Journal of Mathematical Analysis and Applications 263, 264–276 (2001)

doi:10.1006/jmaa.2001.7631, available online at http://www.idealibrary.com on

Lower Semicontinuity of Functionals via the

Concentration-Compactness Principle

Eugenio Montefusco

Dipartimento di Matematica, Università degli Studi di Roma “La Sapienza,”

Piazzale A. Moro 5, 00185 Rome, Italy
E-mail: [email protected]

Submitted by Jean Mawhin

Received June 29, 1999

In this paper we show the weak lower semicontinuity of some classes of function-
als, using the concentration-compactness principle of P. L. Lions. These functionals
involve an integral term, and we do not know whether it can be handled by the
De Giorgi theorem. The semicontinuity result allows us to prove several existence
results for quasilinear elliptic equations.  2001 Academic Press
Key Words: lower semicontinuity; concentration; compactness principle.

1. INTRODUCTION

Some recent papers [1, 7] study a class of functionals whose model is

1
λ
u
xp

F
u = ∇u
xp dx − p dx − f
xu
x dx
p  p  x 

where  is a bounded open subset of
N , 1 < p < N, u belongs to

1 p 
W0
, f belongs to Lp
, and λ is a real positive number.
If λ is sufficiently small the functional is coercive, and one can expect
that there exists a global minimum. Since the Nemitskii operator u
x −→
1 p
u
x/x from W0
 in Lp
 is continuous but not compact, it is not
clear if we can obtain directly the weak lower semicontinuity of the func-
1 p
tional on W0
 by using the De Giorgi theorem (see [5, 6]), so that it
seems that we cannot apply the direct methods of the calculus of variations.
In [7] the authors, using a critical point technique based on the coercivity
and the homogeneity of the functional, show the existence of a global mini-
mum (for λ belonging to the set in which the functional is coercive) without
264
0022-247X/01 $35.00
Copyright  2001 by Academic Press
All rights of reproduction in any form reserved.
 lower semicontinuity of functionals 265

using the direct methods of the calculus of variations. Reference [1] treats
more general problems with an interesting nonvariational method, which
does not require homogeneity, but only coercivity of the quadratic form
associated to the equation. In any case, both papers leave open the ques-
tion of whether the functional is weakly lower semicontinuous.
In this note we want to prove that the functionals
1
λ
u
xp
λ
u = ∇u
xp dx − dx
p  p  xp
 p/p∗
1
λ
p∗
λ
u = ∇u
xp dx − u
x dx
p  p 
1 p
are weakly lower semicontinuous in W0
, provided λ belongs to the
set of
in which the functionals are coercive. Note that both functionals
1 p
have a nonlinear term which is continuous but not compact on W0
.
The proof relies on an idea in [11, 12]: starting from an integral inequality
involving the noncompact term, we are able to describe exactly the behavior
1 p
of the functionals on a weakly convergent sequence in W0
, and so we
show the claim for λ (see Section 2); the same method is extended for
λ in Section 3. The semicontinuity of λ and λ allows us to prove some
existence results for quasilinear elliptic equations. In particular, we are able
to handle functional λ in unbounded domains, so we can show that there
exists a weak solution of the equation
λ
−p u = up−2 u + f in
N 
xp
See Theorem 3.7.
The only assumption we make is

1 < p < N

The main tool in this note is the exact behavior of weakly convergent
sequences in Sobolev spaces. This important result, due to P. L. Lions (see
[11, 12]), is the following.
1 p
Lemma 1.1. Let uk be a weakly convergent sequence in W0
 with
weak limit u. Then we have that

∇uk
xp dx −→ µ

and
∗
uk
xp dx −→ ν
266 eugenio montefusco
Np
in the ∗-weak convergence of measures, where p∗ = N−p
. Then, for some
finite index set J, we have
∗ 
ν = u
xp dx + νj δ x j
j∈J

µ ≥ ∇u
xp dx + µ j δx j (1)
j∈J
p/p∗
νj ≤ µj /S


Moreover, we have µj νj > 0 and xj ⊂ .

2. SOBOLEV INEQUALITY

Lemma 1.1 relies on the Sobolev inequality


1/p∗ 
1/p
p∗ 1 p
u
x dx ≤ ∇u
x dx (2)
 S 1/p 

where S is the best constant in the Sobolev inclusion (see [15]).

P. L. Lions obtains an exact description of the behavior of the weakly
1 p
convergent sequences in W0

N , deriving from (2) a reverse Hölder
inequality, which allows us to write the relations in (1). So we are able to
study the weak lower semicontinuity of the functional
 p/p∗
1
p λ
p∗
λ
u = ∇u
x dx − u
x dx
p  p 

where λ is a real positive number and  ⊂
N is an open bounded set.

Note that, since (2) holds, λ is coercive, provided λ ∈
0 S. Obviously,
if λ is nonpositive the functional is coercive and weakly lower semicontin-
uous, because it is convex.

Theorem 2.1. Assume λ ∈ 0 S; then λ is a weakly lower semicontin-

1 p
uous functional on W0
.

Proof. In this case we cannot apply the classical methods, because the
1 p ∗
embedding W0
 → Lp
 is only continuous and not compact. Using
(1), we are going to study the weak lower semicontinuity of λ .
 lower semicontinuity of functionals 267
1 p
Let uk be a weakly convergent sequence in W0
 and let u be the
weak limit. Set L = lim inf k−→∞ λ
uk . By Theorem 1.1 we have
 
1

L= ∇u
xp dx + µj + nonnegative terms
p  j∈J
 p/p∗
λ
p∗

− u
x dx + νj
p  j∈J
 
p/p∗
1  p∗

≥ λ
u + µ −λ u
x dx + νj
p j∈J j  j∈J

p/p∗ 
p∗
− u
x dx 

p/p∗ p/p∗
Since
1 + t ≤1+t , for all positive real numbers t, it follows that
  p/p∗ 
1  
L ≥ λ
u + µ −λ νj
p j∈J j j∈J
   p∗ /p p/p∗ 
1   µj
≥ λ
u + µj − λ 
p S
∗ ∗
Recalling that
1 + tp ≥ 1 + t p /p , it holds that
/p
 
1   µj
L ≥ λ
u + µ −λ
p j∈J j j∈J
S
 
1 λ 
= λ
u + 1− µ ≥ λ
u
p S j∈J j

since λ ∈ 0 S and µj > 0. The above inequalities show

lim inf λ
uk  ≥ λ
u
k−→+∞
1 p
for every uk weakly convergent to u in W0
, and the claim is proved.
Corollary 2.2. Consider the functional


F
u = λ
u − f
xu
x dx

−1 p
where f belongs to W
. Then there exists a global minimum of the
functional F, provided λ ∈
0 S.
Proof. Since the functional is coercive by (2) and is weakly lower semi-
continuous by the previous theorem, then [5, Chap. 3, Theorem 1.1] proves
the claim.
268 eugenio montefusco
1 p
Remark. Note that λ ∈ C 1
W0

, so that the global minimum
1 p
u ∈ W0
 is a weak solution of the nonlocal problem

p/p∗ −1
∗ ∗
−p u = up up −2
u + f


which is the Euler–Lagrange equation of the functional.

3. HARDY INEQUALITY

In this section we want to repeat the above arguments starting from the
Hardy inequality

u
xp 1

p dx ≤ ∇u
xp dx (3)
 x H 
1 p
where u belongs to W0
,  is an open set in
N containing the origin,
and H =
N−p
p
p is the best constant of the inclusion (see [7]).
1 p
The Hardy inequality proves that the embedding of W0
 in
Lp
 1/xp  is continuous but is as not compact as for the Sobolev
embedding; see [3, 7] for more information.
We are now interested in the behavior of the sequence uk
x/x in
1 p
Lp
, assuming uk −→ u in the weak topology of W0
. We will follow
the ideas in [11, 12] and study the sequence in the space of measures as
above. Now we prove
1 p
Lemma 3.1. Let uk be a weakly convergent sequence in W0
 with
weak limit u. Then we have

∇uk
xp dx −→ µ

and
uk
xp dx
−→ ν
xp

in the ∗-weak convergence of measures. Moreover, it holds that

u
xp dx
ν= + ν 0 δ0
xp (4)
p
µ ≥ ∇u
x dx + µ0 δ0 0 ≤ ν0 ≤ µ0 /H
 lower semicontinuity of functionals 269

Proof. Note that (3) shows that uk
x/x is bounded in Lp
, so
1 p
that the sequence has a weak limit, and, since uk −→ u weakly in W0
,
p
strongly in L
, and a.e. in , it holds that

uk
x u
x
−→
x x

weakly in Lp
.
We are now going to prove an inequality which will allow us to study the
singular part of the weak limit of the sequence in the space of measures.
Assume that 0 is the weak limit of the sequence, because, setting vk =
uk − u, it holds that


φ
xp v
xp 
φ
xp u
xp
φ
xp uk
xp k
dx− dx −→ dx
 xp  xp  xp

for all φ in 
, by the results in [2].

1 p
The sequence uk is bounded in W0
, while the sequence uk /x
p
is bounded in L
, so that the weak-∗ limits of the sequences in the
measure space exist. Given any φ in 
, using the functions φuk in the
Hardy inequality, and passing to the limit, we have

1

φ
xp dν ≤ φ
xp dµ (5)
 H 

where µ and ν are the limit measures of the sequences.

Choosing φ such that 0 ∈ supp
φ, we have that

φ
xp u
xp
φ
xp

k
p dx = uk
xp dx
 x supp
φ xp

Note that the function φ
xp /xp belongs to L∞
supp
φ, and, since
1 p
the inclusion of the Sobolev space W0
 in Lp
 is compact (supp
φ
is bounded), we obtain that

φ
xp

p
u k
x dx ≤ C uk
xp dx −→ 0
supp
φ xp supp
φ

1 p
since uk −→ 0 weakly in W0
 as k goes to infinity. Then ν is a measure
concentrated in 0 and is absolutely continuous with respect to a Dirac mass
(since µ contains Dirac masses; see Lemma 1.1), so it has to be a constant
270 eugenio montefusco

times a Dirac mass at the origin. Fixing a function φ in 
, such that
φ  χB
0 ε , it then follows that

1
ν
B
0 ε ≤ µ
B
0 ε
H
Since ε is arbitrary, we have
µ0
ν0 ≤ 
H
Consider
1
λ
u
xp
λ
u = ∇u
xp dx − dx
p  p  xp

Using the above lemma, we prove the weak lower semicontinuity of the
functional.

Theorem 3.2. Assume λ ∈ 0 H; then λ is a weakly lower semicontin-

1 p
uous functional on W0
.
1 p
Proof. Let uk be a weakly convergent sequence in W0
 and let
u be the weak limit. Set L = lim inf k−→∞ λ
uk . By Lemma 3.1 we have
 
1
p
L= ∇u
x dx + µ0 + nonnegative terms
p 
 
λ
u
xp
− dx + ν 0
p  xp
1
≥ λ
u +
µ − λν0 
p 0
  
1 λ
≥ λ
u + 1− µ0
p H

and the claim follows.

Observe that, if λ is nonpositive, the result trivially holds. Then we have

Corollary 3.3. Consider the functional

F
u = λ
u − f
xu
x dx


where f belongs to Lp
. Then there exists a global minimum of the func-
tional F, provided λ ∈
0 H.
 lower semicontinuity of functionals 271
1 p
Remark. This means that we have found a weak solution u ∈ W0

of the equation
λ
− p u = up−2 u + f (6)
xp

since the functional λ is C 1 .

A natural question could be whether H is an eigenvalue of
λ
−p u = up−2 u
xp
[16] shows that this is not possible in the linear case (p = 2). The author
studies many linear problems in
N , and he proves that there is no eigen-
function belonging to D1 2

N  (the space of measurable functions v having
∇v ∈ Lp

N , up to a constant; see [13]).
For the sake of completeness we want to point out that it is possible to
prove, in a different way, the existence of a weak solution of a more general
problem, whose model is (6). Now we proceed using the ideas of [1], only
we conclude in a slightly different way. Consider the equation
λ
A
u = up−2 u + f
xp (7)
1 p
u∈ W0


and the approximated equations

A
un  = λn
x−p un p−2 un + f

1 p
(8)
un ∈ W0


where n
s = s if 0 ≤ s ≤ n and n
s = n if s > n, f belongs to Lp
,
1 p 
A W0
 −→ W −1 p
, such that A
u = −div
a
x u ∇u, where
a  ×
×
N −→
N is a Carathéodory function such that

a
x s ξξ ≥ αξp

a
x s ξ ≤ c
k
x + sp−1 + ξp−1 

with k belonging to Lp
,

a
x s ξ − a
x s η ·
ξ − η > 0

if ξ ≡ η, and λ ∈
0 αH.
By the results in [8] it follows that there exists a solution un of (8), for
any n. Using un as a test function in the weak formulation of (8), we see
272 eugenio montefusco
1 p
that the sequence un is bounded in W0
. The inequality (3) and
coercivity of A indeed imply

p
αun  1 p ≤ A
un  un 
W0




= λ n
x−p un
xp dx + f
xun
x dx
 
λ p
≤ un  1 p + f Lp
 un W 1 p
 
H W0
 0

Note that the assumption λ < αH implies that
α − λ/H is positive, so

that it holds
 1/
p−1
f Lp

un W 1 p
 ≤ 
0 α − λ/H

Since the sequence is bounded, there exists a subsequence, which we will

1 p
still denote by un , such that un −→ u weakly in W0
, strongly in
p
L
, and a.e. in .

By the assumption on a, a
x un ∇un  is bounded in
Lp
N , so that
it is weakly convergent to a function. Since


 

p−1/p 
1/p
 un
xp−1  un
xp 1
 dx ≤ dx dx ≤ c
  xp   xp  x
p

[4] shows that ∇un → ∇u a.e. in , so we have a
x un ∇un  → a
x u ∇u

in the weak topology of
Lp
N .
Now we choose φ ∈ 
 \ 0  as a test function in the weak formulation
of problem (8); letting n tend to infinity, we find

u
xp−2
a
x u
x ∇u
x∇φ
x dx = λ u
xφ
x dx
  xp


+ f
xu
x dx


because we have proved that the Hardy inclusion is weakly continuous.

In [13] it is proved that a point in
N has zero N-capacity; hence it has
zero p-capacity, because p < N. This means that there exists a sequence
1 p
ψn ⊂ W0
 such that ψn
x ≡ 1 in a neighborhood of the origin,
1 p
0 ≤ ψn
x ≤ 1 in , and ψn −→ 0 strongly in W0
.
 lower semicontinuity of functionals 273

Now in the weak formulation of (8) we use as a test function
1 − ψn v,

1 p
where v ∈ W0
. Recalling the properties of the sequence ψn , letting n
tend to infinity, we find


u
xp−2
a
x u
x ∇u
x∇v
x dx = λ u
xv
x dx
  xp


+ f
xu
x dx

1 p
for any v ∈ W0
. Then we have proved that u is a weak solution of
(7). So we can state the following result.

Theorem 3.4. For every f belonging to Lp
 there exists a solution of
problem (7).
Remark. We emphasize that this method, developed in [1], also holds
for nonvariational equations (the operator A may not be the derivative of
a functional). In the variational case (provided p = 2) is an open question
if the solution u found in such a way is the global minimum of the relative
functional.
By the weak lower semicontinuity of λ (with λ ∈
0 H) it follows that
classical results involving semicontinuous functionals hold. To obtain other
existence results we study some simple problems.
Theorem 3.5. Assume λ ∈
0 H. Then there exists a nontrivial solu-
tion of
λ
−p u = up−2 u + uq−2 u
xp
1 p
u ∈ W0

∗
where 1 < p < q < p .
Proof. Consider the functional λ on the smooth manifold M = u ∈
1 p 
W0
  q1  u
xq dx = 1 . We know that M is weakly closed by
the Rellich–Kondrakov theorem and that λ is coercive and weakly lower
1 p
semicontinuous on W0
. So, by the Lagrange multiplier rule (see [14,
Chap. I, Theorem 2.1]), we have that there exists w ∈ M, which is a con-
strained minimum of λ on M. Then there exists µ ∈
such that w satisfies
λ
−p w − p wp−2 w = µwq−2 w
x
using w as a test function and again using the Hardy inequality, we see
that µ > 0.
So setting u = µ1/
p−q w and using the homogeneity of the equation,
we have a nontrivial solution of the equation. This result is also proved
in [7].
274 eugenio montefusco

Theorem 3.6. Assume λ ∈
0 H, 1 < q < p. Then there exists a non-
trivial solution of
λ
−p u = up−2 u + uq−2 u
xp
1 p
u ∈ W0


Proof. The functional

1
1
u
xp 1

F
u = ∇u
xp dx − dx − u
xq dx
p  p  xp q 
1 p
is well defined, coercive, and weakly lower semicontinuous on W0
,
by the preceding results and the Rellich–Kondrakov theorem. Then there
1 p
exists a global minimum on W0
, which solves in the weak sense the
above equation. We have only to show that the minimum is nontrivial.
1 p
Fix w ∈ W0
 and consider
1
φ
t = F
tw = t p λ
w − t q wLq
 = c1 t p + c2 t q 
q
It is obvious that, for t positive and sufficiently small, φ
t < 0, so that it
follows that inf F < 0. This means that the global minimum of F (i.e., our
solution) is nontrivial, since F
0 = 0.
Now we want to extend the results of Corollary 3.3. More precisely, we
want to prove the compactness of minimizing sequences of the functional
1
λ
u
xp

/
u = ∇u
xp dx − dx − f
xu
x dx
p
N p
N xp
N

where u belongs to D1 p

N  (the space of the measurable functions v hav-

ing ∇v ∈ Lp

N , up to a constant; see [13]), λ ∈
0 H, and f belongs
∗  ∗
to L
p 

N  (because D1 p → Lp

N ).
Inequality (3) holds with  =
N , but the domain is invariant under
the action of the group of translations and the group of dilatations, which
implies a lack of compactness of the minimizing sequences of general inte-
gral functionals. The dilatations produce Dirac masses in the weak limit
of a minimizing sequence, but our preceding arguments (see the proof of
Theorem (3.2)) show that the creation of a delta is not convenient for a min-
imizing sequence of /. So we have to prove that the translations do not
destroy the compactness of a sequence which minimizes the functional /.
P. L. Lions showed in [9, 10] that the translations do not break the com-
pactness of the minimizing sequences if it holds that inf / < inf /∞ , where
/∞
u = limx−→+∞ /
u.
 lower semicontinuity of functionals 275

Assuming that limx−→+∞ f
x = 0, it is easy to see that

1
1 p
/∞
u = ∇u
xp dx = uD1 p

N  
p
N p
so, provided that there exists an open set B ⊂
N in which f
x ≥ δ > 0,
it is trivial that
inf / < 0 = inf /∞ 
Then we have proved
Theorem 3.7. Assume that f
x −→ 0, as x −→ +∞, and that there
exists an open set B ⊂
N , such that f
x ≥ δ > 0 a.e. x ∈ B. Then there
exists a global minimum u of / on D1 p

N . This global minimum is a weak
solution of the equation
λ
−p u = up−2 u + f in
N 
xp
Proof. The above arguments and [9–12] prove that the minimizing
sequence is compact; since the functional / is sufficiently smooth, the
claim follows.

ACKNOWLEDGMENTS

First of all we thank A. Tertikas for suggestions and open questions; moreover, we thank
L. Orsina: chi trova un amico trova un tesoro.

REFERENCES

1. D. Arcoya and L. Boccardo, Some remarks on critical point theory for nondifferentiable
functionals, NoDEA Nonlinear Differential Equations and Appl. 6 (1999), 79–100.
2. H. Brezis and E. Lieb, A relation between pointwise convergence of functions and con-
vergence of functionals, Proc. Amer. Math. Soc. 88 (1983), 486–490.
3. H. Brezis and J. L. Vazquez, Blow-up solutions of some nonlinear elliptic problems, Rev.
Univ. Complutense 10 (1997), 443–469.
4. L. Boccardo and F. Murat, Almost everywhere convergence of the gradients of solutions
for nonlinear elliptic equations with measure data, Nonlinear Anal. 19 (1992), 581–597.
5. B. Dacorogna, “Direct methods in the calculus of variations,” Applied Mathematical
Sciences, Vol. 78, Springer-Verlag, Berlin/New York, 1989.
6. E. De Giorgi, “Teoremi di semicontinuitá nel calcolo delle variazioni,” lecture notes,
Istituto Nazionale di Alta Matematica, 1968–69.
7. J. P. Garcı́a Azorero and I. Peral, Hardy inequalities and some critical elliptic and
parabolic problems, J. Differential Equations 144 (1998), 441–476.
8. J. Leray and J. L. Lions, Quelques résultats de Visik sur le problèmes elliptiques non-
linéaires par les méthodes de Minty et Browder, Bull. Soc. Math. France 93 (1965), 97–107.
276 eugenio montefusco

9. P. L. Lions, The concentration-compactness method in the calculus of variation. The

locally compact case. Part I, Ann. Inst. H. Poincaré Anal. Non Linéaire 1 (1984), 109–145.
10. P. L. Lions, The concentration-compactness method in the calculus of variations. The
locally compact case. Part II, Ann. Inst. H. Poincaré Anal. Non Linéaire 1 (1984), 223–283.
11. P. L. Lions, The concentration-compactness principle in the calculus of variations. The
limit case, part 1, Rev. Mat. Iberoamericana 1 (1985), 145–201.
12. P. L. Lions, The concentration-compactness principle in the calculus of variations. The
limit case, part 2, Rev. Mat. Iberoamericana 1 (1985), 45–121.
13. V. G. Maz’ja, “Sobolev spaces,” Springer Series in Soviet Mathematics, Springer-Verlag,
Berlin/New York, 1985.
14. M. Struwe, “Variational Methods,” Springer-Verlag, Berlin/New York, 1985.
15. G. Talenti, Best constant in Sobolev inequality, Ann. Mat. Pura Appl. 110 (1976), 353–372.
16. A. Tertikas, Critical phenomena in linear elliptic problems, J. Funct. Anal. 154 (1998),
42–66.