INTERNATIONAL CENTER FOR THEORETICAL PHYSICS

TRIESTE

Second School of Nonlinear Functional Analysis

and Applications to Differential Equations

21 April-9 May
1997

Multiplicity of Solutions
for
the p-Laplacian

Ireneo Peral

Departamento de Matemáticas
Universidad Autónoma de Madrid
Madrid, Spain

Miramare-Trieste
Italy
Work partially supported by DGICYT project PB94-0187 MEC Spain
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1 Eigenvalue problem 9
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Existence of eigenvalues by minimax techniques . . . . . . . . . . . . . . . . . . . 9
1.3 Asymptotic behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Isolation of the first eigenvalue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Eigenvalues in the radial case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Bifurcation for the p-Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Multiplicity of solutions: Growth versus Shape 33

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 First results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Critical Problems: The Palais-Smale condition . . . . . . . . . . . . . . . . . . . . 39
2.4 Local results on multiplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.1 The critical case and p∗ > q > p . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.2 The critical case and 1 < q < p . . . . . . . . . . . . . . . . . . . . . . . . 46
2.5 Global results on multiplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.6 Multiple bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.7 Further results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3 What means Growth? 65

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2 Hardy inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3 The Dirichlet problem with singular potential . . . . . . . . . . . . . . . . . . . . 68
3.4 Critical potential and subcritical growth . . . . . . . . . . . . . . . . . . . . . . . 73
3.5 Critical potential and critical growth . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.6 Subcritical potential and critical growth . . . . . . . . . . . . . . . . . . . . . . . 75

1
2

Appendix 77

A The inverse operator of the p-Laplacian 79

B Genus and its properties 85

C The variational principle of Ekeland 87

D The Mountain Pass Theorem 91

E Boundedness of solutions 97

F Two Lemmas by P.L. Lions 103

Bibliography 107
3
4
 5

Preface
This lectures notes are a review of some contributions to the theory of Quasilinear Elliptic
Equations with a reaction term. We study the existence, multiplicity, positivity of solutions,
bifurcation, etc., with respect to the relation between the growth and shape of the zero order
reaction term.

We can propose a quite general class of elliptic problems involving the p-laplacian, for
instance, (
−∆p u ≡ −div (|∇u|p−2 ∇u) = F ′ (u) in Ω,
(P )
u|δΩ = 0,
where 1 < p < N , Ω is a regular bounded domain in IRN , and
F1) F ′ is increasing, F (0) = 0, F regular,
F2) |F ′ (u)| ≤ C(1 + |u|r ), for some r > 0.
Np
If we consider p∗ = , the Sobolev theorem gives the inclusion
N −p
W01,p (Ω) ⊂ Lp .
∗

We say that the problem (P) is subcritical, critical or supercritical if, r < p∗ − 1, r = p∗ − 1
or r > p∗ − 1 respectively. If p ≥ N the problem (P ) is always subcritical.
This type of problems has been studied extensively for p = 2 in many contexts. Thus
in Riemannian Geometry the critical case for p = 2 in the Yamabe problem, (see [93]). In
Astrophysics there appears some supercritical problem, (see [32] for instance). When F (u) = λeu ,
we have the model of Frank-Kamenetstkii for solid ignition, (see [48]).
In recent years there has been an increasing interest in looking at these problems in the
more general framework of quasilinear elliptic equations for which (P) is a model case. In general,
problem (P) can be seen as the stationary counterpart of evolution equations with nonlinear
diffusion. Here it is necessary to say that the growth of the second member is important to get
the blow-up of the solutions of the evolution problems, but it is clearly insufficient to conclude
other properties, such that global existence or stability of the equilibrium states. For such sort
of results it is necessary to know more deeply the behaviour of the associated elliptic problem,
which depends on the growth and also on other properties of the zero order term.
The subcritical case is more or less easy, and we refer to [51] and the references there. The
generality of problem (P) is enough to have a nontrivial project in our hands. The difficulty of
considering problem (P) in its full generality can be understood by the following well known
facts.
6

N +2
1. If Ω is a bounded starshaped domain in IRN , N > 2, p = 2 and F (u) = λu N −2 , λ > 0,
problem (P ) has no positive solution, as was remarked by Pohozaev in his famous paper
[71].
∗ −1
2. Under the same hypotheses about the domain, if we assume 1 < p < N and F (u) = λup ,
λ > 0, Pucci and Serrin show that problem (P) has no positive solution. See [73].

3. It is very easy to prove that if Ω = {x ∈ IRN |0 < a < |x| < b} and the same hypotheses as
in (1) or (2), problem (P) has positive solutions. In fact, if we look for radial solutions, it
is an easy one-dimensional problem. See for instance [61].

4. The behaviour of the problem in (3) when a → 0, is not evident. The answer to this
question was given by Bahri-Coron, [19]. More precisely: If some fundamental group of the
N +2
domain is non trivial then problem (P ) with p = 2 and the zero order term F (u) = λu N −2 ,
λ > 0, has a nontrivial solution.
At this level it seems that the existence of nontrivial solution of (P ) depends strongly on
the topology of the domain.
This idea is not totally exact.

5. In [40] and [36], independently, it is obtained that there are contractible domains for
which the problem in item (1) has a nontrivial solution. In particular, this fact means that
the existence result depends deeply on the geometry of the domain and not only on the
topology.

The rough consequence from the remarks above is the strong dependence of the results on
the domain.

On the other hand we have as starting point the seminal result by Brezis and Nirenberg in
[29] that we can summarize as follows.
Theorem (Brezis-Nirenberg) Consider the problem
( 4

(LP ) −∆u = λ|u|q−2 u + |u| (N −2) u in Ω,

u|δΩ = 0,

where Ω is a bounded domain in IRN and λ > 0. Then,

a) If q = 2 and N ≥ 4, there exists a positive solution of (LP), for all 0 < λ < λ1 , λ1 being the
first eigenvalue of the laplacian in Ω.
 7

b) If q = 2 and N = 3, there exists λ0 depending on Ω such that the problem (LP) admits a
positive solution if 0 < λ0 < λ < λ1 . Moreover if Ω is a ball then the problem has a positive
soltution if and only if λ1 /4 < λ < λ1 , λ1 first eigenvalue of the laplacian.
c) If 2 < q < N2N
−2 and N ≥ 4, the problem (LP) admits a positive solution for all λ > 0. In
dimension N = 3 there exists positive solution for large λ.

The Theorem above means that the action of the term λ|u|q−2 u breaks down the nonexis-
tence obstruction found by Pohozaev [71]. Therefore, Pohozaev’s result depends not only on the
growth of the second member in the equation and on the geometry of the domain, but also on
the precise expression of F .

We will emphasize this last fact in these notes and we will obtain results that depends in a
deep way on the shape of F .

In first place we deal with the typical eigenvalue problems in Chapter 1; we will study
existence, asymptotic behaviour, properties and, as a consequence, some results on bifurcation.
Chapter 2 will be devoted mainly to study the influence of the shape of the zero order term in
the behaviour of the elliptic problems with respect to multiplicity of solutions and bifurcation
properties.
Chapter 3 is devoted to a critical problem related with a Hardy inequality. It seems that
it can be considered as a starting point for a more extensive research. It shows how the depen-
dence on x of the second member, (non autonomous case) changes the concept of growth to be
considered.
General references for different aspect on the topics considered here are the books, [6], [12],
[17], [20], [26], [34], [59], [63], [67], [76], [81], [83], [91], [92] and [94]. But some general results
that will be used often, are collected in several appendix at the end of the lectures notes.

* * *
I would like to thank the organizers of the Second School of Nonlinear Functional Analysis
and Applications to Differential Equations at ICTP of Trieste, Professors Ambrosetti, Chang
and Ekeland, the opportunity that they give to me to teach this course. Also my thanks to
Professors Garcia Azorero and Walias by their helpful comments and missprint correction. Also
my gratitude to J.A. Aguilar by his careful final missprint correction.
Very special thanks to A. Ambrosetti for his friendship and his teaching in these years.

Madrid, February 1997

I. Peral
8
Chapter 1

Eigenvalue problem

1.1 Introduction
This chapter will be devoted to the typical eigenvalue problem for the p-Laplacian, namely, the
problem, (
−∆p u ≡ −div (|∇u|p−2 ∇u) = λ|u|p−2 u, x ∈ Ω,
(1.1)
u = 0 x ∈ ∂Ω,
where 1 < p and Ω ⊂ IRN is a bounded domain. We look for nontrivial solutions, i.e., (u, λ) ∈
W01,p (Ω) × IR, u 6= 0, verifiying the problem in a weak sense.
We call this case the typical eigenvalue problem because if we have a solution (u, λ) to
problem (1.1), then for all α ∈ IR, (αu, λ) is a solution. With different homogeneity in the
second member we get a curve of solutions for each fixed solution to the problem with λ = 1,
simply by a scaling, namely, if u 6= 0 is solution of −∆p u = |u|q−2 u then if α > 0, v = αu verifies
the same equation with λ = αp−q .

1.2 Existence of eigenvalues by minimax techniques

In this section we study the existence of a sequence of eigenvalues for problem (1.1). The result
that we will explain can be found in [51]. The method used in the proof is an adaptation of the
methods in [4], i.e., the Lusternik-Schnirelman theory.
Consider Z
1,p 1
B : W0 (Ω) −→ IR, defined by B(u) = |u|p dx, (1.2)
p
Ω
that is the potential of
b : W01,p (Ω) −→ W −1,p (Ω)
′
defined by b(u) = |u|p−2 u. (1.3)

9
10

It is easy to check that:

1. By the compactness of the Sobolev inclusion, b is compact and uniformly continuous on
bounded sets.

2. As a consequence B is compact.

3. b is odd and B is even.

The idea is to obtain critical points of B(u) on the manifold
Z
1
M = {u ∈ W01,p (Ω) | ( ) |∇u|p dx = α }. (1.4)
p
Ω

For each u ∈ W01,p (Ω) − {0} we find λ(u) > 0 in such a way that λ(u)u ∈ M, i.e.,
 1/p
 pα 
λ(u) =  R p  (1.5)
|∇u| dx
Ω

hence λ : W01,p (Ω) − {0} −→ (0, ∞) is even and bounded on subsets in the complementary of
the origin. Moreover the derivative of λ is
 −(p+1)/p
Z
1 1 1
hλ′ (u), vi = − α p  |∇u|p dx h−∆p u, vi, (1.6)
p p
Ω

for all v ∈ W01,p (Ω) where h·, ·i denotes the duality product. Therefore,

h−∆p u, vi = 0 implies hλ′ (u), vi = 0,

h−∆p u, ui = pα, if u ∈ M
and, for all u ∈ W01,p (Ω) − {0},
 p
1 αp
h−∆p u, ui = h−∆p (λ(u)u), λ(u)ui =
λ(u) (λ(u))p
The next step is to construct a flow in M related to B and the corresponding deformation
lemma that allows us to apply the mini-max theory. (See [76]). By direct computation we get
the derivative of B(λ(u)u) in u ∈ M, that according to (1.6), becomes
hb(u), ui ′
D(u) = b(u) − (−∆p u) ∈ W −1,p (Ω). (1.7)
h−∆p u, ui
 11

We need the tangent component of D(u) to M. Then we will use the duality mapping
1 1
J : W −1,p (Ω) −→ W01,p (Ω),
′
+ ′ = 1,
p p

which gives the possibility to consider J(Du) ∈ W01,p (Ω) and gives its tangent component. It is
′
well known that J verifies for all f ∈ W −1,p (Ω),

i) hf, J(f )i = ||f ||2W −1,p′ (Ω)

ii) ||J(f )||W 1,p (Ω) = ||f ||W −1,p′ (Ω) .

0

(See for instance [26]). Taking into account the properties of J and the previous comments we
define
h−∆p u, J(D(u))i
T : M −→ W01,p (Ω), by T (u) = J(D(u)) − u (1.8)
−h∆p u, ui
and then, T (u) is the tangent component that we are looking for, because

h−∆p u, T (u)i = 0 if u ∈ M,

and also by (1.6)

hλ′ (u), T (u)i = 0 if u ∈ M.
Moreover, by definition, T is odd, uniformly continuous and bounded on M; in particular we
can find a γ0 > 0, t0 > 0, such that

||u + tT (u)|| ≥ γ0 , for all (u, t) ∈ M × [−t0 , t0 ].

As a consequence we define the flow

(
H : M × [−t0 , t0 ] −→ M
(1.9)
(u, t) −→ H(u, t) = λ(u + tT u)(u + tT u).

Then H verifies:

i) Is odd as function of u, namely, H(u, t) = −H(−u, t) for fixed t.

ii) Is uniformly continuous.

iii) H(u, 0) = u for u ∈ M.

The next result shows that H define trajectories on M for which the functional B is
increasing. This is the relevant property of H in order to get a deformation result.
12

Lemma 1.2.1 Let H be defined by (1.9). Then there exists

r : M × [−t0 , t0 ] → IR

such that lim r(u, τ ) = 0 uniformly on u ∈ M, and:

τ →0
Z t
B(H(u, t)) = B(u) + {||Du||2 + r(u, s)}ds
0

for all u ∈ M, t ∈ [−t0 , t0 ].

Proof. We have < B ′ (u), v >=< b(u), v > and H(u, 0) = u hence
Z t ∂
B(H(u, t)) = B(u) + hb(H(u, s)), H(u, s)ids. (1.10)
0 ∂s
But
∂ ∂
H(u, s) = (λ(u + sT (u))(u + sT (u)) =
∂s ∂s
hλ′ (u + sT (u)), T (u)i(u + sT (u)) + λ(u + sT (u))T (u) =
hλ′ (u + sT (u)) − λ′ (u), T (u)i(u + sT (u))+
(λ(u + sT (u)) − λ(u)) T (u) + λ(u)T (u).
If we call,
R(u, s) =
hλ′ (u + sT (u)) − λ′ (u), T (u)i(u + sT (u))+
(λ(u + sT (u)) − λ(u)) T (u)
∂
we have H(u, s) = R(u, s) + T (u). Now, because T is bounded and λ and λ′ are uniformly
∂s
continuous,
lim R(u, s) = 0, uniformly in u ∈ M.
s→0

Hence (1.10) becomes

R
B(H(u, t)) = B(u) + 0t < b(H(u, s)), R(u, s) + T (u) > ds =
R
B(u) + 0t hb(u), T (u)i + r(u, s)ds,

where
r(u, s) = hb(u), R(u, s)i + hb(H(u, s)) − b(u), R(u, s) + T (u)i.
and also r(u, s) → 0 uniformly in u ∈ M, as s → 0. But

hb(u), T (u)i = hD(u), J(D(u))i = ||Du||2W −1,p′ (Ω) ,

 13

and we conclude.
We can now formulate the necessary deformation result. For β > 0, consider the level sets,

Λb = { u ∈ M | B(u) ≥ β } (1.11)

then

Lemma 1.2.2 Let β > 0 be fixed and assume that there exists an open set U ⊂ M, such that
for some constants 0 < δ, 0 < ρ < β, the condition

||D(u)||W −1,p′ (Ω) ≥ δ, if u ∈ Vρ = {u ∈ M| u ∈

/ U, |B(u) − β| ≤ ρ}

is verified. Then, there exists ε > 0, and an operator Hε odd and continuous such that

Hε (Λβ−ε − U ) ⊂ Λβ+ε .

Proof. With t0 and r(u, s) as in the Lemma 1.2.1, consider t1 ∈ (0, t0 ] such that for s ∈ [−t1 , t1 ]

1
|r(u, s)| ≤ δ 2 , for all u ∈ M.
2
Then, for u ∈ Vρ and t ∈ [0, t1 ],
Rt 2
B(H(u, t)) = B(u) + 0 {||Du|| + r(u, s)}ds
Rt 1 2 1
≥ B(u) + 2
0 (δ − δ )ds = B(u) + δ 2 t.
2 2
1
Choosing ε = min{ρ, δ 2 t1 }. If u ∈ Vρ ∩ Λβ−ε , we have |B(u) − β| ≤ ρ and then
4
B(H(u, t1 )) ≥ B(u) + 2ε ≥ β + ε.

By Lemma 1.2.1, fixed u ∈ Vρ , B(H(u, ·)) is increasing in some interval [0, s0 ) ⊂ [0, t1 ). Then
for u ∈ Vε = {u ∈ M| u ∈
/ U, |B(u) − β| ≤ ε} the functional

tε (u) = min{t ≥ 0 | B(H(u, t)) = β + ε}.

is well defined and verifies:

i) 0 < tε (u) ≤ t1 ,
ii) tε (u) is continuous in Vε .
Define
Hε : Λβ−ε − U → Λβ+ε
14

by (
H(u, tε (u)), if u ∈ Vε ,
Hε (u) =
u, if u ∈ Λβ−ε − {U ∪ Vε }.
In this way Hε is odd and continuous.
We will prove the existence of a sequence of critical values and critical points by using a
mini-max argument on the class of sets defined below. For each k ∈ IN consider,

Ck = { C ⊂ M | C compact , C = −C γ(C) ≥ k}, (1.12)

where γ is the genus defined in the Appendix. The main result on existence of eigenvalues is the
following

Theorem 1.2.3 Let Ck be defined by (1.12) and let βk be defined by

βk = sup min B(u). (1.13)

C∈Ck u∈C

Then, βk > 0, and there exists uk ∈ M such that B(uk ) = βk , and uk is a solution to problem
α
(1.1) for λk = .
βk

Proof. The inequality βk > 0 is a consequence of γ(M) = +∞, namely, for all k > 0, Ck 6= ∅,
and given C ∈ Ck , we have min B(u) > 0, hence βk > 0 for all k. Let k be a fixed positive
u∈C
integer. In the general hypotheses there exists a sequence {uj } ⊂ M such that

a) B(uj ) → βk ,
(1.14)
b) D(uj ) → 0.

In fact if not, there must exists δ, ρ > 0, such that

||Du|| ≥ δ if u ∈ { u ∈ M | |B(u) − βk | ≤ ρ}.

We can assume δ < βk and then by the previous deformation Lemma 1.2.2 with U = ∅,
there exists ε > 0 and a odd continuous mapping Hε , such that

Hε (Λβk −ε ) ⊂ Λβk +ε .

From (1.13) there exists Cε ∈ Ck such that B(u) ≥ βk − ε en Cε ; namely, Cε ⊂ Λβk −ε . Then,
B(u) ≥ βk + ε in Hε (Cε ).
Now γ(Hε (Cε )) ≥ k, because γ(Cε ) ≥ k, hence Hε (Cε ) ∈ Ck , that is a contradiction with
the definition of βk .
 15

In this way we find a sequence, {uj }, verifying (1.14) and then, for some subsequence,
uj ⇀ uk in W01,p (Ω). Since B is compact, B(uj ) → βk and by continuity D(uk ) = 0, so we find
α
−∆p uk = λk |uk |p−2 uk , with λk =
βk

To finish this section, we will prove that the sequence of critical values {βk } is infinite,
namely that we have a true sequence of eigenvalues.

Lemma 1.2.4 Let βk be defined in (1.13). Then lim βk = 0 .

k→∞
(As a consequence λk = αβk−1 → ∞ as k → ∞).

Proof.
Consider {Ej } sequence of linear subspaces in W01,p (Ω) such that
i) Ek ⊂ Ek+1 ,
ii) L(∪Ek ) = W01,p (Ω),
iii) dim Ek = k.
Define
β˜k = sup min
c
B(u) (1.15)
C∈Ck u∈C∩Ek−1

where Ekc is the linear and topological complementary of Ek .

Obviously β˜k ≥ βk > 0. So it is sufficient to prove that lim β˜k = 0.
k→0
Now, if for some positive constant, γ > 0, β˜k > γ > 0 for all k ∈ IN , then for each k ∈ IN
there exists Ck ∈ Ck such that
β˜k > min c B(u) > γ,
u∈Ck ∩Ek−1
c
and then there exists uk ∈ Ck ∩ Ek−1 such that

β˜k > B(uk ) > γ.

In this way {uk } ⊂ M, B(uk ) > γ > 0 for all k ∈ IN , hence for some subsequence,

uk ⇀ v weakly in W01,p (Ω),

uk → v in Lp (Ω).
c , implies v = 0.
As a consequence b(v) > γ and this is a contradiction because uk ∈ Ek−1
In the next section we will give a more precise result about the asymptotic behaviour of
the sequence {βk }.
16

1.3 Asymptotic behaviour

Consider the sequence of eigenvalues
Z
λk = αβk−1 , where βk = sup inf |u|p dx, k ∈ IN (1.16)
F ∈Ck (Ω) u∈F
Ω

which has be found in the previous section.

We obtain the following result that gives the asymptotic behaviour of the sequence of
eigenvalues obtained above. The case p = 2 is classical.
Theorem 1.3.1 The sequence {λk }k∈IN of eigenvalues of −∆p defined by (1.16), verifies that
there exist two constants, c(Ω), C(Ω), depending only on the domain Ω, such that

c(Ω)k p/N ≤ λk ≤ C(Ω)k p/N , for all k ∈ IN , (1.17)

where N is the dimension.

Step 1. Dimension N = 1. A fundamental role in proving the estimate is played by the following
Bernstein’s Lemma, which gives the inequality,

λk ≤ Ck p .

Lemma 1.3.2 Let S(x) be a n-degree trigonometric polynomial

n
X
S(x) = aj eiπj .
j=−n

Then: Z Z
1 1 1 1
( |S ′ (x)|p dx) p ≤ n( |S(x)|p dx) p , p≥1
−1 −1

( See [94]).
The reverse inequality is obtained in the following one-dimensional result which proof uses
Fourier Analysis techniques, and completes the case of dimension N = 1.
Lemma 1.3.3 Let Ek be the linear space spanned by

{sin(πjx)|1 ≤ j ≤ k},

and Ekc its topological complement in the Sobolev space W0p (0, 1). Then, there exists a constant
C > 0 such that
1
||u||p ≤ C ||u′ ||p
k
for each u ∈ Ekc .
 17

Proof. Consider (
t, si − 1 ≤ t ≤ 1.
g(t) =
1/t, si |t| > 1.
Then g(t) = w(t),
b namely, is the Fourier transform of w(t) ∈ L1 (IR); in fact |w(t)| ≤ C(1 + t2 )−1
and
lim w(t) 6= lim w(t).
t→0+ t→0−
We show the previous assertions on w. We have by definition
Z ∞ Z 1 Z ∞ sin(xt)
w(x) = 2 g(t) sin(tx)dt = 2 t sin(tx)dt + 2 dt.
0 0 1 t
Integrating by parts we get
Z ∞
sin x 2 cos(xt)
w(x) = 2 2 − dt,
x x 1 t2
and integrating by parts a second time we get for |x| large ,
c
|w(x)| ≈ .
|x|2
For x > 0 small, by the change of variable s = xt we have,
Z x Z ∞
2 sin s
w(x) = 2 s sin sds + 2 ds → π, x → 0.
x 0 x s
Because w is odd we get that w is discontinuous at 0, and moreover that it satisfies in the whole
real line
C
|w(t)| ≤ .
(1 + t2 )
Define wk (t) = kw(kt), then ||wk ||1 = ||w||1 = C < ∞ and
t
g( ) = w
ck (t).
k
For k ≥ 1 we consider the 2-periodization of wk (t) i.e.,
∞
X
w
fk (t)) = wk (t + 2j).
j=−∞

It is clear that ||w

fk ||L1 (−1,1) = C for all k ≥ 1 and, moreover, the Fourier series of w
fk is
∞
X
w
fk (x) = wˆk (j)eπijx ,
j=−∞
18

fk for |j| ≥ k > 1 are

and, by definition, the Fourier coefficients of w

c j k
w
ek (j) = g( ) = .
k j

Call Ω = (0, 1). Then if u ∈ W01,p (Ω) and ū is the odd extension of u to (−1, 1), we get that if
X
ū(x) = aj eijπx ,
j

then its derivative is, X X

ū′ (x) = iπ jaj eijπx = π a′j eijπx .
j j

Now, if u ∈ Ekc we get that

X π X j ′ ijπx
ū(x) = aj eijπx = g( )a e ,
j
k |j|≥k k j

that gives the following identity

k j
b̄
u(j) = g( )ub̄′ (j),
π k
which is equivalent to
k
u=w fk ∗ u′ ,
π
where ∗ means the integral of convolution. As conclusion,
k
||u||p ≤ ||w˜k ||1 ||u′ ||p ≡ c||u′ ||p
π
With this estimate we get,
λk ≥ ck p

Step 2. Dimension N > 1 and Ω = (0, 1)N . The next step is to show the case N > 1 in the
particular case of a product of N intervals that, by scaling, we can assume that Ω = (0, 1)N . We
will consider

{ek = sin(k1 πx1 ) sin(k2 πx2 )... sin(kN πxN ), ki ∈ IN , i = 1, ..., N }

a basis of W01,p (Ω). If on the lattice IN N we take the complement of the cube Qk , of edge k, we
are considering Ekc N , the complement of the k N -dimensional linear space EkN spanned for the
functions with frequencies in such cube.
 19

In fact, we consider for u ∈ Ekc N ,

X
u= em̄ , bm = 0 if m̄ = (m1 , ..., mN ), mi ≤ k, i = 1, ...N.
m̄
Then we prove the corresponding result
Lemma 1.3.4 There exists a constant C > 0, such that if u ∈ Ekc N ,
C
||u||p ≤ ||∇u||p
k
Proof. If u ∈ Ekc N , there exists bm 6= 0 for some m = (m1 , ..., mN ) with mj > k, for some
j = 1, ..., N .
Then by Lemma 1.3.3 we get
Z Z
C ∂u
|u(x1 , ..., xj , ..., xN )|p dxj ≤ | (x1 , ..., xN )|p dxj .
kp ∂xj
The reverse inequality, as above, is an application of Bernstein inequalities. So we conclude.
Assume the linear subspaces in W01,p (Ω) spanned by the basis {ek̄ } defined above and
ordered as follows
i)EkN = L{el̄ }, where ¯l ∈ Qk .
ii) If k N < j < k N +1 , then Ej ⊂ Ekc N ∩ EkN +1 .
In this way the result in Lemma 1.3.4 gives us the result in the Theorem 1.3.1 for an interval.
To read correctly the result put in the horizontal axis the subspace dimension and as ordinate
C
the constant of the inequality, we have for extrapolation the function f (x) = 1/N . Hence we
x
get,
Lemma 1.3.5 Given Ω, a product of N one dimensional intervals in IRN , there exists a constant
C > 0, such that
C C
||u||p ≤ 1/N ||∇u||p , for all u ∈ EK ,
K
where the linear spaces EK are defined and ordered as above.

Finally, the case of a general bounded domain can be proved by comparison. More precisely,
consider Q′ ⊂ Ω ⊂ Q, Q, Q′ cubes in IRN .
By definition we have βk (Q) ≥ βk (Ω) ≥ βk (Q′ ) and then
λk (Q) ≤ λk (Ω) ≤ λk (Q′ ).
Remark 1.3.6 The main open problem is to show that the sequence that we have found contains
all the eigenvalues. A partial answer to this problem is given in [57], for dimension N = 1 and
in [43] for the radial case, that we will explain below.
20

1.4 Isolation of the first eigenvalue

Let us consider the problem (1 < p < ∞):
(
−∆p u = λ|u|p−2 u,
(1.18)
u ∈ W01,p (Ω), u 6≡ 0.

Then according to the previous section, the first eigenvalue is given by

 
Z Z 
λ1 = inf |∇w|p dx : w ∈ W01,p (Ω), |w|p dx = 1 .
 
Ω Ω

The main result in this section is the following Theorem.

Theorem 1.4.1 The first eigenvalue of −∆p is isolated and simple.

Remark 1.4.2 The problem

(
−∆p u = λV (x)|u|p−2 u,
(1.19)
u ∈ W01,p (Ω), u 6≡ 0.

with V ∈ L∞ and Ω with smooth boundary was studied in [15] and was extended in [66] to
general domains. Often the case V ∈ Lq with q < ∞ is useful for the applications.
We can find similar results if V (x) ≥ 0, V (x) ∈ Lq (Ω), and |{x ∈ Ω : V (x) > 0}| = 6 0,
where q ≥ 1 if p > N , q > 1 if p = N , q > N/p > 1 otherwise. The details of this extension can
be found in [2].
Without these hypotheses the behaviour is different and we will give an example in the last
Chapter of this lectures notes.

The proof of Theorem 1.4.1 follows closely the arguments in [66] and is divided in several
Lemmas.

Lemma 1.4.3 λ1 is an eigenvalue, and every eigenfunction u1 corresponding to λ1 does not

changes sign in Ω: either u1 > 0 or u1 < 0.

Proof. Directly from the proof of existence of the first eigenvalue we conclude that there exists
a positive eigenfunction: if v is an eigenfuction also u = |v| is a solution of the minimization
problem and then an eigenfunction. By the strong maximum principle |v| > 0 and then u has
constant sign.
 21

Lemma 1.4.4 i) If p ≥ 2 then

|ξ2 − ξ1 |p
|ξ2 |p ≥ |ξ1 |p + p|ξ1 |p−2 hξ1 , ξ2 − ξ1 i + C(p) , (1.20)
2p − 1

for all ξ1 , ξ2 ∈ IRN .

ii) If p < 2, then

|ξ2 − ξ1 |2
|ξ2 |p ≥ |ξ1 |p + p|ξ1 |p−2 hξ1 , ξ2 − ξ1 i + C(p) , (1.21)
(|ξ2 | + |ξ1 |)2−p

for all ξ2 , ξ2 ∈ IRN .

(C(p) is a constant depending only on p).

(See the proof in [66]).

Lemma 1.4.5 λ1 is simple, i.e. if u, v are two eigenfunctions corresponding to the eigenvalue
λ1 , then u = αv for some α.

Proof. As [15] and [66] point out, if the functions η1 = u − v p u1−p and η2 = v − up v 1−p could
be considered as admisible test function in
Z Z
|∇u|p−2 h∇u, ∇ηi idx = λ1 |u|p−2 uηi , i = 1, 2, (1.22)
Ω Ω

then the result would be obtained by direct calculations.

In sufficiently regular domains it is possible to assume this hypothesis by a convenient use
of Hopf lemma, otherwise, it is necessary to regularize the candidates to test functions. Following
the proof in [66] we consider for ε > 0, uε = u + ε, vε = v + ε and then

upε − vεp vεp − upε

η1 = and η2 = .
up−1
ε vεp−1

Hence the gradient of η1 is

 
vε p vε
∇η1 = 1 + (p − 1)( ) ∇uε − p( )p−1 ∇vε
uε uε

and by symmetry ∇η2 is obtained in a similar way. We substitute in the equations (1.22) and
22

adding we get:
!
R up−1 v p−1
λ1 − (upε − vεp )dx =
Ω up−1
ε vεp−1
    
R vε uε
1 + (p − 1)( )p |∇uε |p + 1 + (p − 1)( )p |∇vε |p dx−
Ω uε vε
 
R vε p−1 uε
p ( ) |∇uε |p−2 + ( )p−1 |∇vε |p−2 h∇uε , ∇vε idx =
Ω uε vε (1.23)
R
(upε − vεp )(|∇ log uε |p − |∇ log vε |p )dx−
Ω
R
−p vεp |∇ log uε |p−2 h∇ log uε , ∇ log vε − ∇ log uε idx−
Ω
R
−p upε |∇ log vε |p−2 h∇ log vε , ∇ log uε − ∇ log vε idx,
Ω

that it is not positive by the convexity of the function f (s) = sp . Moreover the Lebesgue
convergence theorem implies that
Z !
up−1 up−1
lim − (upε − vεp )dx = 0. (1.24)
ε→0+ up−1
ε up−1
ε
Ω

Now, if p ≥ 2, by the estimate (1.20) applied to ξ1 = ∇ log uε , ξ2 = ∇ log vε and viceversa we

get  
R 1 1 p
0 ≤ c(p) p + p |vε ∇uε − uε ∇vε | dx ≤
Ω v ε u ε !
R up−1 up−1
−λ1 − p−1 (upε − vεp )dx.
Ω up−1
ε uε
According to (1.24) we conclude that v∇u = u∇v in Ω, and then u = αv. Similarly, if 1 < p < 2
we use the estimate (1.21) for the same functions as above and we get

R |vε ∇uε − uε ∇vε |2

0 ≤ c(p) (vε + uε )p (uε vε )p dx ≤
Ω ! (vε |∇uε | + uε |∇vε |)2−p
R up−1 v p−1
−λ1 − (upε − vεp )dx,
Ω up−1
ε vεp−1
that finished the proof as in the case p ≥ 2.
 23

To finish the study of the isolation of the first eigenvalue, we extend slightly the results by
Anane [15] and [66] to the case of unbounded potential. We have the following nodal result.

Lemma 1.4.6 If w is an eigenfunction corresponding to the eigenvalue λ, λ > 0, λ 6= λ1 , then

w changes sign in Ω: w+ 6≡ 0, w− 6≡ 0 and

|Ω− | ≥ (λ ||V ||q C p )σ ,

qN
where Ω− = {x ∈ Ω : w(x) < 0}, σ = −2q ′ if p ≥ N , σ = − if 1 < p < N and λ1 is the
qp − N
first eigenvalue for the p-laplacian with weight V in Ω.

Proof. We have by definition λ1 ≤ λ Let u, w be two eigenfunctions corresponding to λ1 and

λ respectively. If w does not change sign, by using the same arguments as above we obtain
Z     
w u
λ1 V (x) − λV (x) (upε − wεp )dx ≤ 0,
wε uε
Ω

then Z
(λ1 − λ) V (x)(up − wp ) dx ≤ 0
Ω
and we arrive to a contradiction if λ > λ1 by taking kw. Then w is positive implies λ = λ1 .
Moreover, if we take w− as a test function, we get
Z
||∇w− ||pp =λ V (w− )p dx ≤ λ ||V ||q || (w− )p ||α |Ω− |1/β ,
Ω

1 1 1
with + + = 1. Now we are considering two cases:
q α β
1. p ≥ N . By Sobolev’s inequality

|| (w− )p ||α = ||w− ||pαp ≤ C||∇w− ||pp , α > 1.

Thus, if we take α = β = 2q ′ then

′
|Ω− | ≥ (λ ||V ||q C p )−2q .

N qN Np
2. 1 < p < N . We take α = ,β= (||(w− )p ||α = ||w− ||pp∗ , where p∗ = ).
N −p qp − N N −p
By Sobolev’s inequality
||∇w− ||pp ≤ λ ||V ||q || w− ||pp∗ ,
24

qp−N qp−N
|Ω− | qN ≤ λ ||V ||q C p ||∇w− ||pp |Ω− | qN .
Hence qN
− qp−N
|Ω− | ≥ (λ ||V ||q C p ) .

The isolation of λ1 is studied to finish this section.

Lemma 1.4.7 λ1 is isolated; that is, λ1 is the unique eigenvalue in [0, a] for some a > λ1 .

Proof. Let λ ≥ 0 be an eigenvalue and v be the corresponding eigenfunction. By the definition

of λ1 (it is the infimum) we have λ ≥ λ1 . Then, λ1 is left-isolated.
We are now arguing by contradiction. We assume there exists a sequence of eigenvalues (λk ),
λk 6= λ1 which converges to λ1 . Let (uk ) be the corresponding eigenfunctions with ||∇uk ||p = 1.
We can therefore take a subsequence, denoted again by (uk ), converging weakly in W01,p , strongly
in Lp (Ω) and almost everywhere in Ω to a function u ∈ W01,p . Since uk = −∆−1p (λk V |uk |
p−2 u ),
k
the subsequence (uk ) converges strongly in W01,p , and subsequently u is the eigenfunction cor-
responding to the first eigenvalue λ1 with norm equals to 1. Hence, by applying the Egorov’s
Theorem ([26] Th. IV.28), (uk ) converges uniformly to u in the exterior of a set of arbitrarily
small measure. Then, there exists a piece of Ω of arbitrarily small measure in which exterior uk
is positive for k large enough, obtaining a contradiction with the conclusion of Lemma 1.4.6

1.5 Eigenvalues in the radial case

In the case of the nonlinear Sturm Liouville problem, N = 1, the corresponding eigenvalue
problem is (
−(|u′ |p−2 u′ )′ = λ|u|p−2 u
(1.25)
u(0) = 0 = u(1),

and can be solved explicitely getting the following results (see [43] and [57]):

1) Problem (1.25) has nontrivial solution if and only if

λ = λk = (πp )p k p ,

where
Z (p−1)1/p dt
πp = 2
0 (1 − sp /(p − 1))1/p
 25

2) Fixed k ∈ IN there exists an unique solution vk to problem (1.25) for λ = λk , such that
||vk || = 1 and vk ∈ Sk+ . Moreover

vk (x) = (−1)m v1 (kx − m), m/k < x ≤ (m + 1)/k,

where v1 is the normalized positive solution corresponding to λ1 .

In this way problem (1.25) is a typical nonlinear eigenvalue problem, and all of its eigen-
values are simple.
Hence the set of solutions of (1.25) is

S = {(µvk , λk ) | µ ∈ IR − {0}, k ∈ IN }.

We will study in some detail the N-dimensional radial eigenvalue problem and we follow
the paper [43] where the results are obtained in a elementary way. In particular we recover the
nonlinear Sturm-Liouville eigenvalue problem for N = 1.

Consider the eigenvalue problem,




 Ap u = rN −1 λ|u(r)|p−2 u(r)
u(1) = 0, (1.26)
 ′
 u (0) = 0,

where Ap u = −(rN −1 |u′ |p−2 u′ )′ .

We will use the following results.

Lemma 1.5.1 The Cauchy problem



 Ap u = rN −1 |u(r)|p−2 u(r),
u(0) = a, (1.27)

 u′ (0) = 0,

has a unique C 1,ν solution u = ua defined in [0, ∞). Moreover, ua has infinitely many zeros in
[0, ∞).

Proof. To be brief we call φp (s) = |s|p−2 s. Then the equation in problem (1.27) can writen as

−(rN −1 φp (u′ ))′ = rN −1 φp (u) (1.28)

26

Then a solution to (1.27) must satisfy

Z r  Z s 
1 1 1
u(r) = a − φp′ φp (u(t))dt ds ≡ Lp (u)(r), + = 1,
0 sN −1 0 p p′

namely u must be a fixed point for the operator Lp . It is very easy to check that for small δ > 0,
Lp verifies the hypotheses of the Schauder’s fixed point theorem. Then we have local solution to
problem (1.27).
We prove the uniqueness following the ideas in [47] where a more general situation is also
considered. If u and v are solutions to 1.27, assume that |u′ (r)| ≤ |v ′ (r)| in some interval. We
call
w(r) = φp (u′ ) − φp (v ′ ),
then w satisfies the equation,

N −1
w′ + w = ψ(r), w(0) = 0,
r
where
ψ(r) = φp (v) − φp (u) ≡ |v(r)|p−2 v(r) − |u(r)|p−2 u(r),
then Z r t r
|w(r)| = | ψ(t)( )N −1 dt| ≤ sup |ψ(t)|,
0 r N t∈[0,r]
and because u and v leaves in the positive real axe we get a local Lipschitz constant for the
second member, namely
|ψ(r)| ≤ M |u(r) − v(r)|,
and then Z r
|ψ(r)| ≤ M |u′ (t) − v ′ (t)|.
0

But
1
|u′ (r) − v ′ (r)| ≤ |w(r)|.
inf |u′ |≤γ≤|v′ | φ′p (γ)
Moreover for a solution
Z r
′ t
φp (u (r)) = − |u(t)|p−2 u(t)( )N −1 dt,
0 r
then Z r t r
φp (γ) > φp (|u′ |) = |u(t)|p−2 u(t)( )N −1 dt ≥ ap−1 ,
0 r 2N
 27

because φp is increasing. Moreover φ′p (γ) ≥ |φp (γ)|µ , µ = 0 if 1 < p ≤ 2 and µ = 1 if p > 2 and
locally. Then we conclude that
r µ
φ′p (γ) ≥ |φp (γ)|µ ≥ (ap−1 )
2N
and then Z r Z r
′ ′ 2N µ
|ψ(r)| = M |u (t) − v (t)|dt ≤ M ( ) |w(t)|dt.
0 0 ap−1 t
In this way we obtain Z r |w(t)|
|w(r)| ≤ Cr dt,
0 tµ
|w(r)|
Calling y(r) = , we get
r
Z r Z s Z r
|w(t)| |w(t)| |y(t)|
|y(r)| ≤ C dt = C dt + C dt,
0 tµ 0 tµ s tµ−1
for all s ∈ [0, r], and by Gronwall inequality

Z R r dt
s |w(t)| (C )
|y(r)| ≤ C dt e s
tµ−1 ,
t µ
0

letting s → 0 we get w(r) = 0.

To see that the solution is defined in [0, ∞), multiply the equation by u′ and integrate by
parts. Then we arrive to the inequality
 Z r 
|u′ (r)|p + |u(r)|p ≤ K(a) 1 + (|u′ (s)|p + |u(s)|p )ds , r ∈ [0, a]
0

and by the Gronwall inequality we conclude that

|u′ (r)|p + |u(r)|p ≤ eKr .

Classical arguments imply that u is defined in the whole interval [0, ∞).
Finally, we will prove that u is oscillatory. We proceed by contradiction. Assume that u
solution of
(rN −1 |u′ |p−2 u′ )′ + rN −1 up−1 = 0 (1.29)
and that for some r0 > 0 u does not vanish on [r0 , ∞). We follow arguments in [69]. Since up−1
is positive for r > r0 ,
(rN −1 |u′ |p−2 u′ )′ < 0 for r > r0
28

and rN −1 |u′ |p−2 u′ is a decreasing function.

In particular, since rN −1 |u′ |p−2 u′ < 0 as u(r) > 0, we have u′ < 0 for r > 0, that is,
u(s) > u(r) for s < r. By integrating (1.29) from r0 to r we obtain,
Rr
rN −1 |u′ (r)|p−2 u′ (r) − r0N −1 |u′ (r0 )|p−2 u′ (r0 ) = − r0 sN −1 up−1 (s)ds ≤
R rN − r0N p−1
−up−1 (r) rr0 sN −1 ds = − u (r).
N
By integrating again from s0 = 2r0 to r (≥ s0 )
p
u(s0 ) p
log ≥ C(r p−1 − s0p−1 ),
u(r)
that is p p
p−1
−r p−1 ))
u(r) ≤ u(s0 )e(C(s0 .
But u′ < 0 and rN −1 |u′ |p−2 u′ is a decreasing function, so rN −1 |u′ |p−2 u′ ≤ −C for large r. By
integrating this inequality from r to ∞, we arrive to a contradiction with the decay of u above,
since
− N −p
u(r) ≥ C1 r p−1 .
So we conclude.
A direct consequence is the following result.
Corollary 1.5.2 Let u 6= 0 a solution of the equation

Ap u(r) = rN −1 |u(r)|p−2 u(r), r ∈ (0, 1)

such that u(r0 ) = 0, then u′ (r0 ) 6= 0.

As a consequence all zeros of u are simple.

Theorem 1.5.3 The problem (1.26) has nontrivial solution if and only if λ belongs to an in-
creasing sequence {λk (p)}. Moreover,
1. For each eigenvalue λk (p) any solution takes the form αvk (r) with α ∈ IR; namely the
multiplicity of each eigenvalue is 1. Moreover vk has exactly (k − 1) simple zeros.
2. Each λk (p) depends continously on p.

Proof. Let u be the solution to the Cauchy problem



 Ap u = rN −1 |u(r)|p−2 u(r),
u(0) = 1, (1.30)

 u′ (0) = 0,
 29

and 0 < ν1 (p) < ν2 (p) < ... < νk (p) < ... the zeros of u, that are simple and, according to
Lemma 1.5.1, lim νk (p) = ∞. We define
k→∞
(
λk = (νk (p))p , k ∈ IN
(1.31)
vk (r) = u(νk (p)r), k ∈ IN .

In this way (λk , vk ) are solution to problem (1.26), namely, λk is an eigenvalue and vk the
corresponding eigenfunction and moreover vk has exactly k − 1 zeros in (0, 1).
We need to prove that they are all the eigenvalues to problem (1.26). Assume µ > 0 is an
eigenvalue and w the corresponding eigenfunction. By the uniqueness result in Lemma 1.5.1 we
have 1
w(r) = w(0)u(µ p r)
and because w(1) = 0 then we have µ = λk (p) for some k ∈ IN and as a conclusion we get,
w(r) = w(0)vk (r).
Now, the solution u to the Cauchy problem (1.30) verifies
Z r  Z s 
1 1 1
u(r) = 1 − φp′ φp (u(t))dt ds ≡ Lp (u)(r), + ′ = 1,
0 sN −1 0 p p
so is continuous in p in the sense of the uniform convergence on bounded intervals. According
to this fact and taking into account that for p fixed the zeros of u are simple, we conclude the
continuous dependence of the eigenvalues on p.

1.6 Bifurcation for the p-Laplacian

Let Ψp,λ be defined in

X = { v ∈ C 1 (Ω) : v(x) = v(|x|), v(1) = 0}

by
Ψp,λ (u) = u − A−1
p (r
N −1
λ|u|p−2 u).
It is no difficult to show that Ψλ,p is a nonlinear compact perturbation of the identity.
Fixed k ∈ IN we call

Sk+ = {v ∈ X | v(r) > 0, if r ∈ [0, ε) and has exactly (k − 1) simple zeros }.

In this way if Sk− = −Sk+ both are open sets in X.

Taking into account the behaviour of the inverse operator of the eigenvalue problem (1.25),
we can use the extension of the Leray-Schauder degree definided in [31], in order to perform a
homotopic deformation in p as is used in [42], to prove the following result.
30

Lemma 1.6.1 Let {λk (p)}k∈IN be the sequence of eigenvalues of (1.25). Consider λ 6= λk (p),
k ∈ IN , then
deg(Φp,λ , Bε (0), 0) = (−1)β , ∀ε > 0,
where β = #{λk (p) | λk (p) < λ}

Proof. If λ < λ1 (p), deg(Φp,λ , Bε (0), 0) = 1 by the variational characterization of the first
eigenvalue. Assume that λ > λ1 (p) and λk (p) < λ < λk+1 (p). Since the eigenvalues depend
continously on q, there exists a continous function

ν : (1, ∞) −→ IR

such that λk (q) < ν(q) < λk+1 (q) and λ = ν(p). Define

T (q, u) = Φq,ν(q) (u) = u − A−1

q (ν(q)|u|
q−2
u).

It is easy to show that T (q, u) is a compact map such that for all u 6= 0, by definition of ν(q),
T (q, u) 6= 0, for all q ∈ [p0 , ∞). Hence the invariance of the degree under homotopies and the
classical result for p = 2 imply (recall λ = ν(p))

deg(Φp,λ , Bε (0), 0) = deg(Φ2,λ̃(2) , Bε (0), 0) = (−1)k .

Remark 1.6.2 It is important to point out that the main point in this proof is the continuity
of the eigenvalues with respect to p.

As an application we found the following variation of the classical Rabinowitz bifurcation

theorem:
Consider the problem


 Ap u = rN −1 (λ|u|p−2 u + g(r, u))
u′ (0) = 0, (1.32)

 u(1) = 0,

where for some α > p,

g(r, u)
lim =M >0
u→0 uα
uniformly in r.
Theorem 1.6.3 From each eigenvalue λk of problem (1.25) it bifurcates an unbounded contin-
uum Ck of solutions to problem (1.32), with exactly k − 1 simple zeros.
 31

Proof. We will be sketchy. The same argument as in the Rabinowitz theorem, [74], taking
into account Lemma 1.6.1, gives that either Ck is unbounded, either it touches (λj , 0). But the
last alternative is impossible because if there exists (µm , um ) → (λj , 0) when m → ∞, um 6= 0
um
and um ∈ / Sj+ , then we could consider vm = , and vm should be a solution of problem,
||um ||

g(r, um )
Ap vm = rN −1 (µm vm + ),
||um ||p−1

and by the properties of g it is easily seen that

g(r, u)
lim = 0.
u→0 ||um ||p−1

By the compactness of A−1p we obtain that for some convenient subsequence vm → w 6= 0 as

m → ∞. Now w verifies the equation Ap w = rN −1 λj |w|p−2 w and ||w|| = 1. Hence w ∈ Sj+
which is an open set in X (observe that any zero of these solutions must be simple), and as a
consequence for some m large enough, vm ∈ Sj+ , and this is a contradiction.
32
Chapter 2

Multiplicity of solutions: Growth

versus Shape

2.1 Introduction
In this chapter the model problem that we will consider is the following,
(
−∆p u = λ|u|q−2 u + |u|r−2 u in Ω,
(2.1)
u|∂Ω = 0,

where Ω ⊂ IRN is a smooth bounded domain, 1 < q, 1 < p < r, λ > 0 and, perhaps, with some
restrictions on λ and q that we will precise in each situation.
Np
We define p∗ = if p < N and p∗ = ∞ if p ≥ N . We have different kind of problems:
N −p
A) With respect to r

1. r < p∗ subcritical case.

2. r = p∗ critical case.

3. r > p∗ supercritical case.

This classification takes into account mainly the growth.

Case 1 is classical and can be studied by critical point methods. We refer to [51] for the
details. See also [70]. In case 2, one of the main difficulties in the application of the critical point
methods is the lack of compactness. Sometimes a local Palais-Smale property can be recovered
and this allows us to solve the problem under convenient conditions. We will explain the main
ideas and methods. Case 3 is different because the study from the point of view of the critical

33
34

point theory has no sense in general. Nevertheless, recently some results are obtained in [33] by
using the critical point techniques. Also in this case, it is impossible to have a general result on
regularity ( by regularity we mean C 1,α ).

B) With respect to the relation between q and p.

1. 1 < q = p.

2. p∗ > q > p.

3. 1 < q < p.

The two first cases can be seen in [51], [53] and [58] and the references therein. The last case
has precedents in [53] and in [11] in the case p = 2.

C) With respect to λ.
In [11] a more deep study of the case 1 < q < p is performed. This study is of global character.
The meaning of global here deals with respect to λ. Namely we have the last classification of
the problems to be considered as:

1. Small λ > 0, i.e., local behaviour.

2. All λ > 0, i.e., global behaviour.

2.2 First results

We will explain two types of results that show the influence of the shape of the zero order term.
First of all, we are interested in obtaining results about the ordering of solutions. Some
previous results can be found in [49] and in [39] for other nonlinearities. We obtain below this
type of result in an elementary way as a consequence of the following lemma by Adimurthi-
Yadava [1].

Lemma 2.2.1 Let 0 ≤ ρ1 (x) ≤ ρ2 (x) be two functions in Lr (Ω) defined on Ω, a bounded domain
with smooth boundary, where r ≥ 1 if p > N , r > N/p if p ≤ N and ρ1 6≡ 0. Let λ1 (ρi ) be the
first eigenvalue corresponding to
(
−∆p ϕ = λρi (x)ϕp−1 , x ∈ Ω, i = 1, 2
(2.2)
ϕ = 0 on ∂Ω.

Then, either λ1 (ρ1 ) < λ1 (ρ2 ) or λ1 (ρ1 ) = λ1 (ρ2 ) and the corresponding eigenfunctions ϕ1 , ϕ2
vanish in K = {x ∈ Ω : ρ2 (x) > ρ1 (x)}.
 35

Proof. It is clear that λ1 (ρ1 ) ≤ λ1 (ρ2 ). Let us suppose that λ1 (ρ1 ) = λ1 (ρ2 ). Since ϕ1 , ϕ2 > 0
and R R
|∇u|p dx |∇ϕ1 |p dx
Ω Ω
λ1 (ρ1 ) = inf R = R ,
ρ1 up dx ρ1 ϕp1 dx
Ω Ω
R R R
|∇u|p dx |∇ϕ2 |p dx |∇ϕ1 |p dx
Ω Ω Ω
λ1 (ρ2 ) = inf R = R ≤ R ,
ρ2 up dx ρ2 ϕp2 dx ρ2 ϕp1 dx
Ω Ω Ω
we get Z
(ρ2 − ρ1 )ϕp1 dx ≤ 0.
Ω
Then, we conclude that ϕ1 ≡ 0 in K. Therefore
R
Z Z |∇ϕ1 |p dx
ρ1 ϕp1 dx = ρ2 ϕp1 dx i.e. Ω
R = λ1 (ρ2 ).
ρ2 ϕp1 dx
Ω Ω Ω

Thus ϕ2 = µϕ1 and ϕ2 ≡ 0 in K.

As a consequence we can prove the following general result:

f (s, λ)
Lemma 2.2.2 Let f : [0, ∞) −→ [0, ∞) be a continuous function verifying that is
sp−1
nondecreasing and
f (s, λ)
∈ L∞
loc ([0, ∞)).
sp−1
Assume that u1 and u2 are bounded solutions of
(
−∆p u = f (u, λ) x ∈ Ω ⊂ IRN , λ > 0
(2.3)
u(x) = 0 x ∈ ∂Ω,

such that 0 < u1 ≤ u2 (0 < u2 ≤ u1 ). Then u1 ≡ u2

Proof. If 0 < u1 ≤ u2 , then

f (ui , λ)
−∆p ui = f (ui , λ) = up−1
i ≡ ρi (x)up−1
i in Ω, i = 1, 2.
up−1
i

with ρ1 (x) ≤ ρ2 (x) by hypothesis, hence u1 and u2 are eigenfunctions for the eigenvalue λ1 (ρ1 ) =
λ1 (ρ2 ) ≡ 1 of the p-Laplacian with weights ρ1 and ρ2 , respectively. By Lemma 2.2.1, u1 and u2
36

vanish in {ρ1 (x) < ρ2 (x)}. By Hopf’s Lemma, u1 , u2 > 0 in Ω; so, {ρ1 (x) < ρ2 (x)} = ∅, i.e.,
u1 ≡ u2 .
Other application of Lemma 2.2.1 is the following uniqueness result for a radial problem
obtained in [1].

Theorem 2.2.3 Consider the problem




 Ap u = rN −1 λ|u|p−2 u + |u|α−2 u
u(1) = 0, (2.4)

 u′ (0) = 0,

where Ap u = −(rN −1 |u′ |p−2 u′ )′ , λ > 0 and 1 < p < α < p∗ . Then the problem (2.4) has a
unique positive solution.

(See the proof in [1].)

The second elementary result that we study here was obtained in [24] and deals with the
existence of positive solution to the following problem,
(
−∆p u = |u|r−2 u + λ|u|q−2 u
(2.5)
u|δΩ = 0,

where we assume:
(H) 1 < q < p < r and λ > 0.
Notice that the growth of the zero order term can be arbitrarily large, but there are a
subdiffusive part, in the sense that 1 < q < p, that produces the behaviour described in the
following result.

Theorem 2.2.4 Consider the problem under the hypothesis (H). Then there exist λ0 > 0 such
that for 0 < λ ≤ λ0 , (2.5) has a positive solution.

Proof. The proof of the theorem is organized in several steps.

Step 1. We find a supersolution to problem (2.5).
Let v be the solution to the Dirichlet Problem
(
−∆p u = λ + 1
u|δΩ = 0.

Then 0 < v < K in Ω. By simplicity of writing we call

F (u) = |u|r−2 u + λ|u|q−2 u.

 37

Define u(x) = T v(x) where T is a constant that will be chosen in such a way that

−∆p u ≥ F (T M ) ≥ F (u)

where M = max {1, kvk∞ }

Now −∆p u = T p−1 (λ + 1) and

F (u) ≡ λT q−1 v q−1 + T r−1 v r−1 ≤ λT q−1 M q−1 + T r−1 M r−1 .

Then, it is sufficient to find T such that

(λ + 1) ≥ λT q−p M q−1 + T r−p M r−1 .

We call
φ(T ) = c1 λT q−p + c2 T r−p ,
with c1 = M q−1 , c2 = M r−1 . Then

lim φ(T ) = lim φ(T ) = ∞,

T →0+ T →∞

because q − p < 0 < r − p; then φ attains a minimum in [0, ∞). By elementary computations we
have that 1
φ′ (T ) ≡ −c3 λT q−p−1 + c4 T r−p−1 = 0 in T0 = c5 λ r−q ,
where c5 = M −1 (r − q)−1 (p − q). To finish it is sufficient that,

φ(T0 ) ≤ λ + 1,

that is r−p
c6 λ r−q < λ + 1,
and here c6 = cp,q M p−1 . Moreover M depends on λ in the correct way because
1 1 1
(N − r(p−1) )
||v||∞ ≤ c|Ω| (λ + 1) p−1
N
where r > p−1 ; then there exists λ0 such that for 0 < λ < λ0 and 1 < q < p < N , u(x) = T0 v is
a supersolution of problem (2.5).
Step 2. Construction of a subsolution to (2.5).
Consider φ1 , positive eigenfunction of the eigenvalue problem,
(
−∆p u = λ1 |u|p−2 u
u|δΩ = 0,
38

such that kφ1 k∞ = 1. We define u(x) = tφ1 (x).Then u is a subsolution if t is small enough
because
λ1 tp−1 φp−1
1 ≤ λtq−1 φp−1
1 ≤ λtq−1 φ1q−1 + tr−1 φ1r−1 .
Now fixed the supersolution u, i.e. T , for t small enough, we get

−∆p u = tp−1 λ1 φ1q−1 ≤ tp−1 λ1 ≤ −∆p u

Then by the weak comparison principle u ≤ u.

End of the proof. By the classical iteration method we get a solution between the sub-
solution and supersolution.

Theorem 2.2.5 Assume that Ω has a smooth boundary. There exists Λ > 0 such that (2.5) has
no positive solution for λ > Λ.

Proof. Consider φ1 > 0 eigenfunction corresponding to the first eigenvalue λ1 , namely,

(
−∆p φ1 = λ1 |φ1 |p−2 φ1
φ1 |δΩ = 0.

The regularity of the boundary of Ω implies that the solutions of the eigenvalue problem are
C 1,α in the closure of Ω. If we assume that (2.5) has a positive solution u for all λ then by Hopf
1
Lemma there exist a t > 0 depending on λ such that t p−1 φ1 ≤ u in Ω.
1
We call ψ = t p−1 φ1 . Pick ε > 0 such that for µ ∈ (λ1 , λ1 + ε) the eigenvalue problem has
no nontrivial solution. The existence of such ε is a consequence of the result of isolation of the
first eigenvalue of the p-laplacian studied in Section 1.4.
r−p
If λ is such that λ1 + ε ≤ bp,q λ r−q where bp,q verifies
r−p
bp,q λ r−q up−1 ≤ λuq−1 + ur−1

then
−∆p ψ = λ1 ψ p−1 ≤ µψ p−1 ≤ µup−1 ≤
r−p
(λ1 + ε)up−1 ≤ bp,q λ r−q up−1 ≤ λuq−1 + ur−1 = −∆p u
But then the problem (
−∆p u = µ|u|p−2 u
u|δΩ = 0,

with µ ∈ (λ1 , λ + ε) must have nontrivial solution and this is a contradiction.

 39

We will consider the following subcritical problem

(
−∆p u = λ|u|q−2 u + |u|α−2 u
(2.6)
u|δΩ = 0

with 1 < q < p < α < p∗ . The associated energy functional is

Z Z Z
1 p λ q 1
J(u) = | u| dx − |u| dx − |u|α dx.
p q α
Ω Ω Ω

If λ is small enough it is possible to show that the solution to problem (2.6), found in the
Theorem 2.2.4, verifies that J(u) < 0. Also it is easy to show that at least for small λ > 0 we
can apply the Mountain Pass Theorem and get the following result.

Theorem 2.2.6 There exist a λ0 such that problem (2.6) has at least two positive solutions for
0 < λ < λ0 .

Comparing the results in this section we notice that the shape of the zero order term plays
a very important role, namely:

1. If p = q the growth is important for existence, and the positive solutions cannot be ordered.
In the radial case the subcritical problem has a unique solution.

2. If 1 < q < p there exists solution without restriction in the growth, the subcritical problem
has a second positive solution that is ordered with the minimum solution.

In the next section we will obtain different results that involve the interplay of shape and
growth.

2.3 Critical Problems: The Palais-Smale condition

The study of existence of solutions to the critical problem (2.1), i.e., the critical case r = p∗ , is
done by looking for critical points of the funcional
Z Z Z
1 λ 1 ∗
J(u) = |∇u|p dx − |u|q dx − |u|p , if q 6= p,
p q p∗
Ω Ω Ω

or, if p = q, by homogeneity, looking for the minimun in the variational problem with constraints,
Z Z
∗
Iλ = inf{ |∇u|p dx − λ |u|p dx | ||u||pp∗ = 1, u ∈ W01,p (Ω)}.
Ω Ω
40

Then the idea is to consider approximate critical points and pass to the limit. (See appendix).
The main difficulty is to have compactness because the Sobolev embedding is not compact.
The concept of Palais-Smale sequence gives form to the meaning of approximate critical
point.
We say that {uj }j∈IN ⊂ W01,p (Ω) is a Palais-Smale sequence if:
′
a) J(uj ) → c, j → ∞; b) J ′ (uj ) → 0 in W −1,p (Ω), j → ∞.
We call to the constant c energy level of the sequence.
We say that J verifies the Palais-Smale condition at the level c if any Palais-Smale sequence
at the level c, admits a strongly convergent subsequence .
Under our hypothesis on J, every Palais -Smale sequence is bounded in W01,p (Ω), but because
of the lack of compactness it is not possible to conclude strong convergence for any subsequence.
The solution of this difficulty depends on a deep analysis of the behavior of the lack of compact-
ness. More precisely, assume that we have a bounded sequence {uj }j∈IN ⊂ W01,p (Ω). Taking a
convenient subsequence we can assume that:
i) uj ⇀ u in W01,p (Ω).
ii) uj → u almost everywhere and in Lr with 1 ≤ r < p∗ .
∗
iii) |uj |p ⇀ dν weakly-∗.
iv) |∇uj |p ⇀ dµ weakly-∗.
where ν and µ are finite measures. We can extend the functions uj by 0 to IRN . To precise the
ideas, assume initially that u ≡ 0. By the Sobolev inequality we have
Z Z
∗ 1 1 1
( |φuj |p dx) p∗ S p ≤ ( |∇(φuj )|p dx) p , for all φ ∈ C0∞ ,
IRN IRN

where S is the best constant in the Sobolev inclusion, i.e.,

S = inf{||∇u||pp | ||u||p∗ = 1, ∇u ∈ Lp }.

From iii) we obtain, Z Z

1 1
p∗ ∗
lim ( |φuj | dx) p∗ =( |φ|p dν) p∗ ,
j→∞
IRN IRN
moreover, by ii) and
|k∇(φuj )kp − kφ∇uj kp | ≤ kuj ∇φkp ,
we obtain that Z Z
p
lim |∇(φuj )| dx = |φ|p dµ,
j→∞
IRN IRN
 41

then we conclude that

Z Z
∗ 1 1 1
( |φ|p dν) p∗ S p ≤ ( |φ|p dµ) p , for all φ ∈ C0∞ .
IRN IRN

The last inequality is a Reverse Hölder Inequality from which we get the following representation
result due to P.L. Lions. (See [64] and appendix.)

Lemma 2.3.1 Let µ, ν be two non-negative and bounded measures on Ω, such that for 1 ≤ p <
r < ∞ there exists some constant C > 0 such that
Z Z
1 1
( |ϕ|r dν) r ≤ C( |ϕ|p dµ) p ∀ϕ ∈ C0∞ .
Ω Ω

Then, there exists {xj }j∈I ⊂ Ω and {νj }j∈I ⊂ (0, ∞), where I is finite, such that:
X X p
ν= νj δ x j , µ ≥ C −p νjr δxj ,
j∈I j∈I

where δxj is the Dirac mass at xj .

In the general case, i.e., when u 6= 0, we can obtain the same conclusion for vj = uj − u,
and then to study the absolutely continous part. We have the following P.L. Lions Lemma. (See
appendix).

Lemma 2.3.2 Let {uj } be a weakly convergent sequence in W01,p (Ω) with weak limit u, and such
that:
i) |∇uj |p → µ weakly-∗ in the sense of measures.
∗
ii) |uj |p → ν weakly-∗ in the sense of the measures.
Then, for some finite index set I we have:
 ∗ P

 1) ν = |u|p + j∈I νj δxj , νj > 0
 P
2) µ ≥ |∇u|p + j∈I µj δxj , µj > 0 , xj ∈ Ω
p

 ∗ µj
 3) νjp ≤ .
S
(See [64] and [65]).
The representation obtained in Lemma 2.3.2, allows us to conjecture how small must be the
energy level for which the singular part of the measures must be 0, and then a local Palais-Smale
condition holds. The calculus of how small must be the energy level is relatively easy.
42

Assume {vj }j∈IN ⊂ W01,p (Ω) a Palais-Smale sequence for the energy level c, i.e.:
′
a) J(vj ) → c; b) J ′ (vj ) → 0, in W −1,p (Ω).

Then the sequence is bounded, and we can apply the Lemma 2.3.2 to some subsequence, i.e., we
obtain,
∗ ∗ P
i) |vj |p ⇀ dν = |v|p + j∈I νj δxj weak-∗.
∗ ∗ P
ii) |∇vj |p ⇀ dµ ≥ |∇v|p + j∈I µj δxj weak-∗.
p/p∗
iii) Moreover, µj /S ≥ νj .

Localizing each singularity by a test function φ with support in B(xj , ε), and by i), ii)
applied to φvj we have,
Z Z Z Z
φdν + λ |v|q φdx − φdµ = lim |∇vj |p−2 vj h∇vj , ∇φidx.
j→∞
Ω Ω Ω Ω

Using the weak convergence, Hölder inequality and taking limits for ε → 0, we get that neces-
sarily νj = µj . Then taking into account iii), either µj = νj ≥ S N/p or µj = νj = 0.
Then if we assume q > p, and there exists a Dirac mass, necessarily we obtain,

c = limj→∞ J(vj ) = limj→∞ (J(vj ) − hJ ′ (vj ), vj i) ≥

1 R
|v|p dx + λ( p1 − 1q )||v||qq + N1 S N/p ,
∗
N
Ω

as a conclusion: N
If there exists a Dirac mass, c ≥ N1 S p .
In a similar way if 1 < q < p we have:
1 N p∗
If there exists a Dirac mass, c ≥ S p − Kλβ , where β = ∗ and K depends on p, q, N
N p −q
and Ω.
As a consequence of the calculus above and the continuity of the inverse of the p-laplacian,
we get the following Lemma.

Lemma 2.3.3 Let {vj } ⊂ W01,p (Ω) be a Palais-Smale sequence for J,

J(vj ) → c
′ 1 1
J ′ (vj ) → 0 in W −1,p (Ω), p + p′ =1

Then, we have:
 43

1 Np
1. If p < q < p∗ , and c < S , there exists a subsequence {vjk } ⊂ {vj } , strongly convergent
N
in W01,p (Ω) .
1 Np p∗
2. If 1 < q < p , and c < S − Kλβ , where β = ∗ and K depends on p, q, N and Ω,
N p −q
then there exists a subsequence {vjk } ⊂ {vj }, strongly convergent in W01,p (Ω).
Outline of the proof. From the previous arguments, we have that the singular part of ν is
identically zero. Therefore, there existsRa subsequenceR {vjk } ⊂ {vj } , such that
∗ ∗
i) vjk → v almost everywhere; ii) |vjk |p dx → |v|p dx.
Ω Ω
∗
So we conclude that vjk → v strongly in Lp . (See [28]).
As a consequence
∗ −2 ∗ −2 ′
λ|vjk |q−2 vjk + |vjk |p vjk → λ|v|q−2 v + |v|p v; in W −1,p (Ω)
and then the continuity of the inverse of the p-laplacian from W −1,p (Ω) to W01,p (Ω) implies the
′

theorem. (See [53] and appendix for more details).

Now assume that p = q. If we have a minimizing sequence for Iλ and uj → u in W01,p (Ω),
by Sobolev inequality,
Z Z Z
|∇uj |p dx − λ |uj |p dx ≥ S − λ |uj |p dx,
Ω Ω Ω

and then Iλ ≥ S − λ||u||pp . As a conclusion:

A sufficient condition to get a nontrivial solution, u 6= 0, is that Iλ < S.
In fact we can prove compactness also with the same hypothesis. More precisely Lemma
2.3.2 gives the following result.
Lemma 2.3.4 If {uj }j∈IN ⊂ W01,p (Ω) is a minimizing sequence for Iλ < S, then, for some
subsequence, uj → u 6= 0 in W01,p (Ω)
(See [51]).

Then Lemmas 2.3.3 and 2.3.4 give the energy level for which we get the compactness. In the
following sections we explain when it is possible to reach this critical level with a Palais-Smale
sequence.

2.4 Local results on multiplicity

In this section we will study local results, namely, for λ small.
44

2.4.1 The critical case and p∗ > q > p

We emphasize the dependence of J on λ writting the action functional as Jλ . We assume in
first place that q > p. If we consider the function gλ (t) = Jλ (tv0 ), we find that it attains its
maximum for t > 0 and lim g(t) = −∞.
t→∞
It is not difficult to prove that Jλ verifies the hypotheses of the Mountain Pass Theorem.

Lemma 2.4.1 Let X be a Banach space and J ∈ C 1 (X, IR) a functional. Let us assume that
there exists r, R > 0, such that:

i) J(u) > r, ∀u ∈ X with kuk = R.

ii) J(0) = 0, and J(w0 ) < r for some w0 ∈ X, with kw0 k > R.

Let us define Γ = {g ∈ C([0, 1]; X) : g(0) = 0, g(1) = w0 }, and

c = inf max J(g(t)).

g∈Γ t∈[0,1]

Then, there exists a sequence {uj } ⊂ X , such that J(uj ) → c, and J ′ (uj ) → 0 in X ′ (dual of
X).

(See [13], [17] or appendix for the proof). As a consequence of the results in Section 2.3 , it is
sufficient to show that
1 Np
sup Jλ (tv0 ) < S , for some v0 ∈ W01,p (Ω).
t≥0 N
N
1
Now, if λ = 0 then supt≥0 J0 (tv0 ) ≥ NS
p . The minimizers of Sobolev inclusion in IRN are,
p−N
Uε (x) = (ε + cp,N |x|p/(p−1) ) p ε > 0.

(See [84]). Define uε (x) = φ(x)Uε (x), vε = uε ||uε ||−1

p∗ for some convenient cuttoff function φ. By
definition we get
1 N
lim sup J0 (tvε ) = S p .
ε→0 t≥0 N
The idea is that the perturbation term λ|u|q−2 u provides a energy level from below the critical
constant.
Now, fixed w0 ∈ W01,p (Ω) we can compute that

lim (sup Jλ (tw0 )) = 0.

λ→∞ t≥0
 45

Then,
There exists λ0 such that for λ > λ0 the problem (2.1) has a nontrivial solution, obtained
as critical point of Jλ by the Mountain Pass Lemma.
But if we want to find a solution for all λ > 0, according with the property of minimization
which define Uε , it seems natural to study if

1 Np
sup Jλ (tvε ) < S .
t≥0 N

We have the following estimates:

N −p
k∇vε kpp = S + O(ε p )

and  p−1 (N −p) p−1 (N −p)

(N −α p ) (N −α p )


 C1 ε p + o(ε p ) if α > p∗ (1 − p1 )
 N −p N −p
α α
||vε ||αα = C1 ε p2 | log ε| + o(ε p2 | log ε|) if α = p∗ (1 − p1 )



 N −p
α N −p
α
C1 ε p2 + o(ε p2 ) if α < p∗ (1 − p1 ),
where the change of behaviour in the last estimate is due to the integrability or not integrability
of Uεα in IRN . For ε > 0 consider gε (t) = Jλ (tvε ). We can estimate directly the value tε for which,

gε (tε ) = max gε (t),

t≥0

getting that Z
∗ −p S
|∇vε |p dx ≥ tpε > .
2
Ω

Therefore, Z
1 Np N −p λ S q
gε (tε ) ≤ S + Cε p − ( ) p∗ −p vεq dx,
N q 2
Ω

and then Z
1 N N −p λ S q
gε (tε ) < S p if Cε p − ( ) p∗ −p vεq dx < 0,
N q 2
Ω

that is equivalent to
N −p p−1 (N − p)
> (N − q ).
p p p
As a conclusion we have the following result.
46

Theorem 2.4.2 a) If p < q < p∗ , there exists λ0 > 0 such that the problem (2.1) has a
nontrivial solution ∀λ ≥ λ0 .
p
b) If max(p, p∗ − ) < q < p∗ , then there exists a nontrivial solution of the problem (2.1),
p−1
∀λ > 0.

(See [53] for details)

Remark 2.4.3 If we assume p2 ≤ N , part b) in Theorem 2.4.2 applies for p < q < p∗ . This
result corresponds for p = 2 to dimension N ≥ 4. ( See [29]). For N = 3 the existence of solution
for all λ > 0 in this case cannot be obtained.

A similar calculation in the case q = p allows us to obtain a sufficient condition for which
Iλ < S. The result is formulated in the following way.

Theorem 2.4.4 Assume p2 ≤ N . Then for 0 < λ < λ1 , first eigenvalue of the p-laplacian, we
have 0 < Iλ < S, and then the problem 2.1 with q = p and r = p∗ has nontrivial solution.

(See [51] for details).

We would like to emphasize the fact that the result in the case q ≥ p depends strongly on
the dimension.

Remark 2.4.5 To obtain positive solution as critical points, we can consider a functional with
the term in uα substitued by (u+ )α , where u+ is the positive part of u.

2.4.2 The critical case and 1 < q < p

Results in the critical case, r = p∗ and 1 < q < p are quite different. The results of existence of
a positive solution are independent of the dimension as we see in section 2.2. Also we can prove
the existence of two positive solutions at least under some hypotheses.

Theorem 2.4.6 Given problem (2.1), with 1 < q < p and r = p∗ , there exists λ0 > 0, such
that, for 0 < λ < λ0 , there exists infinitely many solutions. Moreover at least one of them is
positive. For λ large enough no positive solution exists.

Remark 2.4.7 We find no restrictions on the dimensions and moreover we get infinitely many
solutions, perhaps with change of sign.
 47

We explain briefly the methods of the proof of this statement.

Consider as in the previous section gλ (t) = Jλ (tv0 ). If λ is small enough, gλ (t) attains a
local minimum and a local maximum in t > 0, for 0 < λ < λ0 because, 1 < q < p.
By Sobolev inequality we obtain

Jλ (u) ≥ h(k∇ukp ).

where,
1 1 ∗ λ
h(x) = xp − p∗
xp − Cp,q xq ,
p p∗ S p q

and Cp,q is a constant. We can choose λ0 > 0 such that, if 0 < λ ≤ λ0 , h attains its nonnegative
maximum. Consider R0 and R1 in such a way that h(R0 ) = 0, h(R1 ) = 0, (the point R, where h
attains its maximum, verifies R0 ≤ R ≤ R1 ). We make the following truncation of the functional
Jλ . Take τ : IR+ → [0, 1], nonincreasing and C ∞ , such that τ (x) = 1 if x ≤ R0 , τ (x) = 0 if
x ≥ R1 .
Let ϕ(u) = τ (k∇ukp ). We consider the truncated functional
Z Z Z
˜ 1 1 ∗ λ
J(u) = |∇u|p − |u|p ϕ(u) − |u|q .
p p∗ q
Ω Ω Ω

As a consequence of Lemma 2.3.3 2) and the previous construction, it is not difficult to prove
the following result.

Lemma 2.4.8

1. J˜ ∈ C 1 (W01,p (Ω), IR), is even.

2. If J˜λ (u) ≤ 0, then k∇ukp < R0 , and J(v)

˜ = Jλ (v) for all v in a small enough neighbour-
hood of u.

3. There exists λ0 > 0, such that, if 0 < λ < λ0 , then J˜ verifies a local Palais-Smale condition
for c ≤ 0.

We will use the genus of a symmetric set in W01,p (Ω) defined in the appendix.
It is possible to prove the existence of level sets of J˜ with arbitrarily large genus, more
precisely:

Lemma 2.4.9 Given n ∈ IN , there is ε = ε(n) > 0, such that

γ({u ∈ W01,p (Ω) : J(u)

˜ ≤ −ε}) ≥ n.
48

Proof. Fix n, let En be a n-dimensional subspace of W01,p (Ω). We take un ∈ En , with norm
k∇un kp = 1. For 0 < ρ < R0 , we have:

˜ n ) = F (ρun ) = 1 ρp − 1 ρp∗ ı|u|p∗ − λ ρq ı|u|q

J(ρu
p p∗ q
En is a space of finite dimension; so, all the norms are equivalent . Then, if we define:
R ∗
αn = inf{ |u|p : u ∈ En , k∇un kp = 1} > 0
RΩ
βn = inf{ |u|q : u ∈ En , k∇un kp = 1} > 0
Ω

˜ n ) ≤ 1 ρp − αn ρp∗ − λβn ρq , and we can choose ε (which depends on n ), and

we have J(ρu
p p∗ q
˜
η < R0 , such that J(ηu) ≤ −ε if u ∈ En , and k∇ukp = 1.
Let Sη = {u ∈ W01,p (Ω) : k∇ukp = η}. Sη ∩ En ⊂ {u ∈ W01,p (Ω) : J(u)
˜ ≤ −ε} ; therefore,
by the properties of the genus,

γ({u ∈ W01,p (Ω) : J(u)

˜ ≤ −ε}) ≥ γ(Sη ∩ En ) = n

( See appendix).
The last result is an extension of [21] and permits to prove the existence of critical points by
using a Lusternik-Schnirelmann argument. (See for instance [76]). We get the following results

Lemma 2.4.10 Let

Σk = {C ⊂ W01,p (Ω) − {0}| C is closed, C = −C, γ(C) ≥ k}.

˜
Let ck = inf sup J(u), Kc = {u ∈ W01,p (Ω) : J˜′ (u) = 0, J(u)
˜ = c}, and suppose 0 < λ <
C∈Σk u∈C
λ0 , where λ0 is the same that in Lemma 2.4.8.
Then, if c = ck = ck+1 = ... = ck+r , γ(Kc ) ≥ r + 1.
˜
In particular, the ck ’s are critical values of J.

Proof.
In the proof, we will use the Lemma 2.4.9, and a classical deformation lemma (see [76]).
For simplicity, we call J −ε = {u ∈ W01,p (Ω) : J(u) ≤ −ε} . By lemma 2.4.9, ∀k ∈
IN , ∃ε(k) > 0 such that γ(J −ε ) ≥ k.
Because J is continuous and even, J −ε ∈ Σk ; then, ck ≤ −ε(k) < 0, ∀k. But J is bounded
from below; hence, ck > −∞ ∀k.
Let us assume that c = ck = ... = ck+r . Let us observe that c < 0 ; therefore, J verifies the
Palais-Smale condition in Kc , and it is easy to see that Kc is a compact set.
 49

If γ(Kc ) ≤ r, there exists a closed and symmetric set U , Kc ⊂ U , such that γ(U ) ≤ r.
( We can choose U ⊂ J 0 , because c < 0).
By the deformation lemma, we have an odd homeomorphism η : W01,p (Ω) → W01,p (Ω), such
that η(J c+δ − U ) ⊂ J c−δ , for some δ > 0. (Again, we must choose 0 < δ < −c, because J
verifies the Palais-Smale condition on J 0 , and we need J c+δ ⊂ J 0 ). By definition,

c = ck+r = inf sup J(u).

C∈Σk+r u∈C

Then, there exists A ∈ Σk+r , such that sup J(u) < c + δ ; i.e., A ⊂ J c+δ , and
u∈A

η(A − U ) ⊂ η(J c+δ − U ) ⊂ J c−δ (2.7)

But γ(A − U ) ≥ γ(A) − γ(U ) ≥ k , and γ(η(A − U )) ≥ γ(A − U ) ≥ k.

Then, η(A − U ) ∈ Σk . And this contradicts (2.7); in fact,

η(A − U ) ∈ Σk implies sup J(u) ≥ ck = c.

u∈η(A−U )

With the lemmas above we conclude Theorem 2.4.6. It is necessary only to point out that
there exists at least a positive solution because J˜ has a global minimum and the solution where
J˜ attains its minimum is positive. (See [53]).
With this result the fundamental idea is that all these solutions are related with the compo-
nent of smaller degree of the zero order term, i.e., by λ|u|q−2 u. In fact this same result is obtained
for the problem without the critical term. (See [51]). Moreover, in this subcritical problem, the
positive solution is unique. (See [41] where some ideas by Brezis for p = 2 are generalized).
On the other hand, it is not difficult to show that the solutions obtained above tend to 0
when λ → 0.
The previous remarks, make more relevant the problem of finding a second positive solution.
Such a solution will be find by the Mountain Pass Theorem, and then, when λ → 0 it is not
too hard to prove that converges to a Dirac’s delta. We can say in this sense that this second
positive solution corresponds to the contribution of the critical term.

Theorem 2.4.11 Assume that one of the following hypothesis holds.

2N
1. < p < 3, 1 < q < p.
N +2
2
2. p ≥ 3, p > q > q0 = p∗ − .
p−1
50

Then there exists a constant λ0 > 0 such that if 0 < λ < λ0 then problem (2.1) with q < p <
r = p∗ has at least two positive solutions.

Proof. The proof is much more delicate, then we give only some indications and refer to [54]
for more details.
Step 1.- A local Palais-Smale condition.
Consider c0 = Jλ (u0 ), minimum of the truncated functional. A detailed analysis gives us
that for a Palais-Smale sequence, {uj } ⊂ W01,p (Ω), such that
1 N/p
i) Jλ (uj ) → c < c0 + S ( S being the best Sobolev constant).
N
ii) Jλ′ (uj ) → 0,

and if λ small enough, then, there exists a strongly convergent subsequence.

Step 2.- The Mountain Pass Lemma hyphotesis.
It is possible to show that:
Given v ∈ W01,p (Ω) such that Jλ (v) < c0 there exists a sequence {uj } ∈ W01,p (Ω) such that

i)Jλ′ (uj ) → 0
ii)Jλ (uj ) → c = inf γ∈C supt∈[0,1] Jλ (γ(t))

where C = {γ ∈ C([0, 1], W01,p (Ω)) : γ(0) = 0, γ(1) = v}.

1
Therefore, we must find v ∈ W01,p (Ω) such that c < c0 + S N/p .
N
Step 3.- The energy estimate.
We are looking for an estimate of the type
1 N/p
sup Jλ (u0 + Ruε ) < Jλ (u0 ) + S ,
R>0 N

where uε is a convenient normalization of the Sobolev minimizer for ε small.

For such ε we choose v = u0 + Ruε , with R large enough, such that Jλ (v) < c0 .
We indicate the proof in the case 1) of the theorem.
By using the estimates for the Sobolev minimizers, that u0 is a solution of the problem,
and convenient estimates for Jλ (u0 + v) − Jλ (u0 ), it follows

Jλ (u0 + Ruε ) ≤
Z ∗ Z
Rp Rp ∗
Jλ (u0 ) + |∇uε |p dx − ∗ |uε |p dx+
p p
Ω Ω
N −p
> N p−p .
∗
C1 R γ εβ − C2 Rp −1 ε p , β
 51

For the parameters in the given range we have that the exponent of the negative error term,
is smaller than those of the positive error term. This remark allows us to conclude after some
estimates about the value of Rε where the maximum is attained. (See [54] for all the details).

Remark 2.4.12

1. p > q0 implies the following restriction for the dimension:

p(p − 1)
N > p(1 + ).
2

2. If q ≤ q0 the methods that we use in the proof don’t work.

2N
3. < p is equivalent to p∗ > 2.
N +2
The obstruction in the dimension that appears in the second case of the theorem can be
seen in the following example.
Example.-Assume p = 4, N = 8, q = 2 (and therefore p∗ = 8). It is possible to compute
the integrals in a explicit way (without any loss of accuracy in the estimates; we have an exact
algebraic expression). More precisely, in this case the error term takes the form

Jλ (u0 + Ruε ) ≤
Z Z
R4 R8
c0 + |∇U1 |4 dx − |U1 |8 dx + C1 ε2/3 − C2 ε2/3 + o(ε2/3 ),
4 8
and the constants depend on the L∞ norms of u0 and ∇u0 . Therefore, the sign of the coefficient
of the main error term it is not clear.
Nevertheless, numerical experiments performed by Prof C. Simó, seem to indicate that if
q < p then there is no restriction, while if p = q, it appears the phenomena of the bad dimensions
p2 < N (as in the case p = 2 and N = 3, see [29].)

2.5 Global results on multiplicity

In the work [11] by Ambrosetti, Brezis, and Cerami, it is proved the existence of a second positive
solution in the semilinear case (p=2). Their result is global in the following sense:
Let p = 2, 1 < q < 2, 2 < r ≤ N2N
+2 in problem 2.1 and

Λ = sup{λ | Problem (2.1) has a positive solution}.

For λ ∈ (0, Λ), there is a second positive solution of problem (2.1).

52

This global result is obtained by the application of the results in [30]. (See also [39]).
There is also a multiplicity result by Tarantello [85], when the term uq−1 is replaced by an
inhomogeneous term f (x) with a convenient norm small enough.
Here we will study the case p 6= 2 in the radial case. As a byproduct of the method we
obtain a different proof in the case 1 < p = 2 < (2 + N )/(N − 2)
Consider the radial problem


 Ap u = rN −1 fλ (u), 0 ≤ r < 1,
u′ (0) = 0, (2.8)

 u(1) = 0,

where

Ap u = −(rN −1 |u′ (r)|p−2 u′ (r))′ , fλ (u) = λuq−1 + uα−1 ,

and
Np
1 < q < p < α < p∗ , with p∗ = N −p if p < N and p∗ = ∞ otherwise.

By a solution of (2.8) we mean hereafter u ∈ C(0, 1) which solves (2.8) weakly. In the sequel we
shall make use (also without mentioning them explicitly) of the regularity results of [79], [44]
and [87], namely that if u ∈ C(0, 1) is any weak solution of (2.8), then u ∈ C 1,ν .

In section 1.5 we prove uniqueness of solution for the Cauchy Problem



 Ap u = rN −1 |u(r)|α−2 u(r),
u(0) = a, (2.9)
 ′
 u (0) = 0.

As an application of Lemma 1.6.1 we obtain the following uniqueness results.

Lemma 2.5.1 For λ = 0, problem (2.8) has a unique positive solution v0 .

Proof. Let u1 , u2 be two solutions of (2.8) for λ = 0. Let, for example, u1 (0) ≥ u2 (0) and set

u1 (0)
µ= , R = µ(p−a)/p
u2 (0)

and
r
v2 (r) = µu2 ( ).
R
 53

One checks that v2 satisfies the Cauchy problem,



 A p v2 = rN −1 |v2 (r)|α−2 v2 (r),
v2 (0) = µu2 (0) = u1 (0),

 v ′ (0) = 0.
2

By Lemma 1.6.1 one deduces that v2 (r) = u1 (r), namely

r
u1 (r) = µu2 ( ).
R
As a consequence, u1 (R) = µu1 (1) = 0, where R ≤ 1 because µ ≥ 1. Since u1 > 0 for r ∈ (0, 1),
it follows that R = 1, whence µ = 1. Then u1 (0) = u2 (0) and u1 ≡ u2 by the uniqueness for the
Cauchy Problem.

Lemma 2.5.2 For all a > 0 there exists a unique Ra such that
(
va (Ra ) = 0,
va′ (Ra ) < 0,

where va is the solution of the Cauchy problem (2.9).

Proof. Let a0 ≡ v0 (0) and let
a (p−α)/p
Ra = ( )
a0
a r
ũ(r) = v0 ( ).
a0 Ra
Then a direct computation shows that


 Ap ũ = rN −1 ũα−1
ũ(0) = a,

 ũ′ (0) = 0,

a
so ũ(r) = va (r) by uniqueness, and hence va (Ra ) = u0 (1) = 0. Moreover by Hopf lemma
a0
va′ (Ra ) < 0.
As a consequence of the previous lemma we obtain the following Corollary which is an
extension to our situation of the well known result by Gidas and Spruck, [56].
Corollary 2.5.3 Let u be any solution of
(
Ap u = rN −1 |u|α−2 u, 0 ≤ r < ∞,
u′ (0) = 0, u(r) ≥ 0.
Then u ≡ 0.
54

Proof. If a = u(0) = 0 the result follows trivially by the uniqueness for the Cauchy Problem.
Assume that u > 0. Then by Lemma 2.5.2 there exists Ra where u(Ra ) = 0 and u′ (Ra ) < 0, so
u changes sign, a contradiction.

Remark 2.5.4 An elementary consequence of the uniqueness of the Cauchy Problem is that in
the case N = 1 all the solutions to the problem (2.8) with λ = 0 are symmetric. The reader
could check that if N = 1 the moving planes method also works. (See [50]).

Now we obtain uniform bounds for the solutions of problem (2.8). In first place we have
the radial version of the Theorem 2.2.5.

Lemma 2.5.5 There exists Λ̃ > 0 such that for λ > Λ̃, problem (2.8) has no positive solution.

Set
Λ = sup{λ > 0 : (2.8) has a positive (radial) solution}.
By the previous Lemma we know that Λ < ∞. We know also that Λ > 0 because for λ > 0
small enough we can find a positive solution: it suffices to use critical point theory, see [52], or
by sub and supersolutions, see [24].
Now we get the uniform estimate in the L∞ -norm in the radial case.

Lemma 2.5.6 There exists C > 0 such that

||u||∞ ≤ C,

for all positive (radial) solutions of (2.8) and all λ ∈ [0, Λ].

Proof. We argue by contradiction. Let un be a sequence of (radial) positive solutions of

Ap un = rN −1 fλn (un ), ({λn } ⊂ (0, Λ)), such that ||un ||∞ → ∞. Without loss of generality we
can assume λn → λ.
It is easy to see that for any n the equation implies that un achieves its maximum at the
point r = 0, i.e., un (0) = ||un ||∞ .
Now, we normalize by using the following rescaling:

wn (r) = µβn un (µn r)

where 
 µn= un (0)−(α−p)/p
p
 β = >0
α−p
Obviously, limn→∞ µn = 0.
 55

Such a wn satisfies
  
β(p−q)+p

 A p wn = rN −1 λn µn wnq−1 + wnα−1 , 0 < r < µ−1
n ,

 wn (0) = 1,

wn (µ−1
n ) = 0.

Using Lemma 2.5.2, let R1 be such that v1 (R1 ) = 0 (hence v1 (0) = 1) and choose R̃ > R1 . From
the results of [44] and [87], it follows that wn are uniformly bounded in C 1,ν (0, R̃). Then one
has that
wn → v
in C 1 (0, R̃), up to a subsequence. The limit function v is a solution of (2.8) with λ = 0, v(0) = 1,
such that v(r) > 0 in 0 < r < R̃. This is a contradiction with the choice of R̃.

Remark 2.5.7 The extension of the result in Lemma (2.5.6) for solutions of (2.8) on general
domains and p 6= 2 is an open problem.

We can state the main result in this section.

Theorem 2.5.8 For all λ ∈ (0, Λ) the problem (2.8) admits at least two positive solutions

Proof. Fixed λ0 ∈ (0, Λ) we take µ ∈ (λ0 , Λ). Let u be a positive solution of (2.8) for λ = µ.
Now u is a strict supersolution to the problem (2.8) for λ = λ0 , because λ0 < µ. It is easy to
check that u = εφ1 with ε > 0 small enough and φ1 > 0 a positive eigenfunction, verifies u < u
and is a strict subsolution of the problem (2.8) for λ = λ0 .
Let us set
X = { v ∈ C 1 (Ω) : v(x) = v(|x|), v(1) = 0}
and, for λ ∈ (0, Λ), consider the map Kλ defined on X by setting:

Kλ (v) = (Ap )−1 (fλ (v)).

It is well known (see for example [47]) that Kλ maps X into itself. Moreover

u < Kλ (u), u > Kλ (u).

So, if we define
X = {v ∈ X | u ≤ v ≤ u},
we find that
Kλ0 : X −→ X .
56

By the C 1α estimates in [44] and [87] already refered in the beginning of this section, it follows
that Kλ is compact. In particular,
Kλ0 (X ) ⊂ X

is a compact set in X. The Schauder fixed point theorem implies that there exists a u0 ∈ X
such that Kλ0 (u0 ) = u0 , or, in others words, (2.8) has at least a solution in X for λ = λ0 . If u0
is not the unique fixed point in X we have nothing to prove. Otherwise, u0 can be obtained by
iterations and then u < u0 < u. So by the Hopf lemma we have that u0 is in the interior of X
(in the C 1 topology). Hence there exits ε > 0 such that u0 + εB1 ⊂ X . Here B1 denote the unit
ball in X.
To complete the proof we use topological degree arguments developed in [5]. In fact we have

deg(I − Kλ0 , u0 + εB1 , 0) =

(2.10)
i(Kλ0 , u0 + εB1 , E) = i(Kλ0 , X , X ) = 1,

where we use the permanence and excision properties of the degree. (See [5] Chapter 3, specially
the proof of the Schauder fixed point theorem for the last equality).
On the other hand, by Lemma 2.5.5, we know that for λ > Λ the problem (2.8) has no
positive solutions. By Lemma 2.5.6 we know that for all λ ∈ [0, Λ], (2.8) has no positive solution
such that
||u||∞ ≥ ρ > C.

So by the homotopy invariance of the Leray-Schauder degree we get

deg(I − Kλ0 , ρB1 , 0) = deg(I − KΛ+δ , ρB1 , 0) = 0.

Then by the excision property and (2.10) we infer

deg(I − Kλ0 , ρB1 \ {u0 + εB1 }, 0) =

(2.11)
deg(I − Kλ0 , ρB1 , 0) − deg(I − Kλ0 , u0 + εB1 , 0) = −1

Hence, Kλ0 has another fixed point u1 ∈ ρB1 \ {u0 + εB1 }.

Remark 2.5.9 The fact that Ω is a ball is used only to prove the uniform bounds of Lemma
2.5.6. If p = 2, such a priori estimate holds true for any bounded domain Ω, see [56]. As a
consequence, the proof of Theorem 2.5.8 gives an alternative method to show the subcritical
result in [11]. Actually, the minimal solution in [11] is a local minimum and has index = 1,
while the solution found in [11] as mountain pass critical point has index = −1.
 57

Remark 2.5.10 By using the a priori estimates of [56] and the degree theoretic results of [38],
one can prove the preceding bifurcation result for the second order problem
(
−∆u = λ|u|q−2 u + G′ (u), x ∈ Ω,
(2.12)
u = 0 x ∈ ∂Ω,

where G is possibly not symmetric and Ω a bounded domain.

2.6 Multiple bifurcation

We will look for solutions of the following problem


 Ap u = λh(u) + g(r, u), 0<r<1
u′ (0) = 0, (2.13)

 u(1) = 0,

where p > 1,
1
Ap u = − (rN −1 |u′ |p−2 u′ )′
rN −1
denotes the radial p-Laplace operator, λ ≥ 0 and h ∈ C(IR) and g ∈ C(IR+ × IR) satisfy the
following conditions:
h(s)
(H1 ) lims→0 = 1, with 1 < q < p.
|s|q−2 s
g(r, s)
(H2 ) lims→0 = 0.
|s|p−2 s
Hereafter all the properties of g are assumed to hold uniformly with respect to r ∈ [0, 1].
Let
X = {u ∈ C 1 (B1 ) : u(x) = u(|x|), u(1) = 0}
and set E = IR × X. ( u ∈ X is regular radial function and then u′ (0) = 0). In the sequel by a
solution of (2.13) we mean a pair (λ, u) ∈ E, such that u ∈ C 1,α and satisfies (2.13) in a weak
sense. Our main bifurcation result is:

Theorem 2.6.1 The point (λ, u) = (0, 0) is a bifurcation point for problem (2.13). More pre-
cisely, there are infinitely many unbounded continua (i.e. closed connected set) Γk ⊂ E of solu-
tions of (2.13) branching off from (0, 0), such that
58

1) If (λ, u) ∈ Γk and λ > 0 then u 6= 0.

2) If (λ, u) ∈ Γk , then u has exactly k − 1 simple zeros in the interval (0, 1).
3) There exists a constant ρo > 0 such that if ρ ∈ (0, ρo ], and (λ, u) ∈ Γk with kuk∞ = ρ, then
λ > λ(ρ) > 0.
As an inmediate consequence we get:
Corollary 2.6.2 There exists Λ > 0 such that for all λ ∈ (0, Λ) problem (2.13) has infinitely
many solutions.
Some of the preceding results hold in a greater generality, see Remark 2.6.7 below.
Theorem 2.6.1 cannot be proved using standard bifurcation techniques by linearization.
Actually, even when p = 2 and the differential operator becomes the linear Laplacian ∆, the
nonlinear term h has infinite derivative at u = 0. To overcome this problem we shall employ a
limiting procedure. Define


 h(δ) p−2
 |s| s if |s| ≤ δ,
hδ (s) = δ p−1



h(s) if |s| ≥ δ.
and consider the approximated problems


 Ap u = λhδ (u) + g(r, u), 0 < r < 1,
u′ (0) = 0, (2.14)

 u(1) = 0.

In order to study (2.14) we use the results from [43] studied in Sections 1.5 and 1.6 concerning
to bifurcation from nonlinear eigenvalues of the problem


 Ap u = λ|u|p−2 u, 0 < r < 1
u′ (0)
= 0, (2.15)

 u(1) = 0.

Hence, we have the following result.

Theorem 2.6.3 For each positive integer k there exists an unbounded continuum Sk ⊂ E of
solutions (λ, u) of (2.14) branching off from (µk,δ , 0), where,
δ p−1
µk,δ = · µk ,
h(δ)
and µk is an eigenvalue of problem (2.15). Moreover if (λ, u) ∈ Sk and u 6= 0 then u has exactly
k − 1 zeros in (0, 1).
 59

Remark 2.6.4 The arguments in Section 1.6 permit to handle problem (2.14). More precisely,
setting
δ p−1
µk,δ = · µk ,
h(δ)
one finds that from each (µk,δ , 0) branches off an unbounded continuum Sk,δ of solutions of (2.14)
having exactly k − 1 zeros in (0, 1).

For R > 0 let TR denote the ball of radius R in E. For each fixed k and R let Σk,δ denote the
connected component of Sk,δ ∩ TR that contains (µk,δ , 0). Let us point out that Σk,δ is nonempty
because Sk,δ is unbounded.
We will use the following topological theoretical lemma from [92]:

Lemma 2.6.5 Let {Σn }n∈IN be a sequence of connected sets in a complete metric space E.
Assume that

i) ∪Σn is precompact in E.

ii) lim inf Σn 6= ∅.

Then lim sup Σn is not empty closed and connected.

Here lim inf Σn , respectively lim sup Σn , denotes the set of all x ∈ E such that any neighbourhood
of x meets all but finitely many of Σn , respectively infinitely many Σn .
Take a sequence δn → 0 and let Σk,n = Σk,δn . By classical a priori estimates the set ∪Σk,n
is precompact. Moreover, since µk,δn → 0 as δn → 0, then (0, 0) ∈ lim inf Σk,n . Then Lemma
2.6.5 applies to Σn = Σk,n and hence Γk,R := lim sup(Σk,n ) = lim sup(Sk,δn ∩ TR ) is a not empty
closed and connected set. Moreover it is clear that Γk,R meets TR for all R > 0.
We set Γk = ∪R>0 Γk,R and show in the rest of the section that the Γk satisfy the properties
stated in Theorem 2.6.1.
First of all, it follows directly by the preceding arguments that each Γk is an unbounded
continuum in E and (0, 0) ∈ Γk . It is also clear that any (λ, u) ∈ Γk is a solution of (2.13). Next
we give the proofs of Properties 1 − 2 of Theorem 2.6.1.

Proof of 2.6.1-1). Let δ0 be such that λh(δ0 )δ01−p > µ1 and consider (λ, u) ∈ Σ1,δ , with λ > 0
and δ ∈ (0, δ0 ]. Fixed ǫ > 0 small, by assumptions (H1 − H2 ) there exists c = c(λ) > 0 such that

λhδ (s) + g(r, s) > (µ1 + ǫ)sp−1 , ∀ s ∈ (0, c].

Hence, if kuk∞ ≤ c, u satisfies

Ap u > (µ1 + ε)|u|p−1 u
60

and the usual iterative method yields the existence of a positive solution u ∈ X of the eigenvalue
problem
Ap u = (µ1 + ε)|u|p−1 u,
a contradiction, because µ1 +ε is not the first eigenvalue of (2.15). This shows that if (λ, u) ∈ Σ1,δ ,
with λ > 0 and δ ∈ (0, δ0 ], then kuk∞ > c(λ). Passing to the limit as δ → 0 it follows that if
(λ, u) ∈ Γ1 then kuk∞ ≥ c(λ).
When we consider Γk with k > 1 the argument is similar. If (λ, u) ∈ Σk,δ then there exists at
least one interval Ik with length 1/k where u has constant sign. Therefore if we restrict ourselves
to the interval Ik and we replace µ1 by the first eigenvalue of (2.15) on the interval Ik , then we
get the same contradiction as before.

Proof of 2.6.1-2). By 1) above, we know that if (λ, u) ∈ Γk and λ > 0 then u 6= 0. Let un be
such that (λn , un ) ∈ Σk,n , with λn → λ and un → u. First, let k = 1. Then un > 0 in [0, 1).
Hence u ≥ 0 and the strong Maximum Principle implies that u > 0 in [0, 1).
Next, take k > 1 and let xn < yn , 0 < xn < yn < 1, be two consecutive zeros of un with
xn → ξ and yn → η. Obviously, u(ξ) = u(η) = 0. We claim that ξ 6= η. Otherwise, there exists
a third sequence zn such that u′n (zn ) = 0 and lim zn = ξ. By the Ascoli-Arzelà theorem we find
that u is a solution of
Ap u = λh(u) + g(r, u),
such that u(ξ) = u′ (ξ) = 0. If ξ > 0 this is in contradiction with the Hopf Lemma applied to
the ball |r| ≤ ξ. If ξ = 0 the contradiction arises from the strong maximum principle applied to
a ball |r| ≤ ε, where u(r) ≥ 0. This shows that u has at least k − 1 zeros in (0, 1). On the other
side, if we suppose, as before, that, say, un > 0 in the open interval (xn , yn ), then u > 0 in (ξ, η)
and therefore u cannot have less than k − 1 zeros in (0, 1), proving 2).

Before proving Property 3 we state the following Lemma

Lemma 2.6.6 There exists ρ0 > 0 such that any nontrivial solution u of


 Ap u = g(r, u), 0 < r < 1
u′ (0) = 0, (2.16)

 u(1) = 0.

satisfies ||u||∞ > ρo .

Proof. If not, we can find a sequence un 6= 0 of solutions of (2.16) such that kun k∞ → 0.
Setting vn = un ||un ||−1
∞ one finds that

g(r, un )
Ap vn = . (2.17)
kun kp−1
∞
 61

Using the compactness of A−1p we infer that, up to a subsequence, vn → w in C

1,ν , with w 6= 0.

On the other side, from (2.17) it follows that

Z 1 Z 1 g(r, un ) p N −1
|vn′ |p rN −1 dr = v r dr.
0 0 |un |p−2 un n

According to (H2 ) the last integral tends to 0, a contradiction.

Proof of 2.6.1-3).
Arguing by contradiction, we suppose there exists a sequence (λn , un ) ∈ Σk,n such that
λn → 0, un → u and ||un ||∞ = ρ ≤ ρo . Passing to the limit we find that u 6= 0 is a solution of
(2.16) such that kuk∞ ≤ ρ0 , in contradiction with Lemma 2.6.6. This completes the Proof of
Theorem 2.6.1.

Remark 2.6.7 (i) The preceding arguments show that the existence of the continuum of positive
solutions Γ1 can be established for a problem with the same type of zero order term on a general
bounded domain Ω ⊂ IRN .
(ii) The preceding bifurcation and multiplicity results hold true also for Sturm-Liouville equations

−(a(x)u′ )′ + b(x)u = λh(u) + g(x, u),

with separated boundary conditions, provided h(u) ≃ |u|q−2 u, q < 2, at u = 0, and g is of higher
order.

2.7 Further results

We can give in some cases the global behaviour of the branches obtained in the Section 2.6.
More precisely we have the following results.
a) Subdiffusive nonlinearities.
Let us suppose that h(s) = |s|q−2 s and g(r, s)/|s|q−2 s → γ < 0 (possibly −∞) as |s| → ∞,
uniformly in r ∈ [0, 1]. In particular there exists s0 > 0 such that λh(s) + g(r, s) < 0 for any s
such that |s| > s0 and any λ ≥ 0.
We claim that if (λ, u) is any solution of (2.13) with λ ≥ 0 then kuk∞ ≤ s0 . To prove this fact
let r0 ∈ [0, 1), respectively r1 ∈ (0, 1], be the point where u attaints its maximum, respectively
minimum. From the equation it readily follows that λh(u(r0 )) + g(r0 , u(r0 )) ≥ 0, respectively
λh(u(r1 )) + g(r1 , u(r1 )) ≤ 0. Therefore u(r0 ) ≤ s0 , respectively u(r1 ) ≥ −s0 , as claimed. In
addition if , for example, g(r, s)s < 0 for all s 6= 0 then Γk ⊂ IR+ × {u ∈ X : kuk∞ ≤ s0 } for all
k and hence (2.13) has has infinitely many solutions for all λ > 0.
62

b) Equidiffusive nonlinearities. Consider the eigenvalue problem



 Ap u = λh(u), 0 < r < 1
u′ (0)
= 0, (2.18)

 u(1) = 0.

where h satisfies (H1 ), h > 0 on IR+ and

h(s)
lim = m∞ > 0.
s→+∞ sp−1
In such a case we can apply the results of [7], Section 4, to infer that Γ1 blows up at infinity as
h(s)
λ → λ∞ = µ1 · m−1 ∞ . Similarly, if → m∞ as |s| → ∞, then one could also show that Γk
|s|p−2 s
blows up at infinity as λ → µk · m−1 ∞ . Moreover, if m∞ = 0 then, similarly as in the preceding
point a), (2.18) has infinitely many solutions for all λ > 0.

c) Subcritical problems. Consider problem (2.13) and assume, in addition to (H1 − H2 ), that
g satisfies

(H3 ) g(r, s)s > 0 ∀s 6= 0;

(H4 ) g(r, s) ≃ |s|α−2 s as |s| → ∞, with 1 < q < p and p < α < p∗ , where p∗ = pN/(N − p) if
p < N and p∗ = +∞ if p ≥ N .

In particular, the growth of g is said subcritical, in the sense of the Sobolev embedding. In the
sequel, for the sake of simplicity, we will take h(s) = |s|q−2 s. We will also consider only solutions
u such that u(0) > 0.
The main result in this section is :

Theorem 2.7.1 Suppose that (H1 −H2 −H3 −H4 ) hold. Then, for any k ≥ 1 there exist Λk > 0
and Ck > 0 such that for every (λ, u) ∈ Γk with λ ≥ 0 one has

λ ≤ Λk , kuk∞ ≤ Ck .

In particular for all λ ≥ 0 small, problem (2.13) has infinitely many pairs uk , vk of solutions
with k − 1 zeros.

Remark 2.7.2 The continua emanating from (0, 0) turn back and cross the axis {λ = 0} at a
solution of (2.13) for λ = 0 with k − 1 zeros. In general, Ck → +∞ as k → +∞. We also suspect
that Λk → +∞. If this is true, then (2.13) has infinitely many pairs of solutions for any λ ≥ 0.
 63

The details can be found in [8] where it is described also the behaviour of the branch of
positive solutions Γ+
1 in the critical case. More precisely if


 Ap u = λuq−1 + uα−1 , u(r) > 0, in 0 < r < 1
u′ (0) = 0, (2.19)

 u(1) = 0.

we have

Theorem 2.7.3 Let p < N and suppose 1 < q < p < α ≤ p∗ and that if α = p∗ then either
p = 2 or
2N 2
< p < 3 or p ≥ 3 and p > q > p∗ − . (2.20)
N +2 p−1
holds.

(i) If α < p∗ then Γ+

1 is bounded.

(ii) If α = p∗ then Γ+ ∗ +
1 \ {0, 0} ⊂]0, Λ1 ] × X and blows up at infinity as λ → 0 , in the sense
sepecified above.

(iii) Eq. (2.19) has at least two (positive) solutions for all 0 < λ < Λ∗1 .

(iv) Eq. (2.19) has one (positive) solution for all 0 ≤ λ ≤ Λ∗1 , resp. 0 < λ ≤ Λ∗1 , when α < p∗ ,
resp. α = p∗ .

(v) Eq. (2.19) has no (positive) solution for λ > Λ∗1 .

The same problem in the supercritical case is considered in [3], where the existence of
an unbounded radial solution is proved, and the behaviour of Γ+
1 is discused in terms of the
dimension.
64
Chapter 3

What means Growth?

3.1 Introduction
In this Chapter we analyze problems related to a critical potential, more precisely, problems
related to the operator
λ
Lλ u ≡ −∆p u − p |u|p−2 u.
|x|
We will study equations of the form Lλ u = F (x, u) under suitable hypotheses of regularity and
growth for F .
∗
We will see that if for instance F (x, u) = up −1 , then the Dirichlet Problem in any bounded
λ
starshaped domain containing the origin, has no solution. This means that the term − p |u|p−2 u
|x|
does not regularize as in the autonomous case, previously studied in Chapter 2.
We can also look to this kind of result from the point of view of the eigenvalue problem in a
λ N
extreme case, namely, in the case where the potential − p belongs to Lr if and only if r < ,
|x| p
which is the complementary range of the classical results. ( See for instance the classical book
[63].)
More details and applications of this kind of ideas can be found in [55]. Roughly speaking
all the new problems appear from the lack of compactness in the Hardy inequality that we will
study in the next section.

3.2 Hardy inequality

The main point of this section is to discuss the following classical result, esentially dues to Hardy.
( See [59]). By completeness we include the proof.

65
66

Lemma 3.2.1 Assume 1 < p < N , then if u ∈ W 1,p (IRN ).

u
1. ∈ Lp (IRN ).
|x|
2. (Hardy Inequality)
Z Z
|u|p
dx ≤ CN,p |∇u|p dx
|x|p
IRN IRN
p
with CN,p = ( )p .
N −p
3. The constant CN,p is optimal.

Proof.
Step 1. A density argument allows us to consider only smooth functions u ∈ C0∞ (IRN ).
Under this hypothesis we have the following identity
Z ∞ Z ∞
p d
|u(x)| = − |u(λx)|p dλ = −p up−1 (λx)hx, ∇u(λx)idλ .
1 dλ 1

By using Hölder inequality, it follows that

Z Z
R |u(x)|p ∞ up−1 (λx) x
dx = −p h , ∇u(λx)idxdλ =
IRN
|x|p 1 |x|p−1 |x|
Z IRN Z
R∞ dλ u(y)p−1 ∂u(y) p u(y)p−1 ∂u(y)
−p 1 dy = − dy ≤
λN −1−p IRN |y| p−1 ∂r N −p |y|p−1 ∂r
Z Z IRN
p |u(y)|p ∂u(y) p 1/p
( dy)(p−1)/p ( | | dy) .
N −p |y|p ∂r
IRN IRN

And then we conclude that

Z Z
up (x) p
p
dx ≤ ( )p |∇u(x)|p dx .
|x| N −p
IRN IRN

Step 2. Optimality of the constant . Following the idea of Hardy for the one dimensional
case, we show that the best constant is CN,p = ( N p−p )p .
Given ε > 0, take the radial function
(
AN,p,ε if r ∈ [0, 1],
U (r) = p−N
−ε (3.1)
AN,p,ε r p if r > 1,
 67

where AN,p,ε = p/(N − p + pε), whose derivative is

(
′ 0, if r ∈ [0, 1],
U (r) = −N −ε (3.2)
−r p if r > 1.
By direct computation we get
Z Z Z
U p (x) U p (x) U p (x)
dx = dx + dx =
|x|p B |x|p IRN −B |x|p
IRN
Z 1 Z ∞ 
= ApN,p,ε ωN r N −1−p
dr + r −(1+pε)
dr =
0 1
Z 1 Z
= ApN,p,ε ωN rN −1−p dr + ApN,p,ε |∇U (x)|p dx,
0
IRN

where ωN is the measure of the (N − 1)-dimensional unit sphere. We conclude by letting ε → 0.

Corollary 3.2.2 The same result is true in W 1,p (B), where B is the unit ball in IRN .
Proof. The proof of the first step is the same. The argument of optimality in the case of the
unit ball proceeds by approximation as follows. First, we remark that by the invariance under
dilations the optimal constant has to be the same for any ball. Second, let BR be a ball with
large radius. We take as test function v(x) = ψ(x)U (x) where U is one of the approximate
optimizers explicitly given above and ψ ∈ C0∞ (BR ) is a cutoff function which is identically 1 on
BR−1 with |∇ψ| ≤ m. It is easily seen that for R ≫ 1 the influence of ψ in the calculation of
Step 2 is negligible.

Remark 3.2.3 Sometimes the Hardy inequality in the case p = 2 is known as uncertainty
principle, see [46]. We can read the Hardy inequality saying that the embedding of W 1,p (IRN ) in
Lp with respect to the weight |x|−p is continuous. It is very easy, working a little bit more with
the minimizers that we use in the proof, to see that the inclusion is non compact. This will be
the cause of many of our difficulties.
−1
In the sequel, we will denote λN,p = CN,p .
It will be useful to compare the best constant in the Hardy inequality with the following
approximating eigenvalue problems.
Theorem 3.2.4 Consider λ1 (n) the first eigenvalue to the problem
(
−∆p ψ1 = λWn (x)|ψ1 |p−2 ψ1 , x ∈ Ω ⊂ IRN ,
(3.3)
ψ1 (x) = 0, x ∈ ∂Ω.
68

where Wn (x) = min{|x|−p , n}.

Then λ1 (n) ≥ λN,p , and moreover lim λ1 (n) = λN,p .
n→∞

Proof. The first inequality follows immediatly from the definition of the first eigenvalue by
the Rayleigh quotient. Also, it is easy to see that {λ1 (n)} is a nonincreasing sequence; then
we have to prove that the limit cannot be bigger than λN,p . Assume by contradiction that
lim λ1 (n) = λN,p + ρ.
n→∞
Then, we can choose φ ∈ W01,p (Ω) such that
R
|∇φ|p dx
Ω
R < λN,p + ρ/2.
φp |x|−p dx
Ω
R
|∇φ|p dx
But then λ1 (n) ≤ R Ω p , and this is a contradiction because the last expression has to
φ Wn (x)dx
Ω
be smaller than λN,p + ρ for n large.

3.3 The Dirichlet problem with singular potential

The first result in this section is an easy consequence of the Hardy inequality
Lemma 3.3.1 Consider the nonlinear operator
λ
Lλ u ≡ −∆p u − |u|p−2 u (3.4)
|x|p

in W01,p (Ω). Then

1. If λ ≤ λN,p , Lλ is a positive operator.
2. If λ > λN,p , Lλ is unbounded from below.
Proof. 1) It is obvious from the Hardy inequality . 2) An easy consequence of the optimality
of the constant and a density argument is the existence of φ ∈ C0∞ (Ω) such that hLλ φ, φi < 0.
We can assume that ||φ||p = 1 and then by defining uµ (x) = µN/p φ(µx) we have ||uµ ||p = 1 and
the homogeneity of the operator allows us to conclude that hLλ uµ , uµ i = µp hLλ φ, φi < 0.
Taking into account the previous result, in this section we will study the following problem

 L u ′ 1 1
λ = f (x) ∈ W −1,p (Ω), x ∈ Ω, λ < λN,p , + =1
p p′ (3.5)

u(x) = 0, x ∈ ∂Ω,
 69

where Ω ⊂ IRN is a bounded domain. If 0 ∈ / Ω then we have a classical problem with a bounded
potential. So we will assume hereafter that 0 ∈ Ω.
Consider the energy functional,
Z
J(u) = F (x, u, ∇u)dx
Ω

1 λ up
where F (x, u, ξ) = |ξ|p − − f (x)u.
p p |x|p
The classical results in the Calculus of Variations characterize the weak lower semicontinuity
of J if F (x, u, •) is convex and F verifies a lower bound: positivity, or a lower estimate by a
linear combination of ξ, etc. (See Tonelli [86], Serrin [78], De Giorgi [37], the book by Dacorogna
[34] and the references therein). However, in our case these usual hypotheses are not fulfilled.
Variational approach.- The energy functional,
Z Z Z
1 λ up
J(u) = |∇u|p dx − dx − f udx,
p p |x|p
Ω Ω Ω

by the Hardy inequality, is continuous, Gateaux differentiable and coercive, namely, there exist
constants γ > 0 and c ∈ IR such that
Z
J(u) ≥ γ |∇u|p dx − c.
Ω

Hence by the Variational Principle of Ekeland we can find a sequence {un }n∈IN such that

J(un ) → inf J, and J ′ (un ) → 0, as n → ∞.

As usually we say that {un }n∈IN is a Palais-Smale sequence. (See [45] and Appendix). The coer-
civity of J implies the boundedness of {un } in W01,p (Ω), so we have that for some subsequence:

i) ∇un ⇀ ∇u in Lp ,

ii) un converges in Lp and a.e.,

up−1
n
iii) λ are bounded as Radon measures and converges weakly in L1 .
|x|p
Under these hypotheses we can apply the convergence theorem by Boccardo and Murat in [25].
The results proved by Boccardo and Murat are much more general and we refer to the paper
[25] for the details.
70

Lemma 3.3.2 Let {un }n∈IN verifying the problems

(
−∆p un = gn + fn , in Ω
(3.6)
un ∈ W01,p (Ω).

Assume that

1) un ⇀ u weakly in W01,p (Ω) and un → u in Lp (Ω).

′
2) fn → f in W −1,p (Ω).

3) gn ⇀ g weak-∗ in the sense of the measures.

Then, ∇un → ν in (Lq (Ω))N for 1 < q < p.

Moreover if 3) is replaced by

3′ ) gn ⇀ g weak-∗ in L1 (Ω),

then
Tk (un ) → Tk (u) in W01,p (Ω)
for all k > 0, where Tk (s) = s if |s| ≤ k and Tk (s) = ks/|s| if |s| ≥ k.

Proof. We take as test function Tk (un − u) ∈ W01,p (Ω). Then

R
h|∇un |p−2 ∇un − |∇u|p−2 ∇u, ∇Tk (un − u)idx =
Ω R (3.7)
− h|∇u|p−2 ∇u, ∇Tk (un − u)idx + hgn , Tk (un − u)i + hfn , Tk (un − u)i
Ω

Fixed k, by 1) we have that

hfn , Tk (un − u)i → 0
and Z
h|∇u|p−2 ∇u, ∇Tk (un − u)idx → 0
Ω
as n → ∞.
From 2) we have |hgn , Tk (un − u)i| ≤ Ck.
Then fixed k we get
Z
lim sup h|∇un |p−2 ∇un − |∇u|p−2 ∇u, ∇Tk (un − u)idx ≤ Ck.
n→∞
Ω
 71

Now the inequalities in Appendix A imply that the functions

en (x) = h|∇un |p−2 ∇un − |∇u|p−2 ∇u, ∇Tk (un − u)i
are non negative and uniformly bounded in L1 (Ω). Take 0 < θ < 1 and split Ω in
Snk = {x ∈ Ω| |un − u| ≤ k}, Gkn = {x ∈ Ω| |un − u| > k}
Hölder inequality provides the following estimate
R R R
eθn dx = θ
k en dx
Sn + θ
Gkn en dx ≤
Ω
 θ R θ
R
k
Sn en dx |Snk |1−θ + Gkn en dx |Gkn |1−θ .

Now, fixed k, |Gkn | → 0 as n → ∞ and from the uniform boundedness in L1 we get

Z
lim sup eθn dx ≤ (Ck)θ |Ω|1−θ ,
n→∞
Ω

and letting k → 0 we get that eθn → 0 strongly in L1 . To finish the first part we use the
inequalities for the p-laplacian obtained in appendix A.1 and we conclude.
If we assume now 3′ ), the idea is to get the same type of estimate but with θ = 1. We take
as test function Tk (un − u) and proceed in a similar way. We refer to the reader to [25] for the
details.
In our context the Lemma 3.3.2 will be read as follows.
Lemma 3.3.3 Let {un }n∈IN verifying (i), (ii) and (iii) by the Variational Principle of Ekeland.
Then for some subsequence,
unj → u, in W01,q (Ω), q < p
and
Tk (un ) → Tk (u) in W01,p (Ω)
for all k > 0, where Tk (s) = s if |s| ≤ k and Tk (s) = ks/|s| if |s| ≥ k.
According with the previous Lemma, and by a density argument, we can prove the required
compactness property. We will call such a compactness property singular Palais-Smale condition
in the sense of the following Lemma:
Lemma 3.3.4 Consider {un }n∈IN the Palais-Smale sequence obtained above.
Then {un }n∈IN satisfies the singular Palais-Smale condition, namely, there exists a subse-
quence {unj }j∈IN such that
unj → u, in W01,q (Ω), q < p.
72

An inmediate consequence of Lemma (3.3.4) is that u is a solution of our problem in the sense
of distributions. Moreover by density and taking into account that u ∈ W01,p (Ω), we conclude
that u is solution in the sense of W01,p (Ω).
Finally the homogeneity of the problem implies that u is a minimum for J. Consider
Z
1 1
J(uk ) − hJ ′ (uk ), uk i = ( − 1) f uk ,
p p
Ω

where in the last term we can pass to the limit by weak W01,p (Ω)-convergence. Therefore
1
inf J = lim J(uk ) = lim (J(uk ) − hJ ′ (uk ), uk i) =
k→∞
Z k→∞ p
1 1 ′
( − 1) f u = J(u) − hJ (u), ui = J(u).
p p
Ω

We would like to point out that this approach solves the minimization problem, namely, the
solution is obtained as a minimum of J. Also it is interesting to emphasize that this approach
will be used to study problems with unbounded energy functionals in the following sections.

Remark 3.3.5
I) The uniqueness in the case p = 2 is obvious. Also it is easy to prove that in the linear case
(p = 2) the sequence of minimizers converges strongly in the Sobolev space.
II) If p > 2 the uniqueness is in general not true as the following argument shows (see [42]).
Assume B ⊂ Ω a ball and consider u0 ∈ C02 (Ω) such that u = k > 0 in B. Consider f (x) = Lλ u
′
with λ < λN,p . In this way f ∈ W −1,p (Ω) and then we can find a solution v by minimization of
J as above. Now it is clear that such v 6= u0 , because u0 cannot be a minimum for J: in fact,
taking into account that p > 2, we see that
 
Z Z p−2
|u 0 |
hJ ′′ (u0 )z, zi = (p − 1)  |∇u0 |p−2 |∇z|2 dx − λ p
z 2 dx .
|x|
Ω Ω

Hence, in particular if z ∈ C0∞ (Ω) is such that supp (z) ⊂ B,

Z
|u0 |p−2 2
hJ ′′ (u0 )z, zi = −λ(p − 1) z dx < 0,
|x|p
Ω

and then u0 is not a minimum for J.

III) The uniqueness in the case 1 < p < 2 seems to be an open problem.
IV) The strong convergence of the Palais-Smale sequence if p 6= 2, is unknown. The main
difficulty is classical, we start with a weakly convergent sequence and the dependence in the
gradient is nonlinear.
 73

3.4 Critical potential and subcritical growth

In this section we will try to study the same kind of problems as in Chapter 2, but including
λ
the critical potential .
|x|p
We will use again the variational approach to study the case of unbounded funcionals,
more precisely the existence of solution via the Mountain Pass Lemma. (See the Appendix). For
instance the following result holds.

Theorem 3.4.1 Consider the Dirichlet problem

(
−∆p u = λ|u|p−2 u|x|−p + |u|α−2 u, λ < λN,p , p < α < N p/(N − p),
u|∂Ω = 0.

Then there exists at least a positive solution u ∈ W01,p (Ω).

Proof. The proof of this theorem follows closely the previous variational approach; instead of
minimizing and using the variational principle of Ekeland, the geometry of the energy functional
allows us to use the Mountain Pass Lemma of Ambrosetti-Rabinowitz. In fact since λ < λN,p
then Z Z Z
1 p λ |u|p 1
J(u) ≡ |∇u| dx − dx − |u|α dx ≥
p p |x|p α
R Ω R Ω Ω
γ |∇u|p dx − C(p, α)( |∇u|p dx)q/p .
Ω Ω

Then we find the required Palais Smale sequence, which is easy to show has to be bounded in
W01,p (Ω). Using again Lemma 3.3.3, we get the compactness result, in the sense of the singular
Palais Smale condition defined above. Then we can find a function u ∈ W01,p (Ω) which is the
strong limit in W01,q (Ω), q < p, and therefore is a solution in the sense of distributions. Since
u ∈ W01,p (Ω), by density we get that u is a weak solution.
Finally, by homogeneity, and taking into account that we have strong convergence in Lα ,
we can see that u 6≡ 0:

c = lim J(un ) =
n→∞
1
lim (J(un ) − hJ ′ (un ), un i) =
n→∞
Zp Z
1 1 1 1
lim ( − ) |un |α dx = ( − ) |u|α dx,
n→∞ p α p α
Ω Ω

and we conclude.
74

3.5 Critical potential and critical growth

We will use the method developed by Pohozaev to show how our problem with critical Sobolev
exponent is not regularized by the term in |x|−p up−1 . This is a deep difference with the au-
tonomous case studied in the Chapter 2. In this sense, the following nonexistence result shows,
from a different point of view, the critical character of the problem.
Lemma 3.5.1 Consider the problem


 up−1
−∆p u = λ + γf (u), x ∈ Ω, λ > 0,
|x|p (3.8)

 u(x) = 0, on ∂Ω,

where Ω ⊂ IRN is bounded, starshaped with respect to the origin, f is a continuous function and
Z u
N −p
γ(N F (u) − uf (u)) ≤ 0, F (u) = f (s)ds.
p 0

Then (3.8) has no positive solution u ∈ W01,p (Ω).

Proof. We will use a Pohozaev type identity. The idea consists on multiply the equation by
hx, ∇ui and integrate by parts. (Observe that the regularity of u does not suffice to justify this
calculus directly but as Pohozaev points out in [71] the results are valid also for weak solutions;
an argument of approximation which justifies the previous observation can be seen for instance
in [57]. See also [73].)
Z Z
p−1 N −p
( ) |∇u|p hx, νidσ + ( ) |∇u|p dx =
p ∂Ω p
Z Z Ω
N −p up
λ( ) dx + γN F (u)dx,
p |x|p
Ω Ω

where ν is the outwards normal to ∂Ω. On the other hand, multiplying the equation by u and
integrating Z Z Z
up
|∇u|p dx = λ dx + γ uf (u)dx. (3.9)
|x|p
Ω Ω Ω
Both identities give
Z Z
p−1 N −p
( ) |∇u|p hx, νidσ = γ (N F (u) − uf (u))dx.
p ∂Ω p
Ω

The conclusion is now obvious.

Notice the contrast of Lemma 3.5.1 with the results in Chapter 2 and in the papers [29]
[68] in the case p = 2, and in [53] for p 6= 2, where the autonomous case is considered.
 75

Remark 3.5.2 In the case Ω = IRN , p = 2 and f (u) = u(N +2)/(N −2) an existence result can
be seen in [64], Th.I.3, pg. 179. However the previous Lemma proves that this doubly critical
problem has no positive solution in bounded starshaped domains. This means that the term with
the potential cannot be seen as a lower order perturbation of the term with critical Sobolev
exponent, although this is the case in terms of the growth in u.

3.6 Subcritical potential and critical growth

We will prove that in fact the potential λ|x|−p is critical in the sense of the remark above by
considering λ|x|−q with 0 < q < p.
N
It is known that for the potential λ|x|−q , which belongs to Lr for some r > , there exists
p
a first isolated and simple eigenvalue λ1 , for the corresponding Dirichlet problem. We will show
that this subcritical potential produces a similar effect to the case q = 0, in the sense that
the lack of compactness arises only from the highest power term. Moreover there are also bad
dimensions, p < N < p2 − (p − 1)q as in the autonomous case. This interval coincides with the
known result in the case q = 0, and decreases when q → p, dissappearing the solution in the
limit case q = p, according to Lemma 3.5.1.
Theorem 3.6.1 Consider the Dirichlet problem


 |u|p−2 u ∗
−∆p u = λ + |u|p −2 u, λ < λ1
|x| q (3.10)

 u|∂Ω = 0, 0 < q < p, 1 < p < N, p∗ = N p/(N − p).

If N ≥ p2 − (p − 1)q, then there exists at least a positive solution u ∈ W01,p (Ω).

Proof. The arguments are similar to those in [53] and then we will be sketchy. The geometry
of the energy functional, J, satisfies the requirements of the Mountain Pass Theorem as can be
checked following the same calculations as in Theorem 3.4.1. Then by using the concentration-
compactness method by P.L. Lions, see [64], we get a local Palais-Smale condition. More precisely,
let S the optimal constant in the Sobolev embedding and given a Palais-Smale sequence such
that
S N/p ′
J(uk ) → c < , J ′ (uk ) → 0, in W −1,p (Ω)
N
there exists a convergent subsequence. The only thing to check (and this is the main point), is
the existence of a Palais-Smale sequence at this subcritical energy level. To get this particular
sequence it is well known that it is sufficient to find a direction vε ∈ W01,p (Ω) for which
S N/p
sup J(tvε ) < c0 < , (3.11)
t>0 N
76

because in this way the minimax critical value in the Mountain Pass Theorem verifies c < c0 .
The natural election is
uε φ(x)
vε = , where uε = p N −p
||uε ||p∗ (ε + |x| p−1 ) p

are the minimizers of the Sobolev inclusion with a convenient truncation φ ∈ C0∞ to adapt their
support to Ω. As usual we compute:
N −p
1. ||∇vε ||pp ≈ S + c1 ε p .

R vεp p2 −(p−1)q−p
2. q
dx ≈ c2 ε p if N > p2 − (p − 1)q.
Ω |x|
R vεp N −p
3. dx ≈ c2 ε p | log ε| if N = p2 − (p − 1)q.
Ω |x|q
R vεp N −p
4. q
dx ≈ c2 ε p if N < p2 − (p − 1)q.
Ω |x|
Following the proofs in the previous Chapter, the decay as ε → 0 of ||∇vε ||pp must be faster than
R vεp
the decay of q
dx, to get the inequality (3.11) and this is true if N ≥ p2 − (p − 1)q. So we
Ω |x|
conclude. (See also the proof of Theorem 3.3 in [53]).

Remark 3.6.2 Something more can be said in the case N < p2 − (p − 1)q. Following the same
argument as in the proof of Theorem 3 in [16] weR find that there exists a positive solution to
problem (3.10) if λ ∈ (λ1 − A, λ1 ), where A = S( |x|−N q/p dx)−p/N (S is the optimal Sobolev
Ω
constant).

Remark 3.6.3 A good exercise to the reader is to study different combinations of potentials
and nonlinearities and describe the complete fauna.
 77

Appendix
78
Appendix A

The inverse operator of the

p-Laplacian

We will study the p-Laplacian operator defined by

∆p u ≡ div (|∇u|p−2 ∇u) 1<p<∞

that appears in several situations. The p-Laplacian is the paradigmatic example of degener-
ated/singular quasilinear elliptic operator. Notice that if p = 2 it becomes the classical Laplace
operator. In this section we deal mainly with the Dirichlet problem,
(
−∆p u = f (x) if x ∈ Ω
(A.1)
u|∂Ω = 0 if x ∈ ∂Ω,

where Ω ∈ IRN , f ∈ W −1,p (Ω), p′ = p/(p − 1). The boundary condition will be understood as
′

u ∈ W01,p (Ω).
We have the following elementary result, as application of the classical Calculus of Varia-
tions.
Theorem A.0.4 Assume Ω ⊂ IRN a bounded domain and f ∈ W −1,p (Ω), then problem A.1
′

has a solution u ∈ W01,p (Ω) in the weak sense, namely

Z
{h|∇u|p−2 ∇u, ∇φi − f φ}dx = 0, ∀φ ∈ W01,p (Ω). (A.2)
Ω

Proof. The functional

Z Z
1
J(u) = p
|∇u| dx − f (x)udx u ∈ W01,p (Ω)
p Ω Ω

79
80

is Gateaux diferentiable and the Euler equation of J coincide with the partial differential equa-
tion under study. Then, we look for solution to problem A.1 as critical points of J. But, we
have
i) J is coercive since
1
J(u) ≥ ||∇u||pp − ||f ||W −1,p′ (Ω) ||∇u||p
p
1 ′
≥ ||∇u||pp − Ckf kpW −1,p′ (Ω) .
2p
ii) J is lower weakly semicontinuous because the first term is the norm in W01,p (Ω) and the
second one is continuous. Hence we can apply the abstract minimization result to J to find a
minimum.
We are interested in the properties of the inverse operator

(−∆p )−1 : W −1,p (Ω) −→ W01,p (Ω)

′

and we need the following inequalities. (See [80]).

Lemma A.0.5 Let x, y ∈ IRN and h·, ·i the standard scalar product in IRN . Then


 cp |x − y|p , if p ≥ 2
p−2 p−2
h|x| x − |y| y, x − yi ≥ |x − y|2 (A.3)

 cp , if 1 < p < 2.
(|x| + |y|)2−p
Proof. By homogeneity we can assume that |x| = 1 and |y| ≤ 1. Morover by choosing a
convenient basis in IRN we can assume
q
x = (1, 0, ..., 0), y = (y1 , y2 , 0, ..., 0), and y12 + y22 ≤ 1.

i) Case 1 < p < 2. It is clear that the inequality is equivalent to the next one
 q 2−p
  
  1− y12 + y22
1 −
y1  (1 − y1 ) +
y22


2−p  ≥ C. (A.4)

y12 + y22
2−p
2 y12 + y22 2 (1 − y1 )2 + y22

But 
 y1
y1 1− ≥ (p − 1)(1 − y1 ), if 0 ≤ y1 ≤ 1
1 − q 2−p ≥ |y1 |2−p
 1 − y1 ≥ (p − 1)(1 − y1 ), if y1 ≤ 0,
2 2
y1 + y 2

then 2−p
2 (1
+ y 1 + y2 ) 2
(p − 1){(1 − y1 ) + y22 } ≥ p − 1.
(1 − y1 )2 + y22
 81

ii) Case p ≥ 2. The inequality is equivalent to prove that

 

p−2
p−2
1− y1 y12 + y22 2 (1 − y1 ) + y22 y12 + y22 2

p ≥ C.
(1 − y1 )2 + y22 2

|y| hx, yi
Denote t = and s = then, we must show that the function
|x| |x||y|

1 − (tp−1 + t)s + tp
f (t, s) = p ,
(1 − 2ts + t2 ) 2
∂f
is bounded from below. Direct calculation shows that fixed t, = 0 if
∂s
tp−2 + 1
1 − (tp−1 + t)s + tp = (1 − 2ts + t2 ),
p
then for the critical s for f we have
tp−2 + 1 1
f (t, s) = p−2 ≥
p (1 − 2ts + t2 ) 2

1 tp−2 + 1 1 tp−2 + 1 1
p−2
≥ min ≥ .
p (t + 1) p 0≤t≤1 (t + 1)p−2 2p

The main properties of −∆p and (−∆p )−1 are summarized in the following theorem.
Theorem A.0.6 Let Ω ⊂ IRN a bounded domain.
A) ∆p : W01,p (Ω) → W −1,p (Ω) is uniformly continuous on bounded sets.
′

B) (−∆p )−1 : W −1,p (Ω) → W01,p (Ω), and is continuous.

′

C) The composed operator

(−∆p )−1 : W −1,p (Ω) → W01,p (Ω) ֒→ Lq (Ω)

′

pN
is compact if 1 ≤ q < .
N −p
Proof.
A) Consider C ⊂ W01,p (Ω) bounded set, i.e.,

kukW 1,p (Ω) < M, if u ∈ C.

0
82

Then for u, v ∈ C we get

R
k − ∆p u − (−∆p v)kW −1,p′ (Ω) = sup h∆p u − ∆p v, ∇φidx ≤
kφk 1,p =1 Ω
W (Ω)
 R 0


 cp sup Ω |∇u|p−2 + |∇v|p−2 |∇u − ∇v||∇φ|dx, if p ≥ 2

 kφk 1,p =1
W (Ω)
≤ 0 R



cp sup |∇u − ∇v|p−1 |∇φ|dx, if 1 < p < 2
 kφk 1,p =1 Ω
W (Ω)
0

then by Hölder inequality we conclude that

(
2Cp M p−2 k∇u − ∇vkp , if p ≥ 2,
|| − ∆p u − (−∆p v)||W −1,p′ (Ω) ≤
Cp M k∇u − ∇vkp , if p ≤ 2.

B) We need to prove uniqueness of solution to the Dirichlet problem A.1. Let u1 , u2 ∈ W01,p (Ω)
be solutions to the problems
−∆p u1 = f1 , −∆p u2 = f2 ,
then
h−∆p u1 − (−∆p u2 ), (u1 − u2 )i = hf1 − f2 , u1 − u2 i ,
where the first term must be understood as the duality product. By the previous lemma

h−∆p u1 − (−∆p u2 ), (u1 − u2 )i =

R
h|∇u1 |p−2 ∇u1 − |∇u2 |p−2 ∇u2 , ∇(u1 − u2 )idx ≥
Ω  R

 Cp |∇(u1 − u2 )|p dx, if p ≥ 2
 Ω
≥ R |∇(u1 − u2 )|2

 Cp dx, if 1 < p < 2,

Ω (|∇u1 | + |∇u2 |)2−p

then if p ≥ 2 R
|∇(u1 − u2 )|p dx ≤ Dp ||f1 − f2 ||W −1,p′ (Ω) ||∇(u1 − u2 )||p
Ω
1 1
||∇(u1 − u2 )||p ≤ Dpp−1 ||f1 − f2 ||W
p−1
−1,p′ (Ω) ,

in particular, if f1 ≡ f2 implies u1 ≡ u2 .
If 1 < p < 2
Z
|∇(u1 − u2 )|2
dx ≤ Cp ||f1 − f2 ||W −1,p′ (Ω) ||u1 − u2 ||W 1,p (Ω) ,
(|∇u1 | + |∇u2 |)2−p 0
Ω
 83

and then by Hölder inequality

R
|∇(u1 − u2 )|p dx =
Ω
R |∇(u1 − u2 )|p p(2−p)

p(2−p)
(|∇u1 | + |∇u2 |) 2 dx ≤
Ω (|∇u1 | + |∇u2 |) 2
!p ! 2−p
R |∇(u1 − u2 )|2 2
R p
dx (|∇u1 | + |∇u2 |)p dx .
Ω (|∇u1 | + |∇u2 |)2−p Ω

Hence !1
p
R
|∇(u1 − u2 )|p dx
Ω
 2−p ≤ Cp ||f1 − f2 ||W −1,p′ (Ω) .
||u1 ||W 1,p (Ω) + ||u2 ||W 1,p (Ω)
0 0

In this way we have obtained simultaneously the existence of the inverse operator (uniqueness)
and its continuity.

C) Is a direct consequence of B) and the Sobolev embedding

Lemma A.0.7 (Weak comparison Principle). Let Ω be a bounded domain in IRN with smooth
boundary. Let u1 , u2 ∈ W01,p (Ω) satisfy
(
−∆p u1 ≤ ∆p u2 in the weak sense in Ω
u1 ≤ u2 on ∂Ω

then u1 ≤ u2 in Ω.

Proof. Take as test function φ = max{u1 − u2 , 0} that is nonnegative and belong to W01,p (Ω)
by hypothesis. So we get
Z
h|∇u1 |p−2 ∇u1 − |∇u2 |p−2 ∇u2 , (∇u1 − ∇u2 )idx ≤ 0,
{u1 >u2 }

and we conclude taking into account the Lemma A.0.5.

We will use the following extension of the Hopf Lemma, that can be found in [77].
Lemma A.0.8 Let Ω be a bounded domain with smooth boundary. If u ∈ C 1 (Ω) ∩ W01,p (Ω) and
verifies 
 p−2 ∇u) ≥ D′
 −∆p u ≡ −div(|∇u| 0 in
u > 0 in Ω

 u|∂Ω = 0
84

∂u
then < 0 on ∂Ω . Here n is the outward normal to ∂Ω.
∂n
Proof. We will assume that the domain Ω verifies the interior sphere condition.
Consider x0 ∈ ∂Ω and a interior ball tangent to ∂Ω, namely,

Br (y) ⊂ Ω, B̄r (y) ∩ ∂Ω = {x0 }.

Define ( p−N

v(x) = A|x − y| p−1 + B N 6= p

A log |x − y| + B, N = p,
where A = [2(N −p)/(p−1) − 1]−1 r(N −p)/(p−1) , B = [1 − 2(N −p)/(p−1) ]−1 if p 6= N , and A =
log r
−(log 2)−1 , B = if p = N . The function v verifies:
log 2
i) v(x) ≡ 1 in ∂B 2r (y) and v(x) ≡ 0 on ∂Br (y),

ii) 0 < v(x) < 1 if x ∈ Br (y) − B 2r (y) and

|∇v(x)| > c > 0, for some positive constant c.

But u(x) > 0 in Ω, then

τ = inf{u(x)| x ∈ ∂B 2r (y)} > 0,
and putting τ v = w we find that w satisfies


 −∆p w(x) = 0, if x ∈ Br (y) − B̄ r2 (y)
w(x) = τ, if x ∈ ∂B r2 (y) and


w(x) = 0, if x ∈ ∂Br (y).

Now w ≤ u on the boundary of the ring Br (y) − B r2 (y) and ∆p w ≤ ∆p u in Ω in the weak sense.
Hence, the weak comparison principle implies that w ≤ u in Br (y) − B r2 (y). Taking into account
that u(x0 ) = w(x0 ) = 0, then

∂u u(x0 − tn) w(x0 − tn) ∂w ∂v

(x0 ) = lim ≤ lim = (x0 ) = τ (x0 ) < 0.
∂n t→0 t t→0 t ∂n ∂n
Appendix B

Genus and its properties

We introduce in this section the classical idea of genus due to Krasnoselskii, as a tool to
measure in a convenient way the size of the sets.

Given X a Banach space, we consider the class

Σ = {A ⊂ X | A closed , A = −A}. (B.1)

Definition B.0.9 Then the genus, γ, is defined as follows.

γ : Σ −→ IN ∪ {∞}
A −→ γ(A)

where
γ(A) = min{k ∈ IN | ∃ϕ ∈ C(A, IRk − {0}), ϕ(x) = −ϕ(−x)}. (B.2)
If the minimum does not exist, then we define γ(A) = +∞.

Notice that if C ⊂ X is such that C ∩ (−C) = ∅ and we define A by A = C ∪ (−C) then

A ∈ Σ and moreover γ(A) = 1. It is sufficient define φ(x) = 1 if x ∈ C and φ(x) = −1 if x ∈ −C.
In this way φ ∈ C(A, IR − {0}).
As a consequence we have that if A1 ∈ Σ and γ(A1 ) > 1 then A1 contains infinitely many
different points.
This elementary remark shows how to use the genus.
In general to calculate the exact genus of a set is a difficult task. Often it is sufficient to
have some estimates which can be obtained by comparison with sets whose genus is known, for
instance, with respect to spheres. In this way the following results are very useful.

85
86

P
Lemma B.0.10 Let A, B ∈ . Then:

1. If there exists f ∈ C (A, B), odd, then γ(A) ≤ γ(B).

2. If A ⊂ B , then γ(A) ≤ γ(B).

3. If there exists an odd homeomorphism between A and B, then γ(A) = γ(B).

4. If S N −1 is the sphere in IRN , then γ(S N −1 ) = N .

5. γ(A ∪ B) ≤ γ(A) + γ(B).

6. If γ(B) < +∞, then γ(A − B) ≥ γ(A) − γ(B).

7. If A is compact, then γ(A) < +∞, and there exists δ > 0 such that γ(A) = γ(Nδ (A))
where Nδ (A) = {x ∈ X : d(x, A) ≤ δ}.

8. If X0 is a subspace of X with codimension K , and γ(A) > K, then A ∩ X0 6= ∅.

For the proof and more details on this topic we refer to [76] and the references therein.
Appendix C

The variational principle of Ekeland

The idea of the ε-variational principle of Ekeland is the following: Assume f lower semicontinuous
real valued function defined on the metric space (M, d) such that f (x) ≥ β for all x ∈ M.
The principle deals with the construction of minimizing sequences with some control, precisely,
sequences verifying
inf {f (x)} + ε > f (xε ),
x∈M

and
f (y) ≥ f (xε ) − εd(xε , y).
We can read geometrically this conditions saying that for all ε > 0 the graph of f is above of
this cone. See the original paper [45]; we follows here the proof in [67].

Theorem C.0.11 Let M be a complete metric space and let

φ : M → (−∞, ∞]

a proper function such that

i) φ(y) ≥ β,

ii) φ lower semicontinuous.

Given ε > 0 and u ∈ M such that

φ(u) ≤ inf φ + ε,
M

then there exists v ∈ M such that:

1. φ(u) ≥ φ(v),

87
88

2. d(u, v) ≤ 1,

3. If v 6= w ∈ M then φ(w) ≥ φ(v) − εd(v, w).

Proof. Fixed ε > 0, we define the following order relation on M, we say

w ≤ v, if and only if φ(w) + εd(w, v) ≤ φ(v).

Consider u0 = u and by recurrence define the sequence {un }, as follows: for n ∈ IN we take

Sn = {w ∈ M : w ≤ un },

choosing un+1 ∈ Sn such that

1
φ(un+1 ) ≤ inf {φ} +
,
Sn n+1
we get that un+1 ≤ un and Sn+1 ⊂ Sn . The lower semicontinuity of φ implies that Sn is a closed
set. Now, if w ∈ Sn+1 , we have w ≤ un+1 ≤ un and then,
1 1
εd(w, un+1 ) ≤ φ(un+1 ) − φ(w) ≤ inf {φ} + − inf {φ} = .
Sn n + 1 Sn n+1
2
Namely, if we call diameter(Sn+1 ) = δn+1 , (δn+1 ) ≤ , so, limn→∞ δn+1 = 0. The com-
ε(n + 1)
pleteness of M implies ∩∞ n=1 Sn = {v} for some v ∈ M. But, in particular, v ∈ S0 , hence
v ≤ u0 = u, i.e., φ(v) ≤ φ(u) + εd(u, v) ≤ φ(u) and
φ(u) − φ(v)
d(u, v) ≤ ≤ ε−1 (inf {φ} + ε − inf {φ}) = 1.
ε M M

Then, d(u, v) ≤ 1.
To get 3), suppose that w ≤ v. Then for all n ∈ IN , w ≤ un , i.e.,

w ∈ ∩∞
n=1 Sn

and then w = v.
So we conclude that if w 6= v then φ(w) ≥ φ(v) − εd(v, w).
The results that we explain below show how to use the variational principle to find critical
points of functionals.

Corollary C.0.12 Let X be a Banach space, and ϕ : X → IR a differentiable lower bounded

function in X . Then for all ε > 0 and for all u ∈ X such that

ϕ(u) ≤ inf ϕ + ε
X
 89

there exists v ∈ X verifying

ϕ(v) ≤ ϕ(u),
ku − vkX ≤ ε1/2 ,
kϕ′ (v)kX ′ ≤ ε1/2 .

1
Proof. In Theorem C.0.11 take M = X , φ = ϕ, ε > 0, λ = , d = k · k. We obtain v ∈ X
ε1/2
such that
ϕ(v) ≤ ϕ(u)
ku − vkX ≤ ε1/2
and for all w 6= v
ϕ(w) > ϕ(v) − ε1/2 k(v − w)k.
Taking in particular w = v + th with t > 0 and h ∈ X , khk = 1, then

ϕ(v + th) − ϕ(v) > −ε1/2 t

that implies −ε1/2 ≤ hϕ′ (v), hi for all h ∈ X , khk = 1, hence

kϕ′ (v)kX ′ ≤ ε1/2 .

Corollary C.0.13 If X and ϕ are as in Corollary C.0.12. Then, for all minimizing sequence
of ϕ, {uk } ⊂ X there exists a minimizing sequence {vk } ⊂ X such that

ϕ(vk ) ≤ ϕ(uk )
kuk − vk kX → 0 k → ∞
kϕ′ (vk )kX ′ → 0 k → ∞

1
Proof. If ϕ(uk ) → c = inf X ϕ consider εk = ϕ(uk ) − c if is positive and εk = if ϕ(uk ) = c.
k
Then for εk we take vk that gives the Corollary C.0.12.
In the sense of the previous Corollary we find almost minimum for ϕ. The problem now
is to get almost critical points of different types, for instance mountain pass type critical points
in the sense of [13]. Here we point out how the variational principle of Ekeland separates the
geometric aspect of the problem of finding critical points of functional from the analytical one,
namely, we will need some compactness property to complete the search of critical points.
90
Appendix D

The Mountain Pass Theorem

Seeking by the completeness of the exposition we will give a proof of the classical Mountain Pass
Theorem by Ambrosetti and Rabinowitz. (See [13]).
We will precise the notation as follows,
H1 X is a Banach space, K ⊂ IRN a compact metric space, and K0 ⊂ K a closed set.
H2
χ : K0 −→ X
is a continuous function.
H3 Define
M = {g ∈ C(K, X ) : g(s) = χ(s), s ∈ K0 }.
We recall the definition of subdifferential in the case of the norm in X ,

∂kx0 k = {p ∈ X ′ : kx0 k − kxk ≤ p(x0 − x), ∀x ∈ X }

(X ′ dual space of X .
We will use the following abstract result.

Lemma D.0.14 Let X be a Banach space, then for the norm

k · k : X → IR

we have
∂kx0 k = {p ∈ X ′ : p(x0 ) = kx0 k, kpkX ′ = 1}

Proof. Given x0 ∈ X by Hahn-Banach theorem there exists p ∈ X ′ such that p(x0 ) = kx0 k
and kpkX ′ = 1. For such p we have

p(x − x0 ) = p(x) − p(x0 ) ≤ kpkX ′ kxk − kx0 k ≤ kxk − kx0 k,

91
92

hence kx0 k − kxk ≤ p(x0 − x), namely, p ∈ ∂kx0 k. Reciprocally if p ∈ ∂kx0 k, we get

kx0 k − kxk ≤ p(x0 − x) ∀x ∈ X .

And in particular for x = λx0 , λ > 0, we find kx0 k(1−λ) ≤ (1−λ)p(x0 ), so (p(x0 )−kx0 k)(1−λ) ≥
0 for all λ ∈ IR+ . Then p(x0 ) = kx0 k. Moreover kx0 k − kxk ≤ p(x0 ) − p(x) ≤ kx0 k − p(x) implies
p(x) < kxk and then kpkX ′ ≤ 1 since p(x0 ) = kx0 k and kpkX ′ = 1.

Corollary D.0.15 Let C(K, IR) with the supremum norm. Denote by M(K) the Radon measures
on K. Then
Z
∂kf k∞ = {µ ∈ M(K) : µ ≥ 0, dµ = 1, supp(µ) ⊂ {t : f (t) = kf k∞ }}.
K

Proof. It is clear that the Radon measures verifying the conditions above are the subdifferential
of the norm. More precisely, by Lemma D.0.14 we have

i) If the conditions are satisfied then kµk = 1 and

Z Z
f (x)dµ = kf k∞ dµ = kf k∞ ,
K suppµ

R
ii) If kµk = 1 and K f (x)dµ = kf k∞ then µ > 0, and

supp(µ) ⊂ {s| f (s) = ||f ||∞ }.

Theorem D.0.16 Consider X , K, K0 and M being as in H1 − H2 − H3. Let u : X → IR a

functional verifying:

1) u is continuous,

2) u is Gateaux diferentiable and, moreover,

u′ : X −→ X ′

is continuous from the strong to weak-∗ topologies.

 93

consider (
α = inf ϕ∈M maxs∈K u(ϕ(s))
α1 = maxχ(K0 ) u,
and assume that α > α1 . Then for all ε > 0 and for all ϕ ∈ M such that maxs∈K u(ϕ(s)) ≤ α+ε
there exists vε ∈ X such that
α − ε ≤ u(vε ) ≤ maxs∈K u(ϕ(s))
d(vε , ϕ(K)) ≤ ε1/2
ku′ (vε )k ≤ ε1/2 .
Proof. We can assume that 0 < ε < α − α1 . Let ϕ ∈ M verifiying
max u(ϕ(s)) ≤ α + ε.
s∈K

Define
I : M → IR, by I(c) = max u(c(s))
s∈K
for c ∈ M. By hypothesis α = inf c∈M I(c) > α1 . Moreover I is lower semicontinuous, because if
kck − ck∞ → 0 we have
I(c) = u(c(s0 )) = lim u(ck (s0 )) ≤ lim I(ck ).
k→∞ k→∞

By the Ekeland variational principle applied to I : M → IR, given ε > 0 and ϕ ∈ M such that
I(ϕ) ≤ α + ε there exists cε ∈ M such that
(i) I(cε ) ≤ I(ϕ) ≤ α + ε.
(ii) I(c) ≥ I(cε ) − ε1/2 kc − cε k∞ if c ∈ M.
(iii) kcε − ϕk∞ ≤ ε1/2 .
Consider γ : K → X such that γ(K0 ) = 0. For h 6= 0, h ∈ IR from (ii) we conclude that
I(cε + γh) − I(cε ) ≥ −ε1/2 |h|kγk∞ , namely
1
I(cε + γh) − I(cε ) ≥ −ε1/2 kγk∞ .
|h|
By definition of I,
I(cε + γh) − I(cε ) = max u(cε + γh) − max u(cε )
= maxs∈K {u(cε (s)) + hhu′ (cε (s)), γ(s)i + o(h)} − maxs∈K u(cε (s)).
To simplify the printing we write
f (s) = u(cε (s)) and g(s) = hu′ (cε (s)), γ(s)i.
We have by hypothesis that f ∈ C(K, IR); also g ∈ C(K, IR) because,
94

a) By Banach-Alaoglu theorem ∂kf k∞ is weak-∗ compact.

b) By hypothesis if vn → v en X then

hu′ (vn ), yi → hu′ (v), yi

for all y ∈ X .

c) If u′ (vn ) ⇀ u′ (v) weak-∗ in X ′ and yn → y in X then

hu′ (vn ), yn i → hu′ (v), yi, as n → ∞.

Then the inequalities above can be written as

1
−ε1/2 kγk∞ ≤ limh→0 {I(cε + hγ) − I(cε )}
h
1
= limh→0 {k(f + hg)k∞ − kf k∞ }
R h
≤ sup{ 01 gdµ : µ ∈ ∂kf k∞ }.
R ∗
Define F : ∂kf k∞ → IR by F (µ) = K gdµ and consider a weak-∗ convergent sequence µn ⇀ µ
in ∂kf k∞ , then Z Z
F (µn ) = gdµn → gdµ = F (µ),
K K
so, the supremum is attained. As a consequence we have
Z
1/2
−e kγk∞ ≤ max{ {u′ (cε (s)), γ(s)}dµ : µ ∈ ∂kf k∞ }.
K

Consider Γ = {γ ∈ C(K, X ) : γ(s) = 0 if s ∈ K0 , kγk∞ ≤ 1}. Then,

Z
−ε1/2 ≤ −ε1/2 kγk∞ ≤ inf {max{ {u′ (cε (s)), γ(s)}dµ : µ ∈ ∂kf k∞ }}.
γ∈Γ K

Take {µn (γ)} ⊂ ∂kf k∞ such that fixed γ,

Z
{u′ (cε (s)), γ(s)}dµn (γ) −→ M (γ), n → ∞
K

where Z
M (γ) = max { {u′ (cε (s)), γ(s)}dµ},
u∈∂kf k∞ K

Then by choosing a convenient weak-∗ convergent subsequence µn (γ) ⇀ µ(γ) we have

Z
{u′ (cε (s)), γ(s)}dµ(γ) = M (γ).
K
 95

Then by Lemma D.0.15,

Z
1/2
−ε ≤ inf M (γ) = inf {u′ (cε (s)), γ(s)}dµ(γ) ≤
γ∈Γ γ∈Γ K
Z Z
′
max inf {u (cε (s)), γ(s)}dµ = max (− ku′ (cε (s))kX ′ dµ) =
µ∈∂kf k∞ γ∈Γ K µ∈∂kf k∞ K

− min{ku′ (cε (s))kX ′ : s ∈ {t ∈ K : u(cε (t)) = kf k∞ }}

Namely
min ku′ (cε (s))kX ′ ≤ ε1/2
s∈{t∈K:u(cε (t))=kf k∞ }

In other words, there exists sε ∈ K such that, if vε = cε (sε ), then:

i) u(vε ) = maxs∈K u(cε (s)) = I(cε ) ≤ inf I + ε1/2 ,

ii) ku′ (vε )k ≤ ε1/2 ,

iii) u(vε ) ≥ α1 ,

iv) d(vε , ϕ(K)) ≤ ε1/2 .

The classical condition of compactness is the condition of Palais-Smale that we formalize

in the following definition.

Definition D.0.17 Let X be a Banach space and U : X −→ IR a functional Gateaux diferen-

tiable. U verifies the Palais-Smale condition to the level c ∈ IR if and only if for all sequence
{xk } ⊂ X verifying

i) U (xk ) → c as k → ∞,

ii) U ′ (xk ) → 0 in X ′ for k → ∞

there exists a subsequence xkj → x in X .

A sequence verifying i) and ii) is called a Palais-Smale sequence for U .

With this notions we can formulate the Montain Pass Theorem.

Theorem D.0.18 Let X be a Banach space and U : X → IR a continuous functional verifiying:

i) U is Gateaux differentiable,
96

ii) U ′ is continuous from X to X ′ with the weak-∗ topology.

Assume that there exists u0 , u1 ∈ X , and a ball Br , centered in u0 and with radius r such that
if u1 ∈ X − Br and if |x − u0 | = r then

inf U (x) > max{U (u0 ), U (u1 )}

∂Br

Consider
Γ = {g ∈ C([0, 1], X ) : g(0) = u0 , g(1) = u1 }
and
c = inf sup U (f (s)).
f ∈Γ s∈[0,1]

If U satisfies the Palais-Smale condition at the level c, then c is a critical value for U and
c > max{U (u0 ), U (u1 )}.

Proof. Take K = [0, 1], K0 = {0, 1}, χ(0) = u0 , χ(1) = u1 and M = Γ; since c0 =
inf ∂Br U (x) > max{U (u0 ), U (u1 )} = c1 and because all path g ∈ Γ has nonempty intersec-
tion with ∂Br , Theorem D.0.16 implies that for a given sequence {εk } with εk → 0 we can find
a sequence {xk } ⊂ X such that

a) U (xk ) → c,

b) U ′ (xk ) → 0,

and then by the Palais-Smale condition for the level c, there exists a subsequence such that:

c = lim U (xk ) = U (x),

k→∞

0 = lim U ′ (xk ) = U ′ (x).

k→∞

Namely, x is a critical point at level c.

Appendix E

Boundedness of solutions

The boundedness of the solutions in the case p ≥ N is a consequence of Morrey’s theorem and
the classical bootstrapping.
One of the few general results for problems with critical growth is the following one on C 1,α
regularity. We will concentrate on the following problem
(
−∆p u ≡ −div (|∇u|p−2 ∇u) = f (x, u) in Ω,
(E.1)
u|δΩ = 0,

where Ω ⊂ IRN is a smooth bounded domain, 1 < p < N and f satisfies

Np
|f (x, u)| ≤ C(1 + |u|r ), with r + 1 ≤ p∗ = , (E.2)
N −p
that is the critical Sobolev exponent. The regularity of the solution of E.1 for r < p∗ − 1 is a
consequence of the results in the paper by Serrin [79] about the L∞ estimates and the results
by Di Benedetto, [44] and Tolksdorf, [87], for the C 1,α regularity. The case r = p∗ − 1 is much
more delicate and will be obtained below, while for the supercritical case the result is not true:
in general in the supercritical case a weak solution is not bounded.
The idea of the L∞ estimate can be found in the argument used by Trudinger in [88] for
Yamabe’s Problem. The main point is to use some nonlinear test functions in the line of the
classical Moser method.
If we have that the solutions of E.1 are bounded then the regularity C 1,α is a consequence
of the results in [44] or [87]. I would like to remark that a different (Schauder) aproach to the
regularity of problem E.1, can be seen in Guedda-Véron, [58].

The first result in this section is the following theorem which comes from Stampacchia
classical work [81].

97
98

Theorem E.0.19 Let u ∈ W01,p (Ω) be the weak solution of

(
N
−∆p u = g, in Ω, g ∈ W −1,q (Ω), q > p−1 (E.3)
u|∂Ω = 0,

then u ∈ L∞ (Ω).

Proof. We can write g = div F , F = (f1 , ..., fN ) where fi ∈ Lq (Ω). Since u is the weak solution
to (E.3) we can write
Z Z
|∇u| p−2
h∇u, ∇vidx = hF, ∇vi, ∀v ∈ W01,p (Ω) (E.4)
Ω Ω

For k > 0 we consider the test function



 u − k if u > k
v = sign(u)(|u| − k) = 0 if u = k (E.5)

 u + k if u < k

then u = v + ksign(u) and uxi = vxi in A(k) = {x ∈ Ω | |u(x)| > k}, v = 0 in Ω − A(k) and
v ∈ W01,p (Ω). For this election of test function (E.4) becomes
Z Z
p
|∇v| dx = hF, ∇vidx,
A(k)
Ω

and by Hölder inequality

Z Z !1/q Z !1/p
q p 1−( p1 + 1q )
hF, ∇vidx ≤ |f | |∇v| |A(k)| ,
A(k) A(k)
Ω

here |C| denotes the Lebesgue measure of the measurable set C. Then we have

Z !1−1/p Z !1/q
p q
|∇v| dx ≤ |F | |A(k)|1−(1/p+1/q)
A(k) A(k)

and by Sobolev inequality we obtain

Z !p/p∗ Z !1/q
p∗ q
S |v| dx ≤ |F | |A(k)|1−(1/p+1/q) . (E.6)
A(k) A(k)
 99

Notice that for 0 < k < h, A(h) ⊂ A(k) and then

∗
R ∗
1/p∗
|A(h)|1/p (h − k) = p
A(h) (h − k) dx ≤
R 1/p∗ R 1/p∗ (E.7)
p∗ p∗
A(h) |v| dx ≤ A(k) |v| dx .

From (E.6) and (E.7) we obtain

p∗ /(p−1)
||F ||q 1 p∗ ( p1 − (p−1)q
1
|A(h)| ≤ |A(k)| .
S p∗ /p (h − k)p ∗

Because the hypothesis q > N/(p − 1),

1 1
p∗ ( − ) > 1.
p q(p − 1)
Now we have the following
Claim: Assume φ : [0, ∞) −→ [0, ∞) is a nonincreasing function such that if h > k > k0 ,
for some α > 0, β > 1,
c
φ(h) ≤ [φ(k)]β .
(h − k)α
αβ
Then φ(k0 + d) = 0, where dα = c2 β−1 [φ(k0 )]β−1 .

In our case φ(h) = |A(h)|, α = p∗ and

1 1
β = p∗ ( − ) > 1.
p q(p − 1)
We have φ(0) = |Ω|. By the Claim we get that φ(d) = 0 for
1/(p−1)
||F ||q ( p1 − q(p−1)
1
)− p1∗
d=c |Ω| ,
S 1/p
namely
1/(p−1)
||F ||q ( p1 − q(p−1)
1
)− p1∗
||u||∞ ≤ c |Ω| .
S 1/p
To finish we need to prove the Claim.
d
Proof of the claim: Given d as above, define dn = d0 + d − 2n
By recurrence we have that
φ(k0 ) α
φ(kn ) ≤ −nµ
, µ= ,
2 1−β
100

then
0 = lim φ(kk ) ≥ φ(k0 + d) ≥ 0,
n→∞
as we want to prove.
The second result in this section is the following adaptation of the announced result by
Trudinger, [88].

Theorem E.0.20 Let u ∈ W01,p (Ω) be a solution of E.1. If f verifies E.2, then u ∈ L∞ (Ω).

Proof. We assume r + 1 = p∗ . We can assume also that u ≥ 0, in other case we can do the
argument for the positive and negative parts of u by a minor change in the test function. We
define for each l > 0 and β > 1,
(
uβ , if u ≤ l
F (u) =
βlβ−1 (u − l) + lβ , if u > l

and (
u(β−1)p+1 , if u ≤ l
G(u) =
β((β − 1)p + 1)l(β−1)p (u − l) + l(β−1)p+1 , if u > l.
It is easy to show the following properties
i) G(u) ≤ uG′ (u).
ii) For some constant independent of l, c[F ′ (u)]p ≤ G′ (u).
iii) up−1 G(u) ≤ C[F (u)]p with C independent of l.
iv) If u ∈ W01,p (Ω), then F (u), G(u) ∈ W01,p (Ω) because F and G are Lipschitz functions.
We choose β > 1 such that βp < p∗ . For this choice, consider the test function

ξ = η p G(u),

where η ∈ C0∞ will be fixed below.

Then the equation gives us
Z Z
p−2 p
|∇u| h∇u, ∇(η G(u))idx = f (x, u)η p G(u)dx.
Ω Ω

For ε > 0 and by direct computation on the left hand side, the previous identity becomes
R
|∇u|p η p G′ (u)dx ≤
ΩR R R
ε |∇u|p η p G′ (u)dx + C |∇η|p (up−1 G(u))dx + f (x, u)η p G(u)dx,
Ω Ω Ω
 101

where we use the property i) above, Hölder and Young inequalities. Then, by choosing ε small
enough, R
|∇u|p η p G′ (u)dx ≤
Ω R R ∗ −p
≤ C1 |∇η|p [F (u)]p dx + C2 η p up [F (u)]p dx + C3 |Ω|,
Ω Ω
∗
where we use that G(u)f (x, u) ≤ C3 (1 + up −1 G(u)) (for l > 1), motivated by the hypothesis
E.2, and the property iii).
The left hand side can be estimated by using the property ii), i.e., G′ ≥ c[F ′ ]p ,
Z Z
|∇u|p η p G′ (u)dx ≥ c |ηF ′ (u)∇u|p dx,
Ω Ω

but then, R
|∇(ηF (u))|p dx ≤
Ω R R ∗ −p
≤ C4 |∇η|p [F (u)]p dx + C5 η p up [F (u)]p dx + C6 |Ω|.
Ω Ω
By the Sobolev inequality we get
! p
p∗
R ∗ ∗
[F (u)]p η p dx ≤
ΩR R ∗ −p
C4 |∇η|p [F (u)]p dx + C5 η p up [F (u)]p dx + C6 |Ω|.
Ω Ω

Given xo ∈ Ω, we choose η ∈ C0∞ such that if supp (η) = B(xo , R), then
p −p∗ 1
||u||Lp∗ (B(x ≤ .
o ,R)) 2C5
Then, by Hölder inequality,
 p/p∗
Z Z
∗ 1 ∗ ∗
C5 η p up −p [F (u)]p dx ≤  η p [F (u)]p dx .
2
Ω Ω

Finally we get the inequality

Z Z
p∗ p∗
η [F (u)] dx ≤ C5 |∇η|p [F (u)]p dx + C6 |Ω|.
Ω Ω

Taking limits for l → ∞ we conclude

 p/p∗
Z Z
p∗ βp∗
 η u dx ≤ C(η) uβp dx + C6 |Ω|
Ω Ω
102

∗
and, because βp < p∗ , we have u ∈ Lβploc .
Then, u is a solution of the equation

−∆p u = a(x)up−1 + b(x),

where 
 0 if u < 1
a(x) = f (x, u)
 if u ≥ 1
up−1
and (
0 if u > 1
b(x) =
f (x, u) if u ≤ 1.
N
Therefore, a(x) ∈ Lr , with r > , and b ∈ L∞ . We conclude, by Theorem E.0.19, that u ∈ L∞ ,
p
and moreover, u ∈ C α by Theorem 1 in [79].
Appendix F

Two Lemmas by P.L. Lions

By the sake of completness we will include in this appendix the proof of two lemmas by P.L.
Lions. (See [64] and [65]). These results have been used in section 2.3. The first one is a real
variable result that allows us to get precise representation of measures related by a reverse
Hölder inequality.

Lemma F.0.21 Let µ, ν be two non-negative and bounded measures on Ω, such that for 1 ≤
p < r < ∞ there exists some constant C > 0 such that,
Z Z
1 1
( |ϕ|r dν) r ≤ C( |ϕ|p dµ) p ∀ϕ ∈ C0∞ . (F.1)
Ω Ω

Then, there exist {xj }j∈J ⊂ Ω and {νj }j∈I ⊂ (0, ∞), where I is finite, such that:
X X p
ν= νj δ x j , µ ≥ C −p νjr δxj , (F.2)
j∈I j∈I

where δxj is the Dirac mass at xj .

Proof. By the reverse Hölder inequality (F.1), the measure ν is absolutely continuous with
respect to µ. As a consequence there exists f ∈ L1 (Ω), f ≥ 0, such that ν = f µ. Also by (F.1)
we have,
ν(A) ≤ cq0 (µ(A))q/p ,
for any borelian set A ⊂ Ω. In particular, f ∈ L∞ (Ω).
On the other hand the Lebesgue decomposition of µ with respect to ν gives us

µ = gν + σ, where g ∈ L1 (dν), g ≥ 0

103
104

and σ is a bounded positive measure, singular respect to ν, namely, if K is the support of σ

then ν(K) = 0.
Now consider (F.1) applied to the test function

φ = g 1/(q−p) χ{g≤n} ψ

we obtain
R 1 R 1
( |ϕ|r dν) r = ( g r/(r−p) |ψ|r χ{g≤n} dν) r
Ω Ω
R 1 R 1 (F.3)
≤ C( g 1+p/(r−p) |ψ|p χ{g≤n} dν) p ≤ C( g r/(r−p) |ψ|p χ{g≤n} dν) p .
Ω Ω

Hence calling dνn = g r/(r−p) χ{g≤n} dν the following reverse Hölder inequality holds
Z Z
1 1
( |ψ|r dνn ) r ≤ C( |ψ|p dνn ) p ,
Ω Ω

and in particular for each borelian A ⊂ Ω we have

(νn (A))1/r ≤ C(νn (A))1/p .

Taking into account that p < r then either νn (A) = 0 or

−( p1 − r1 )
νn (A) ≥ c = δ > 0.

As a consequence, given a point x ∈ Ω, either νn ({x}) = 0, or νn ({x}) ≥ δ > 0 and this means
that νn is a linear combination of Dirac masses, that necessarily must be finite because νn is
bounded. Taking limits as n → ∞ we conclude that
X
ν= νj δ x j .
j∈I

Moreover the inequality (F.1) applied in each xj gives,

X p
µ ≥ C −p νjr δxj .
j∈I

The second result is, roughly speaking, a description of how the lack of compactness happens
in the Sobolev inclusion. More precisely we have.
 105

Lemma F.0.22 Let {uj } be a weakly convergent sequence in W01,p (Ω) with weak limit u, and
such that:

i) |∇uj |p → µ weakly-∗ in the sense of measures.

∗
ii) |uj |p → ν weakly-∗ in the sense of the measures.

Then, for some finite index set I we have:

 ∗ P

 1) ν = |u|p + j∈I νj δxj , νj > 0
 P
2) µ ≥ |∇u|p + j∈I µj δxj , µj > 0 , xj ∈ Ω
p

 ∗ µj
 3) νjp ≤ .
S
Proof. Given any φ ∈ C0∞ (Ω), by Sobolev inequality we have,
  1∗  1
Z p Z p
∗ ∗ 1
 |φ|p |un |p dx S p ≤  |∇(φun )|p dx . (F.4)
Ω Ω

We write vn = un − u and by the convergence results in [28], we have,

Z Z Z
∗ ∗ ∗ ∗ ∗ ∗
|φ|p |un |p dx − |φ|p |vn |p dx → |φ|p |u|p dx, n → ∞.
Ω Ω Ω

Moreover vn ⇀ 0 as n → ∞ then applying the Sobolev inequality to φvn and taking limits we
arrive to a reverse Hölder inequality as in Lemma F.0.21, so we have the representation
∗ X
ν = |u|p + νj δ x j .
j∈I

Now by (F.4) and taking into account that un → u in Lp , we obtain,

  1∗  1  1
Z p Z p Z p
∗ 1
 |φ|p dν  S p ≤  |φ|p dµ +  |∇φ|p |u|p dx , (F.5)
Ω Ω Ω

for all φ ∈ C0∞ (Ω). Consider a φ ∈ C0∞ (Ω) such that 0 ≤ φ ≤ 1, φ(0) = 1 and supported in the
x − xj
unit ball B ⊂ IRN . Fixed j ∈ I and ε > 0, we consider φ( ). For ε small enough we have,
ε
Z !1
p
1
p∗
1 1
−p xj − x p p
νj S ≤ µ(B(xj , ε)) +
p p ε |∇φ( )| |u| dx
B(xj ,ε) ε
106

and by Hölder inequality,

!1 ! 1  1
Z p Z p∗ Z N
xj − x p p ∗ xj − x N 
ε−p |∇φ( )| |u| dx ≤ |u|p dx ε −1 
|∇φ( )| dx ,
B(xj ,ε) ε B(xj ,ε) ε
Ω

then ! p
1 Z p∗
1 1
p∗ p∗
νj S ≤ µ(B(xj , ε)) + c
p p |u| dx
B(xj ,ε)

and taking limit for ε → 0 we obtain.

p
∗
µj ≥ νjp S, j ∈ I,

that can be read in an equivalent way as

X p
∗
µ≥ νjp Sδxj ≡ µ1 .
j∈I

But because un ⇀ u in W01,p (Ω), µ ≥ |∇u|p and |∇u|p is orthogonal to µ1 , we conclude.

Bibliography

[1] Adimurthi, S.L. Yadava, An Elementary Proof of the Uniqueness of Positive Radial
Solutions of a Quasilinear Dirichlet Problem. Arch. Rational Mech. Anal. 127 (1994),
219-229.

[2] J. A. Aguilar, I. Peral, On an elliptic equation with exponential growth Rend. Semi.
Mat. Univ. Padova Vol 96 (1996), 143-175.

[3] J. A. Aguilar, I. Peral, Positive radial solutions of quasilinear equations involving su-
percritical growth Preprint U.A.M. 1997.

[4] H. Amann, Lusternik-Schnirelman Theory and Nonlinear eigenvalue problems Math.

Ann. 199 (1972), 55-72.

[5] H. Amann, Fixed point equations and nonlinear eigenvalue problems in ordered Banach
spaces SIAM Review, Vol. 18 No 4 (1976), 620-709.

[6] A. Ambrosetti, Critical points and nonlinear variational problems Memoire No 49,
Societe Mathematique de France Tome 120, Fascicule 2, 1992.

[7] A. Ambrosetti, J. Garcia , I. Peral, Multiplicity of solutions for semilinear and quasi-
linear elliptic problems, Jour. Funct. Anal. 137 (1996), 219-242.

[8] A. Ambrosetti, J. Garcia, I. Peral, Quasilinear Equations with a Multiple Bifurcation

To appear in Diff. and Int. Equations.

[9] A. Ambrosetti, P. Hess, Positive Solutions of Asymptotically Linear Elliptic Problems

Jou. Math. Anal. and Applic. 73, No 2 (1980), 411-422.

[10] A. Ambrosetti, D. Arcoya, B. Buffoni, Positive solutions for some semi-positone prob-
lems via bifurcation theory Differential and Integral Equations, Vol 7 No 3. (1994),
655-663.

107
108

[11] A. Ambrosetti, H. Brezis, G. Cerami, Combined Effects of Concave and Convex Non-
linearities in some Elliptic Problems Journal of Functional Anal. 122, No 2 (1994),
519-543.

[12] A. Ambrosetti, G. Prodi, A primer of Nonlinear Analysis Cambridge Studies in Ad-

vanced Mathematics No 34, Cambridge University Press, 1995.

[13] A. Ambrosetti, P.H. Rabinowitz, Dual variational methods in critical point theory and
applications. Jour. Funct. Anal. 14 (1973), 349-381.
∗ −1
[14] A. Ambrosetti, M. Struwe, A note on the problem −∆u = λu+u|u|2 Manusc. Math.
54 (1986), 373-379.

[15] A. Anane, Simplicité et isolation de la première valeur propre du p-laplacien avec poids
C.R. Acad. Sci. Paris. t. 305, Série I. (1987), 725-728.

[16] G. Arioli, F. Gazzola, Some results on p-Laplace equations with a critical growth term
Research Report in Mathematics No 14, 1996. Dept. Math. Stockholm University.

[17] J.P. Aubin, I. Ekeland, Applied Nonlinear Analysis Ed. John Wiley, 1984.

[18] P. Baras, J. Goldstein, The heat equation with a singular potential Trans. Amer. Math.
Soc Vol 294 (1984), 121-139.

[19] 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. Vol XLI
(1988), 253-294.

[20] T. Bartsch, M. Willem, On an elliptic equation with concave and convex nonlinearities
Procee. A.M.S. 123 (1995), 3555-3561.

[21] V. Benci, D. Fortunato, Bifurcation from the essential spectrum for odd variational
operators Confer. Sem. Mat. Univ. Bari 178 (1981).

[22] M. F. Bidaut-Veron Local and global behaviour of solutions of quasilinear elliptic equa-
tions of Emden-Fowler type Arch. Rat. Mech. Anal. 107 (1989), 293-324.

[23] L. Boccardo, T. Gallouët, Nonlinear Elliptic Equations with right hand side measures,
Commun. in P.D.E. Vol 17 No 3,4 (1992), 641-655.

[24] L. Boccardo, M. Escobedo, I. Peral, A Dirichlet problem involving critical exponent

Nonlinear Analysis T.M.A. Vol 24 (11)(1995), 1639-1648.
 109

[25] L. Boccardo, F. Murat, Almost Everywhere Convergence of the Gradients of Solutions

to Elliptic and Parabolic Equations, Nonlinear Analysis T.M.A. Vol 19 No 6 (1992),
581-597.

[26] H. Brezis, Analyse Fonctionelle. Théorie et Applications Ed. Masson, 1983.

[27] H. Brézis, Some variationals problems with lack of compactness Proc. Berkeley Sym.
Nonlinear Functional Analysis, 1985.

[28] H. Brézis, E. Lieb. A relation between pointwise convergence of functions and conver-
gence of functionals Proc. American Math. Soc. 88 (1983), 486-490.

[29] H. Brezis, L. Nirenberg, Positive solutions of nonlinear elliptic equations involving

critical Sobolev exponents Comm. Pure App. Math. 36 (1983), 437-477.

[30] H. Brezis, L. Nirenberg, H 1 versus C 1 local minimizers C.R.A.S. Paris t.317 (1993),
465-472.

[31] F.E. Browder, W.F. Petryshyn, Approximation methods and the generalized topological
degree for nonlinear mapping in Banach spaces, Jou. Funct. Anal., 3 (1969), 217-245.

[32] S. Chandrasekar, An introduction to the study of stellar structure Ed. Dover Publ. Inc.
(1985).

[33] X. Cabre, P. Majer, Truncation of nonlinearities in some supercritical elliptic problems

Preprint 1996.

[34] B. Dacorogna, Direct Methods in the Calculus of Variations, Springer-Verlag, 1989.

[35] L. Damascelli, Comparison Theorems for some Quasilinear Degenerate Elliptic Opera-
tors and Aplications to Symmetry and Monotonicity Results To appear in Annal. I.H.P.
Anal. Nonlineaire.

[36] E.N. Dancer, A Note on an Equation with Critical Exponent Bull. London Math. Soc.
Vol 20 (1988), 600-602.

[37] E. De Giorgi, Teoremi di semicontinuita nel calcolo delle variazioni Ins. Naz. di Alta
Matematica, Roma 1968.

[38] D.G. De Figueiredo, P.L. Lions, R. Nussbaum, A priori estimates and existence of
positive solutions of semilinear elliptic equations J. Math. Pures et Appl., 61 (1982),
41-63.
110

[39] D.G. De Figueiredo, On the existence of multiple ordered solutions of nonlinear eigen-
value problems Nonlinear Analysis T.M.A. 11 (1987), 481-492.
N +2
[40] W.Y. Ding, Positive solutions of −∆u = u N −2 on contractible domains Jou. P.D.E.
Vol 2 No. 4 (1989), 83-88.

[41] I. Dı́az , J.E. Saa, Existence et unicité de solutions positives pour certaines équations
elliptiques quasilinéaires C.R.A.S. de Paris t. 305, Serie I (1987), 521-524.

[42] M. Del Pino, M. Elgueta, R. Manasevich, A Homotopic Deformation along p of a Leray-

Schauder Degree Result and existence for (|u′ |p−2 u′ ) + f (t, u) = 0, u(0) = u(T ) = 0
Journal of Differential Equations, 80 (1989), 1-13.

[43] M. Del Pino, R. Manasevich, Global Bifurcation from the Eigenvalues of the p-Laplacian
Journal of Differential Equations, 92 (1991), 226-251.

[44] E. Di Benedetto, C 1,α local regularity of weak solutions of degenerate elliptic equations
Nonlinear Analysis. Theory, Methods and Applications 7, No 8 (1983), 827-850.

[45] I. Ekeland, On the variational principle J. Math. Anal.Appl. Vol 47 (1974), 324-353.

[46] C. L. Fefferman, The uncertainty principle Bull A.M.S. Vol 9, No 2 (1983), 129-206.

[47] B. Franchi, E. Lanconelli, J. Serrin, Existence and uniqueness of nonnegative solutions

of quasilinear equations in IRN (to appear in Advances in Mathematics)

[48] D.A. Frank-Kamenetskii, Diffusion and heat transfer in chemical kinetics Ed. Plenum
Press, 1969.

[49] H. Fujita, On the nonlinear equations ∆u + eu = 0 and vt = ∆u + eu , Bull. A.M.S. 75

(1969), 132-135.

[50] B. Gidas, Y.M. Ni, L. Nirenberg, Symmetry and Related Properties via the Maximum
Principle Commun. Math. Phys. 68 (1979), 209-243.

[51] J. Garcia Azorero, I. Peral Alonso, Existence and nonuniqueness for the p-Laplacian:
Nonlinear Eigenvalues. Commun. in PDE 12(1987), 1389-1430.

[52] J. Garcia Azorero, I. Peral Alonso, Comportement asymptotique des valeurs propres du
p-laplacien C.R. Acad. Sci. de Paris t. 307, Série I. (1988), 75-78.

[53] J. Garcia Azorero, I. Peral Alonso, Multiplicity of solutions for elliptic problems with
critical exponent or with a non-symmetric term. Transactions of the Amer. Math. Soc.
323 No 2 (1991), 877-895.
 111

[54] J. Garcia Azorero, I. Peral Alonso, Some results about the Existence of a Second Positive
Solution in a Quasilinear Critical Problem Indiana University Math. Jou. 43 No 3
(1994), 941-957.

[55] J. Garcia Azorero, I. Peral Alonso, Hardy Inequalities and some Critical Elliptic and
Parabolic Problems Preprint 1996.

[56] B. Gidas, J. Spruck, Global and Local Behavior of Positive Solutions of Nonlinear
Elliptic Equations Commun. Pure Appl. Math. XXXIV (1981), 525-598.

[57] M. Guedda, L. Veron, Bifurcation phenomena associated to the p-Laplace operator

Trans. Amer. Math. Soc., 310 No 1 (1988), 419-431.

[58] M. Guedda, L. Veron, Quasilinear Elliptic Equations Involving Critical Sobolev Expo-
nents Nonlinear Anal TMA 13 (8) (1989), 879-902.

[59] G. Hardy, J.E. Littlewood, G. Polya, Inequalities Cambridge University Press, 1934

[60] H. Hofer, A note on the topological degree at a critical point of mountain pass-type
Proc. A.M.S. 90, No 2 (1984), 309-315.

[61] J. Kazdan, F. Warner, Remarks on some Quasilinear Elliptic Equations Comm. on

Pure Appl. Math. Vol 38 (1975), 557-569.

[62] S Kesavan, F. Pacella, Symmetry of positive solutions of a quasilinear elliptic equation

via isoperimetric inequalities Applicable Anal. 54 (1994), 27-37.

[63] O.A. Ladyzhenskaya, N.N. Ural’tseva, Linear and Quasilinear Elliptic equations Aca-
demic Press, 1968.

[64] P.L. Lions, The concentration-compactness principle in the calculus of variations. The
limit case, part 1 Rev. Matemática Iberoamericana Vol 1 No 1 (1985), 145-201.

[65] P.L. Lions, The concentration-compactness principle in the calculus of variations. The
limit case, part 2. Rev. Matemática Iberoamericana Vol 1 No 2 (1985), 45-121.

[66] P. Lindqvist, On the equation ∆p u + λ|u|p−2 u = 0 Proc. A.M.S. 109-1 (1990), 157-164.

[67] J. Mawhin, M. Willem, Critical Point Theory and Hamiltonian Systems Springer Ver-
lag, 1989

[68] F. Merle, L.A. Peletier, Positive solutions of elliptic equations involving supercritical
growth Proc. of the Royal Soc. Edinburgh Vol 118 A (1991), 49-62.
112

[69] W.M. Ni, J. Serrin, Existence and non-existence theorems for ground states of quasi-
linear PDEs Acca. Naz. Lincei, Convegni del Lincei 77 (1986), 231-257.

[70] M. Otani, Existence and Nonexistence of Nontrivial Solutions of Some Nonlinear De-
generate Equations Jou. Funct. Analysis 76 (1988), 140-159.

[71] S.I. Pohozaev, Eigenfunctions of the equation ∆u + λf (u) Soviet Math. 5 (1965), 1408-
1411.

[72] S. Pohozaev, On the Eigenfunctions of Quasilinear Elliptic Problems, Mat. Sbornik Vol
82 (124) No 2 (1970), 171-188.

[73] P. Pucci, J. Serrin, The structure of the critical set in the mountain pass theorem Trans.
A.M.S. 299 (1987), 115-132

[74] P. Rabinowitz, Some Global Results for Nonlinear Eigenvalue Problems Jou. Funct.
Analysis (1971), 487-513

[75] P.H. Rabinowitz, Theorie du degree topologique et applications a des problemes aux
limites non lineaires Lectures of Paris VI and Paris XI 1973 writen by H. Berestycki.
Publications de Universite VI Laboratoire d’Analyse Numerique. Paris, 1975.

[76] P. Rabinowitz, Minimax Methods in Critical Point Theory with Applications to Differ-
ential Equations CBMS Regional Conference No 65 A.M.S.

[77] S. Sakaguchi, Concavity properties of solutions to some degerate quasilinear elliptic

Dirichlet Problems. Ann. Scuola Normale Sup. di Pisa Serie 4 Vol 14 No 3 (1987),
403-421.

[78] J. Serrin, On a fundamental theorem of the calculus of variations Acta Math. 102(1959),
1-32.

[79] J. Serrin, Local behavior of solutions of quasi-linear elliptic equations Acta Math. 111
(1964), 247-302.

[80] J. Simon, Regularité de la solution d’une equation non lineaire dans IRN Lectures Notes
in Math. No. 665 P. Benilan editors Springer Verlag, 1978.

[81] G. Stampacchia, Equations Elliptiques du Second Ordre a Coefficients Discontinus Les

Presses de L’Universite de Montreal, 1966.

[82] M. Struwe, A global compactness result for elliptic boundary problems involving limiting
nonlinearities Math. Z. 187 (1984), 511-517.
 113

[83] M. Struwe, Variational Methods. Applications to Nonlinear Partial Differential Equa-

tions and Hamiltonian Systems Springer Verlag, 1990.

[84] G. Talenti, Best constant in Sobolev inequality Ann. Math. Pura Appl. 110 (1976),
353-372.

[85] G. Tarantello, On nonhomogeneous elliptic equations involving critical Sobolev Expo-

nents Annales I.H.P., Analyse Nonlinéaire 9 (1992), 281-304.

[86] L. Tonelli, Fondamenti di calcolo delle variazioni Zanichelli, 1921

[87] P. Tolksdorff, Regularity for a more general class of quasilinear elliptic equations Jou.
of Diff. Equations 51 (1984), 126-150

[88] N. Trudinger, Remarks concerning the conformal deformation of Riemannian structures

on compact manifolds. Ann. Scuola Normale Su. Pisa. 22 (1968), 265-274.

[89] K. Uhlenbeck, Regularity for a class of nonlinear elliptic systems Acta Math. 111
(1977), 219-240.

[90] Z. Q. Wang, On a superlinear elliptic equation Ann. Inst. Henri Poincare 8 No 1 (1991),
43-57

[91] M. Willem, Minimax Theorems Birkhäuser, 1996.

[92] G.T. Whyburn, Topological Analysis, Princeton University Press, Princeton, New Jer-
sey, 1964

[93] H. Yamabe, On the deformations of Riemannian structures on compact manifolds Os-

aka Math. Jou. Vol 12 (1960), 21-37.

[94] A. Zygmund, Trigonometric Series, Cambridge University Press, 1977