www.elsevier.com/locate/anihpc
Breaking of resonance and regularizing effect of a first order quasi-linear term in some elliptic equations
✩Boumediene Abdellaoui
a, Ireneo Peral
b,∗, Ana Primo
baDépartement de Mathématiques, Université Aboubekr Belkaïd, Tlemcen, Tlemcen 13000, Algeria bDepartamento de Matemáticas, U. Autonoma de Madrid, 28049 Madrid, Spain Received 9 November 2006; received in revised form 14 June 2007; accepted 14 June 2007
Available online 17 October 2007
Abstract
In this article we study the problem
(P)
⎧⎨
⎩
−u+ |∇u|q=λg(x)u+f (x) inΩ,
u >0 inΩ,
u=0 on∂Ω,
with 1q2 andf, gare positive measurable functions. We give assumptions ongwith respect toqfor which for allλ >0 and allf∈L1,f 0, problem (P) has a positive solution. In particular we focus our attention ong(x)=1/|x|2to prove that the assumptions ongare optimal.
©2007 Elsevier Masson SAS. All rights reserved.
Résumé
Dans cet article nous étudions le problème
(P)
⎧⎨
⎩
−u+ |∇u|q=λg(x)u+f (x) inΩ,
u >0 inΩ,
u=0 on∂Ω,
où 1q2 etf, gsont des fonctions mesurables positives. Nous donnons des hypothèses surgdépendant deqtelles que pour toutλ >0 et pour toutf ∈L1,f 0, le problème (P) a une solution positive. Nous portons une attention particulière au cas g(x)=1/|x|2pour montrer que les hypothèses surgsont optimales.
©2007 Elsevier Masson SAS. All rights reserved.
MSC:35D05; 35D10; 35J20; 35J25; 35J70; 46E30; 46E35
Keywords:Quasilinear elliptic equations; Existence and nonexistence; Regularization; Resonance
✩ Work partially supported by projects MTM2004-02223, MEC, Spain and CCG06-UAM/ESP-0340, CAM-UAM.
* Corresponding author.
E-mail addresses:boumediene.abdellaoui@uam.es (B. Abdellaoui), ireneo.peral@uam.es (I. Peral), ana.primo@uam.es (A. Primo).
0294-1449/$ – see front matter ©2007 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.anihpc.2007.06.003
1. Introduction
This paper is devoted to obtain existence and nonexistence results for nonlinear elliptic equations of the form −u+ |∇u|q=λg(x)u+f (x) inΩ,
u >0 inΩ,
u=0 on∂Ω,
(1.1) whereΩ ⊂RN is an open bounded domain, 1q 2 and λ∈R. In the whole of this work, we suppose thatf andgare positive measurable functions with some summability assumptions that we will specify in each case. For λ≡0, equations of the form (1.1) have been widely studied in the literature. We refer to [1,4,5,9,7,8,17,19] and the references therein. The caseq >2 andf Lipschitz has been studied in [22] with quite different methods. The case λ >0 and the presence of the term|∇u|qin the right-hand side of the equation has been recently studied in [2]. In [10]
it is proved that for allλ >0 the equation
−u=λ u
|x|2+f (x) inΩ⊂RN, N3 and 0∈Ω, (1.2)
has in general no solution for a positivef ∈L1(Ω). However, in [3], by adding the gradient term|∇u|2on the left- hand side of Eq. (1.2), we find a positive solution for all λ >0 and for all positivef ∈L1(Ω). This means that if q=2 andg(x)= |x|−2in problem (1.1), the term|∇u|2produces a strong regularizing effect. The new feature in this paper is to find the assumptions ongin terms ofqin order to solve problem (1.1) for allλ >0 and for allf ∈L1(Ω), f 0. We can reformulate the main objective of the paper as follows: fixedgfind theoptimalq for which for all λ >0 and for all positive functionf ∈L1(Ω), problem (1.1) has a positive solution. Precisely we prove the existence of solutionfor allλ >0and for allf ∈L1(Ω),f 0, ifg0 is anadmissible weightin the sense of (2.1) below.
This condition, in general, also becomes optimal. The previous result also proves that the absorption term |∇u|q is sufficient to break down any resonant effect of the linear zero order term. Forf ∈L1(Ω), we say thatuis a weak solution to problem (1.1) ifu∈W01,q(Ω),g(x)u∈L1(Ω)and
Ω
u(−φ) dx+
Ω
|∇u|qφ dx=λ
Ω
g(x)uφ dx+
Ω
f (x)φ dx, ∀φ∈C0∞(Ω).
Since we consider solutions to (1.1) with data inL1(Ω), we refer to [16] and [15] for a complete discussion about this framework. The paper is organized as follows. Section 2 is devoted to prove existence results. In Subsection 2.1, fixed 1< q2 we prove the main result, that is, ifgis an admissible weight then for allλ >0 and for allf ∈L1(Ω), f 0, there existsu∈W01,q(Ω)positive solution to problem (1.1). This is the result of Theorem 2.3.
In Subsection 2.2, we obtain conditions ongandλsuch that for all 1< q2, there exists solution to problem (1.1) provided that f is in a suitable class associated explicitly to g. Due to the presence of the Laplace operator and the linear termλg(x)u, then the natural condition is to assume (2.12). In Subsection 2.3, we give some results on uniqueness of solution. In Section 3, we consider the Hardy potential to prove nonexistence results that show the optimality of the condition required for existence in Theorems 2.3 and 2.9. These nonexistence results are related to the classical Hardy inequality,
Ω
|∇u|2dxΛN
Ω
u2
|x|2dx, for allu∈C∞0 (Ω)whereΛN=
N−2 2
2
(see for instance [18]). More precisely by settingg(x)= |x|−2, we prove nonexistence results for a very weak solution if 1qNN−1 andλ > ΛN, the Hardy constant. Next we prove the optimality of the summability assumption onf in Theorem 2.9. Finally in Subsection 3.1 we study the potentialg(x)= |x|−α, 1< α < N+22. Givenua measurable function we consider thek-truncation ofudefined by
Tk(u)= u, |u|k, k|uu|, |u|> k.
2. Existence of weak positive solutions
In this section we prove the existence of positive solution to problem (1.1) according to a relation betweenq, λ, g andf. In the first part, we prove that for allλ >0 and for allf ∈L1(Ω),f 0, there exists a positive solution to (1.1), provided thatgsatisfies some assumptions related toq. In the second part, we study the class of dataf for which, for allqin[1,2]and under some conditions ong, there exists positive solution for small values ofλ.
2.1. Admissible weights: Global existence result
Let consider in (1.1) a fixedq, 1< q2, then we say thatgis anadmissible weightin the sense of (2.1) below, g0, g∈L1(Ω) and C(g, q)= inf
φ∈W01,q(Ω)\{0}
(
Ω|∇φ|qdx)q1
Ωg|φ|dx >0. (2.1)
Remark 2.1.It is clear that ifgsatisfies (2.1), theng∈W−1,q(Ω)∩L1(Ω),q =q−q1. Moreover (2.1) implies that (a)
Ωg|u|dx <∞for allu∈W01,q(Ω).
(b) DefiningT:W01,q(Ω)→Rby T , u ≡
Ω
gu dx, (2.2)
thenT is a linear continuous form onW01,q(Ω). That means that there existsF=(f1, f2, . . . , fN)∈(Lq(Ω))N such thatg= −div(F ) and then
T , u ≡
Ω
gu dx=
Ω
F ,∇udx.
As a consequence, the duality product is equivalent to the first integral and TW−1,q (Ω)= F(Lq (Ω))N.
Proposition 2.2.Assume thatgsatisfies(2.1)and letT be defined by(2.2). Considergn(x)=min{g(x), n}and the corresponding linear continuous form
Tn:W01,q(Ω)→R, u→Tn(u)=
Ω
gnu dx.
The following statements hold.
(i) Tn→T strongly inW−1,q(Ω).
(ii) Assume thatun uweakly inW01,q(Ω),un0andun→ua.e., thengnun→gustrongly inL1(Ω).
Proof. (i) As in Remark 2.1 we also have that Tn, u =
Ω
gnu dx,
that is, the duality is realized by the integral and moreover C(gn, q)= inf
φ∈W01,q(Ω)\{0}
(
Ω|∇φ|qdx)1q
Ωgn|φ|dx >0.
Notice that
TnW−1,q = sup
u
W1,q 0 (Ω)1
Tn, u sup
u
W1,q 0 (Ω)1
Ω
gnu
sup
u
W1,q 0 (Ω)1
Ω
g|u|TW−1,q .
Then {Tn}n∈N, up to a subsequence, converges weakly in W−1,q(Ω). Since for all u∈W01,q(Ω), the Lebesgue theorem proves that
gn|u| →g|u| strongly inL1(Ω), then{Tn}n∈N T therefore,
TW−1,q lim inf
n→∞ TnW−1,q lim sup
n→∞ TnW−1,q TW−1,q , hence,
Tn→T , strongly inW−1,q (Ω).
(ii) According to the strong convergence proved in (i), we have that ifun uweakly inW01,q(Ω), then, Tn, un =
Ω
gnundx→
Ω
gu dx= T , u asn→ ∞.
If we assume that, moreover, un 0 and un→u a.e., then by using a well-known result in [12] we obtain that gnun→gustrongly inL1(Ω). (See also [21], Theorem 1.9, page 21.) 2
The main result of this section is the following one.
Theorem 2.3.Assume that1< q2and letgbe a positive function such that(2.1)holds, then for allλ0and for allf ∈L1(Ω),f0, there existsu∈W01,q(Ω),u0, that solves problem(1.1)in the sense of distributions.
To prove Theorem 2.3 we start by proving the result in some particular cases and then we proceed by approximation ofgandf. Notice that since 1< q2, thenN2 Nq, therefore the following first approach is quite natural.
Theorem 2.4. Assume that f, g∈Lr(Ω) are positive functions withr > Nq, then for all λ0, there exists u∈ W01,2(Ω)∩L∞(Ω)a weak positive solution to problem(1.1).
Proof. Step 1: For every fixedk >0, considerv∈W01,2(Ω)∩L∞(Ω)such that−v=λkg(x)+f (x)inΩ and denote Mk = vL∞. Notice thatv is bounded by the assumptions on gandf and by standard elliptic estimates (see [24]). It follows that zero is a subsolution andvis a supersolution to problems
−wn+1+|∇1wn|q
n|∇wn|q =λg(x)Tkwn+f (x), wn∈W01,2(Ω), wn0
(2.3) for alln∈N. By a simple variation of the arguments used in [8] and [4], there exists a sequence of nonnegative minimal solutions {wn}to problems (2.3). It follows that−wnλkg(x)+f (x)= −v, hence by the weak comparison principle we conclude that 0wnvM, uniformly inn, in particular,wn∈W01,2(Ω)∩L∞(Ω). Takingwnas test function in (2.3) we obtain,
Ω
|∇wn|2dx+
Ω
Hn(∇wn)wndx=λ
Ω
g(x)Tkwnwndx+
Ω
f (x)wndx,
whereHn(s)= |s|q/(1+1n|s|q). It is easy to check that there exists a positive constantCsuch that
Ω
|∇wn|2dxC(f, g, Ω, k) uniformly inn,
therefore, up to subsequenceswn uk weakly inW01,2(Ω). By weak-* convergence inL∞(Ω)we also have that uk∈W01,2∩L∞(Ω)andukMk. Next we investigate the equation satisfied byuk. To do that we prove the following claim.
Convergence claim.wn→uk strongly inW01,2(Ω).
Proof of the convergence claim. We follow closely the argument used in [5]. Considerφ(s)=se14s2, which satisfies φ(s)− |φ(s)|12. Takingφ(wn−uk)as test function in (2.3),
Ω
∇wnφ(wn−uk)∇(wn−uk) dx+
Ω
Hn(∇wn)φ(wn−uk) dx
=λ
Ω
g(x)Tkwnφ(wn−uk) dx+
Ω
f (x)φ(wn−uk) dx. (2.4)
Aswn ukweakly inW01,2(Ω), then a direct computation shows that
Ω
∇wnφ(wn−uk)∇(wn−uk) dx=
Ω
∇(wn−uk)2φ(wn−uk) dx+o(1).
Sinceq2, it is well known that∀ε >0 there exists a nonnegative constantCεsuch that
sqεs2+Cε, s0. (2.5)
Hence the second term in the left-hand side can be estimated in the following way,
Ω
Hn(∇wn)φ(wn−uk) dxε
Ω
|∇wn|2φ(wn−uk)dx+C(ε)
Ω
φ(wn−uk)dx
=ε
Ω
|∇wn− ∇uk|2φ(wn−uk)dx−ε
Ω
|∇uk|2φ(wn−uk)dx +2ε
Ω
∇wn∇ukφ(wn−uk)dx+C(ε)
Ω
φ(wn−uk)dx.
Sincewn ukweakly inW01,2(Ω)and by the fact that|φ(wn−uk)| →0 almost everywhere (and inL2(Ω)), then it follows
(i)
Ω|∇uk|2|φ(wn−uk)|dx→0 asn→ ∞, (ii)
Ω∇wn∇ukφ(wn−uk) dx→0 asn→ ∞. Therefore, passing to the limit asntends to∞, we have
Ω
Hn(∇wn)φ(wn−uk) dxε
Ω
|∇wn− ∇uk|2φ(wn−uk)dx+o(1).
Moreover, it is clear that the right-hand side in (2.4) goes to zero asn→ ∞. Sinceφ(s)− |φ(s)|>12, choosingε1 we conclude that
1 2
Ω
|∇wn− ∇uk|2dx
Ω
φ(wn−uk)−εφ(wn−uk)|∇wn− ∇uk|2dxo(1), whencewn→uk inW01,2(Ω)and the claim is proved. Moreover, from (2.5) it follows that
Hn(∇wn)c1|∇wn|2+c2.
By the claim, we have in particular the almost everywhere convergence of the gradients and therefore we conclude that
Hn(∇wn)→ |∇uk|q inL1(Ω).
Hence we find thatuksolves problem
−uk+ |∇uk|q=λg(x)Tkuk+f (x) inΩ, uk∈W01,2(Ω). (2.6) Step 2. We claim that there existsM >0 such thatukL∞(Ω)Mfor allk >0.
First of all we prove that{uk}is uniformly bounded inLa(Ω)for alla >1.
Consider
λ1(θ, g)= inf
φ∈C0∞(Ω), φ=0
Ω|∇φ|θdx
Ωg|φ|θdx.
Sinceg∈Lr(Ω)withr >Nq, it follows thatλ1(θ ) >0 for allθq.
Sinceuk∈L∞(Ω), then usinguak, witha2, as a test function in (2.6), we get a
Ω
uak−1|∇uk|2dx+
Ω
uak|∇uk|qdxλ
Ω
guak+1dx+
Ω
f uadx.
Thus 4a (2+a)2
Ω
|∇u
a 2+1 k |2dx+
q a+q
q
Ω
|∇u
a q+1
k |qdxλ
Ω
guak+1dx+
Ω
f uadx.
Using Hölder, Young and Poincaré inequalities there results that
Ω
gua+1dx=
Ω
g2(aa+q)ua2g21ua2+1g
q 2(a+q)dx
Ω
gua+qdx a
2(a+q)
Ω
gua+2dx 1
2
Ω
g dx q
2(a+q)
ε
Ω
gua+qdx+ε
Ω
gua+2dx+C(ε)
Ω
g dx
ε
λ1(q, g)
Ω
|∇u
a q+1
k |qdx+ ε λ1(2, g)
Ω
|∇u
a 2+1
k |2dx+C(ε, g), whereεis a positive constant that will be chosen later. On the other hand we have
Ω
f uadx
Ω
u2∗(a2+1)dx 2∗(a/2a
+1)
Ω
f
N (a+2) 2(N+a)dx
2(NN (a+2)+a)
C(f, S)
Ω
|∇ua2+1)|2dx a+a2
ε
Ω
|∇ua2+1)|2dx+C(f, S, ε),
whereSis the Sobolev constant and 2∗=N2N−2. Therefore, putting together the above inequalities it follows that 4a
(1+a)2 − ελ λ1(2, g)−ε
Ω
|∇u
a 2+1 k |2dx+
q a+q
q
− ελ λ1(q, g)
Ω
|∇u
a q+1
k |qdxC(f, g, S, ε).
Choosingεsmall enough we conclude that
Ω
|∇u
a 2+1 k |2dx+
Ω
|∇u
a q+1
k |qdxC,
whereC is independent ofk. Hence using Sobolev inequality we obtain thatukLa(Ω)C(a, λ, f, g)for alla >1.
Whence{guk+f}Lr(Ω)is uniformly bounded ink, for somer > N2. The uniform boundedness is a consequence of classical results about elliptic regularity, see [24].
Therefore ifk > M,u≡ukand is a solution to problem (1.1). 2
Remark 2.5.Notice that ifq2, then the passage to the limit in theconvergence claimabove can be performed in a different way; indeed using a compactness result by Boccardo–Murat in [6], the gradients converge almost every- where, therefore using Vitali theorem we get the strong convergence of the gradient.
However, we prove theconvergence claimwith arguments valid in a more general setting, which are needed to obtain strong convergence in the next theorems.
In the following result, we continue considering a weight with extra summability, but a generalf∈L1(Ω).
Theorem 2.6.Assume thatf, g are positive functions,f ∈L1(Ω),f 0and g∈Lr(Ω)withr > Nq, then for all λ0, problem(1.1)has a positive solutionu∈W01,q(Ω).
Proof. Consider a sequencefn∈L∞(Ω)such thatfn↑f inL1(Ω). Thanks to Theorem 2.4, there exists a sequence of positive bounded functions{un}, solutions to problems
⎧⎨
⎩
−un+ |∇un|q=λg(x)un+fn(x) inΩ, un>0 inΩ, un=0 on∂Ω,
un∈W01,2(Ω)∩L∞(Ω).
(2.7)
Notice that, in particularunsolves a problem of the form−un=Fn∈L1(Ω), thenunis the unique entropy solution to this problem. As a consequence we can useTk(un)as a test function. See for instance [24] or [16].
TakingTkunas test function in (2.7), it follows that
Ω
|∇Tkun|2dx+
Ω
|∇un|qTkundx=λ
Ω
g(x)unTkundx+
Ω
fn(x)Tkundx inΩ.
DefineΨk(s)=s
0Tk(t )1qdt, that explicitly is,
Ψk(s)=
⎧⎨
⎩
q q+1s
q+1
q ifs < k,
q q+1k
q+1
q +(s−k)k1q ifs > k,
then by using the assumptions ongandf, it follows that for allε >0, there existsCε>0 such that
Ω
|∇Tkun|2dx+
Ω
|∇Ψkun|qdxkελ
Ω
g(x)undx q
+λkC(ε)+kfnL1
εkλ
C(q, g)
Ω
|∇un|qdx+λkC(ε)+kfnL1. (2.8)
Hence, as in the second step of the proof of Theorem 2.4,
Ω
|∇un|qdx
Ω
|∇Tkun|2dx+k
{unk}
|∇un|qdx+Cq|Ω|
εkλ
C(q, g)
Ω
|∇un|qdx+λkC(ε)+Cq|Ω| +kfL1,
then un u weakly inW01,q(Ω) and Tkun Tku weakly in W01,2(Ω). It is clear by the assumption on g that gun→gustrongly in L1(Ω). LetGk(s)=s−Tk(s)and defineψk−1(s)=T1(Gk−1(s)), thenψk−1(un)|∇un|q
|∇un|qχ{unk}. As in [5], usingψk−1(un)as a test function in (2.7) there results that
Ω
∇ψk−1(un)2dx+
Ω
ψk−1(un)|∇un|qdx=
Ω
λg(x)un+fn(x)
ψk−1(un) dx.
Since{un}is uniformly bounded inLp(Ω),∀pq∗,it follows that
x∈Ω, such thatk−1< un(x) < k→0, x∈Ω, such thatun(x) > k→0 ask→ ∞, uniformly inn. Thus we conclude
klim→∞
{unk}
|∇un|qdx=0, uniformly inn. (2.9)
We follow the same arguments as in the proof of theconvergence claimin Theorem 2.4. Takeφ(Tk(un)−Tk(u))as a test function in (2.7), then
Tkun→Tku strongly inW01,2(Ω). (2.10)
To finish the proof, it is sufficient to show that
|∇un|q→ |∇u|q strongly inL1(Ω).
Since the sequence of gradients converges a.e. inΩ, we only have to show that is equi-integrable and to apply Vitali’s theorem. LetE⊂Ω be a measurable set. Then,
E
|∇un|qdx
E
|∇Tkun|qdx+
{unk}∩E
|∇un|qdx.
From (2.10) it follows that for allk >0,Tk(un)→Tk(u)strongly inW01,p(Ω)for allp2. In particular we obtain the strong convergence forp=q. Hence the integral
E|∇Tk(un)|qdxis uniformly small if|E|is small enough. On the other hand, thanks to (2.9) we obtain that
{unk}∩E
|∇un|qdx
{unk}
|∇un|qdx→0 ask→ ∞uniformly inn.
The equi-integrability of|∇un|qfollows immediately, and the proof is completed. 2
Proof of Theorem 2.3. We consider the truncationgn(x)=min{g(x), n} ∈L∞(Ω). Due to Theorem 2.6, there exists a sequence of positive functions{un}such thatunsolves
⎧⎨
⎩
−un+ |∇un|q=λgn(x)un+f (x) inΩ, un>0 inΩ, un=0 on∂Ω,
un∈W01,q(Ω).
(2.11)
SinceTkun∈W01,q(Ω)∩L∞(Ω), then we can useTkunas a test function in (2.11), it follows that
Ω
|∇Tkun|2dx+
Ω
|∇Ψkun|qdxλ
Ω
gn(x)Tkunundx+
Ω
f (x)Tkundx
kλ
Ω
gn(x)undx+k
Ω
f (x) dx.
Since
Ω
|∇Ψkun|qdx
{unk}
|∇Ψkun|qdxk
{unk}
|∇u|qdx,
then as above
Ω
∇Tk(un)2dx+k
{unk}
|∇un|qdxkελ
Ω
gn(x)undx q
+k
Ω
f (x) dx+λkC(ε, Ω).
Therefore by (2.1) we have that
Ω
|∇un|qdx kελ C(g, q)
Ω
|∇un|qdx+k
Ω
f (x) dx+λkC(ε, q, Ω),
henceun uweakly inW01,q(Ω). We have thatun0 and by Sobolev and Rellich theorems, up to a subsequence, we obtain thatun→ua.e. Then we apply Proposition 2.2 to obtain that
gnun→gu strongly inL1(Ω).
Therefore to finish the proof it is sufficient to show thatun→ustrongly inW01,q(Ω). This fact follows by proving thatTkun→Tkustrongly inW01,2(Ω)and using Vitali’s theorem as in the previous steps. 2
Remarks 2.7.
(1) The solution of problem (1.1) obtained in Theorem 2.3 is also an entropy solution in the sense that we can take in problem (1.1) test functions of the formTk(u−v)withv∈W1,2(Ω)∩L∞(Ω).
(2) The same existence result holds iff is a bounded positive Radon measure such thatf ∈L1(Ω)+W−1,2(Ω)(fis absolutely continuous with respect to the classical capacity). These results are obtained with some minor technical changes, i.e. the result follows using the same approximation arguments. See [15] to find the precise meaning of solution in this framework and equivalent definitions. By the regularity theory of renormalized solutions we easily get that ifuis a positive solution to problem (1.1), thenu∈W01,q(Ω)∩W01,p(Ω)for all 1p <NN−1.
(3) The existence result obtained in Theorem 2.3 in particular means that resonance phenomenon doesn’t occur if we add|∇u|q as an absorption term. Without the gradient term, positive solution exists just by assuming thatλis less than the infimum of the spectrum of the operator−with the corresponding weightgand under a suitable condition onf.
As a direct application of Theorem 2.3 we obtain the following well known result.
Corollary 2.8.Assume thatλ=0, then problem(1.1)has an entropy positive solution for all0f ∈L1(Ω)and for all1q2.
In fact we can considerg∈L∞(Ω)in Theorem 2.3 and then pass to the limit using a priori estimates asλgoes to 0.
A direct proof of this particular existence result can be obtained using truncation argument. See [5]. We will analyze in particular the Hardy potential in section 3 in order to prove the optimality of the assumptions in Theorem 2.3.
2.2. Existence of solution for allq∈ [1,2]and smallλ
In this section we will find general conditions ongandf that assure the existence of solution for allq2. The presence of the linear termλg(x)umotivates the following assumption ong,
g0, g≡0 andg∈L1(Ω)withλ1(g,2)= inf
φ∈W01,2(Ω)\{0}
Ω|∇φ|2dx
Ωg|φ|2dx >0. (2.12)
It is easy to check that under assumption (2.12), for everyλ < λ¯ 1(g,2)there exists a uniqueϕ∈W01,2(Ω), positive weak solution to the problem
−ϕ= ¯λg(x)ϕ+g(x) inΩ, ϕ=0 on∂Ω. (2.13)
The main result in this section is the following one.
Theorem 2.9. Assume that 0< λ < λ1(g,2)and let ϕ be the solution to problem (2.13) with λ <λ < λ¯ 1(g,2).
Suppose thatf is a positive function such that
Ωf ϕ dx <∞, then there existsu∈W01,q(Ω)positive solution to problem(1.1)and moreover,
Ω|∇u|pdx <∞,∀pqifqNN−1andp <NN−1, on the contrary case.
Proof. By using Theorem 2.4, we find a sequence{un}of positive solutions to the approximated problems
−un+ |∇un|q=λgn(x)un+fn(x) inΩ,
un∈W01,q(Ω), (2.14)
withgn(x)=min{g(x), n} ∈L∞(Ω)andfn(x)=min{f (x), n} ∈L∞(Ω). It is clear that
Ωfn(x)ϕ dx < C uni- formly inn. By a duality argument we obtain
Ω
un(−ϕ) dx+
Ω
|∇un|qϕ dx=λ
Ω
gn(x)unϕ dx+
Ω
fn(x)ϕ dx.
Therefore, (¯λ−λ)
Ω
g(x)unϕ dx+
Ω
g(x)undx+
Ω
|∇un|qϕ dx
Ω
fn(x)ϕ dx
Ω
f (x)ϕ dxC,
and hence we conclude that
Ω
g(x)unϕ dx C
λ¯−λ uniformly inn,
Ω
g(x)undxC uniformly inn,
Ω
|∇un|qϕ dxC uniformly inn.
TakingTkunas a test function in (2.14), we have that
Ω
|∇Tkun|2dx+
Ω
|∇un|qTkundxλk
Ω
gn(x)undx+k
Ω
f (x) dx.
Thus 1k
Ω|∇Tkun|2dxCandTkun Tkuweakly inW01,2(Ω). In particular fork=1 we obtain that
Ω
|∇un|q
Ω
|∇T1un|2dx+C(q, Ω)+
Ω
|∇un|qT1undxC,
thereforeun uweakly inW01,q(Ω). Ifq >NN−1 andg∈Lm(Ω)withm >N2 org(x)= |x|−2the existence result is a consequence of the previous theorem. On the contrary, if qNN−1, then using the regularity result for entropy solutions obtained in [16] (see also [15] for the case of Radon measures), we obtain an extra regularity, that is,un u inW01,p(Ω)for allp <NN−1. We claim thatgn(x)un→g(x)ustrongly inL1(Ω). To prove the claim we consider the sequence{wn} ∈W01,2(Ω)of solutions to the problem,
−wn=λgn(x)wn+fn(x) inΩ,
wn=0 on∂Ω. (2.15)
Since λ < λ1(g,2), then using the assumption on f we obtain that wnw everywhere and wn w weakly in W01,p(Ω), for allp < NN−1, withwthe unique entropy solution to problem−w=λg(x)w+f. Notice that solu- tions to (2.14) are subsolutions to (2.15), hencegnungnwngw. Thanks to the dominated convergence theorem, gn(x)un→g(x)uinL1(Ω)and we conclude. To finish we have just to prove the strong convergence of|∇un|q to
|∇u|qinL1(Ω). The proof is done in two steps. First we start by proving the strong convergence ofTk(un)toTk(u) inW01,2(Ω), which follows by applying toTk(un)andTk(u)the techniques used in the proof ofconvergence claimin
Theorem 2.4. Then to get the main convergence we use Vitali’s Lemma and a suitable test function as in the last step in the proof of Theorem 2.6. 2
Remarks 2.10.
(1) Notice that ifg(x)= |x|−2, then the regularity required onf in Theorem 2.9, depends onλ. See [10].
(2) In Section 3 we will show that the condition on the integrability off in Theorem 2.9 is, in general, optimal.
2.3. Some remarks on the uniqueness
We consider the caseq=2 for which a change of variables allows us to find a comparison principle and uniqueness.
Lemma 2.11.Letu∈W01,2(Ω)be such thatu=0and−u+ |∇u|20, then there exist constantsC, R >0such thatuCinBR(0)⊂Ω.
Proof. The change of variablesv=1−e−utransforms−u+ |∇u|20 into−v0 withv=0. It is sufficient to apply the classical maximum principle of the Laplacian operator and we conclude. 2
Lemma 2.12.Assume thatf ∈L1(Ω)is a positive function andgsatisfies(2.1). Then for allλ >0the problem
⎧⎨
⎩
−u+ |∇u|2=λg(x)u+f (x) inΩ,
u >0 inΩ,
u=0 on∂Ω,
(2.16) has at most a positive solutionu∈W01,2(Ω).
Proof. Ifu∈W01,2(Ω)is a solution to (2.16) thenv=1−e−usatisfies 0v1 inΩ and it is a solution to
−v=λg(x)(1−v)log(1−1v)+(1−v)f (x) inΩ,
v=0 on∂Ω. (2.17)
Define
H (x, v)= λg(x)(1−v)log(1−1v)+(1−v)f (x), if 0v <1,
0, ifv1.
By a direct computation we find that H (v,x)v is a nonincreasing function invforv0, then by similar arguments as in [11] we conclude thatvis the unique solution to (2.17). Therefore,uis the unique solution to (2.16). 2
Remark 2.13.Using the same ideas of Lemma 2.12 we get a comparison principle for a sub- and super-solution to problem (2.16), namely ifu1is a nonnegative subsolution andu2is a nonnegative supersolution such thatu1, u2∈ W1,2(Ω)andu2u1on∂Ω, thenu2u1inΩ.
We will prove the following comparison principle, that we will use below, without change of variables.
Theorem 2.14. Assume that1q < NN−1 and 0f ∈L1(Ω). Letv, u be two nonnegative functions such that u, v∈W1,q(Ω),u, v∈L1(Ω)and
−u+ |∇u|qf (x) inΩ,
−v+ |∇v|qf (x) inΩ,
vu on∂Ω,
(2.18) thenvuinΩ.
Proof. Considerw=v−u, then it is clear thatw∈W1,q(Ω),w0 on∂Ωandw∈L1(Ω). In order to conclude, it is sufficient to prove thatw+=0. It follows that
−w+ |∇v|q− |∇u|q0.