View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Elsevier - Publisher Connector

J. Math. Pures Appl. 80, 9 (2001) 919–940

 2001 Éditions scientifiques et médicales Elsevier SAS. All rights reserved
S0021-7824(01)01211-9/FLA

BOUNDED AND UNBOUNDED SOLUTIONS FOR

A CLASS OF QUASI-LINEAR ELLIPTIC PROBLEMS
WITH A QUADRATIC GRADIENT TERM

Lucio BOCCARDO a , Sergio SEGURA DE LEÓN b ,

Cristina TROMBETTI c
a Dipartimento di Matematica, Università di Roma I, P.le Aldo Moro 2, 00185 Roma, Italy
b Departament d’Anàlisi Matemàtica, Universitat de València, Dr. Moliner 50, 46100 Burjassot, València, Spain
c Dipartimento di Matematica e Applicazioni, Università di Napoli, Via Cintia, 80126 Napoli, Italy

Manuscript received 30 July 2000

A BSTRACT . – Our aim in this article is to study the following nonlinear elliptic Dirichlet problem:

−div[a(x, u) · ∇u] + b(x, u, ∇u) = f, in Ω; u = 0, on ∂Ω;

where Ω is a bounded open subset of R N , with N > 2, f ∈ Lm (Ω). Under wide conditions on functions
a and b, we prove that there exists a type of solution for this problem; this is a bounded weak solution for
m > N/2, and an unbounded entropy solution for N/2 > m
2N/(N + 2). Moreover, we show when this
entropy solution is a weak one and when can be taken as test function in the weak formulation. We also
study the summability of the solutions.  2001 Éditions scientifiques et médicales Elsevier SAS
Keywords: Bounded and unbounded solutions, Quasi-linear elliptic problems, Quadratic gradient term

0. Introduction and assumptions

This paper is devoted to study existence and regularity of solutions of an elliptic problem
whose model example is the following:


(0.1) −div α(u)∇u = β(u)|∇u|2 + f, in Ω; u = 0, on ∂Ω;

where Ω is a bounded open subset of R N , with N > 2, f ∈ Lm (Ω), m being bigger than
2N/(N + 2), and α and β are two positive continuous functions√ / L1 ([0, +∞[) ∪
satisfying α ∈
L (] − ∞, 0]) and β/α ∈ L (R). For instance, α(s) = 1/ 1 + s and β(s) = 1/ (1 + s 2 )3 , or
1 1 2
α(s) = e|s| and β(s) = e|s| /(1 + s 2 ).
In some recent papers problems similar to (0.1) have been considered. Without lower order
terms and α(s) = (1 + |s|)−θ , with 0  θ < 1, a priori estimates can be found in [1,5] and [6],
existence and regularity results in [5] and [6] and in a limit case in [24], and uniqueness is shown
in [19] . On the other hand, problem (0.1) may be seen as the Euler equation of a real functional
when α  = −2β; from this point of view has been studied in [12].

E-mail address: [email protected] (L. Boccardo).

920 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

The existence of weak solutions of quasilinear elliptic equations with lower order terms having
quadratic growth with respect to the gradient has been studied in some papers in the last years.
We point out that, in those papers, the L∞ estimates on the solutions are proved thanks to the
presence of a zero-order term ([10] and [11]) or thanks to a sign condition (easy situation).
Also the previous existence results of unbounded solutions depends on a sign condition on the
quadratic term ([3,4,7] and [9]). In [14] a limit case is studied. In [25] the existence of bounded
solutions is obtained when the datum f is “small” and the assumptions on α and β are different.
In this paper we do not use the presence of a zero-order term nor a sign condition in order to
show the existence of bounded or unbounded solutions (it depends on the summability of the
data). On the other hand, we shall use the assumptions (H4) and (H5) below.
Finally, when the function α is bounded but the equation has the same lower order term, this
problem is dealt in [20], where existence and uniqueness for L1 -data are studied. Moreover the
L∞ -estimate of Section 2 below is applied in that paper to obtain a kind of uniqueness.
Two remarks concerning the difficulties in dealing with this problem are in order. First of all,
observe that no bounds are assumed on function α, and so classical methods do not apply. Indeed,
the operator defined by −div(α(u)∇u) does not satisfy the Leray–Lions conditions (see [18])
since, on the one hand, we suppose no growth limitation on function α so that it can happen
−div(α(u)∇u) ∈ / H −1 (Ω) for all u ∈ H01 (Ω) and, on the other hand, it is not coercive; see [19]
for an explicit example. To deal with that equation we will obtain the solution by approximation,
getting a sequence of approximated solutions (un )n which converge to the solution u. To avoid
some troubles in this convergence process, we will consider the convergence of another sequence
(A(un ))n , function A being the primitive of α such that A(0) = 0. The other remark is about
the lower order term: it does not satisfy the “right” sign condition, since β is positive. Thus, it
appears the problem of getting the a priori estimates. This hindrance is overcome by considering
test functions of exponential type as in [20]; see Lemma 2.2 below.
We next state our assumptions more precisely. We shall study the following nonlinear elliptic
problem:
 
(0.2) −div a(x, u) · ∇u + b(x, u, ∇u) = f, in Ω; u = 0, on ∂Ω.
2
Here Ω is a bounded open subset of RN , with N > 2, and a : Ω × R → RN and
b : Ω × R × RN → R are Carathéodory functions satisfying the following hypotheses:
 
(H1) sup a(x, s) ∈ L∞ (Ω), for all k > 0,
|s|
k
 
(H2) a(x, s) · ξ · ξ
α(s)|ξ |2 ,
 
(H3) b(x, s, ξ )  β(s)|ξ |2 ,
α and β being positive continuous functions such that



(H4) α∈/ L1 [0, +∞[ ∪ L1 ]−∞, 0] ,

and
β
(H5) ∈ L1 (R).
α
So that, defining
s
β(τ )
γ (s) = dτ
α(τ )
0
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 921

and
s
A(s) = α(τ ) dτ,
0

we have that the function γ is bounded, while

 
lim A(s) = +∞.
s→±∞

On the other hand, f ∈ Lm (Ω), with m
2N/(N + 2); however, we will not deal with the limit
case m = N/2. Recall that in the classical case, that is, when the function a(x, s) · ξ satisfies
the Leray–Lions conditions; if f ∈ LN/2 (Ω), then the solution belongs to an exponential Orlicz
space.
The plan of this article is as follows. The next section is on notations. In Section 2,
problem (0.2) is studied when f ∈ Lm (Ω), with m > N/2. We define weak solution for (0.2) and
get an L∞ -estimate which allows us to obtain a weak solution of (0.2). Section 3 is devoted to
prove existence of an entropy solution of (0.2) when f ∈ Lm (Ω), with 2N/(N + 2)  m < N/2.
In this section we also see some conditions to guarantee that that entropy solution is actually
a weak solution, and to obtain an energy type equality. Moreover, we end this section studying
the summability of a solution. Finally, in Section 4, we will present an alternative approach in
finding a priori estimates for solutions of problem (0.2), based on the symmetrization techniques.

1. Notations

Some notations are used throughout this paper. Ω ⊂ RN will denote an open bounded set, |A|
Lebesgue measure of A ⊂ Ω and ci positive constants which only depend on the parameters of
our problem.
For k > 0 we define the truncature at level ±k as Tk (s) = (−k) ∨ [k ∧ s]; we also consider
Gk (s) = s − Tk (s) = (|s| − k)+ sign(s).
1,p
Following [2], we introduce T0 (Ω) as the set of all measurable functions u : Ω → R such
that Tk u ∈ W0 (Ω) for all k > 0. We point out that T0 (Ω) ∩ L∞ (Ω) = W0 (Ω) ∩ L∞ (Ω).
1,p 1,p 1,p
1,p
For a measurable function u belonging to T0 (Ω), a gradient can be defined: it is a
measurable function which is also denoted by ∇u and satisfies ∇Tk u = (∇u)χ{|u|<k} for all k > 0
(see [2], Lemma 2.1).

2. Existence of bounded solutions

In this section we begin by defining weak solution of problem (0.2), then a new type of L∞ -
estimates is obtained in Theorem 2.3 (we remark again that we do not suppose the sign condition
b(x, s, ξ )s
0 for all s ∈ R ). As a consequence we have an existence result for (0.2).
The interest for getting L∞ -estimates for a problem as (0.2) is well known. The simplest
examples deal with Euler’s equations of functionals defined on H01 (Ω) or assume b(x, s, ξ )s
0
for all s ∈ R. A different approach (without that sign condition) was introduced by [11] for
studying problem (0.2) when, for instance, b(x, s, ξ ) = λs − |ξ |2 , with λ > 0.
Next, let us define weak solution for (0.2).
922 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

D EFINITION 2.1. – We will say that a function u ∈ T01,2 (Ω) is a weak solution of (0.2) if
a(x, u) · ∇u ∈ L2 (Ω), b(x, u, ∇u) ∈ L1 (Ω) and
  
 
(2.1) a(x, u) · ∇u · ∇ϕ + b(x, u, ∇u)ϕ = fϕ
Ω Ω Ω

holds for all ϕ ∈ H01 (Ω) ∩ L∞ (Ω).

We will also say that ϕ may be taken as test function in (2.1) if the functions [a(x, u) · ∇u] · ∇ϕ,
b(x, u, ∇u)ϕ and f ϕ belong to L1 (Ω), and (2.1) holds true.

From now on in this section we assume

f ∈ Lm (Ω), with m > N/2.

L EMMA 2.2. – Let u be a weak solution of (0.2) and let v ∈ H01 (Ω) ∩ L∞ (Ω).
(i) If v
0, then eγ (u) v can be taken as test function in (2.1) and
 
 
e γ (u)
a(x, u) · ∇u · ∇v  eγ (u) f + v.
Ω Ω

(ii) If v  0, then e−γ (u) v may be taken as test function in (2.1) and
 
 
e−γ (u) a(x, u) · ∇u · ∇v  e−γ (u) f − (−v).
Ω Ω

(iii) If Φ is a locally Lipschitz continuous and increasing real function such that Φ(0) = 0
and Φ(u) may be taken as test function in (2.1), then there exists c1 > 0 satisfying
 
 
(2.2) α(u)Φ  (u)|∇u|2  c1 f Φ(u).
Ω Ω

Proof. – (i) Let v
0. First observe that we have eγ (Tk u) ∈ H 1 (Ω) ∩ L∞ (Ω), for every k > 0.
Thus, v ∈ H01 (Ω) ∩ L∞ (Ω) implies eγ (Tk u) v ∈ H01 (Ω) ∩ L∞ (Ω) and so it is an admissible test
function in (2.1). Taking it so, it yields
 
β(Tk u) γ (Tk u)    
e v a(x, u) · ∇u · ∇Tk u + eγ (Tk u) a(x, u) · ∇u · ∇v
α(Tk u)
Ω Ω
 
(2.3) + b(x, u, ∇u)e γ (Tk u)
v = fe γ (Tk u)
v.
Ω Ω

Now we are going to study these integrals. Note that

 
β(Tk u) γ (Tk u)   β(u) γ (u)  
e v a(x, u) · ∇u · ∇Tk u = e v a(x, u) · ∇u · ∇u;
α(Tk u) α(u)
Ω {|u|<k}
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 923

so that, by the positivity of α and β, and by (H2), the integrand function is nonnegative. Hence,
applying the monotone convergence theorem, we have
 
β(Tk u) γ (Tk u)   β(u) γ (u)  
lim e v a(x, u) · ∇u · ∇Tk u = e v a(x, u) · ∇u · ∇u.
k→∞ α(Tk u) α(u)
Ω Ω

On the other hand, the functions [a(x, u) · ∇u] · ∇v, b(x, u, ∇u)v and f v are summable, and the
functions eγ (Tk u) are bounded in L∞ (Ω); so Lebesgue’s dominated convergence theorem may
be applied in the remaining integrals. Thus, letting k tend to ∞ in (2.3), we obtain
 
β(u) γ (u)    
e v a(x, u) · ∇u · ∇u + eγ (u) a(x, u) · ∇u · ∇v
α(u)
Ω Ω
 
+ b(x, u, ∇u)eγ (u)v = f eγ (u) v
Ω Ω

and so eγ (u)v may be taken as test function in (2.1).

Finally, since
 
β(u) γ (u)  
e v a(x, u) · ∇u · ∇u + b(x, u, ∇u)eγ (u)v
0,
α(u)
Ω Ω

by (H2) and (H3); it follows that

  
 
e γ (u)
a(x, u) · ∇u · ∇v  f e γ (u)
v  f + eγ (u) v.
Ω Ω Ω

(ii) Let v  0 and consider now e−γ (Tk u) v ∈ H01 (Ω) ∩ L∞ (Ω) as a test function in (2.1) and
reason in the same way as above.
(iii) Let Φ be a locally Lipschitz continuous and increasing real function such that Φ(0) = 0.
Assume also that Φ(u) can be taken as test function in (2.1) and let k > 0. Observe first that
Φ(Tk u) = Tmax{Φ(k),−Φ(−k)} Φ(Tk u) ∈ H01 (Ω) ∩ L∞ (Ω). Thus, taking v = Φ(Tk u+ ) in (i) and
v = −Φ(Tk u− ) in (ii), it yields
 
 

eγ (u) Φ  (u) a(x, u) · ∇u · ∇u  eγ (u) f + Φ Tk u+
{0
u<k} Ω

and
 
 
e−γ (u) Φ  (u) a(x, u) · ∇u · ∇u  e−γ (u) f − Φ(Tk u− ).
{0u>−k} Ω

Having in mind (H2) and that e±γ (u) are bounded, we add up both formulae obtaning c1 > 0
such that
  
   

α(u)Φ (u)|∇u|  c1
2  f Φ(Tk u)  c1 f Φ(u)

{|u|<k} Ω Ω
holds true for all k > 0. Therefore, we get (2.2) from Fatou’s lemma. Observe that, since Φ(u)
may be taken as test function in (2.1), f Φ(u) ∈ L1 (Ω). ✷
924 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

T HEOREM 2.3. – (i) Suppose that α(s)
λ for some λ > 0. If u is a weak solution of (0.2)
which may be taken as test function, then u∞  c2 , where c2 > 0 only depends on λ, on m,
on the norm of f in Lm (Ω) and on the parameters of (0.2): that is, on N , on Lebesgue measure
of Ω, and on the function γ .
(ii) If u is a weak solution of (0.2) such that A(u) may be taken as test function, then
A(u)∞  c3 , where c3 > 0 only depends on m, on the norm of f in Lm (Ω) and on the
parameters of (0.2); thus, u∞  max{A−1 (c3 ), −A−1 (−c3 )}.
Proof. – (i) Taking Φ(s) = Gk (s) in (2.2), it yields
  
 
λ |∇u| 2
α(u)|∇u|  c4
2 f Gk (u)
{|u|>k} {|u|>k} Ω

for some c4 > 0 and consequently

 
   
∇Gk (u)2  c4 f Gk (u).
λ
Ω Ω

By applying Stampacchia’s L∞ -regularity procedure (see [21]), it follows that there exists c5 > 0
satisfying u∞  c5 .
(ii) The proof is similar to (i), we only have to take Φ(s) = Gk (A(s)) in (2.2) and reason
in the same way. The last assertion is a consequence of being A strictly increasing and
lims→±∞ |A(s)| = +∞. ✷
T HEOREM 2.4. – There exists u ∈ H01 (Ω) ∩ L∞ (Ω) which is a weak solution of (0.2).
Proof. – We obtain the solution u by approximation: consider the following sequence of
problems:
 
(2.4) −div an (x, un ) · ∇un + bn (x, un , ∇un ) = f, in Ω; un = 0, on ∂Ω;

where an (x, s) = a(x, Tn s), αn (s) = α(Tn s), βn (s) = αn (s) β(s)
α(s) and


 
 
bn (x, s, ξ ) = min Tn βn (s)|ξ |2 , max −Tn βn (s)|ξ |2 , b(x, s, ξ ) .

Let us see some simple properties of these functions. Note that, by (H1), the function |an | is
bounded from above; thus there exists Λn > 0 such that |an (x, s)|  Λn . Moreover, since α is
continuous, there exists λn > 0 such that αn (s)
λn and so αn ∈ / L1 ([0, +∞[) ∪ L1 (]−∞, 0]).
Then
 
λn |ξ |2  αn (s)|ξ |2  an (x, s) · ξ · ξ  Λn |ξ |2 .
On the other hand, the function bn is bounded and
 

bn (x, s, ξ )  Tn βn (s)|ξ |2  βn (s)|ξ |2 .

We also observe that βn /αn = β/α ∈ L1 (R) and

 β
βn  max α(s): |s|  n ,
α
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 925

s
so that βn ∈ L1 (R). Furthermore, consider An (s) = 0 αn (τ ) dτ , and note that the functions
defined by these formulae are continuous and strictly increasing. We finally point out that if
|s|  n, then an (x, s) = a(x, s), αn (s) = α(s), βn (s) = β(s) and An (s) = A(s).
Applying the classical result by Leray and Lions [18], for each n ∈ N, there exists a weak
solution un ∈ H01 (Ω), which is an admissible test functions in the weak formulation of (2.4).
Hence, Lemma 2.2 and Theorem 2.3 can be applied. By Theorem 2.3(i), we have un ∈ L∞ (Ω)
and so An (un ) ∈ H01 (Ω) ∩ L∞ (Ω) for all n ∈ N. Applying now Theorem 2.3(ii), there is
c6 > 0 such that un ∞  max{A−1 −1 −1
n (c6 ), −An (−c6 )} and, since the sequences (±An (±c6 ))n
converge to ±A−1 (±c6 ), it follows that (un )n is bounded in L∞ (Ω). An easy consequence is
that αn (un ) = α(un ) for n big enough, so that the sequence (αn (un ))n is bounded from below by
a positive number, say µ. Likewise, the sequence (an (x, un ))n is bounded from above, by (H1);
we also point out that
 
µ|ξ |2  αn (un )|ξ |2  an (x, un ) · ξ · ξ
for all ξ ∈ RN .
Taking Φ(s) = s in Lemma 2.2(iii), there is c7 > 0 such that
 
αn (un )|∇un |2  c7 |f un |
Ω Ω

and so
  
µ |∇un |2  αn (un )|∇un |2  c8 |f | = c9 .
Ω Ω Ω

This estimate proves that (un )n is bounded in H01 (Ω).

Hence, up to subsequences, (un )n
converges weakly; moreover, Rellich–Kondrachov’s theorem implies that we may also assume
that converges almost everywhere in Ω. Let u be that limit; then u ∈ H01 (Ω) ∩ L∞ (Ω),

(2.5) un + u weakly in H01 (Ω)

and
(2.6) un → u almost everywhere in Ω.
Next, we will see that
(2.7) un → u in H01 (Ω).
To prove this, take v = (un − u)+ in Lemma 2.2(i) and v = −(un − u)− in Lemma 2.2(ii) to get
 
  +
(2.8) e γ (un )
an (x, un ) · ∇un · ∇(un − u)  eγ (un ) |f |(un − u)+
Ω Ω

and
 
 
(2.9) − e−γ (un ) an (x, un ) · ∇un · ∇(un − u)−  e−γ (un ) |f |(un − u)− .
Ω Ω

If we denote δ = µe− sup |γ | , then

926 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

 2
δ ∇(un − u)
Ω
 
   
=δ ∇(un − u)2 + δ ∇(un − u)2
{un −u0} {un −u<0}

 
 eγ (un ) an (x, un ) · ∇(un − u) · ∇(un − u)+
Ω

 
− e−γ (un ) an (x, un ) · ∇(un − u) · ∇(un − u)−
Ω
 
   
= eγ (un ) an (x, un ) · ∇un · ∇(un − u)+ − eγ (un ) an (x, un ) · ∇u · ∇(un − u)+
Ω Ω
 
   
− e−γ (un ) an (x, un ) · ∇un · ∇(un − u)− + e−γ (un ) an (x, un ) · ∇u · ∇(un − u)− .
Ω Ω
Applying (2.8) and (2.9), it yields

 
∇(un − u)2
Ω
 
1 1  
 eγ (un ) |f |(un − u)+ − eγ (un ) an (x, un ) · ∇u · ∇(un − u)+
δ δ
Ω Ω
 
1 −γ (un ) − 1  
+ e |f |(un − u) + e−γ (un ) an (x, un ) · ∇u · ∇(un − u)− .
δ δ
Ω Ω

Now, taking into account that un , an (x, un ) and e±γ (un ) define bounded sequences, it follows
from (2.5), (2.6) and Lebesgue’s convergence theorem that the right-hand side converges to 0.
Therefore, (2.7) is proved and, as a consequence, there exists a subsequence (still denoted by
(un )n ) such that
(2.10) ∇un → ∇u almost everywhere in Ω.
In order to take limits in the weak formulation of (2.4), we next show that

(2.11) bn (x, un , ∇un ) → b(x, u, ∇u) in L1 (Ω).

Since, by (2.6) and (2.10), we already know that bn (x, un , ∇un ) → b(x, u, ∇u) almost
everywhere in Ω, it is enough to see the equi-integrability of this sequence and then apply
Vitali’s convergence theorem. Observe that βn (un ) = β(un ) for n big enough, so that the
sequence (βn (un ))n is bounded, that is, there is c10 > 0 such that βn (un )∞  c10 . Thus,
βn (un )|∇un |2  c10 |∇un |2 and so |bn (x, un , ∇un )|  c10 |∇un |2 . Finally, the equi-integrability
of (|∇un |2 )n , which follows from (2.7), implies that of (bn (x, un , ∇un ))n . Hence, (2.11) is
proved.
Let ϕ ∈ H01 (Ω) ∩ L∞ (Ω), then
  
 
(2.12) an (x, un ) · ∇un · ∇ϕ + bn (x, un , ∇un )ϕ = fϕ
Ω Ω Ω
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 927

for all n ∈ N, by the weak formulation of (2.4). Having in mind that the sequence (an (x, un ))n is
bounded in L∞ (Ω), it follows from (2.5), (2.6) and (2.11) that we may pass to the limit in (2.12)
obtaining that u is a weak solution of (0.2). ✷

3. Existence of unbounded solutions

In this section, we assume f ∈ Lm (Ω), with 2N/(N + 2)  m < N/2. We first define entropy
solution and then prove that there is an entropy solution of problem (0.2) through a convergence
process which involves the sequence (An (un ))n . Furthermore, we will see when this entropy
solution is a weak solution and when we may take u as a test function in the weak formulation
of the Dirichlet problem (0.2).
D EFINITION 3.1. – We will say that a function u ∈ T01,2 (Ω) is an entropy solution of (0.2) if
b(x, u, ∇u) ∈ L1 (Ω) and
  
 
a(x, u) · ∇u · ∇Tk [u − ϕ] + b(x, u, ∇u)Tk [u − ϕ] = f Tk [u − ϕ]
Ω Ω Ω

holds for all ϕ ∈ H01 (Ω) ∩ L∞ (Ω).

∗∗
T HEOREM 3.2. – There exists u ∈ T01,2 (Ω) such that A(u) ∈ H01 (Ω) ∩ Lm (Ω),
b(x, u, ∇u) ∈ L1 (Ω) and it is an entropy solution of (0.2).
Proof. – Consider, for instance, the following approximating problems:
 
(3.1) −div a(x, un ) · ∇un + b(x, un , ∇un ) = Tn f, in Ω; un = 0, on ∂Ω.

We know, by Theorem 2.4, that there exists un ∈ H01 (Ω) ∩ L∞ (Ω), which is a weak solution
of (3.1). We shall see that this sequence (un )n converges, in some sense, to the entropy solution.
The proof will be based on four lemmata.
L EMMA 3.3. – There exists c1 > 0 such that

   (N+2)/N
 
(3.2) ∇A(un )2  c1 |f | 2N/(N+2)

{|A(un )|>k} {|A(un )|>k}

for all k
0 and all n ∈ N.

In particular, taking k = 0, the sequence (A(un ))n is bounded in H01 (Ω).
Proof. – Taking Φ(s) = Gk (A(s)) in Lemma 2.2(iii), one deduces
 
  

∇Gk (A(un ))2  c2 (Tn f )Gk A(un ) .
Ω Ω

It follows from this, and Hölder’s and Sobolev’s inequalities, that

 

 

∇Gk A(un ) 2  c2 |Tn f | · Gk A(un ) 
Ω Ω
928 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

 1/2∗   (N+2)/2N

 ∗
 c2 Gk A(un ) 2 · |Tn f |2N/(N+2)
Ω {|A(un )|>k}
 1/2   (N+2)/2N


 c3 ∇Gk A(un ) 2 · |f | 2N/(N+2)

Ω {|A(un )|>k}

so that Lemma 3.3 is proved. ✷

As a consequence of (3.2), we can extract a subsequence (still denoted by (un )n ), such that
A(un ) + w weakly in H01 (Ω). Moreover, by Rellich–Kondrachov’s theorem, we may assume
that A(un ) → w almost everywhere in Ω. Since A is strictly increasing, defining u = A−1 (w)
we get a measurable function u such that

(3.3) un → u almost everywhere in Ω

and
(3.4) A(un ) + A(u) weakly in H01 (Ω).
∗∗
Using the results of [8], from inequality (3.2), it follows that A(u) ∈ Lm (Ω).
L EMMA 3.4. –
(3.5) A(un ) → A(u) in H01 (Ω).
In other words,

N
(3.6) α(un )∇un → α(u)∇u in L2 (Ω) .
Proof. – To begin with we have to see two facts. The first one is to prove that



(3.7) ∇Tk A(un ) − Th A(u) 2  ωn,h + ηn (h, k),
Ω

where limn→∞ ηn (h, k) = 0 for any h, k > 0 and limh→∞ limn→∞ ωn,h = 0.
To see it, we apply Lemma 2.2(i) with v = Tk (A(un ) − Th (A(un )))+ to obtain

 
+
eγ (un ) a(x, un ) · ∇un · ∇Tk A(un ) − Th A(u)
Ω


+
(3.8)  eγ (un ) |Tn f |Tk A(un ) − Th A(u)
Ω

and Lemma 2.2(ii) with v = −Tk (A(un ) − Th (A(un )))− to get


 
−
− e−γ (un ) a(x, un ) · ∇un · ∇Tk A(un ) − Th A(u)
Ω


−
(3.9)  e−γ (un ) |Tn f |Tk A(un ) − Th A(u) .
Ω

Now, let δ = e− sup |γ | and observe that (H2) implies |ξ |2  α(s) [a(x, s) · ξ ] · ξ ,
1
so that
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 929



δ ∇Tk A(un ) − Th A(u) 2
Ω
 
a(x, un )


δ · ∇Tk A(un ) − Th A(u) · ∇Tk A(un ) − Th A(u)
α(un )
{A(un )−Th A(u)0}
 
a(x, un )


+δ · ∇Tk A(un ) − Th A(u) · ∇Tk A(un ) − Th A(u)
α(un )
{A(un )−Th A(u)<0}
 
a(x, un )

+
 e · ∇Tk A(un ) − Th A(u) · ∇Tk A(un ) − Th A(u)
γ (un )
α(un )
Ω
 
−γ (un ) a(x, un )


−
− e · ∇Tk A(un ) − Th A(u) · ∇Tk A(un ) − Th A(u) .
α(un )
Ω

Then



∇Tk A(un ) − Th A(u) 2
Ω

1  
+
 eγ (un ) a(x, un ) · ∇un · ∇Tk A(un ) − Th A(u)
δ
Ω
 
1 a(x, un )
+
− eγ (un ) · ∇Th A(u) · ∇Tk A(un ) − Th A(u)
δ α(un )
Ω

1  
−
− e−γ (un ) a(x, un ) · ∇un · ∇Tk A(un ) − Th A(u)
δ
Ω
 
1 a(x, un )
−
+ e−γ (un ) · ∇Th A(u) · ∇Tk A(un ) − Th A(u) .
δ α(un )
Ω

Thus, taking into account (3.8) and (3.9), there exists c4 > 0 such that



∇Tk A(un ) − Th A(u) 2
Ω



 c4 |Tn f | · Tk A(un ) − Th A(u) 
Ω

|a(x, un )|   

+ c4 ∇Th A(u) · ∇Tk A(un ) − Th A(u) 
α(un )
Ω
 
  |a(x, un )|   

 c4 |f | · A(un ) − Th A(u) + c4 ∇Th A(u) · ∇Tk A(un ) − Th A(u) 
α(un )
Ω Ω
= ωn,h + ηn (h, k).
We next go on estimating these integrals. On the one hand, it follows from (3.4) and the
2N
Sobolev imbedding that A(un ) + A(u) weakly in L N−2 (Ω) and from (3.3) that A(un ) → A(u)
930 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

almost everywhere in Ω. Thus,


 

A(un ) − Th A(u)  + A(u) − Th A(u)  2N
weakly in L N−2 (Ω).

2N
Since we also have f ∈ L N+2 (Ω), it yields
 

 

lim |f | · A(un ) − Th A(u)  = |f | · A(u) − Th A(u) ,
n→∞
Ω Ω

so that limh→∞ limn→∞ Ω |f | · |A(un ) − Th (A(u))| = 0.

On the other hand, note that

|a(x, un )|   

∇Th A(u) · ∇Tk A(un ) − Th A(u) 
α(un )
Ω

|a(x, un )|   

= ∇Th A(u) · ∇Tk A(un ) − Th A(u) 
α(un )
{|A(un )−Th A(u)|<k}

|a(x, un )|   

= ∇Th A(u) · ∇Tk A(un ) − Th A(u) ,
α(un )
{|un |<A−1 (k+h)}

and, in this integration set, the function |a(x, un)|/α(un ) is bounded, by (H1). Hence, we also
deduce from (3.4) that

|a(x, un )|   

lim ∇Th A(u) · ∇Tk A(un ) − Th A(u) 
n→∞ α(un )
Ω

|a(x, u)|   

= ∇Th A(u) · ∇Tk A(u) − Th A(u)  = 0.
α(u)
Ω

Therefore, (3.7) is proved.

The second claim is easier. We will see that, for each h > 0,



(3.10) ∇Gk A(un ) − Th A(u) 2  εk (h),
Ω

where limk→∞ εk (h) = 0.

To do this, fix h > 0 and observe that

  
A(un ) − Th A(u)  > k implies A(un ) > k − h.

Then
 

 

∇Gk A(un ) − Th A(u) 2 = ∇ A(un ) − Th A(u) 2
Ω {|A(un )−Th (A(u))|>k}



 ∇ A(un ) − Th A(u) 2
{|A(un )|>k−h}
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 931
 
   
2 ∇A(un )2 + 2 ∇A(u)2 .
{|A(un )|>k−h} {|A(un )|>k−h}

Hence, by (3.2), we obtain




∇Gk A(un ) − Th A(u) 2
Ω
  (N+2)/N 
 
 c5 |f | 2N/(N+2)
+2 ∇A(u)2
{|A(un )|>k−h} {|A(un )|>k−h}

and so it converges to 0 when k goes to infinity, uniformly on n.

Now, having in mind (3.2), (3.7) and (3.10); we will proceed to prove (3.5). Let ε > 0 be fixed.
Consider h > 0 satisfying


   2
(3.11) ∇A(u) − ∇Th A(u) 2 = ∇A(u)2 < ε
2 16
{|A(u)|>h}

and limn→∞ ωn,h < ε2 /8, ωn,h being that of (3.7). Since h have already been choosen,
claim (3.10) implies that there is k > 0 such that


 
 2
(3.12) ∇Gk A(un ) − Th A(u) 2 = ∇Gk A(un ) − Th A(u) 2 < ε
2 16
Ω

for all n ∈ N. Recall that limn→∞ ωn,h < ε2 /8 and, also by claim (3.7), we have
limn→∞ ηn (h, k) = 0; so n0 ∈ N can be found such that if n
n0 , then ωn,h < ε2 /8 and
ηn (h, k) < ε2 /8. Hence,


 
 2
(3.13) ∇Tk A(un ) − Th A(u) 2 = ∇Tk A(un ) − Th A(u) 2 < ε
2 4
Ω

for all n
n0 . Finally, since




A(un ) − A(u) = Tk A(un ) − Th A(u) + Gk A(un ) − Th A(u) − A(u) − Th A(u) ,

it follows from (3.11), (3.12) and (3.13) that

 
∇A(un ) − ∇A(u) < ε + ε + ε = ε
2 2 4 4

for all n
n0 . Therefore, A(un ) → A(u) strongly in H01 (Ω) and Lemma 3.4 is completely
proved. ✷
L EMMA 3.5. – The sequence (un )n has the following properties:
(i) ∇un → ∇u in measure in Ω so that, up to a subsequence,

(3.14) ∇un → ∇u almost everywhere in Ω.

932 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

(ii) For every k > 0,

(3.15) T k un → T k u in H01 (Ω).
As a consequence, Tk u ∈ H01 (Ω) for all k > 0 and so u ∈ T01,2 (Ω).
Proof. – (i) It follows from (3.6) that α(un )∇un → α(u)∇u in measure in Ω. Since (3.3)
implies 1/α(un ) → 1/α(u) in measure in Ω, we conclude that ∇un → ∇u in measure.
(ii) Let k > 0 and, taking into account the continuity of α, denote αk = min{α(s): |s|  k}.
Note that (3.6) may be written as

α(un ) α(u)
N
∇un → ∇u in L2 (Ω) .
αk αk
From this observation and the inequalities

α(un )2 α(un )2
|∇Tk un |2  |∇T k u n | 2
 |∇un |2 ,
αk2 αk2
we deduce that the sequence (|∇Tk un |2 )n is equi-integrable. Hence, it follows from (3.14) and
Vitali’s convergence theorem that (3.15) holds true. ✷
L EMMA 3.6. –
(3.16) β(un )|∇un |2 → β(u)|∇u|2 in L1 (Ω)
and it follows from (H3), (3.3) and (3.14), that

(3.17) b(x, un , ∇un ) → b(x, u, ∇u) in L1 (Ω).

Proof. – Observe that, by (3.3) and (3.14), β(un )|∇un |2 → β(u)|∇u|2 almost everywhere
in Ω. Thus, to get (3.16) we only have to check the equi-integrability of the sequence and apply
Vitali’s convergence theorem.
To proof the equi-integrability, consider the function Φ(s) = γ (Gk (s) + k(sign s)) −
γ (k(sign s)) and observe that then α(s)Φ  (s) = β(s)χ{|s|k} . Thus, taking Φ in Lemma 2.1(iii),
there is c6 > 0 such that
 
β(un )|∇un |  c6
2
|f |.
{|un |k} {|un |k}
Let ε > 0 and fix k > 0 satisfying

ε
β(un )|∇un |2 <
2
{|un |k}

for all n ∈ N. Letting

βk > max{β(s): |s|  k}, by (3.15), there exists δ > 0 such that if n ∈ N
and |A| < δ, then A |∇Tk un |2 < 2βε k . Therefore,
  
β(un )|∇un |2 = β(un )|∇un |2 + β(un )|∇un |2
A A∩{|un |<k} A∩{|un |k}
 
ε ε
 βk |∇Tk un |2 + β(un )|∇un |2 < βk + = ε,
2βk 2
A {|un |k}
as required. ✷
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 933

End of the proof of Theorem 3.2. – We only have to show that u is an entropy solution of (0.2).
Note first that u ∈ T01,2 (Ω), by Lemma 3.5, and that b(x, u, ∇u) ∈ L1 (Ω), by Lemma 3.6.
Let ϕ ∈ H01 (Ω)∩L∞ (Ω) and consider v = Tk [un −ϕ] as test function in the weak formulation
of (3.1). Then
  
 
a(x, un ) · ∇un · ∇Tk [un − ϕ] + b(x, un , ∇un )Tk [un − ϕ] = (Tn f )Tk [un − ϕ].
Ω Ω Ω

We point out that, in the first integral, we may consider the integration set as {|un | < ϕ∞ + k}
without loss of generality; observe that, in this integration set, the functions |a(x, un)| are
bounded in L∞ (Ω), by (H1). Now, it follows from (3.3), (3.14), (3.15) and (3.17) that we may
take limit as n tends to +∞ obtaining
  
 
a(x, u) · ∇u · ∇Tk [u − ϕ] + b(x, u, ∇u)Tk [u − ϕ] = f Tk [u − ϕ]
Ω Ω Ω

and proving that u is an entropy solution of (0.2).

P ROPOSITION 3.7. – Let u be an entropy solution of (0.2). If a(x, u) · ∇u ∈ L2 (Ω), then u is
a weak solution.
As a consequence, u is a weak solution if |a(x, s)|  λα(s) for some λ > 0.
Proof. – Let v ∈ H01 (Ω) ∩ L∞ (Ω) and take ϕ = Th u − v in the formulation of entropy solution
of (0.2). Then
  
 
a(x, u) · ∇u · ∇Tk [u − Th u + v] + b(x, u, ∇u)Tk [u − Th u + v] = f Tk [u − Th u + v].
Ω Ω Ω

Now, as h goes to ∞, Lebesgue’s dominated convergence theorem implies

  
 
a(x, u) · ∇u · ∇Tk v + b(x, u, ∇u)Tk v = f Tk v.
Ω Ω Ω

Finally, it is enough to take k > v∞ to obtain

  
 
a(x, u) · ∇u · ∇v + b(x, u, ∇u)v = f v.
Ω Ω Ω

Hence, u is a weak solution. The second part follows from ∇A(u) = α(u)∇u ∈ L2 (Ω). ✷
P ROPOSITION 3.8. – Suppose that β(s)|s|  λα(s) for some λ > 0. Let u be an entropy
solution of (0.2). If f u ∈ L1 (Ω), then u may be taken as test function in the weak formulation
of (0.2) obtaining the following energy type equality:
  
 
a(x, u) · ∇u · ∇u + b(x, u, ∇u)u = f u.
Ω Ω Ω
934 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

Proof. – First of all, we will prove that [a(x, u) · ∇u] · ∇u ∈ L1 (Ω). Let h > k >
j > 0. Observe that it follows from β(s)/α(s)  λ/|s| and the continuity of β/α that
γ  = β/α ∈ L∞ (Ω); consequently the functions eγ (u)Tj u+ and e−γ (u) Tj u− belong to H01 (Ω) ∩
L∞ (Ω). Taking ϕ = Th (u) − eγ (u) Tj (u+ ) in the entropy formulation of (0.2), we get
 
     
a(x, u) · ∇u · ∇Tk [u − Th u] + a(x, u) · ∇u · ∇Tk u − Th u + eγ (u)Tj u+
{u
0} {u>0}

 
+ b(x, u, ∇u)Tk u − Th u + eγ (u) Tj u+
Ω

 
= f Tk u − Th u + eγ (u)Tj u+ .
Ω

The first two integrals are nonnegative: so that we may drop the first one and apply Fatou’s lemma
to the other as h tends to infinity. In the others two integrals we may apply Lebesgue’s dominated
convergence theorem as h goes to infinity. Thus, it yields
  
       
a(x, u) · ∇u · ∇Tk eγ (u) Tj u+ + b(x, u, ∇u)Tk eγ (u) Tj u+  f Tk eγ (u) Tj u+ .
{u>0} Ω Ω

Since eγ (u)Tj u+ is a bounded function, when k is big enough, we obtain

 
γ (u) 
   
e γ (u)Tj u a(x, u) · ∇u · ∇u + eγ (u) a(x, u) · ∇u · ∇Tj u+
+

Ω Ω

+
+ b(x, u, ∇u)e γ (u)
Tj u
Ω

 f eγ (u)Tj u+ .
Ω

Reasoning in the same way as Lemma 2.2(i), we deduce that there is c7 > 0 such that
 
   
(3.18) a(x, u) · ∇Tj u+ · ∇Tj u+  c7 f T j u + .
Ω Ω

On the other hand, taking ϕ = Th (u) + e−γ (u) Tj (u− ) in the entropy formulation of (0.2) and
having now in mind Lemma 2.2(ii), arguments similar to that above imply that there exists c8 > 0
such that
 
 
(3.19) a(x, u) · ∇Tj u · ∇Tj u  c8 |f Tj u− |.
− −

Ω Ω

Adding up (3.18) and (3.19), we conclude that there exists c9 > 0 such that
 
 
a(x, u) · ∇Tj u · ∇Tj u  c9 |f Tj u|,
Ω Ω
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 935

and so, letting j tend to infinity, by Fatou’s lemma in the left-hand side and Lebesgue’s theorem
in the right one, we deduce that
 
 
a(x, u) · ∇u · ∇u  c9 |f u|.
Ω Ω

Hence, it follows from f u ∈ L1 (Ω) that [a(x, u) · ∇u] · ∇u ∈ L1 (Ω).

Taking ϕ = 0 in the entropy formulation of (0.2), it yields
  
 
(3.20) a(x, u) · ∇Tk u · ∇Tk u + b(x, u, ∇u)Tk u = f Tk u.
Ω Ω Ω

Now, observe that the following inequalities hold:

|f Tk u|  |f u| ∈ L1 (Ω),

   
a(x, u) · ∇Tk u · ∇Tk u  a(x, u) · ∇u · ∇u ∈ L1 (Ω)
and
   
b(x, u, ∇u)u  β(u)|u| · |∇u|2  λα(u)|∇u|2  λ a(x, u) · ∇u · ∇u ∈ L1 (Ω).

Therefore, by Lebesgue’s dominated convergence theorem, it follows from (3.20) that

  
 
a(x, u) · ∇u · ∇u + b(x, u, ∇u)u = f u,
Ω Ω Ω

as desired. ✷
Up to now, we have not really seen the summability of the solution u, but that of A(u). In the
last result, we study this regularity of the solution when α and β are concrete functions.
P ROPOSITION 3.9. – Let θ > −1 and suppose that α(s) = (1 + |s|)θ and β(s) =
(1 + |s|)θ−1−ε , with ε > 0.
If u is an entropy solution of (0.2), then u ∈ Lr (Ω) and |∇u| ∈ Lq (Ω), where
r = Nm(1 + θ )/(N − 2m) and q = min[2, Nm(1 + θ )/(N − m(1 − θ ))]; thus, u belongs to
a Sobolev space if m
N/(N + 1 + θ (N − 1)).
Moreover, u can be taken as test function in the weak formulation of problem (0.2) if
m
N(2 + θ )/(N(1 + θ ) + 2).
Proof. – Obviously, α(s) = (1 + |s|)θ implies |A(s)| = 1+θ 1
(1 + |s|)1+θ . Since
∗∗
A(u) ∈ L (Ω), by Theorem 3.2, we have (1 + |u|)
m 1+θ ∈ L Nm/(N−2m) (Ω) and then
u ∈ LNm(1+θ)/(N−2m) (Ω).
We now pass to see the regularity of the gradient ∇u. On the one hand, θ
0 implies α(s)
1
and so |∇u|  α(u)|∇u|. Moreover, since α(u)|∇u| = |∇(A(u))| ∈ L2 (Ω), by Theorem 3.2, it
follows that |∇u| ∈ L2 (Ω). We point out that then Nm(1 + θ )/(N − m(1 − θ ))  2. On the other
hand, if −1 < θ < 0, then we may follow the arguments of [6], Lemma 2.3 to obtain that

Nm(1 + θ )
q = min 2, .
N − m(1 − θ )
936 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

Next, considering m
N(2 + θ )/(N(1 + θ ) + 2), we have

m Nm(1 + θ )
 ,
m−1 N − 2m

so that u ∈ Lm (Ω) and f u ∈ L1 (Ω). By Proposition 3.8, we deduce that u may be taken as test
function. Observe also that when θ
0,

N(2 + θ ) 2N

N(1 + θ ) + 2 N + 2

and hence, in this case, u can always be taken as test function.

4. A priori estimates by symmetrization

An alternative approach in finding a priori estimates for solutions of problem (0.2), like the
ones obtained in Theorem 2.3 and in Lemma 3.3, is based on the so called symmetrization
techniques which go back to the papers by Talenti [22,23] and which have been widely used
in similar contexts (see, for example, [1,15]).
In this section we show how it is possible to use such an approach pointing out that in this way
we can obtain estimates also when f belongs to intermediate spaces (see Remarks 4.5 and 4.8).
We first recall the definition of decreasing rearrangement of a function u. If u is a measurable
function in an open bounded set Ω ⊂ RN , we denote by µu the distribution function of u:
 
µu (t) =  x ∈ Ω: |u| > t , t
0.

The decreasing rearrangement u∗ (s) of u is defined by:

  
u∗ (s) = sup t > 0: µu (t) > s , s ∈ 0, |Ω| ,

and the spherically decreasing rearrangement u# (x) of u is defined by:


u# (x) = u∗ CN |x|N , x ∈ Ω #,

where Ω # is the ball centered at the origin having the same measure of Ω and CN denotes the
measure of the unit ball of RN .
We recall also that for every 1  p < +∞ it results

 |Ω|
 
∗ p
u(x)p dx = u (s) ds
Ω 0

and if u ∈ L∞ (Ω)
uL∞ = u∗ L∞ = u∗ (0).
For an exhaustive treatment of rearrangements we refer, for example, to [17] or [23].
We recall Bliss inequalities (see [22]), which are used to obtain estimates of Lp -norms of
solutions in terms of Lq -norms of the datum f .
 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 937

L EMMA 4.1. – If ϕ(r) is positive for 0 < r < +∞ and 1 < p < q, then
+∞ r q  +∞ q/p
1
ϕ(s) ds r −1+q/p dr  A(N, p, q) ϕ(r)p dr ,
r
0 0 0
+∞ +∞ q  +∞ q/p
p −1+p+p/q
ϕ(s) ds dr  B(N, p, q) ϕ(r) r dr .
0 r 0
Now we prove an a priori estimate for a solution u of (0.2).
P ROPOSITION 4.2. – Let u ∈ L∞ (Ω) be a weak solution of (0.2) in the sense of 2.1, then
there exists a constant M1 > 0 such that:

|Ω|  r 
 ∗ 1 ∗


(4.1) A(u) (s)  M1 2/N
f (σ ) dσ dr, s ∈ 0, |Ω| .
N 2 CN r 2−2/N
s 0

Proof. – Applying Lemma 2.2(iii) to the function Φ(s) = Th [A(s) − Tt (A(s))] we get:
  



α 2 (u)|∇u|2  M1 f Th A(s) − Tt A(s)   M1 h |f |,
{t <|A(u)|
t +h} Ω {|A(u)|>t }

where M1 depends only on the data (M1 = e2 sup |γ | ).

Setting w = A(u) we have:
 
1
(4.2) |∇w|2  M1 |f |.
h
{t <|w|
t +h} {|w|>t }

Schwartz inequality, Fleming–Rishel formula (see [16]) and the isoperimetric inequality give
(see [13])
t
−µw (t)
1  M1 2/N
f ∗ (r) dr
N 2 CN µw (t)2−2/N
0
from which, using the properties of rearrangement (see [22]) we get inequality (4.1).
Remark 4.3. – If we denote by v the solution of the problem

(4.3) −:v = f # in Ω # ; v=0 on ∂Ω # ;

the function v is given by:

|Ω|  s 
−2/N
(4.4) v(x) = N −2 CN s 2/N−2 f ∗ (σ ) dσ ds.
CN |x|N 0

From Lemma 4.2 we deduce that

 ∗

A(u) (s)  M1 v ∗ (s), s ∈ 0, |Ω| ,
938 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

where v ∗ is the decreasing rearrangment of the solution v of (4.3).

Clearly if f ∈ Lm (Ω) with m > N/2 then v is bounded and its maximum value is v(0).
The following two corollaries give uniform estimates on the Lp -norms of solutions of (0.2).
C OROLLARY 4.4. – Let u ∈ L∞ (Ω) be a solution of (0.2) in the sense of 2.1, then if m > N/2
there exists a constant M2 > 0 such that
  
uL∞  max A−1 (M2 ), A−1 (−M2 ) ,

where M2 = ( M1 |Ω|
2/N−1/m
−1
2m−N f L ), A
m
2/N
m is the inverse function of A and M1 is the constant
NCN
defined in Proposition 4.2.
Proof. – Setting w = A(u), inequality (4.1) and Hölder inequality imply:
|Ω| r
   ∗ 1
A(u)  A(u) (0)  M1 f ∗ (σ ) dσ
L∞ (Ω) 2/N
N 2 CN r 2−2/N
0 0
|Ω|  r 1/m
M1 1 ∗ 
 2/N
f (σ ) dσ m
r 1/m
N 2 CN r 2−2/N
0 0
M1 |Ω| 2/N−1/m m
(4.5)  f Lm = M2 .
NCN
2/N 2m − N
Remark 4.5. – We observe that by inequality (4.5) the estimate on the L∞ norm of A(u), and
then on the L∞ norm of u, holds true if

|Ω| r
1
2/N
f ∗ (σ ) dσ < +∞.
N 2 CN r 2−2/N
0 0

The above condition is satisfied if f belongs to the Lorentz space L(N/2, 1).
C OROLLARY 4.6. – Let u ∈ L∞ (Ω) be a solution of (0.2) in the sense of Definition 2.1 then,
if 2N/(N + 2)  m < N/2, there exists a constant M3 > 0 such that:
 
A(u) q  M3 f Lm ,
L

where q = m∗∗ .
Proof. – From inequality (4.1) we have:
 |Ω| |Ω| r  q 1/q
     1
A(u) q =  A(u) ∗  q  M1 ∗
f (σ ) dσ r −2+2/N
dr ds .
L L 2/N
N 2 CN
0 s 0

Using Bliss and Hardy inequality (see [22]) we get

 
A(u) q  M3 f Lm . ✷
L

The following result gives an estimate of A(u) in H01 (Ω).

 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940 939

P ROPOSITION 4.7. – Let u ∈ L∞ (Ω) be a solution of (0.2) in the sense of Definition 2.1, then
A(u) belongs to H01 (Ω) and there exists a constant M4 such that
 
A(u) 1  M4 f  2N/(N+2) .
H L
0

Proof. – Setting w = A(u), from estimate (4.2) we get:

 µw (t )
d
− |∇w|  M1 2
f ∗ (σ ) dσ.
dt
{|w|>t } 0

Moreover it results that (see [22])

 +∞  2
−µw (t)−1+1/N d

|∇w|  2
1/N
|∇w| 2
−dµw (t) .
NC dt
Ω 0 N {|w|>t }

Setting f¯(t) = 1 t
t 0 f ∗ (s) ds ∀t > 0 using Bliss inequality we get:
 +∞ µw (t ) 2
−µw (t)−1+1/N ∗


|∇w| 2
 M12 1/N
f (σ ) dσ −dµw (t)
NCN
Ω 0 0
|Ω| r 2
M12 ∗
= 2/N
r 1/N−1
f (σ ) dσ dr
N 2 CN
0 0
|Ω|
M12
2 dr
= 2/N
r (2+N)/2N f¯(r)  M4 f 22N/(N+2)
N 2 CN r
0
and the thesis follows. ✷
Remark 4.8. – We observe that the estimate on the H01 norm of A(u) holds if

|Ω|

2 dr
r (2+N)/2N f¯(r) < +∞.
r
0

The above condition is satisfied if f belongs to the Lorentz space L(2N/(N + 2), 2).

Acknowledgements

Part of this work was done while the second author visited the Dipartimento di Matematica of
the Università di Roma I, and he would like to thank the very warm hospitality enjoyed there.
The second author has partially been supported by the Spanish DGICYT, Proyecto PB98-1442.

REFERENCES

[1] A. A LVINO, V. F ERONE , G. T ROMBETTI, A priori estimates for a class of non uniformly elliptic
equations, Atti Sem. Mat. Fis. Univ. Modena 46 (1998) 381–391.
940 L. BOCCARDO ET AL. / J. Math. Pures Appl. 80 (2001) 919–940

[2] P H . B ÉNILAN, L. B OCCARDO, T H . G ALLOUËT , R. G ARIEPY, M. P IERRE , J.L. VÁZQUEZ , An L1

theory of existence and uniqueness of solutions of nonlinear elliptic equations, Ann. Scuola Norm.
Sup. Pisa Cl. Sci. 22 (1995) 241–273.
[3] A. B ENSOUSSAN, L. B OCCARDO, F. M URAT , On a nonlinear partial differential equation having
natural growth terms and unbounded solutions, Ann. Inst. H. Poincaré Anal. Non Linéaire 5 (1988)
347–364.
[4] L. B OCCARDO, Positive solutions for some quasilinear elliptic equations with natural growth, Rend.
Accad. Lincei 11 (2000) 31–39.
[5] L. B OCCARDO, H. B RÉZIS, Some remarks on a class of elliptic equations with degenerate coercivity,
Preprint.
[6] L. B OCCARDO, A. DALL’AGLIO, L. O RSINA, Existence and regularity results for some elliptic
equations with degenerate coercivity, Atti Sem. Mat. Fis. Univ. Modena 46 (1998) 51–81.
[7] L. B OCCARDO, T. G ALLOUET , Strongly nonlinear elliptic equations having natural growth terms and
L1 data, Nonlinear Anal. 19 (1992) 573–579.
[8] L. B OCCARDO, D. G IACHETTI, Ls -regularity of solutions of some nonlinear elliptic problems, Pre-
print.
[9] L. B OCCARDO, F. M URAT , J.P. P UEL , Existence de solutions non bornées pour certaines équations
quasi-lineaires, Portugal. Math. 41 (1982) 507–534.
[10] L. B OCCARDO, F. M URAT , J.P. P UEL , Résultats d’existence pour certains problèmes elliptiques
quasilinéaires, Ann. Scuola Norm. Pisa 11 (1984) 213–285.
[11] L. B OCCARDO, F. M URAT , J.M. P UEL , L∞ estimate for some nonlinear elliptic partial differential
equations and application to an existence result, SIAM J. Math. Anal. 2 (1992) 326–333.
[12] L. B OCCARDO, L. O RSINA, Existence and regularity of minima for integral functionals non coercive
in the energy space, Ann. Scuola Norm. Sup. Pisa Cl. Sci. 25 (1997) 95–130.
[13] E. D E G IORGI, Su una teoria generale della misura (r − 1)-dimensionale in uno spazio ad r
dimensioni, Ann. Mat. Pura Appl. 36 (1954) 191–213.
[14] V. F ERONE , F. M URAT , Nonlinear problems having natural growth in the gradient: an existence result
when the source term is small, to appear in Nonlinear Anal.
[15] V. F ERONE , M.R. P OSTERARO, On a class of quasilinear elliptic equations with quadratic growth in
the gradient, Nonlinear Anal. 20 (1993) 703–711.
[16] W. F LEMING, R. R ISHEL , An integral formula for total gradient variation, Arch. Math. 11 (1960)
218–222.
[17] B. K AWOHL , Rearrangements and Convexity of Level Sets in P.D.E., Lecture Notes in Math.,
Vol. 1150, Springer, Berlin, 1985.
[18] J. L ERAY, J.L. L IONS, Quelques résultats de Višik sur les problèmes elliptiques non linéaires par les
méthodes de Minty–Browder, Bull. Soc. Math. France 93 (1965) 97–107.
[19] A. P ORRETTA, Uniqueness and homogenization for a class of non coercive operators in divergence
form, Atti Sem. Mat. Fis. Univ. Modena 46 (1998) 915–936.
[20] S. S EGURA DE L EÓN, Existence and uniqueness for L1 data of some elliptic equations with natural
growth, to appear in Adv. Differential Equations.
[21] G. S TAMPACCHIA, Le problème de Dirichlet pour les équations elliptiques du second ordre
à coefficients discontinus, Ann. Inst. Fourier, Grenoble 15 (1965) 189–258.
[22] G. TALENTI, Elliptic equations and rearrangements, Ann. Scuola Norm. Sup. Pisa (4) 3 (1976) 697–
718.
[23] G. TALENTI, Linear elliptic P.D.E.’s: level sets, rearrangements and a priori estimates of solutions,
Boll. Un. Mat. Ital. (6) 4-B (1985) 917–949.
[24] C. T ROMBETTI, Existence and regularity for a class of non-uniformly elliptic equations in two
dimension, Differential Integral Equations 13 (2000) 687–706.
[25] C. T ROMBETTI, Non-uniformly elliptic equations with natural growth in the gradient, Preprint.