ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC

GRADIENT TERM AND A CHANGING SIGN DATUM

DANIELA GIACHETTI, FRANCESCO PETITTA AND SERGIO SEGURA DE LEÓN

Abstract. In this paper we study a singular elliptic problem whose model is


−∆u = |∇u|2 + f (x) , in Ω ;
|u|θ
u = 0 , on ∂Ω ;

where θ ∈ (0, 1) and f ∈ Lm (Ω), with m ≥ N 2

. We do not assume any sign
condition on the lower order term, nor assume the datum f has a constant
sign.
We carefully define the meaning of solution to this problem giving sense to
the gradient term where u = 0, and prove the existence of such a solution. We
also discuss related questions as the existence of solutions when the datum f
is less regular or the boundedness of the solutions when the datum f ∈ Lm (Ω)
with m > N 2
.

1. Introduction
The systematic study of second order equations having a gradient term with
natural growth was initiated by Boccardo, Murat and Puel in the 80’s of last century
(see [11], [12] and [13]). This gradient term also depends on the solution, for instance
it can be written as g(u)|∇u|2 , but always in a continuous way. Recently existence
of solutions of problems whose model is
( 2
−∆u = |∇u|
|u| θ + f (x) , in Ω ;
(1.1)
u = 0, on ∂Ω ;
where θ > 0 and Ω is a bounded open set in RN , has attracted the attention of
several authors (see for example [2], [4], [5], [6], [7], [8], [16], [3], [1]; other related
problems are studied in [10] and [17].) The problem presents a lower order term
which is singular in the u-variable and has a natural (quadratic) growth in the ∇u-
variable. The interest in studying this kind of problems relies, first of all, on the
fact that the equation looks like a simplified version of the formal Euler’s equation
for a functional of the type
Z Z
α−1 2
I[u] = |u| u|∇u| − fu
Ω Ω
with α ∈ (0, 1).
Another motivation occurs by considering equations of the type
ut − ∆(|u|m−1 u) = |∇u|2 + f ,

Key words and phrases. Dirichlet problem, gradient term, singularity at zero, data with non-
constant sign
2010 Mathematics Subject Classification: 35J75, 35J25.
1
2 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

with m > 1, which represents a model of gas flow in porous media. If we consider
steady states solutions and we perform a change of unknown |u|m−1 u = v, we get
an equation with singular behaviour in v, with quadratic growth in the ∇v-variable.
The papers we quoted before deal with different situations depending on the
exponent θ of the singularity, on the sign and size of the lower order term. Existence
and nonexistence of solutions in H01 (Ω) or Hloc 1
(Ω), depending on the regularity
of the datum f (x) (which can induce bounded or unbounded solutions) and other
related questions are considered. Anyway, all the previous known results are strictly
confined to the case of nonnegative data f (x), since they are mainly based on the
strong maximum principle. In other words, the sign of the datum guarantees that
the possible solutions do not cross the singularity; this is due to the fact that u ≡ 0
is, in a certain sense, a subsolution to the problem.
Dealing with data that do not have constant sign adds then some new extra
difficulties to the study of this kind of equations. First of all, since the method of
sub/supersolutions does not apply in this case, we need both to obtain new a priori
estimates and to perform a deeper analysis near the singularity u = 0 to study the
2
singular quotient |∇u||u|θ
.
Moreover, a very basic remark on the meaning of the solution is in order. Re-
ferring again to the model problem (1.1) and to the case f ≥ 0, we observe that
the definition of solution is completely clear if u > 0 in Ω. In our situation, where
f can change its sign, the solution u can vanishes inside Ω. This fact is not only
a possibility, it really occurs as shown in Proposition 4.2 below. If we look for
H01 (Ω)–solutions, an indeterminate quotient appears since, by Stampacchia’s the-
orem, |∇u| = 0 on the set {u = 0}. Therefore, we have to carefully define the
meaning of solution and it is done in Definition 2.1 and Lemma 2.2 below. There,
we introduce a suitable notion of solution that ensures us that u ∈ H01 (Ω) and
|∇u|2
|u|θ
∈ L1 (Ω).
In the present paper, we present a complete account on the existence of finite
energy solutions for problems modelled by (1.1) with general, possibly changing-
sign, data f ∈ Lm (Ω), m ≥ N2 and θ ∈ (0, 1). We will obtain a priori estimates by
means of a generalized Cole–Hopf change of unknown. Recall that, if a lower order
term appears in the form g(u)|∇u|2 , test functions involving terms like exp(γ(u)),
where γ(s) is a primitive function of g(s), are often used in order to get a priori
estimates (see [14], [19] and [18]). Observe that, if θ ∈ (0, 1), then the function
g(s) = |s|1θ is an L1 –function near the singularity s = 0 so that exp(γ(s)) is well–
defined. Obviously, this fact does not occur if θ ≥ 1.
Nevertheless, we point out that our restriction on θ is not technical: indeed, even
if θ = 1 and f ≥ 0, solutions do not belong, in general, to H01 (Ω) anymore, nor the
gradient term to L1 (Ω), as shown in [3]. In other words, if the singularity is too
strong (e.g. |s|1θ , with θ ≥ 1), then there is no room for a solution of finite energy
to satisfy the boundary condition and the solution must loose its regularity.
The paper is organized as follows. Section 2 is devoted to the hypotheses and
the statements of the results. Section 3 deals with the proof of the main theorem.
Section 4 contains further results on boundedness of solutions in the case f ∈
Lm (Ω), m > N2 and on stability with respect to the lower order term; it also
provides examples and possible extensions.
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 3

2. Hypotheses and statements of results

Let us state our main assumptions. Let Ω be an open bounded set in RN (N ≥ 3).
We will deal with the following problem
(
−div(a(x, u, ∇u)) = b(x, u, ∇u) + f (x) , in Ω ;
(2.2)
u = 0, on ∂Ω .
The function
a(x, s, ξ) : Ω × R × RN → RN
satisfies the Carathéodory conditions (i.e. a(x, ·, ·) is continuous for a.e. x ∈ Ω and
a(·, s, ξ) is measurable for any s ∈ R, ξ ∈ RN ) and there exist some constants α > 0
and ν > 0 such that
(2.3) a(x, s, ξ) · ξ ≥ α|ξ|2 ,
(2.4) |a(x, s, ξ)| ≤ ν|ξ|,
(2.5) (a(x, s, ξ) − a(x, s, η)) · (ξ − η) > 0 ;
N
for all ξ, η ∈ R , with ξ 6= η, for all s ∈ R and for almost all x ∈ Ω.
The function
b(x, s, ξ) : Ω × R\{0} × RN → R
also satisfies the Carathéodory conditions and there exists a nonnegative continuous
function g : R\{0} → [0, +∞) such that
(2.6) |b(x, s, ξ)| ≤ g(s)|ξ|2 ;
for all ξ ∈ RN , for all s ∈ R\{0} and for almost all x ∈ Ω. Moreover,
(2.7) lim g(s) = 0
|s|→∞
Λ
and there exist constants Λ, s0 > 0 and θ ∈ (0, 1) such that g(s) = |s|θ
for all
|s| ≤ s0 .
Remark 2.1. We explicitly observe that, without loss of generality, we can choose g
to be nonincreasing in [0, +∞[ and to be nondecreasing in ]−∞, 0]. Indeed, changing
the value of s0 if necessary, it is not difficult to define a continuous g : R\{0} → R
satisfying the same hypotheses of g and moreover
• g(s) ≥ g(s) for all s ∈ R.
• g is nonincreasing in [0, +∞[ and nondecreasing in ] − ∞, 0].
As far as the datum f is concerned, it satisfies
N
(2.8) f (x) ∈ Lm (Ω), m≥ ,
2
while no sign condition is assumed (cfr. with [4], [8], [16] and references therein).
Let us point out that, under the general assumption (2.7), the summability
requested to f is optimal as showed in [18]. In Section 4 we will show how this
assumption can be relaxed depending on the behaviour of the lower order term.
We remark that, as we look for solutions u ∈ H01 (Ω), the equation in (2.2)
involves an indeterminate quotient on {u = 0}, since |b(x, u, ∇u)|χ{|u|≤s0 } ≤
Λ
|∇u|2 and |∇u| = 0 on the set {u = 0}, by Stampacchia’s Theorem. To
|u|θ
clarify this situation, we define
4 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

θ
Definition 2.2. If u and |u|1− 2 belong to H01 (Ω), we define
|∇u|2 4 θ

θ
= 2
|∇ |u|1− 2 |2 .
|u| (2 − θ)
|∇u|2
Observe that, by definition, always belongs to L1 (Ω). Moreover, as a
|u|θ
consequence of Stampacchia’s Theorem, we obtain
|∇u|2
= 0 a.e. in {u = 0} .
|u|θ
As a consequence of (2.6), we may extend b(x, s, ξ) to s = 0 (only when s = u and
ξ = ∇u) and define
(2.9) b(x, u, ∇u) = 0 a.e. in {u = 0} .
1
Hence, b(x, u, ∇u) ∈ L (Ω).
Remark 2.3. We would like to explicitly stress that solutions satisfying |{u =
0}| > 0 can actually occur. For instance consider the function defined in B2 (0), the
ball of radius 2 of RN , by
− 1
(
e 1−|x|2 , if |x| ≤ 1 ;
w(x) =
0, if 1 < |x| ≤ 2 .

An easy computation (using that θ < 1) shows that there exists f ∈ C ∞ (B 2 (0))
such that w solves ( 2
−∆w = |∇w|
|w|θ
+ f , in B2 (0) ;
w = 0, on ∂B2 (0) .

Definition 2.4. A weak solution to problem (2.2) is a function u ∈ H01 (Ω) satis-
θ
fying |u|1− 2 ∈ H01 (Ω) (so that b(x, u, ∇u) ∈ L1 (Ω)) and
Z Z Z
a(x, u, ∇u) · ∇v = b(x, u, ∇u)v + f v,
Ω Ω Ω

for any v ∈ H01 (Ω) ∞

∩ L (Ω).
In order to check that a function u ∈ H01 (Ω) is actually solution to problem (2.2),
θ
we will have to see |u|1− 2 ∈ H01 (Ω). To this aim the following simple claim will
be applied. Although its proof is similar to that of Lemma 2.1 in [17], we sketch
it for the sake of completeness. Here and below we will use the following auxiliary
functions: for any s ∈ R we consider the standard truncation function defined by
Tk (s) = max(−k, min(s, k)), while we denote Gk (s) = s − Tk (s).
Lemma 2.5. Let u ∈ H01 (Ω). If g(u)|∇u|2 is integrable on {u 6= 0}, then
θ
|u|1− 2 ∈ H01 (Ω) .
Moreover, b(x, u, ∇u) is integrable on Ω, and
Z Z
b(x, u, ∇u) = b(x, u, ∇u) .
Ω {u6=0}
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 5

Proof. Observe that

|∇u|2 1
Z Z
≤ g(u)|∇u|2 ≤ C ,
{0<|u|≤s0 } ( n1 + |u|)θ Λ {0<|u|≤s0 }

for all n ∈ N. In other words,

Z  1 1− θ2  1 1− θ2  2
∇ + Ts0 (|u|) − ≤C,
Ω n n
 1− θ2  1− θ2
for all n ∈ N. Hence, n1 + Ts0 (|u|) − n1 is bounded in H01 (Ω) and, up
to subsequences, there exists v ∈ H01 (Ω) such that
1 1− θ2  1 1− θ2
+ Ts0 (|u|) − ⇀v
n n
weakly in H01 (Ω). Obviously, passing to a subsequence if necessary, we get
1 1− θ2  1 1− θ2
1− θ2
v(x) = lim + Ts0 (|u(x)|) − = Ts0 (|u(x)|) a.e. in Ω ,
n→∞ n n

1− θ2
so that Ts0 (|u|) ∈ H01 (Ω). In particular, since 0 < θ < 1, using the chain rule
for Sobolev spaces we have
2

1− θ2  2−θ 2
θ
1− θ2
∇Ts0 (|u|) = ∇ Ts0 (|u|) = Ts0 (|u|) 2 ∇ Ts0 (|u|) ,
2−θ
a.e. on Ω.
Note that, on account of Stampacchia’s theorem, the two functions which appear at
the right-hand side and at the left-hand side are both a.e. zero on the set {u = 0}
and so we get

1− θ2 2 − θ ∇Ts0 (|u|)
∇ Ts0 (|u|) =
θ a.e. on Ω.
2 T (|u|) 2
s0

1− θ2 θ

Therefore, denoting k = s0 , we have just seen that Tk |u|1− 2 ∈ H01 (Ω).
Moreover,  
θ
θ ∇|u|
∇Tk |u|1− 2 = 1 − χ{|u|<s0 } , a.e. on Ω.
2 |u| θ2
θ

Since Gk |s|1− 2 defines a Lipschitz continuous function, it follows from u ∈
θ

H01 (Ω) that Gk |u|1− 2 ∈ H01 (Ω) and
 
1− θ2

θ ∇|u|
∇Gk |u| = 1− χ{|u|≥s0 } , a.e. on Ω.
2 |u| θ2
θ θ
θ

Therefore, |u|1− 2 = Tk |u|1− 2 + Gk |u|1− 2 ∈ H01 (Ω) and
2
|∇u|2

1− θ2 2 θ
∇|u| = 1− .
2 |u|θ
|∇u|2
Hence, the term is well–defined and belongs to L1 (Ω). As a consequence,
|u|θ
g(u)|∇u|2 is well–defined in {u = 0}, where g(u)|∇u|2 = 0 a.e., and g(u)|∇u|2 ∈
L1 (Ω). Thus, by assumption (2.6), recalling (2.9), we deduce the second assertion
of our Lemma. 
6 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

Our main result is the following

Theorem 2.1. There exist a weak solution u ∈ H01 (Ω) to problem (2.2)

3. Proof of Theorem 2.1

3.1. Approximating Problems. We shall take approximating problems without
singularities. To this end, we will consider truncating continuous functions bn of b.
Since b is not assumed to be an even function with respect to s, our truncation will
not be standard. So, for any n ∈ N, we define the following bounded sequence of
functions
(
1+t 1 1−t −1 t
(3.10) bn (x, s, ξ) := 2 b(x, n , ξ) + 2 b(x, n , ξ) , if s = n , |t| ≤ 1 ;
b(x, s, ξ) , if |s| > n1 ;

for any ξ ∈ RN , and a.e. x ∈ RN . Moreover, let fn := Tn (f ) and consider

(
−div(a(x, un , ∇un )) = bn (x, un , ∇un ) + fn (x) , in Ω ;
(3.11)
un = 0 , on ∂Ω .
A bounded weak solution to problem (3.11) does exist as proved in [18]. That is
there exists un ∈ H01 (Ω) ∩ L∞ (Ω) such that
Z Z Z
(3.12) a(x, un , ∇un ) · ∇v = bn (x, un , ∇un )v + fn v,
Ω Ω Ω

for any v ∈ H01 (Ω) ∞

∩ L (Ω).
For n ≥ 1/s0 , we define the following auxiliary functions:
(
Λnθ , if |s| ≤ n1
(3.13) gn (s) :=
g(s) , otherwise,

s s
1
Z Z
(3.14) γn (s) = gn (σ) dσ, and Ψn (s) = e|γn (σ)| dσ.
α 0 0

Observe that γn (s) is Lipschitz continuous, while Ψn (s) is locally Lipschitz contin-
uous and it satisfies
(3.15) |Ψn (s)| ≥ |s| , for all s ∈ R .
Moreover, thanks to (2.6),
(3.16) |bn (x, s, ξ)| ≤ gn (s)|ξ|2 .
Of course, there is some connection among all these functions, which we want to
highlight. Let
1 s
Z Z s
γ(s) = g(σ) dσ, and Ψ(s) = e|γ(σ)| dσ ,
α 0 0
so that it also holds
(3.17) |Ψ(s)| ≥ |s| , for all s ∈ R .
Observe that
(3.18) 0 ≤ γn (s)sign (s) ≤ γ(s)sign (s) , for all s ∈ R and all n ∈ N ,
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 7

and that we also have

Λ θ
0 ≤ |γ(s)| − |γn (s)| ≤ 1−θ
,
αn 1−θ
Λ θ
for all s ∈ R. It follows from lim = 0 that
n→∞ αn1−θ 1 − θ
e|γn (s)| ≤ e|γ(s)| ≤ Cn e|γn (s)| , ∀ s ∈ R,
(3.19)
|Ψn (s)| ≤ |Ψ(s)| ≤ Cn |Ψn (s)| , ∀ s ∈ R,
where Cn satisfies limn→∞ Cn = 1. On the other hand, since g vanishes at infinity,
by L’Hôpital’s rule we have
e|γ(s)|
(3.20) lim = 0,
|s|→∞ |Ψ(s)|

so that for any ε > 0 there exists a constant C such that

e|γ(s)| ≤ ε|Ψ(s)| + C , ∀ s ∈ R.
Thanks to (3.19), we deduce that, given ε > 0, there is an only constant C satisfying
(3.21) e|γn (s)| ≤ ε|Ψn (s)| + C , ∀ n ∈ N, s ∈ R ;
we will use this kind of bound in what follows.
Let us specify some useful notation we will use from now on. If not differently
stated, the symbol C will indicate a positive constant, only dependent on the data,
whose value may change line by line. Moreover, the symbol ω(ε), ω(n) will denote
any quantity that vanishes as the argument goes to its natural limit (that is ε → 0,
n → ∞).
3.2. Estimate on both Ψn (un ) and un in H01 (Ω). We take e|γn (un )| Ψn (un ) as
test in (3.12) to obtain, using (3.16) and the fact that both γn (un ) and Ψn (un )
have the same sign as un
Z Z
|γn (un )| 2
gn (un )e |Ψn (un )||∇un | + α e2|γn (un )| |∇un |2
Ω Ω
Z Z
≤ gn (un )e|γn (un )| |Ψn (un )||∇un |2 + |f |e|γn (un )| |Ψn (un )| ,
Ω Ω
that is,
Z Z
(3.22) α |∇Ψn (un )|2 ≤ |f |e|γn (un )| |Ψn (un )| .
Ω Ω
Using first (3.21) and then Young’s inequality, we get
Z Z Z
(3.23) |f |e|γn (un )| |Ψn (un )| ≤ ε |f ||Ψn (un )|2 + C |f ||Ψn (un )|
Ω Ω
Z Ω Z
2
≤ 2ε |f ||Ψn (un )| + C |f | .
Ω Ω
Now, by Hölder’s inequality, the summability of f and Sobolev’s inequality, we
obtain (choosing a suitable ε)
Z Z (N −2)/N α
Z
2 2N (N −2)
2ε |f ||Ψn (un )| ≤ 2εkf kN/2 |Ψn (un )| ≤ |∇Ψn (un )|2 ,
Ω Ω 2 Ω
8 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

which, by (3.23), implies

α
Z Z Z
|f |e|γn (un )| |Ψn (un )| ≤ |∇Ψn (un )|2 + C |f | .
Ω 2 Ω Ω

Going back to (3.22) we deduce

Z
(3.24) |∇Ψn (un )|2 ≤ C , for all n ∈ N ,
Ω

and
Z
(3.25) |f |e|γn (un )| |Ψn (un )| ≤ C , for all n ∈ N .
Ω

Moreover, Young’s inequality implies

1 1
Z Z Z
|γn (un )| 2|γn (un )|
|f |e ≤ |f |e + |f | ,
Ω 2 Ω 2 Ω
which, due to (3.21), becomes
Z
(3.26) |f |e|γn (un )| ≤ C , for all n ∈ N .
Ω

On the other hand, notice that, by (3.21) again,

2N
(3.27) the sequence e|γn (un )| is bounded in L N −2 (Ω) .
Thus, since
|∇un |2 ≤ e2|γn (un )| |∇un |2 = |∇Ψn (un )|2 ,
we also have
kun kH01 (Ω) ≤ C , for all n ∈ N .
Therefore, up to subsequences, there exists u ∈ H01 (Ω) such that un ⇀ u weakly
in H01 (Ω), un → u strongly in L2 (Ω) and a.e. on Ω.

3.3. Estimate of bn (x, un , ∇un ) in L1 (Ω). Here we want to prove an L1 –bound

for the lower order term bn (x, un , ∇un ). We take (e|γn (un )| − 1)sign (un ) as test
function in (3.12) and we use (2.3) to get
Z
gn (un )e|γn (un )| |∇un |2
Ω
Z Z
≤ |bn (x, un , ∇un )|(e|γn (un )| − 1) + |f |(e|γn (un )| − 1)
Ω Ω
Z Z
≤ gn (un )|∇un |2 (e|γn (un )| − 1) + |f |(e|γn (un )| − 1) ,
Ω Ω

that implies, using (3.16) and (3.26),

Z Z Z
2
(3.28) |bn (x, un , ∇un )| ≤ gn (un )|∇un | ≤ |f |e|γn (un )| ≤ C .
Ω Ω Ω
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 9

3.4. Near the singularity. Here we want to prove that, for any ε > 0
Z
(3.29) lim sup |bn (x, un , ∇un )| = 0 .
ε→0 n {|un |≤ε}

To this end, consider the function


|γn (un )|


 e
 − 1 sign un , if |un | ≤ ε ;
v = e|γn (ε)| − 1 , if un > ε ;
1 − e|γn (−ε)| ,

if un < −ε ;


and observe that, by (3.19),

|v| ≤ max{e|γn (ε)| − 1, e|γn (−ε)| − 1} ≤ max{e|γ(ε)| − 1, e|γ(−ε)| − 1} ,
that is |v| ≤ ω(ε) uniformly in n. Choosing v as test function in (3.12) and applying
(2.3), we obtain
Z
e|γn (un )| gn (un )|∇un |2
{|un |≤ε}
1
Z
≤ gn (un )e|γn (un )| a(x, un , ∇un ) · ∇un
α {|un |≤ε}
Z Z
≤ |bn (x, un , ∇un )| |v| + |f | |v| .
Ω Ω

Hence, by (3.16),
Z Z
(3.30) |bn (x, un , ∇un )| ≤ e|γn (un )| gn (un )|∇un |2
{|un |≤ε} {|un |≤ε}
hZ Z i
≤ ω(ε) |bn (x, un , ∇un )| + |f | .
Ω Ω
Since the terms in brackets are uniformly bounded, by the previous step, it yields
(3.29).

3.5. Far from the singularity. Here we want to prove

Z
(3.31) lim sup |bn (x, un , ∇un )| = 0.
k→∞ n {|un |>k}

We consider
e|γn (Gk (un ))| − 1 sign un

as test function in (3.12); applying (2.3) and (2.6) we obtain

Z
gn (Gk (un ))e|γn (Gk (un ))| |∇un |2
{|un |>k}
Z Z
|bn (x, un , ∇un )| e|γn (Gk (un ))| − 1 + |f | e|γn (Gk (un ))| − 1



≤
{|un |>k} {|un |>k}
Z Z
≤ gn (un )e|γn (Gk (un ))| |∇un |2 − |bn (x, un , ∇un )|
{|un |>k} {|un |>k}
Z
|f | e|γn (Gk (un ))| − 1


+
{|un |>k}
10 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

As we said in Remark 2.1, we may assume that g is nondecreasing on ] − ∞, 0]

and nonincreasing on [0, +∞[. It follows from the inequality g(un ) ≤ g(Gk (un )) on
{|un | > k} that we may cancel two terms, and so
Z Z
(3.32) |bn (x, un , ∇un )| ≤ |f | (e|γn (Gk (un ))| − 1) .
{|un |>k} {|un |>k}

Having in mind (3.25), we set C = supn Ω |f | |Ψn (un )|e|γn (un )| . Then, due to
R

(3.15),
Z Z
|bn (x, un , ∇un )| ≤ |f | e|γn (un )|
{|un |>k} {|un |>k}
1 C
Z
≤ |f | |Ψn (un )| e|γn (un )| ≤ ,
min{|Ψn (k)|, |Ψn (−k)|} {|un |>k} k
which gives (3.31).
3.6. Strong convergence of truncations. Here we want to prove that, for each
k > 0, ∇Tk (un ) strongly converges to ∇Tk (u) in Lp (Ω; RN ).
First we take eγn (un ) (Tk (un ) − Tk (u))+ as test function in (3.12), to get
1
Z
a(x, un , ∇un ) · ∇un gn (un )eγn (un ) (Tk (un ) − Tk (u))+
α Ω
Z
+ eγn (un ) a(x, un , ∇un ) · ∇(Tk (un ) − Tk (u))+
Ω
Z Z
≤ gn (un )|∇un |2 eγn (un ) (Tk (un ) − Tk (u))+ + |f |eγn (un ) (Tk (un ) − Tk (u))+ ,
Ω Ω
that is, using (2.3) and simplifying,
Z
(3.33) eγn (un ) a(x, un , ∇un ) · ∇(Tk (un ) − Tk (u))+
Ω
Z
≤ |f |eγn (un ) (Tk (un ) − Tk (u))+ .
Ω
The right hand side of the previous inequality goes to zero as n diverges since
N 2N
f ∈ L 2 (Ω), the sequence eγn (un ) is bounded in L N −2 (Ω), by (3.27), and Tk (un )
2N
converges to Tk (u) strongly in L N −2 (Ω), due to the pointwise convergence. So that
we can write
Z
eγn (un ) a(x, un , ∇un ) · ∇(Tk (un ) − Tk (u))+
{|un |≤k}
Z .
γn (un )
≤ ω(n) + e a(x, un , ∇un ) · ∇Tk (u)
{un >k}

Now observe that

|eγn (un ) a(x, un , ∇un )| ≤ νe|γn (un )| |∇un | = ν|∇Ψn (un )| ,
it implies, thanks to the estimate on Ψn (un ),
Z
eγn (un ) a(x, un , ∇un ) · ∇Tk (u) = ω(n).
{un >k}
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 11

Therefore, gathering together the previous estimates we have

Z
(3.34) eγn (un ) a(x, un , ∇un ) · ∇(Tk (un ) − Tk (u))+ ≤ ω(n) .
{|un |≤k}

On the other hand, since |a(x, un , ∇Tk (u))| ≤ ν|∇Tk (u)| ∈ L2 (Ω), the sequence
γn (un )
e χ{|un |≤k} is uniformly bounded in L∞ (Ω) and Tk (un ) ⇀ Tk (u) weakly in
H01 (Ω), it follows that
Z
(3.35) eγn (un ) a(x, un , ∇Tk (u)) · ∇(Tk (un ) − Tk (u))+ = ω(n) .
{|un |≤k}

Now we can subtract (3.34) and (3.35) to obtain

Z
eγn (un ) (a(x, un , ∇Tk (un ))−a(x, un , ∇Tk (u)))·∇(Tk (un )−Tk (u))+ ≤ ω(n) .
{|un |≤k}

Recall that, by (3.18), we have that |γn (s)| ≤ max{γ(k), −γ(−k)} for all s ∈ [−k, k]
and consequently inf eγn (s) ≥ min{e−γ(k) , eγ(−k) } > 0. Applying this fact and
{|s|≤k}
the monotonicity condition (2.5), we deduce
Z
(a(x, un , ∇Tk (un )) − a(x, un , ∇Tk (u))) · ∇(Tk (un ) − Tk (u))+ ≤ ω(n) .
{|un |≤k}
Hence, we get
Z
(a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))) · ∇(Tk (un ) − Tk (u))+
Ω
Z
≤ ω(n) + a(x, k, ∇Tk (u)) · ∇Tk (u) = ω(n) ,
{un >k}

the last equality is due to Lebesgue’s Theorem and the following inequalities
Z Z
0≤ a(x, k, ∇Tk (u)) · ∇Tk (u) ≤ ν |∇Tk (u)|2 .
{un >k} {un >k}

To the deal with the negative part, we may follow a similar argument, using now
−e−γn (un ) (Tk (un ) − Tk (u))− as test function in (3.12). Adding both, the positive
and the negative part, we obtain that
Z
(3.36) (a(x, Tk (un ), ∇Tk (un )) − a(x, Tk (un ), ∇Tk (u))) · ∇(Tk (un ) − Tk (u))
Ω
tends to 0 as n goes to ∞. A result by Browder (see [15] or [13]), implies that
∇Tk (un ) → ∇Tk (u) , strongly in L2 (Ω; RN ) .
A diagonal argument now supplies us the pointwise convergence of the gradients
(3.37) ∇un (x) → ∇u(x) , a.e. in Ω .
Three important consequences of this fact are
(3.38) a(x, un (x), ∇un (x)) → a(x, u(x), ∇u(x)) , a.e. in Ω ,
(3.39) bn (x, un (x), ∇un (x)) → b(x, u(x), ∇u(x)) , a.e. in {u 6= 0} ,
(3.40) gn (un (x))|∇un (x)|2 → g(u(x))|∇u(x)|2 , a.e. in {u 6= 0} .
It follows from this last convergence, (3.28) and Fatou’s Lemma, that
g(u(x))|∇u(x)|2 ∈ L1 ({u 6= 0}) ,
12 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

θ
from where, thanks to Lemma 2.5, we obtain |u|1− 2 ∈ H01 (Ω), b(x, u, ∇u) ∈ L1 (Ω)
and
Z Z
(3.41) b(x, u, ∇u) = b(x, u, ∇u) .
{u6=0} Ω

3.7. Equi–integrability of bn (x, un , ∇un ). Consider a measurable set E ⊂ Ω and

δ > 0. Applying (3.29) and (3.31), given δ > 0, we may find ε, k > 0 satisfying
δ
Z Z
|bn (x, un , ∇un )| + |bn (x, un , ∇un )| ≤ .
{|un |<ε} {|un |>k} 2
Thus, it yields
δ
Z Z
|bn (x, un , ∇un )| ≤ + |bn (x, un , ∇un )|
E 2 E∩{ε≤|un |≤k}
δ δ
Z Z
≤ + g(un )|∇un |2 ≤ + sup g(s) |∇Tk (un )|2 ,
2 E∩{ε≤|un |≤k} 2 ε≤|s|≤k E

and, when |E| is small enough, the last term becomes less than 2δ since ∇Tk (un )
converges strongly in L2 (Ω; RN ). Therefore, the sequence bn (x, un , ∇un ) is equi–
integrable.

3.8. Passage to the limit. In order to prove that u is a weak solution to (2.2),
we fix v ∈ H01 (Ω) ∩ L∞ (Ω) and consider it as test function in (3.12). Then
Z Z Z
(3.42) a(x, un , ∇un ) · ∇v = bn (x, un , ∇un )v + fn v .
Ω Ω Ω

It is easy to pass to the limit in the last term, but two facts are needed to handle
the other terms. On the one hand,
a(x, un , ∇un ) ⇀ a(x, u, ∇u)
weakly in L2 (Ω; RN ). This is due to our estimate of un in H01 (Ω), (2.4) and (3.38).
So that we may pass to the limit in the second order term.
On the other hand, the previous step and (3.39) imply
bn (x, un , ∇un ) → b(x, u, ∇u) , strongly in L1 ({u 6= 0}) .
This fact has as consequence that we may pass to the limit in the gradient term;
indeed, given δ > 0 and using (3.29), we may find ε > 0 satisfying
Z
kvk∞ |bn (x, un , ∇un )| < δ/2 ,
{|un |≤ε}

for all n ∈ N. Then, it follows from Fatou’s Lemma that

Z
kvk∞ |b(x, u, ∇u)| < δ/2 .
{|u|≤ε}∩{u6=0}

Hence, applying the previous estimates,

Z Z
bn (x, un , ∇un )v + b(x, u, ∇u)v < δ ,
{|un |≤ε} {|u|≤ε}∩{u6=0}
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 13

from where it yields

Z Z
bn (x, un , ∇un )v − b(x, u, ∇u)v
Ω {u6=0}
Z Z
≤ bn (x, un , ∇un )v − b(x, u, ∇u)v + δ .
{|un |≥ε} {|u|≥ε}

Since v ∈ L∞ (Ω) and the sequence bn (x, un , ∇un )χ{|un |≥ε} is equi–integrable, to
see that the absolute value of the right hand side tends to 0, we only have to check
the pointwise convergence. We split

bn (x, un , ∇un )χ{|un |≥ε} = bn (x, un , ∇un )χ{|un |≥ε}∩{|u|>ε}

+ bn (x, un , ∇un )χ{|un |≥ε}∩{|u|<ε} + bn (x, un , ∇un )χ{|un |≥ε}∩{|u|=ε} ,

the first term converges pointwise to b(x, u, ∇u)χ{|u|>ε} (observe that is equal to
b(x, u, ∇u)χ{|u|≥ε} by (2.6) and Stampacchia’s Theorem), while the second one
tends to 0. Regarding the third one we have

|bn (x, un , ∇un )χ{|un |≥ε}∩{|u|=ε} | ≤ gn (un )|∇un |2 χ{|u|=ε} → g(u)|∇u|2 χ{|u|=ε}

that vanishes by Stampacchia’s Theorem. Thus, we deduce that

Z Z
lim bn (x, un , ∇un )v = b(x, u, ∇u)v ,
n→∞ {|un |≥ε} {|u|≥ε}

and, therefore,
Z Z
lim sup bn (x, un , ∇un )v − b(x, u, ∇u)v ≤ δ .
n→∞ Ω {u6=0}

Since δ > 0 is arbitrary and having (3.41) in mind, we obtain

Z Z Z
lim bn (x, un , ∇un )v = b(x, u, ∇u)v = b(x, u, ∇u)v.
n→∞ Ω {u6=0} Ω

Passing to the limit in (3.42), we have proved that u is a solution to problem

(2.2).

4. Further remarks, extensions and examples

4.1. Remarks on the estimates satisfied by u. We explicitly point out that
the solution we have found satisfies many of the estimates proved to un in the proof
of Theorem 2.1. For instance, it is easy to see that
Z Z
(4.43) |b(x, u, ∇u)| ≤ |f |e|γ(u)|
Ω Ω

holds. Indeed, observe that in (3.28) we have proved

Z Z
|bn (x, un , ∇un )| ≤ |f |e|γn (un )|
Ω Ω
14 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

for all n ∈ N. Taking into account that un and ∇un pointwise converge to u and
∇u, respectively, we apply in the left hand side Fatou’s Lemma to obtain
Z Z
|b(x, u, ∇u)| = |b(x, u, ∇u)|
Ω {u6=0}
Z Z
≤ lim |bn (x, un , ∇un )| ≤ lim |bn (x, un , ∇un )| .
n→∞ {u6=0} n→∞ Ω

On the other hand, it follows from (3.27), Hölder’s inequality and the pointwise
convergence that e|γn (un )| → e|γ(u)| strongly in LN/(N −2) (Ω). Thus, it follows from
f ∈ LN/2 (Ω) that Z Z
|f |e|γ(u)| = lim |f |e|γn (un )| .
Ω n→∞ Ω
Other inequalities that also hold true are
Z hZ Z i
(4.44) |b(x, u, ∇u)| ≤ ω(ε) |b(x, u, ∇u)| + |f |
{|u|≤ε} Ω Ω
Z Z
(4.45) |b(x, u, ∇u)| ≤ |f | e|γ(u)| ,
{|u|≥k} {|u|≥k}

letting n go to infinity in (3.30) and (3.32), respectively. There are other type of
estimates that can be adapted, namely, those which appear in the proof of the
strong convergence of truncations. For instance, it follows from (3.33) that
Z
(4.46) eγ(u) a(x, u, ∇u) · ∇(Tk (u) − Tk (w))+
Ω
Z
≤ |f |eγ(u) (Tk (u) − Tk (w))+
Ω

holds for every w ∈ H01 (Ω).

4.2. Bounded solutions. Throughout this paper, we have assumed that f belongs
to LN/2 (Ω); if the datum has a greater summability, the boundedness of the solution
is guaranteed.
N
Proposition 4.1. Assume that f ∈ Lm (Ω), with m > 2. Then there exists a
bounded weak solution to problem (2.2).
To prove it, consider again the function given by Gk (s) = s − Tk (s) and take
e|γn (un )| Gk Ψn (un )

as test function in (3.12). Since this function lives far from the singularity, we may
now follow the proof of Theorem 3.1 in [18] and deduce that kΨn (un )k∞ is bounded
by a constant that only depends on the function g and the parameters m, kf km ,
N , and |Ω|. Hence, Ψ(u) ∈ L∞ (Ω) and, by (3.17), u ∈ L∞ (Ω).

4.3. Stability with respect to the lower order term. In this subsection we
provide a stability result with respect to perturbations of the lower order term.
The result is important by his own; moreover, in the next subsection we show, as a
consequence of this result, that there always exist solutions with no constant sign.
Let
bρ (x, s, ξ), b(x, s, ξ) : Ω × R\{0} × RN → R
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 15

be Carathéodory functions satisfying

lim bρ (x, s, ξ) = b(x, s, ξ)
ρ→0

for all ξ ∈ RN , for all s ∈ R\{0} and for almost all x ∈ Ω. Moreover, for fixed
ρ > 0, there exist nonnegative functions gρ , g : R\{0} → [0, +∞) such that
(4.47) |bρ (x, s, ξ)| ≤ gρ (s)|ξ|2 , |b(x, s, ξ)| ≤ g(s)|ξ|2 ;
for all ξ ∈ RN , for all s ∈ R\{0} and for almost all x ∈ Ω; and there exist constants
Λ
Λρ , Λ ≥ 0, s0 > 0 and θρ , θ ∈ (0, 1) such that gρ (s) = |s|θρρ and g(s) = |s|Λθ for all
|s| ≤ s0 . We assume that, as ρ → 0, θρ → θ and Λρ → Λ.
These hypotheses imply that γρ (s) → γ(s) and Ψρ (s) → Ψ(s) uniformly on
[−s0 , s0 ], where γρ and Ψρ are the auxiliary functions associated with each gρ . We
also assume that
(1) gρ (s) → g(s) local uniformly on (−∞, −s0 ] ∪ [s0 , +∞) as ρ goes to ∞ .
e|γρ (s)|
(2) lim = 0, uniformly with respect to ρ .
|s|→+∞ Ψρ (s)

The last condition seems to be a little cumbersome. A simple case where it is

certainly satisfied is when gρ (s) = g(s) for |s| large enough. We have essentially
applied in this way in the proof of Theorem 2.1, and so will be used in the example
of the following subsection.
Due to our assumptions, we can derive that, for every ǫ > 0 there exists C > 0,
not depending on ρ, satisfying
(4.48) e|γρ (s)| ≤ ǫΨρ (s) + C , for all s ∈ R .
Finally, consider f ∈ Lm (Ω) with m ≥ N2 and uρ as the solution to problem
(
−div(a(x, uρ , ∇uρ )) = bρ (x, uρ , ∇uρ ) + f (x) , in Ω ;
(4.49)
uρ = 0 , on ∂Ω ,
given in Theorem 2.1.
Theorem 4.1. There exists u ∈ H01 (Ω) such that (up to subsequences)
uρ ⇀ u , weakly in H01 (Ω) ,
uρ → u , a.e. in Ω ,
∇uρ → ∇u , a.e. in Ω ,
and u is a weak solution of problem
(
−div(a(x, u, ∇u)) = b(x, u, ∇u) + f (x) , in Ω ;
u = 0, on ∂Ω ,
Moreover, bρ (x, uρ , ∇uρ ) strongly converges to b(x, u, ∇u) in L1 (Ω).
Sketch of the Proof. The proof of this result is based on a careful adaptation of the
same steps in the proof of Theorem 2.1. The key point is that, thanks to (4.48), the
estimate on Ψn (un ) in the proof of Theorem 2.1 does depend on α, |Ω|, kf k N2
L (Ω)
and SN , but it does not depend on ρ.
Now a remark concerning the test functions used in the proof is in order. It is not
clear that, in each step, we may take the corresponding test function. The reason
16 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

lies in the singularity at 0 of functions gρ that does not hold in the approximating
functions gn . To overcome this difficulty, we can apply the estimates deduced in
Subsection 4.1.
So that, by arguing as in the proof of Theorem 2.1 we easily obtain the bound
in H01 (Ω) for uρ and so, up to subsequences, a weak limit u ∈ H01 (Ω) is found.
Moreover, we also obtain, as in (3.25) and (3.26), that
Z
|f |e|γρ (uρ )| |Ψρ (uρ )| ≤ C
Ω
Z
|f |e|γρ (uρ )| ≤ C ,
Ω

C being a positive constant not depending on ρ. From this last fact and (4.43), we
derive the estimate of bρ (x, uρ , ∇uρ ) in L1 (Ω). It follows from the estimate (4.44)
that, for ǫ > 0, Z
bρ (x, uρ , ∇uρ ) ≤ ω(ǫ) .
{|uρ |≤ǫ}

The lower order term can be studied far from the singularity by using (4.45) and
so, for k > 0, we get
C
Z
bρ (x, uρ , ∇uρ ) ≤ ,
{|uρ |≥k} k
where C is a positive constant non depending on ρ. We can also apply estimates
like (4.46) to prove the strong convergence of Tk (uρ ) to Tk (u) and, by a diagonal
argument, deduce that ∇uρ tends to ∇u pointwise. The only actual difference relies
in proving the equi–integrability of the lower order term where we use again the
local uniform convergence of gρ to prove that
Z
sup gρ (s) |∇Tk (uρ )|2 ≤ Cǫ,k ω(|E|).
ǫ≤s≤k E

This way we get the equi–integrability of the lower order term and this allow us to
pass to the limit in the weak formulation for uρ and to conclude the proof. 

4.4. Example of a sign-changing solution. It is worth to give an example of a

solution which changes his sign. For the sake of simplicity we take as a model the
problem
(
−∆u = g(u)|∇u|2 + f (x) , in Ω ;
(4.50)
u = 0, on ∂Ω ,
with a nonnegative g satisfying the same assumptions as in (2.6) and f ∈ L∞ (Ω).
The proof is based on the maximum principle together with the stability result
given in Theorem 4.1.
Proposition 4.2. There exist g and f such that the solution of problem (4.50) has
no constant sign.
Proof. Let us fix a g satisfying our assumptions and such that g(s) = 0 for |s| ≥ s1
for some s1 > s0 . Consider v ∈ C02 (Ω) such that v changes his sign. Then, by the
maximum principle, the function
f := −∆v
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 17

changes his sign. Now consider u as the solution, given by Theorem 2.1, of problem
(4.50). Since g is nonnegative, it follows that u turns out to satisfy

−∆u ≥ f, u ∈ H01 (Ω),

in D′ (Ω). So that by comparison, u ≥ v. In particular there exists a set E ⊂ Ω of

positive measure such that u > 0 on E. Now, suppose by contradiction that u ≥ 0
on Ω and, for any fixed ρ, consider the family of problems
(
−∆uρ = ρg(uρ )|∇uρ |2 + f (x) , in Ω ;
(4.51)
uρ = 0 , on ∂Ω .

Reasoning as before we deduce that, for any ρ, uρ ≥ v on Ω. In particular we can

assume uρ ≥ 0 on Ω since, if this is not the case, the proof is concluded with f and
ρg(s) as data.
Therefore, applying Theorem 4.1 we can deduce that

ρg(uρ )|∇uρ |2 → 0, in L1 (Ω),

as ρ goes to zero, and, since the solution to the limit problem is unique, we get

0 ≤ uρ −→ v a.e. on Ω

which is a contradiction since v changes his sign.



4.5. Weakening the hypotheses on g. Throughout this paper we have assumed

that g(s) → 0 as |s| → +∞. However, this hypothesis can be changed by being g
bounded, if kf kN/2 is small enough. We remark that we only apply that g(s) → 0
to obtain (3.21) and it is just used (in an essential way) to deduce an estimate of
Ψn (un ) in H01 (Ω).

Proposition 4.3. Assume, instead of (2.7), that there exists M > 0 satisfying
lim sup|s|→∞ g(s) ≤ M and, besides (2.8), that kf kN/2 < MαS 2 , SN denoting the
N
Sobolev constant. Then there exists a weak solution to problem (2.2).

Proof. Consider the same approximating problems (3.11). To check the estimate of
Ψn (un ) in H01 (Ω), first observe that condition lim sup|s|→∞ g(s) ≤ M implies that
there exists a constant C > 0 such that

e|γn (s)| ≤ M |Ψn (s)| + C , ∀ n ∈ N, s ∈ R;

to see it, just recall the argument used to derive (3.21).

Taking e|γn (un )| Ψn (un ) as test function in (3.12) and dropping nonnegative terms
we also obtain
Z Z
(4.52) α |∇Ψn (un )|2 ≤ |f |e|γn (un )| |Ψn (un )| .
Ω Ω

Then we reason as follows. Hölder’s and Sobolev’s inequalities imply

Z Z
2 2
M |f ||Ψn (un )| ≤ M SN kf kN/2 |∇Ψn (un )|2 .
Ω Ω
18 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

Thus, (4.52) becomes

Z Z Z
α |∇Ψn (un )|2 ≤ M |f ||Ψn (un )|2 + C |f ||Ψn (un )|
Ω Ω
ZΩ Z
2 2
≤ M SN kf kN/2 |∇Ψn (un )| + C |f ||Ψn (un )|
Ω Ω
and it yields
Z Z
2 2
(α − M SN kf kN/2 ) |∇Ψ(un )| ≤ C |f ||Ψn (un )| .
Ω Ω
It easily follows the estimate of Ψn (un ) in H01 (Ω).
Next we may follow the same proof that the one of Theorem 2.1. 
4.6. Taking less regular data. In this subsection, we will assume extra hypothe-
ses on g that allow us to consider less regular data. In the following result, we will
assume
- There exists
(4.53) lim g(s)|s| .
|s|→∞

- There exist constants λ > 0 and M ≥ 0 satisfying

e|γ(s)|
(4.54) lim =M.
|s|→∞ |s|λ

Remark 4.4. Condition (4.54) seems a little bit strange, since it is not a direct
assumption on g. Let us see what is the behaviour of g to satisfy this condition.
(1) Conditions (4.53) and (4.54) imply that lim|s|→∞ g(s)|s| = αλ.
|γ(s)|
λ
(2) If g(s) = |s| for all s ≥ s0 , then (4.53) and (4.54) hold, since e|s|λ/α is
constant.
(3) One could think that condition (4.54) holds for every function g satisfying
lim|s|→∞ g(s)|s| = λα. As the function given by g(s) = λα 1
|s| + |s| log |s| (for s
large enough) shows, it is not true.
(4) The limit occurring in (4.54) vanishes, when it exists, for every function g
such that lim|s|→∞ g(s)|s| < αλ.
(5) In some cases function g satisfies condition (4.54) for all λ (and so M = 0).
Obviously, this is the case when g is summable at infinity. An instance of a
non summable function satisfying condition (4.54) for all λ is the function
1
given by g(s) = |s| log |s| , for s large enough.

Proposition 4.5. Assume, instead of (2.7), that (4.53) and (4.54) hold and, in-
 ∗ ′
stead of (2.8), that f ∈ Lm (Ω), with m = 2 2λ+1
(λ+1)
= N2N (λ+1)
+2(2λ+1) . Then there
exists a weak solution to problem (2.2).
Proof. In this case, we have to change (3.20) by
e|γn (s)|
lim = M 1/(λ+1) (λ + 1)λ/(λ+1)
|s|→∞ |Ψn (s)|λ/(λ+1)

and so (3.21) becomes

(4.55) e|γn (s)| ≤ C|Ψn (s)|λ/(λ+1) + C for all s ∈ R .
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 19

Z
This inequality is used to estimate f e|γn (un )| Ψn (un ) as follows. By Hölder’s
Ω
inequality, we obtain
Z Z Z
|f |e|γn (un )| |Ψn (un )| ≤ C |f ||Ψn (un )|(2λ+1)/(λ+1) + C |f ||Ψn (un )|
Ω Ω Ω
Z
(2λ+1)/(λ+1)
≤ Ckf km kΨn (un )k2∗ +C |f ||Ψn (un )| .
Ω
2λ+1
Since λ+1 < 2, we may apply Young’s inequality to get
(2λ+1)/(λ+1)
Ckf km kΨn (un )k2∗ 2(λ+1)
≤ ǫkΨn (un )k22∗ + C(ǫ)kf km .
Then, taking e|γn (un )| Ψn (un ) as test function in (3.12), we deduce
Z Z
2 2 2(λ+1)
α |∇Ψn (un )| ≤ ǫkΨn (un )k2∗ + C(ǫ)kf km +C |f ||Ψn (un )| ,
Ω Ω

from where estimates on both Ψn (un ) and un in H01 (Ω)

are obtained. Moreover,
∗
the sequence e|γn (un )| is bounded in L2 (λ+1)/λ (Ω), due to (4.55). Next we may
follow the same proof that the one of Theorem 2.1. 

Observe that N2N (λ+1) 2N

+2(2λ+1) goes to N/2 as λ goes to +∞, while it converges to N +2
as λ goes to 0 that correspond to the case of an integrable g. Thus, the previous
Proposition along with Theorem 2.1 and the following result show that there is
continuity with respect to the summability of the datum.
Proposition 4.6. Assume, instead of (2.7), that g ∈ L1 (R) and, instead of (2.8),
that f ∈ Lm (Ω), with m = (2∗ )′ = N2N
+2 . Then there exists a weak solution to
problem (2.2).
We may easily obtain estimates on both Ψn (un ) and un in H01 (Ω), having in mind
that we now have e|γn (s)| ≤ C for all s ∈ R and this implies, taking e|γn (un )| Ψn (un )
as test function in (3.12), that
Z Z
2
α |∇Ψn (un )| ≤ C |f ||Ψn (un )| .
Ω Ω
The proof now follows the same steps that the one of Theorem 2.1.
Let us finally remark that, in this case in which g ∈ L1 (R), we may want to take
less regular data up to m = 1 by readapting the arguments in [19]. This is certainly
possible, but this would bring us out of our framework of finite energy solutions.

4.7. Lower order terms satisfying a sign condition. In this last subsection,
we deal with a lower order term having the sign condition. Our aim is to show
how the behavior of these type of lower order terms allow us to choose an even less
regular datum f .
For the sake of simplicity, we will consider the model problem
(
−div(a(x, u, ∇u)) + g(u)|∇u|2 = f (x) , in Ω ;
(4.56)
u = 0, on ∂Ω ,
where g satisfies
(4.57) λ|s|1−θ ≤ g(s)s ≤ Λ|s|1−θ for all s ∈ R .
20 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

For a more general lower order term, we would change this condition by
λ|s|1−θ |ξ|2 ≤ −b(x, s, ξ)s ≤ Λ|s|1−θ |ξ|2 for all s ∈ R .
2∗ ′
Proposition 4.7. Assume that (4.57) holds. If f ∈ Lm (Ω), with m =


θ , then
there exists a weak solution to problem (4.56).
Proof. For fixed n we define the continuous functions

g(s) ,
 if |s| ≥ n1 ;
(4.58) gn (s) := ng(1/n)s , if 0 ≤ s < n1 ;
−ng(−1/n)s , if − n1 < s ≤ 0 ;



note that these functions satisfy the same sign condition of g, namely, gn (s)s ≥ 0.
We consider the approximating problems
(
−div(a(x, un , ∇un )) + gn (un )|∇un |2 = Tn (f (x)) , in Ω ;
(4.59)
un = 0 , on ∂Ω .
By [18] (or, alternatively, by applying Theorem 2.1 and Proposition 4.1), we may
find a bounded weak solution un to problem (4.59).
To obtain an estimate on un in H01 (Ω), we first take T1 (un ) as test function.
Dropping nonnegative terms, we get
Z Z Z
α |∇T1 (un )|2 ≤ Tn (f )T1 (un ) ≤ |f | .
Ω Ω Ω
Hence,
1
Z Z
(4.60) |∇un |2 ≤ |f | .
{|un |≤1} α Ω

Now we take (ǫ + |un |)θ Tk (u k

n)
as test function. Disregarding nonnegative terms,
it yields
θ Tk (un ) Tk (un )
Z Z
2
(ǫ + |un |) gn (un )|∇un | ≤ |f |(ǫ + |un |)θ .
Ω k Ω k
Letting ε and k go to 0, we obtain
Z Z
|un |θ |gn (un )||∇un |2 ≤ |f ||un |θ ,
Ω Ω
from here, using (4.57) and the definition of gn , we deduce
1
Z Z
(4.61) |∇un |2 ≤ |f ||un |θ .
{|un |>1} λ Ω
Putting together (4.60) and (4.61), it yields
Z Z
|∇un |2 ≤ C |f ||un |θ + C ,
Ω Ω
from where, using first the Hölder inequality and then the Sobolev one, an estimate
of un in H01 (Ω) can be obtained.
From now on, the proof runs as that of Theorem 2.1 with a suitable simplification.
In order to reproduce the Steps 3.3, 3.4, 3.5 and 3.6 in the proof of Theorem 2.1,
we argue as follows. Consider the following auxiliary function:
1 s
Z
γn (s) = gn (σ) dσ ,
α 0
 ELLIPTIC EQUATIONS HAVING A SINGULAR QUADRATIC GRADIENT TERM 21

and observe that γn (s) ≥ 0 for all s ∈ R.

(1) We take Tk (u
k
n)
as test function and then let k tend to 0 to prove the
1
L –estimate on the lower order term.
(2) We consider

−γn (un )


 1−e
 sign un , if − ε ≤ un ≤ ε ;
v = 1 − e−γn (ε) , if un > ε ;
 −γn (−ε)

e − 1, if un < −ε
as test function to control the singularity on the set {|un | < ε}.
(3) We choose T1 (Gk (un )) (with Gk (s) = s − Tk (s) as before) as test function
to handle the set where un is large. This way we obtain
Z Z
|gn (un )||∇un |2 ≤ |f | .
{|un |>k+1} {|un |>k}

(4) We consider eγn (un ) (Tk (un ) − Tk (u))+ and −e−γn (un ) (Tk (un ) − Tk (u))− as
test functions to check the strong convergence of ∇Tk (un ) in L2 (Ω; RN ).
This is enough to prove that the limit u is a weak solution to (4.56). 
∗
Let us observe that m = ( 2θ )′ converges to 1 as θ goes to 0. That is, in the
limit, we recover the classical nonsingular result of [9].

5. Acknowledgements
We wish to thank the referees for carefully reading this article and suggesting us
some improvements. The second and third authors are partially supported by the
Spanish PNPGC project, reference MTM2008- 03176.

References
[1] B. Abdellaoui, D. Giachetti, I. Peral and M.Walias: “Elliptic problems with nonlinear
terms depending on the gradient and singular on the boundary”, Nonlinear Analysis, 74
(2011), 1355–1371.
[2] D. Arcoya, S. Barile and P.J. Martı́nez-Aparicio: “Singular quasilinear equations
with quadratic growth in the gradient without sign condition”, J. Math. Anal. Appl., 350
(2009), no. 1, 401–408.
[3] D. Arcoya, L. Boccardo, T. Leonori and A. Porretta: ”Some elliptic problems with
singular natural growth lower order terms”, J. Differential Equations, 249 (2010), no. 11,
2771–2795.
[4] D. Arcoya, J. Carmona, T. Leonori, P.J. Martı́nez-Aparicio, L. Orsina and F. Pe-
titta: “Existence and nonexistence of solutions for singular quadratic quasilinear equa-
tions”, J. Differential Equations, 246 (2009), no. 10, 4006–4042.
[5] D. Arcoya, J. Carmona and P.J. Martı́nez-Aparicio: “Elliptic obstacle problems with
natural growth on the gradient and singular nonlinear terms”, Adv. Nonlinear Stud., 7
(2007), 299–317.
[6] D. Arcoya and P.J. Martı́nez-Aparicio: “Quasilinear equations with natural growth”,
Rev. Mat. Iberoamericana, 24 (2008), no. 2, 597–616.
[7] D. Arcoya and S. Segura de León: “Uniqueness of solutions for some elliptic equa-
tions with a quadratic gradient term”, ESAIM: Control, Optimization and the Calculus of
Variations, 16 (2010), 327-336.
[8] L. Boccardo: “Dirichlet problems with singular and gradient quadratic lower order terms”,
ESAIM Control Optim. Calc. Var., 14 (2008), no. 3, 411–426.
[9] L. Boccardo and T. Gallouët: “Strongly nonlinear elliptic equations having natural
growth terms and L1 data”; Nonlinear Anal., 19 (1992), 573–579.
22 DANIELA GIACHETTI, FRANCESCO PETITTA, SERGIO SEGURA DE LEÓN

[10] L. Boccardo, T. Leonori, L. Orsina and F. Petitta: “Quasilinear elliptic equations

with singular quadratic growth terms”, Comm. Contemp. Math., 13 No. 4 (2011) , 607–642.
.
[11] L. Boccardo, F. Murat and J.P. Puel: “Existence de solutions non bornées pour
certaines équations quasi-linéaires”, Portugal. Math., 41 (1982), 507–534.
[12] L. Boccardo, F. Murat and J.P. Puel: “Résultats d’existence pour certains problèmes
elliptiques quasilinéaires”, Ann. Scuola Norm. Sup. Pisa Cl. Sci., (4) 11 no. 2, (1984),
213–235.
[13] L. Boccardo, F. Murat and J.P. Puel: “Existence of bounded solutions for nonlinear
elliptic unilateral problems”, Ann. Mat. Pura Appl. 152 (1988), 183-196.
[14] L. Boccardo, S. Segura de León and C. Trombetti: “Bounded and unbounded solu-
tions for a class of quasi-linear elliptic problems with a quadratic gradient term”, J. Math.
Pures Appl., 80, 9, (2001) 919–940.
[15] F.E. Browder: “Existence theorems for nonlinear partial differential equations”, ”Global
Analysis” (Proc. Sympos. Pre Math., vol XVI, Berkeley, California, 1968), Amer. Math.
Soc. (1970), 671–688.
[16] D. Giachetti and F. Murat: “An elliptic problem with a lower order term having singular
behaviour”, Boll. Unione Mat. Ital. (9) 2 (2009), no. 2, 349–370.
[17] D. Giachetti and S. Segura de León: “Quasilinear stationary problems with a quadratic
gradient term having singularities” to appear in J. London Math. Soc.
[18] A. Porretta and S. Segura de León: “Nonlinear elliptic equations having a gradient
term with natural growth” J. Math. Pures Appl., (9) 85 no. 3, (2006), 465–492.
[19] S. Segura de León: “Existence and Uniqueness for L1 data of some Elliptic Equations
with Natural Growth” Adv. Diff. Eq., 8 no. 11, (2003), 1377–1408.

Daniela Giachetti
Dipartimento di Scienze di Base e Applicate per l’ Ingegneria,
Sapienza, Università di Roma,
Via Scarpa 16, 00161 Roma, Italia.
E-mail address: [email protected]

Francesco Petitta
Dipartimento di Scienze di Base e Applicate per l’ Ingegneria,
Sapienza, Università di Roma,
Via Scarpa 16, 00161 Roma, Italia.
E-mail address: [email protected]

Sergio Segura de León

Departament d’Anàlisi Matemàtica, Universitat de València,
Dr. Moliner 50, 46100 Burjassot, València, Spain.
E-mail address: [email protected]