. Problems with nearly critical growth
	6.10. The main results
	6.11. The weak limit
	6.12. Proof of the main results
	6.13. Mountain-pass critical values

	7. The Singularly Perturbed Case, I
	7.1. The del Pino-Felmer penalization scheme
	7.2. Energy estimates and concentration
	7.3. Proof of the main result
	7.4. A few related open problems

	8. The Singularly Perturbed Case, II
	8.1. Penalization and compactness
	8.2. Two consequences of the Pucci-Serrin identity
	8.3. Energy estimates
	8.4. Proofs of the main results

	9. Nonexistence Results
	9.1. A general Pucci-Serrin type identity
	9.2. The approximation argument
	9.3. Non-strict convexity in some particular cases
	9.4. The splitting case
	9.5. The one-dimensional case
	9.6. Non-existence results

	References