diffusive logistic equations in R with indefinite weight
Dimitri Mugnai ∗
Dipartimento di Matematica e Informatica Universit` a di Perugia
Via Vanvitelli 1, 06123 Perugia - Italy tel. +39 075 5855043, fax. +39 075 5855024,
e-mail: dimitri.mugnai@unipg.it Nikolaos S. Papageorgiou
Department of Mathematics National Technical University Zografou Campus, Athens 15780 Greece
e-mail: npapg@math.ntua.gr
Abstract
We consider a diffusive p–logistic equation in the whole of R
Nwith absorption and an indefinite weight. Using variational and truncation techniques we prove a bifurcation theorem and describe completely the bifurcation point. In the semilinear case p = 2, under an additional hypothesis on the absorption term, we show that the positive solution is unique.
Keywords: diffusive p–logistic equation, positive solution, Hardy’s inequal- ity, bifurcation theorem, nonlinear regularity, indefinite weight.
2000AMS Subject Classification: 35J20, 35J70, 35P30
1 Introduction
In this paper we are interested in non existence, existence and uniqueness (in the semilinear case p = 2) of positive solutions for the following p–logistic type equation with absorption defined in the whole of R N (N > p):
(P λ )
( −∆ p u(z) = λ β(z)u p−1 − f (z, u(z))
in R N , u(z) → 0 as kzk → ∞, u > 0,
∗
Research supported by the M.I.U.R. project Metodi Variazionali ed Equazioni Differen- ziali Nonlineari.
1
where λ > 0, k · k is the Euclidean norm in R N and ∆ p denotes the p–Laplace differential operator defined by ∆ p u = div(kDuk p−2 Du) for all u ∈ W 1,p (R N ).
Moreover, β is a sign changing weight function and f : R N × R → R is a Carath´ eodory perturbation (i.e., for all x ∈ R the map z 7→ f (z, x) is measur- able, and for a.e. z ∈ R N the map x 7→ f (z, x) is continuous).
Problems of this type arise as the steady state equation of population dy- namics models (for example, see [3], [10], [11], [12], [13], [14], [26]). In this case the unknown u corresponds to the density of a population, the weight β corresponds to the birth rate of the population if self–limitation is ignored and the perturbation −f expresses the fact that the population is self–limiting. The function β is in general sign changing and the region where β is positive (resp.
negative) the population, ignoring self–limitation, has positive (resp. negative) birth rate. Since u describes the population density, we are interested only in positive solutions of problem (P λ ), such solutions corresponding to possible steady state distributions of the population.
The result we prove establishes a relation between the diffusion parameter λ > 0 (extent of diffusion) and the existence of a nontrivial steady state popula- tion density. Thus, the population will prosper in those regions where the birth rate β is positive, and if the diffusion is small (i.e. if λ > 0 is large) nontrivial steady state solutions are possible, even if the population is subject to some overall disadvantage (i.e. β is predominantly negative). On the other hand, if the diffusion is big (i.e. λ > 0 is small), the population is not safely protected in small regions of disadvantage and a steady state solution cannot exist.
Problem (P λ ) settled in a bounded region with Neumann boundary condi- tion was studied by Cantrell–Cosner–Hutson [4] and Umezu [25], and exten- sions to the p–Laplacian Dirichlet case can be found in the works of Dong [7], Garcia Melian–Sabina de Lis [15], Guedda–Veron [17], Kamin–Veron [19] and Papageorgiou–Papalini [21]. A related Neumann problem can be found also in Cardinali–Papageorgiou–Rubbioni [5]. Diffusive logistic equations in the whole of R N were studied by Afrouzi–Brown [2], who dealt with the semilinear equa- tion (i.e. p = 2). More precisely, they considered the equation
−∆u = λβu − u 2
in R N
with β : R N → R a smooth function bounded above. Our semilinear result here (see Theorem 9) extends the result of Afrouzi–Brown [2], since we do not assume that β is smooth and bounded above, and our absorption term −f (z, x) is more general than the one considered therein.
2 The bifurcation parameter
Following Szulkin–Willem [24], we impose the following conditions on the weight function β:
H(β): β ∈ L ∞ loc (R N ), β + = β 1 + β 2 6= 0, with β 1 ∈ L N/p (R N ) ∩ L 1 (R N ) (1 < p < N ) and
kzk p β 2 (z) → 0 as kzk → ∞, lim
z→ˆ z kz − ˆ zk p β 2 (z) = 0 for all ˆ z ∈ R N ,
kzk
p(N −1)Nβ 2 (z) → 0 as z → 0.
Here, as usual, β + = max{β, 0} and β − = max{−β, 0}.
Moreover, in view of the Bifurcation Theorems 7 and 9, we also assume the following condition:
H(β) 2 :
( (rˆ a(r)) 1/(p−1) ∈ L 1 (1, ∞) if 1 < p ≤ 2, or r
(p−2)N +1p−1ˆ a(r) ∈ L 1 (1, ∞) if p ≥ 2, where we have set
ˆ
a(r) := sup
kzk=R
β + (z).
Now, for any r > 0, let B r = z ∈ R N : kzk < r . We introduce the following quantity:
λ ˆ 1 (r) = inf n Z
B
rkDuk p dz : z ∈ W 0 1,p (B r ), Z
B
rβ|u| p dz = 1 o
. (1)
From Szulkin–Willem [24] we know that ˆ λ 1 (r) is the principal eigenvalue of the nonlinear eigenvalue problem
(E)
( −∆ p u = λβ|u| p−2 u in B r ,
u = 0 on ∂B r .
From (1) it is clear that the map r 7→ ˆ λ 1 (r) is decreasing on (0, ∞) and so we can define
λ ˆ ∗ = lim
r→∞
λ ˆ 1 (r). (2)
In the following we will need the so called Hardy’s inequality, an extension of the Poincar´ e inequality to the case of singular potentials:
Z
R
N|u(z)|
kzk
p
dz ≤ p p (N − p) p
Z
R
NkDuk p dz (3)
for all u ∈ D 1,p (R N ) = C c ∞ (R N ), the closure being taken in the norm kuk = kDuk p .
Proposition 1. If hypotheses H(β) hold, then ˆ λ ∗ > 0.
Proof. For r > 0, let u ∈ W 0 1,p (B r ) be such that R
B
rβ|u| p dz = 1. We have 1 =
Z
B
rβ|u| p dz = Z
B
r(β + − β − )|u| p dz
≤ Z
B
rβ + |u| p dz = Z
B
rβ 1 |u| p dz + Z
B
rβ 2 |u| p dz.
(4)
By hypotheses H(β), we have that β 1 ∈ L N/p (R N ), and by the Sobolev Embed- ding Theorem we have that u ∈ L p
∗(B r ), where p ∗ = N p/(N − p) is the critical Sobolev exponent. Note that
p N + p
p ∗ = p
N + N − p
N = 1. (5)
So, using H¨ older’s and Sobolev’s inequalities, we have Z
B
rβ 1 |u| p dz ≤ kβ 1 k L
N/p(B
r) kuk p L
p∗(B
r
) ≤ c 1 kβ 1 k L
N/p(B
r) kDuk p L
p(B
r
,R
N) (6) for some c 1 > 0. From the assumptions on β 2 in H(β), given ε > 0, we can find r 0 = r 0 (ε) > δ = δ(ε) > 0 such that
kzk p β 2 (z) ≤ ε for all z ∈ R N , kzk ≥ r 0 (7) and
kzk
p(N −1)Nβ 2 (z) ≤ ε for all z ∈ B δ . (8) Let r > r 0 be such that 1/r < δ and set Ω 1 = B r \ B r
0. We have
Z
Ω
1β 2 |u| p dz ≤ ε Z
Ω
1|u| p
kzk p dz (by (7))
≤
p
N − p
p ε
Z
Ω
1kDuk p dz (by (3))
= c 2 ε Z
Ω
1kDuk p dz
(9)
with c 2 = p
N −p
p
> 0.
Now let Ω 2 = B δ − B 1/r , so that Z
Ω
2β 2 |u| p dz ≤ ε Z
Ω
2|u| p kzk
p(N −1)Ndz (by (8))
≤ ε
Z
Ω
21 kzk
p(N −1)N! N/p
dz
p/N
kuk p L
p∗(Ω
2
)
(by H¨ older’s inequality, see (5))
≤ c 3 ε Z δ
1/r
1
s N −1 s N −1 ω N ds
! p/N
kDuk p L
p(B
r,R
N)
by Sobolev’s inequality and for some c 3 > 0
= c 4 ε
δ − 1
r
p/N Z
B
rkDuk p dz for some c 4 > 0;
(10)
Here ω N denotes the measure of the unit sphere in R N .
Next let Ω 3 = B r
0\ B δ . The compactness of Ω 3 and hypotheses H(β) imply that we can find closed balls B ρ
k(z k ) m
k=1 such that in correspondence of ε > 0 chosen above, there holds
kz − z k k ≤ ρ k ⇒ kz − z k k
p(N −1)Nβ 2 (z) ≤ ε. (11) From (11) it follows that we can find ρ > 0 such that for all k ∈ {1, . . . , m}
we have ρ < ρ k and
kz − z k k ≤ ρ ⇒ kz − z k k
p(N −1)Nβ 2 (z) ≤ ε
m . (12)
Setting U = ∪ m k=1 B ρ (z k ), we have Z
B
ρ(z
k)
β 2 |u| p dz ≤ ε m
Z
B
ρ(z
k)
|u| p kz − z k k
p(N −1)Ndz (by (12))
≤ ε m
"
Z
B
ρ(z
k)
1 kz − z k k
p(N −1)N!# p/N
kuk p L
p∗(B
r
)
by H¨ older’s inequality and (5)
≤ c 5 ε m
"
Z
B
ρ1 kzk N −1 dz
# p/N
kDuk p L
p(B
r,R
N)
by Sobolev’s inequality
≤ c 6 ε
m kDuk p L
p(B
r,R
N) ,
(13)
for some c 5 > 0 and with c 6 = (ω N ρ) p/N .
Adding these inequalities over k ∈ {1, . . . , m} we obtain Z
U
β 2 |u| p dz ≤ c 6 εkDuk p L
p(B
r,R
N) . (14)
Finally, if z ∈ Ω 3 \ U , then we can find k ∈ {1, . . . , m} such that ρ < kz − z k k ≤ ρ k .
Hence from (11) it follows that
β 2 (z) ≤ ε ρ
p(N −1)N. (15)
Therefore, we have Z
Ω
3\U
β 2 |u| p dz ≤ ε ρ
p(N −1)NZ
Ω
3\U
|u| p dz (see (15))
≤ c 7 εkDuk p L
p(B
r
,R
N) ,
(16)
where c 7 > 0 is a constant depending on r 0 and ρ, but not on r.
Now, we return to (15) and use (6), (9), (10), (14) and (15), obtaining 1 ≤ h
c 1 kβ 1 k L
N/p(B
r) + c 2 ε + c 4 ε
δ − 1
r
p/N
+ (c 6 + c 7 )ε i Z
B
rkDuk p dz.
Since u ∈ W 0 1,p (B r ) with R
B
rβ|u| p dz = 1 was arbitrary, from (1) it follows that
h
c 1 kβk L
N/p(B
r) + c 2 ε + c 4 ε
δ − 1
r
p/N
+ (c 6 + c 7 )ε i −1
≤ ˆ λ 1 (r).
Passing to the limit as r → ∞ in the previous inequality, we obtain 0 < h
c 1 kβk L
N/p(R
N) + c 2 ε + c 4 εδ p/N + (c 6 + c 7 )ε i −1
≤ ˆ λ ∗ , as claimed.
Remark 1. For Proposition 1, the requirement β 2 ∈ L ∞ loc (R N ) is sufficient, but in the future, namely for Proposition 2 and its consequences, we shall use β ∈ L ∞ loc (R N ).
We end up this section recalling that, if Y is a Banach space, Y ∗ is its topological dual and E : Y → Y ∗ is a map, we say that E is of type (S) + if
y n * y in Y and lim sup
n→∞
hE(y n ), y n − yi Y
∗,Y ≤ 0
(by h , i Y
∗,Y we denote the duality brackets for the pair (Y ∗ , Y )), then y n → y in Y (for example, see Gasinski–Papageorgiou [16]).
3 The bifurcation theorem
The hypotheses on the absorption term f are:
H(f ): f : R N × R → R is a Carath´eodory function such that for a.e. z ∈ R N there holds f (z, 0) = 0, f (z, x) > 0 for all x > 0 and
(i): for every ρ > 0 there exists a ρ ∈ L ∞ loc (R N ) ∩ L (p
∗)
0(R N ) (1/p ∗ + 1/(p ∗ ) 0 = 1) such that
f (z, x) ≤ a ρ (z) for a.e. z ∈ R N and all 0 ≤ x ≤ ρ;
(ii): there exists η ∈ L ∞ (R N ) such that ϑ = ess inf
R
N(η − β) > 0 and
lim inf
x→∞
f (z, x)
x p−1 ≥ η(z) uniformly for a.e. x ∈ R N ; (iii): we have
lim
x→0
+f (z, x)
x p−1 = 0 uniformly for a.e. z ∈ R N .
Remark 2. Since we are interested in positive solutions and hypotheses H(f ) concern the positive semiaxis R + = [0, ∞) only, by truncating f if necessary, we may (and will) suppose that f (z, x) = 0 for a.e. z ∈ R N and all x ≤ 0.
Examples of functions satisfying H(f ) are easy to construct; as an example, if β + ∈ L ∞ (R N ) ∩ L (p
∗)
0(R N ), let us present
f (z, x) =
( β + (z)x p−1 if x ≥ 1, β + (z)x ℘−1 if x < 1, where ℘ is any number strictly greater than p.
In what follows a fundamental rˆ ole will be played by the suitably weighted space
V = n
u ∈ D 1,p (R N ) : kuk p V = kDuk p L
p(R
N,R
N) + Z
R
Nβ − |u| p dz < ∞ o , which turns out to be a reflexive Banach space, and in which we will look for solutions to problem (P λ ).
Proposition 2. If hypotheses H(β), H(β) 2 , H(f ) hold and λ > ˆ λ ∗ > 0, then problem (P λ ) has a positive solution u 0 ∈ C 1 (R N ) ∩ V such that
u 0 (z) > 0 for all z ∈ R N and u 0 (z) → 0 as kzk → ∞.
Proof. The proof is rather long, and will need several intermediate steps.
By virtue of hypothesis H(f )(ii), given ε > 0, there exists M = M (ε) > 0 such that
f (z, x) ≥ (η(z) − ε)x p−1 for a.e z ∈ R N and all x ≥ M . (17) Let ξ > M . For any r > 0 and any h ∈ W 0 1,p (B r ), h ≥ 0, we have
Z
B
r(β(z)ξ p−1 − f (z, ξ))h(z) dz
≤ Z
B
r(β(z)ξ p−1 − η(z)ξ p−1 + εξ p−1 )h(z) dz (see (17))
≤ ξ p−1 (ε − ϑ) Z
B
rh dz (see hypothesis H(f )(ii)).
(18)
Choosing ε ∈ (0, ϑ), from (18) we infer that Z
B
r(β(z)ξ p−1 − f (z, ξ))h(z) dz ≤ 0 for all h ∈ W 0 1,p (B r ), h ≥ 0. (19) Now, we consider the following truncation for the reaction term:
g(z, x) =
0 if x ≤ 0,
β(z)x p−1 − f (z, x) if 0 ≤ x ≤ ξ, β(z)ξ p−1 − f (x, ξ) if x > ξ.
(20)
Of course, g is a Carhath´ eodory function. We set G(z, x) = R x
0 g(z, s) ds for all (z, x) ∈ B r × R, and we consider the C 1 functional ψ λ : W 0 1,p (B r ) → R defined by
ψ λ (u) = 1
p kDuk p p − λ Z
B
rG(z, u(z)) dz for all u ∈ W 0 1,p (B r ).
It is clear from (20) that ψ λ is coercive. Moreover, exploiting the compact embedding of W 0 1,p (B r ) into L p (B r ), we can easily check that ψ λ is sequentially weakly lower semicontinuous. Therefore, by the Weierstrass Theorem, we can find u r 0 ∈ W 0 1,p (B r ) such that
ψ λ (u r 0 ) = inf n
ψ λ (u) : u ∈ W 0 1,p (B r ) o
= m r . (21)
Lemma 3. u r 0 6= 0.
Proof. By virtue of hypothesis H(f )(iii), given ε > 0, we can find δ = δ(ε) > 0 such that
f (z, x) ≤ εx p−1 for a.e. z ∈ B r and all x ∈ [0, δ], so that
F (z, x) ≤ ε
p x p for a.e. z ∈ B r and all x ∈ [0, δ]. (22) The ordered Banach space C 0 1 (B r ) = n
u ∈ C 1 (B r ) : u = 0 on ∂Ω has a positive cone given by C + r = n
u ∈ C 0 1 (B r ) : u(z) ≥ 0 for all z ∈ B r
o
, and this cone has a nonempty interior given by
int C + r = n
u ∈ C + r : u(z) > 0 for all z ∈ B r and ∂u ∂ν < 0 for all z ∈ ∂B r o , where ν(z) denotes the unit outward normal to ∂B r at z.
Let ˆ u r ∈ int C + r be the principal eigenfunction of the eigenvalue problem (E) (see Szulkin–Willem [24] for the existence and Vazquez [27] for the regularity).
We can find t ∈ (0, 1) so small that tˆ u r (z) ∈ [0, δ] for all z ∈ B r . Then ψ λ (tˆ u r ) = t p
p kDˆ u r k p L
p(B
r,R
N) − λt p p
Z
B
rβ ˆ u p r dz + λ Z
B
rF (z, tˆ u r ) dz (see (20))
≤ (ˆ λ 1 (r) − λ)t p p
Z
B
rβ ˆ u p r dz + λt p
p εkˆ u r k p p (see (22)).
(23) But R
B
rβ ˆ u p r dz = 1, and for r > 0 large enough we have ˆ λ 1 (r) < λ (see Proposi- tion 1 and recall that λ > ˆ λ ∗ ). So, if in (23) we choose ε > 0 sufficiently small, then we have ψ λ (tˆ u r ) < 0, and by (21)
m r = ψ λ (u r 0 ) < ψ λ (0) = 0,
and therefore u r 0 6= 0, as claimed.
Lemma 4. u r 0 ∈ int C + r , and in addition u r 0 ∈ [0, ξ] = n
u ∈ W 0 1,p (B r ) : 0 ≤ u(z) ≤ ξ a.e. in B r o . Proof. From (21) we have (ψ λ ) 0 (u 0 ) = 0, that is
A(u r 0 ) = λN g (u r 0 ), (24) where A : W 0 1,p (B r ) → W −1,p
0(B r ) is the nonlinear map defined by
hA(u), yi = Z
B
rkDuk p−2 (Du, Dy) R
Ndz for all u, y ∈ W 0 1,p (B r ),
and N g (u)(·) = g(·, u(·)) for all u ∈ W 0 1,p (B r ).
On (24) we first act with −(u r 0 ) − ∈ W 0 1,p (B r ), and recalling (20), we obtain Z
B
rkD(u r 0 ) − k p dz = 0,
that is u r 0 ≥ 0 and u r 0 6= 0 (from above).
Next we act on (24) with (u r 0 − ξ) + ∈ W 0 1,p (B r ), obtaining hA(u r 0 ), (u r 0 − ξ) + i = λ
Z
B
rg(z, u r 0 )(u r 0 − ξ) + dz
= λ Z
B
r(βξ p−1 − f (z, ξ))(u r 0 − ξ) + dz (by (20))
≤ 0 (by (19)), that is
Z
{u
r0>ξ}
kDu r 0 k p dz ≤ 0.
Thus, either |{u r 0 > ξ}| N = 0, where | · | N denotes the Lebesgue measure in R N , or u r 0 is constant in the set {u r 0 > ξ}. Nonlinear regularity theory (see, for example, Gasinski–Papageorgiou [16, pp. 737–738]) implies that u r 0 ∈ C 0 1 (B r ), so that the second possibility is neglected, and thus u r 0 ≤ ξ.
Therefore, we have proved that
u r 0 ∈ [0, ξ].
Then (20) implies that (24) becomes
A(u r 0 ) = λβ(u r 0 ) p−1 − λN f (u r 0 ), with N f (u)(·) = f (·, u(·)) for all u ∈ W 0 1,p (B r ), that is
( −∆ p u r 0 (z) = λβ(u r 0 (z)) p−1 − f (z, u r 0 (z))) in B r ,
u r 0 = 0 on ∂B r . (25)
Then
∆ p u r 0 (z) ≤ λ(kβ + k L
∞(B
r) u r 0 (z) p−1 + f (z, x)) a.e. in B r . (26) The assumptions on f say that there exists C, δ > 0 such that
0 ≤ f (z, x) ≤
1 + C
δ
|x| p−1 for a.e. z ∈ B r and all x ≥ 0. (27) Then, from (26) and (27) we have
∆ p u r 0 (z) ≤ λ
kβ + k L
∞(B
r) +
1 + C
δ
u r 0 (z) p−1 a.e. in B r , and so u r 0 ∈ int C + r , see Vazquez [27] and Pucci–Serrin [22].
Lemma 5. Problem (25) has a smallest solution u belonging to [0, ξ] ∩ int C + r . Proof. Set
S ξ r = n
u ∈ W 0 1,p (B r ) : u is a nontrivial solution of (25) with u ∈ [0, ξ] o . We have just shown above that S ξ r 6= ∅ and that S ξ r ⊂ int C + r . Using Lemma 4.3 of Filippakis–Kristaly–Papageorgiou [9], we infer that S ξ r is downward di- rected, i.e., if u, v ∈ S ξ r , then there exists y ∈ S ξ r such that y ≤ min{u, v}. We now show that S ξ r has a minimal element. To this purpose, let C ⊂ S ξ r be a chain (i.e., a totally ordered subset of S ξ r ). From Dunford–Schwartz [8, p. 336]
we know that we can find a sequence {u n } n≥1 ⊂ C such that inf
n≥1 {u n } n≥1 = inf C.
Evidently, {u n } n≥1 ⊂ W 0 1,p (B r ) is bounded, and so we may assume without loss of generality that
u n * u in W 0 1,p (B r ), u n → u in L p (B r ) and a.e. in B r as n → ∞. (28) Since u n ∈ S ξ r for any n ∈ N, we have
A(u n ) = λβu p−1 n − λN f (u n ) for all n ≥ 1. (29) On (29) we act with u n − u ∈ W 0 1,p (B r ) and then pass to the limit as n → ∞.
By (28) we immediately obtain
n→∞ lim hA(u n ), u n − ui = 0,
and so, A being of type (S) + (see, for example, [16]), we get that
u n → u in W 0 1,p (B r ). (30)
Thus, if in (29) we pass to the limit as n → ∞ and use (30), we obtain
A(u) = λβu p−1 − λN f (u). (31)
By (28) we have that u ∈ [0, ξ]; we now show that u 6= 0. If u = 0, then from (30) we would have that ku n k = ku n k W
1,p0
(B
r) → 0 as n → ∞, and so, up to a subsequence, u n → 0 a.e. in B r .
Let y n = u n /ku n k, n ≥ 1. Then ky n k = 1 for all n ≥ 1 and so we may assume that
y n * y in W 0 1,p (B r ), y n → y in L p (B r ) and a.e. in B r as n → ∞. (32) From (29) we have
A(y n ) = λβy p−1 n − λ N f (u n )
ku n k p−1 , n ≥ 1. (33)
Acting on (33) with y n − y ∈ W 0 1,p (B r ) and passing to the limit as n → ∞, by H(f )(iii) and (32) we obtain
n→∞ lim hA(y n ), y n − yi = 0, and thus
y n → y in W 0 1,p (B r ), so that kyk = 1 and y ≥ 0. (34) From H(f )(i) it is clear that n N
f
(u
n) ku
nk
p−1o
n≥1 ⊂ L p
0(B r ) is bounded, and so we may assume that ku N
f(u
n)
n
k
p−1* h in L p
0(B r ) as n → ∞. Moreover, hypothesis H(f )(iii) implies h = 0. Therefore, if in (33) we pass to the limit as n → ∞, we obtain
A(y) = λβy p−1 . Equivalently,
( −∆ p y = λβy p−1 in B r ,
y = 0 on ∂B r ,
that is (λ, y) solve problem (E), and so y ∈ C 0 1 (B r ) must be nodal, since λ >
ˆ λ 1 (r) (for example, see [6, Theorem 3.2]).
This contradicts (34) and this proves that u 6= 0. As before, via the nonlinear maximum principle (see [22], [27]), we have that u ∈ int C + r , and so u ∈ S ξ r . Since C was an arbitrary chain in S ξ r , from the Kuratowski–Zorn Lemma we infer that S ξ r has a minimal element u ∗ ∈ S ξ r ⊂ int C + r . If u ∈ S ξ r , since S ξ r is downward directed and due to the minimality of u ∗ , we have u ∗ ≤ u, and so u ∗ ∈ int C + r is the smallest element of S ξ r . This proves the lemma.
Now, let r n ↑ ∞ be such that ˆ λ 1 (r 1 ) < λ, and hence ˆ λ 1 (r n ) < λ for all n ≥ 1. Let u n 0 = u r 0
n, n ≥ 1. Extending by zero outside B r
n, we obtain a function ¯ u n 0 ∈ W 1,p (R N ). From the claim above we know that
0 ≤ ¯ u n 0 ≤ ξ in R N for all n ≥ 1. (35)
We also know that A(¯ u 2 0 ) = λβ(¯ u 2 0 ) p−1 − λN f (¯ u 2 0 ) in W 0 1,p (B r
2) ∗ . In view of the first part of the proof, we make the truncation
g(z, x) =
0 if x ≤ 0,
β(z)x p−1 − f (z, x) if 0 ≤ x ≤ ¯ u 2 0 (z), β(z)¯ u 2 0 (z) p−1 − f (x, ¯ u 2 0 (z)) if x > ¯ u 2 0 (z).
In this way we obtain a solution of (26) with r = r 1 which is located in the order interval [0, ¯ u 2 0
B
r1
]. But ¯ u 2 0 B
r1
= u 1 0 is the smallest solution of (26) (with r = r 1 ) in [0, ξ] ⊃ [0, ¯ u 2 0
B
r1
] by (35). Thus ¯ u 1 0 ≤ ¯ u 2 0 . Continuing this way, by induction it follows that
0 < ¯ u n 0 ≤ ¯ u n+1 0 in R N for all n ≥ 1, (36) and thus there exists
u 0 = lim
n→∞ u ¯ n 0 in R N .
For every open and bounded domain Ω ⊂ R N with C 2 boundary, there exists n 0 ∈ N such that Ω ⊂ B r
nfor all n ≥ n 0 ; from the nonlinear regularity theory (see, for example [16, p. 738]) we know that there exists α ∈ (0, 1) and M 0 > 0 such that
¯ u n 0
Ω ¯ ∈ C 1,α ( ¯ Ω) and k¯ u n 0
Ω ¯ k C
1,α( ¯ Ω) ≤ M 0 for all n ≥ n 0 . (37) From the compact embedding of C 1,α ( ¯ Ω) into C 1 ( ¯ Ω) and (36), (37), we have
¯ u n 0
Ω ¯ → u 0| ¯ Ω in C 1 ( ¯ Ω), so that
u 0 ∈ C 1 (R N ) and u 0 > 0 in R N . We also know that
A(u n 0 ) = λβ(u n n ) p−1 − λN f (u n 0 ) in W 0 1,p (B r
n) ∗ for all n ≥ 1. (38) In (38) we act with u n 0 ∈ W 0 1,p (B r
n) and obtain
Z
B
rnkDu n 0 k p dz = λ Z
B
rnβ(u n 0 ) p dz − λ Z
B
rnf (z, u n 0 )u n 0 dz
≤ λ Z
B
rnβ(u n 0 ) p dz (see H(f ))
≤ λ Z
B
rnβ + (u n 0 ) p dz
= λ Z
B
rn(β 1 + β 2 )(u n 0 ) p dz (see H(β)).
(39)
We have Z
B
rnβ 1 (u n 0 ) p dz ≤ ξ p Z
B
rn|β 1 |dz (by (35))
≤ ξ p Z
R
N|β 1 |dz
= c 9 = c 9 (ξ, kβk 1 ) > 0 (see H(β)).
(40)
Moreover, from the estimates found for (13) in the proof of Proposition 1, we have
Z
B
rnβ 2 (u n 0 ) p dz ≤ c 10 ε Z
B
rnkDu n 0 k p dz (41)
for some c 10 > 0.
We return to (39) and use (40), (41), obtaining Z
B
rnkDu n 0 k p dz ≤ c 9 + c 10 ε Z
B
rnkDu n 0 k p dz for all n ≥ 1. (42)
Now choose ε ∈ (0, 1) so small that c 10 ε < 1. Then from (42) it follows that Z
B
rnkDu n 0 k p dz ≤ c 11 for some c 11 > 0 and all n ≥ 1. (43)
But we know that D ¯ u n 0 |B
rn
= Du n 0 and D ¯ u n 0 |R
N\B
rn= 0 for all n ≥ 1, see, for example, Kesavan [20, p. 69,70]. Therefore, from (43) we have
kD¯ u n 0 k L
p(R
N,R
N) ≤ c 11 for all n ≥ 1. (44) We also have
0 ≤ Z
R
NkD¯ u n 0 k p dz = λ Z
R
Nβ(¯ u n 0 ) p dz − λ Z
R
Nf (z, ¯ u n 0 )¯ u n 0 dz ≤ λ Z
R
Nβ(¯ u n 0 ) p dz by H(f ), so that
λ Z
R
Nβ − (u n 0 ) p dz ≤ λ Z
R
Nβ + (u n 0 ) p dz. (45) From the estimates in the proof of Proposition 1 and (44), (45) it follows
λ Z
R
Nβ + (u n 0 ) p dz ≤ c 12 for some c 12 > 0 and all n ≥ 1. (46)
Then from (44) and (46), we infer that n
¯ u n 0 o
n≥1 ⊂ V is bounded, and so we may assume that
¯
u n 0 * u 0 in V. (47)
Let ¯ A : V → V ∗ be the nonlinear map defined by h ¯ A(y), vi V
∗,V =
Z
R
NkDyk p−2 (Dy, Dv) R
Ndz for all y, v ∈ V.
From Szulkin–Willem [24] (see also Huang [18]) we know that ¯ A is of type (S) +
in V . By the Claim above, we know that
A(¯ ¯ u n 0 ) = λβ(¯ u n 0 ) p−1 − λN f (¯ u n 0 ) in V ∗ for all n ≥ 1. (48) Acting on (48) with ¯ u n 0 − u 0 ∈ V , we obtain
h ¯ A(¯ u n 0 ), ¯ u n 0 − u 0 i = λ Z
R
Nβ(¯ u n 0 ) p−1 (¯ u n 0 − u 0 ) dz − λ Z
R
Nf (z, ¯ u n 0 )(¯ u n 0 − u 0 ) dz.
(49) By Szulkin–Willem [24], Lemma 4.2, we have
Z
R
Nβ(¯ u n 0 ) p−1 (¯ u n 0 − u 0 ) dz
= Z
R
Nβ + (¯ u n 0 ) p−1 (¯ u n 0 − u 0 ) dz − Z
R
Nβ − (¯ u n 0 ) p−1 (¯ u n 0 − u 0 ) dz → 0
(50)
as n → ∞. Moreover, for any r > 0
Z
R
Nf (z, ¯ u n 0 )(¯ u n 0 − u 0 )dz
≤ Z
R
Na ξ |¯ u n 0 − u 0 | dz (see (35) and H(f )(i))
= Z
B
ra ξ |¯ u n 0 − u 0 | dz + Z
R
N\B
ra ξ |¯ u n 0 − u 0 | dz.
(51)
Evidently, by H(f ))(i) and the compact embedding of V in L 1 loc (R N ), we have
n→∞ lim Z
B
ra ξ |¯ u n 0 − u 0 | dz = 0. (52) Moreover, given ε > 0, using H¨ older’s inequality, we get
Z
R
N\B
ra ξ |¯ u n 0 − u 0 | dz ≤ k¯ u n 0 − u 0 k L
p∗(R
N\B
r)
Z
R
N\B
ra (p ξ
∗)
0dz
! 1/(p
∗)
0≤ c 13 ε (53) for r > 0 large enough and some c 13 > 0.
Therefore, if in (51) we pass to the limit as n → ∞ and use (52) and (53), we obtain
n→∞ lim Z
R
Nf (z, ¯ u n 0 )(¯ u n 0 − u 0 ) dz
≤ c 13 ε, and since ε > 0 was arbitrary, we conclude that
n→∞ lim Z
R
Nf (z, ¯ u n 0 )(¯ u n 0 − u 0 ) dz = 0. (54)
We now return to (49), pass to the limit as n → ∞ and use (50) and (54), obtaining
n→∞ lim h ¯ A(¯ u n 0 ), ¯ u n 0 − u 0 i V
∗,V = 0, and since ¯ A is of type (S) + , we finally get
¯
u n 0 → u 0 in V. (55)
Note that for h ∈ V we have
|hN f (¯ u n 0 ) − N f (u 0 ), hi V
∗,V |
≤ Z
B
r[f (z, ¯ u n 0 ) − f (z, u 0 )]h dz
+ Z
R
N\B
r[f (z, ¯ u n 0 ) − f (z, u 0 )]h dz
. (56) From the continuity of the Nemitsky operator on the bounded domain B r , we have
n→∞ lim Z
B
r[f (z, ¯ u n 0 ) − f (z, u 0 )]h dz = 0. (57) On the other hand, via H¨ older’s inequality, as in (53), we obtain
Z
R
N\B
r[f (z, ¯ u n 0 ) − f (z, u 0 )]h dz
≤ c 14 ε (58)
for r > 0 large enough, for all n ≥ 1 and some c 14 > 0.
Thus, if in (56) we let n → ∞ and we use (57), (58) and the fact that ε > 0 was arbitrary, we conclude that
n→∞ lim hN f (¯ u n 0 ) − N f (u 0 ), hi V
∗,V = 0 for all h ∈ V, that is
N f (¯ u n 0 ) * N ∗ f (u 0 ) in V ∗ . (59) From (55) it follows that
Z
R
Nβ − (¯ u n 0 ) p−1 hdz → Z
R
Nβ − u p−1 0 hdz for all h ∈ V, while, invoking [24, Lemma 4.2], we have
Z
R
Nβ + (¯ u n 0 ) p−1 h dz → Z
R
Nβ + u p−1 0 h dz for all h ∈ V, so that
Z
R
Nβ(¯ u n 0 ) p−1 h dz → Z
R
Nβu p−1 0 h dz for all h ∈ V. (60) Therefore, if in (48) we pass to the limit as n → ∞ and we use (55), (59) and (60), obtaining
A(u ¯ 0 ) = λβu p−1 0 − λN f (u 0 ) in V ∗ ,
and so u 0 ∈ C 1 (R N ) ∩ V is a solution of (P λ ) and
u 0 > 0 in R N . Finally, we show that lim
kzk→∞ u 0 (z) = 0. For this, it is enough to prove that the problem
( −∆ p v = λβ + u p−1 0 ∈ L ∞ loc (R N ) in R N ,
v > 0 and v(z) → 0 as kzk → ∞ (61)
has a bounded solution. Indeed, if this is the case, we find
−∆ p v = λβ + u p−1 0 ≥ λβ + (¯ u n 0 ) p−1 ≥ λβ(¯ u n 0 ) p−1 − f (z, ¯ u n 0 ) = −∆ p u ¯ n 0 . Moreover, by (36), we know that 0 ≤ ¯ u n 0 ≤ ξ and ¯ u n 0 = 0 at infinity for every n ∈ N, so that ¯ u n 0 ≤ v by Theorem 3.5.1 and the following Remark in [22]; hence the claim follows by (55). Thus we are reduced to consider problem (61) and to show that it admits a solution vanishing at infinity. But finding a bounded solution of (61) is equivalent to finding a bounded one for
( −∆ p w = λβ + u p−1 0 w q−1 in R N ,
w > 0 and w(z) → 0 as kzk → ∞ (62)
with q < p, see Abdellaoui-Peral [1, Theorem 4.6]. Hence, let us prove that (62) actually admits a C 1 solution vanishing at infinity, as required. For this, it is enough to apply a result of Santos [23, Theorem 1.1], whose assumptions (H 1 ) and (H 2 ) are satisfied with a = β + u p−1 0 and b ≡ 0, while hypothesis (H 3 ) is a consequence of H(β) 2 and of the fact that u 0 is bounded.
Next, we prove that for λ ∈ [0, ˆ λ ∗ ] problem (P λ ) has no positive solutions.
Proposition 6. If hypotheses H(β) and H(f ) hold and λ ∈ (0, ˆ λ ∗ ], then prob- lem (P λ ) has no positive solutions.
Proof. Suppose that (P λ ) has a positive solution u. Then, as usual, u ∈ C 1 (R N ) ∩ V and
−∆ p u(z) = λ β(z)u(z) p−1 − f (z, u(z)) in R N . By definition of solution and by assumption, we have
Z
R
NkDuk p dz = λ Z
R
N[βu p − f (z, u)u] dz < λ Z
R
Nβu p dz (see H(f )). (63)
Now, choose ζ ∈ C c 1 (R N ) such that 0 ≤ ζ ≤ 1, ζ
{kzk≤1} = 1 and ζ
{kzk≥2} = 0.
Let ζ n (z) = ζ z n and define u n = ζ n u. Then u n (z) → u(z) for all z ∈ R N .
Since u ∈ D 1,p (R N ), from the Sobolev Embedding theorem, we have that u ∈
L p
∗(R N ). Since 0 ≤ u n ≤ u for any n ≥ 1, from the Lebesgue Dominated Convergence Theorem we infer that
u n → u in L p
∗(R N ). (64)
Consider the ring R n = n
z ∈ R N : n < kzk < 2n o
. We have kDu n − Duk L
p(R
N,R
N) ≤ k(ζ n − 1)Duk L
p(R
N,R
N) + kuDζ n k L
p(R
N,R
N)
≤ k(ζ n − 1)Duk L
p(R
N,R
N) + kuk p L
p∗(R
N) kDζ n k N −p L
N(R
N) (65) by H¨ older’s inequality.
Since Du ∈ L p (R N , R N ) and ζ n (z) → 1 for all z ∈ R N as n → ∞, from the Lebesgue Dominated Convergence Theorem we have
k(ζ n − 1)Duk L
p(R
N,R
N) → 0 as n → ∞. (66) Moreover, we have
kDζ n k L
N(R
n) = kDζ n k L
N(R
N,R
N) = kDζk L
N(R
N,R
N) < ∞ (67) and
kuk L
p∗(R
n) → 0 as n → ∞, since u ∈ L p
∗(R N ). (68) From (67) and (68) it follows that
kuk p L
p∗(R
n