Quasi uniformity for the abstract Neumann antimaximum principle and applications with a
priori estimates
Genni Fragnelli Dipartimento di Matematica Universit` a degli Studi di Bari Via E. Orabona 4, 70125 Bari Italy
email: fragnelli@dm.uniba.it Dimitri Mugnai
Dipartimento di Matematica e Informatica Universit` a degli Studi di Perugia Via Vanvitelli 1, 06123 Perugia Italy
email: mugnai@dmi.unipg.it
Abstract
In this paper we prove a general result giving the maximum and the antimaximum principles in a unitary way for linear operators of the form L + λI, provided that 0 is an eigenvalue of L with associated constant eigenfunctions. To this purpose, we introduce a new notion of “quasi”–
uniform maximum principle, named k–uniform maximum principle: it holds for λ belonging to certain neighborhoods of 0 depending on the fixed positive multiplier k > 0 which selects the good class of right–hand–
sides. Our approach is based on a L
∞− L
pestimate for some related problems. As an application, we prove some generalization and new re- sults for elliptic problems and for time periodic parabolic problems under Neumann boundary conditions.
Keywords: Maximum principle, antimaximum principle, uniform maximum principle with fixed multiplier, polyharmonic operators, periodic parabolic prob- lems
2000AMS Subject Classification: Primary 35B50; Secondary 35J25, 35J40,
35K10, 34C25
1 Introduction and abstract setting
Let us start considering the easy problem
( ∆u + λu = f (x) in Ω,
Bu = 0 on ∂Ω, (1.1)
where Ω is a bounded domain of R N , N ≥ 1, B denotes Dirichlet or Neumann boundary conditions and f ∈ L p (Ω). Denoting by λ 1 the first eigenvalue of −∆
under the corresponding homogeneous boundary conditions Bu = 0, a classical result is the following maximum principle:
(MP): if λ < λ 1 , then for any f ≥ 0 the associated solution u ≤ 0 in Ω.
On the other hand, a related stronger version, namely the strong maximum principle, holds:
(SMP): if in addition f 6= 0, then u < 0 in Ω.
We immediately observe that these results present two main features: they are uniform in λ and no special functional space of right hand sides f is involved.
Moreover, it is now well know that jumping after λ 1 changes the situation a lot: indeed, Cl´ement and Peletier in [8] proved the following antimaximum principle:
(AMP): for any f ≥ 0 in L p (Ω), p > N , there exists δ = δ(f ) > 0 such that if u solves (1.1) with λ ∈ (λ 1 , λ 1 + δ), then u ≥ 0 in Ω.
They also showed that under Neumann boundary conditions a uniform anti- maximum principle, i.e.
(UAMP): in (AMP) it is possible to take δ independent of f holds only when N = 1.
Therefore, it is clear that, in general, (AMP) is not uniform, while (MP) is.
Moreover, (MP) needs no special requirements on p, while (AMP) needs p > N , as showed in [30]: under Dirichlet boundary conditions, there exists f ∈ L N (Ω) with f > 0 such that for all λ > λ 1 different from an eigenvalue of −∆ in H 0 1 (Ω), the solution of (1.1) changes sign.
After [8] a large number of refinements of (UAMP) have been established.
We recall [6] for higher order ODE’s with periodic boundary conditions, while
for general second order PDE’s with Neumann or Robin boundary conditions
we mention [20] (where again it is proved that (UAMP) holds only if N = 1),
while polyharmonic operators in low dimensions (say for all those dimensions
for which the natural Sobolev space containing weak solutions is embedded in
C 0 (Ω)) are considered in [9], [10], [23]. In [30] it is showed that the condition
p > N in [8] is sharp for the validity of an antimaximum principle when L = ∆
under Dirichlet boundary conditions: in this context, taking a right–hand side
f ∈ L 2 (Ω), forces to assume N = 1 also for the validity of (AMP) without any
uniformity. Finally, we recall a local extension in R N proved in [12] and some extensions in presence of the p–Laplace operator in [5] and [21].
Recently, Campos–Mawhin–Ortega in [7] show that it is possible to state maximum and antimaximum principles in a unitary way, at least in some cases.
Roughly speaking, having in mind (1.1) with Lu = u ′′ and Neumann boundary conditions, so that λ 1 = 0, they start with the following definition of maximum principle:
Definition 1.1. Given λ ∈ R \ {0}, we say that the operator L + λI satisfies a maximum principle if for every f ∈ L 1 (Ω) the equation
Lu + λu = f, u ∈ Dom(L) ⊂ C 0 (Ω) (1.2) has a unique solution with λu ≥ 0 for any f ≥ 0. Moreover, the maximum principle is said to be strong if λu(x) > 0 for any x ∈ Ω whenever f ≥ 0 and f (x) > 0 in a subset of Ω with positive measure.
It is clear that the definition above covers the case of a “classical” maximum principle when λ < λ 1 = 0 and a “classical” antimaximum principle when λ > 0;
more precisely, we remark that for λ > 0 Definition 1.1 includes a (UAMP) tout court.
The main result in [7] sounds as the following Theorem 1.2 ([7]). There exist λ − and λ + such that
−∞ ≤ λ − < 0 < λ + ≤ +∞
and L + λI satisfies a maximum principle if and only if λ ∈ [λ − , 0) ∪ (0, λ + ].
Moreover the maximum principle is strong if λ ∈ (λ − , 0) ∪ (0, λ + ).
The basic ingredient of [7] is an L ∞ − L 1 estimate for solution–datum of the form
kuk L
∞(Ω) ≤ Mkfk L
1(Ω) ,
which is common and natural for ODE’s and for the wave equation in 1D. On the other hand, for the classical theory of elliptic problems like (1.1), by the classical Agmon–Douglis–Nirenberg result ([1]), if Ω is smooth enough, say of class C 2 , a more natural setting would be a W 2,p − L p estimate of the form
kuk W
2,p(Ω) ≤ Mkfk L
p(Ω) , (1.3) since data belonging to L 1 are not the good ones to perform a standard varia- tional approach (we refer to [3] for this case). For example, the easy problem
( ∆u + λu = f in Ω,
∂u
∂ν = 0 on ∂Ω, (1.4)
where Ω is a bounded domain of R N , N > 1, cannot be handled by Theorem 1.2
if f ∈ L 1 (Ω), since L 1 is not contained in the dual of the Sobolev space where
weak solutions are sought.
On the other hand, by Morrey’s Theorem, if N < 2p, then W 2,p (Ω) ֒→
C 0,α (Ω), for some α ∈ (0, 1), so that (1.3) implies kuk C
0(Ω) ≤ Mkfk L
p(Ω) , where, of course, kuk C
0(Ω) = max Ω |u| = kuk L
∞(Ω) .
In this paper we want to combine the spirit of [7] with the result stated in the other papers cited above showing two possible generalizations of Theorem 1.2. To this purpose, we first prove that, although a (UAMP) cannot hold (as for the Laplace operator with Neumann conditions in dimension greater than 1), a “quasi–(UAMP)” does (in the sense of k–(UMP) introduced in Definition 1.4 below), thus extending to any p ∈ (1, ∞) the sufficient condition proved for the case p = 2 in [16]. Moreover, with an additional hypothesis, we characterize completely the set of λ’s for which a strong k–(UMP) holds. To do this, we will also show some a priori estimates of independent interest, and some of which are in the spirit of bifurcation from 0. Finally, we also show the sharpness of the exponent p > N/2 for problem (1.4). In this context, we believe that our results can help understanding better all the situation.
We start introducing the abstract framework. Let Ω ⊂ R N be a bounded domain with C 2 boundary ∂Ω; a positive and finite measure µ is given in Ω, and we write L q (Ω) := L q (Ω, µ) for any q ∈ [1, ∞].
From now on, we fix p ∈ (1, ∞) and for any f ∈ L p (Ω) we set f := 1
µ(Ω) Z
Ω
f dµ and f := f − f. ˜ We also set
L p := n
f ∈ L p (Ω) : f = 0 o
and C := C ˜ 0 (Ω) ∩ L p ; finally, fixed k > 0, we introduce the cone
F k p := n
f ∈ L p (Ω) : k ˜ f k L
p(Ω) ≤ kkfk L
1(Ω)
o and
P k p := n
f ∈ F k p : f ≥ 0 o , i.e. P k p is the sub–cone of positive functions in F k p .
Remark 1.3. It is clear that any f ∈ L p (Ω) belongs to a suitable F k p and to F ℓ p for any ℓ ≥ k. Moreover, ∪ k F k p ⊂ L 1 (Ω) with strict inclusion.
We now consider a linear operator L : Dom(L) ⊂ C 0 (Ω) → L p (Ω) satisfying the following properties:
Ker(L) = n
constant functions o
, Im(L) = L p , (1.5)
the problem Lu = ˜ f has a unique solution ˜ u ∈ ˜ C and ∃ M = M(L) > 0 such that k˜uk C
0(Ω) ≤ Mk ˜ f k L
p(Ω) .
(1.6)
We remark that these requirements are the natural extensions to our setting of the assumptions made in [7]. Therefore, having in mind Definition 1.1, we give the following uniform maximum principle with fixed multiplier k > 0.
Definition 1.4. Fix k > 0. Given λ ∈ R \ {0}, we say that the operator L + λI satisfies a uniform maximum principle with multiplier k, k–(UMP) for short, if for every f ∈ F k p equation (1.2) has a unique solution u ∈ Dom(L) ⊂ C 0 (Ω)
(1.7) with λu ≥ 0 when f ≥ 0. We say that a strong k–(UMP) holds if λu(x) > 0 for any x ∈ Ω whenever f ≥ 0 and f(x) > 0 in a subset of Ω having positive measure.
Remark 1.5. Again as in Definition 1.1, the case λ < 0 corresponds to a clas- sical maximum principle, while the case λ > 0 corresponds to an antimaximum principle, which is “almost” uniform, since f belongs to F k p and not to the whole of L p (Ω).
Our first result concerns the existence of a neighborhood U k of 0 such that U k \ {0} ⊂ n
λ ∈ R : L + λI satisfies a k–(UMP) o . More precisely:
Theorem 1.6. Assume that (1.5) and (1.6) hold and fix k > 0. Then there exists Λ = Λ(k) > 0 such that L + λI satisfies a k–(UMP) if λ ∈ [−Λ, Λ] \ {0}.
Moreover a strong k–(UMP) holds if λ ∈ (−Λ, Λ) \ {0}.
Remark 1.7. As it will be clear from the proof, Λ = Λ(k) does not depend on p, so that the result is uniform also with respect to the integrability exponent.
As a corollary of the previous result, we obtain the validity of an (AMP), also in a quasi–uniform way, for problem (1.4) with f in L p (Ω), p > N/2, already established in [7] for f in L 2 (Ω) and N = 1, 2, 3. On the other hand, we underline the fact that our somehow “standard” everywhere approach is quite natural for antimaximum principles, and so we cannot extend our results to an almost everywhere fashion, typical of a (MP) also for more general PDE’s and differential inequalities, possibly on Riemannian manifolds (see the recent [2], [27]) or in anisotropic settings ([13]).
Let us note that, in some sense, any function f defines a whole set of
“good” right–hand–sides, in the sense that a crucial rˆole is played by the ratio k ˜ f k L
p(Ω) /kfk L
1(Ω) .
Moreover, as a direct consequence of Theorem 1.6, in contrast to the Dirichlet
case considered in [30], we get the following result:
Corollary 1.8. Concerning problem (1.4), L N is not sharp for the validity of the antimaximum principle.
Actually, this fact was already noticed in [20, Remark 4.5] and in [10, Propo- sition 2], were the sufficiency of the condition “p > N/2” was stated for the va- lidity of the (AMP), but not for any form of uniformity, as we give in Theorem 1.6. Moreover, the following result holds:
Proposition 1.9. In problem (1.4), L N/2 (Ω) is sharp for the validity of the antimaximum principle.
Proof. It follows the lines of [30] with obvious adaptations, so we will be sketchy, though we seize this opportunity to make precise a small step in the proof of the Dirichlet case which contained a misleading missprint. Assume Ω = B 1 , the unit ball, and take
f (x) = 1
|x| 2 (2 − log |x|)
N − 2 + 1
2 − log |x|
∈ L N/2 (Ω).
Calculations show that v(x) = log(2 − log |x|) solves ∆v = −f. Now set w λ = u λ + χv, where χ is a smooth function with χ(x) = 1 if |x| ≤ 1/2, χ(x) = 0 if
|x| ≥ 7/8, and u λ solves problem (1.4). Thus w λ solves the problem ( ∆w = −λw + (1 − χ)f + v∆χ + 2Dχ · Dv + λχv in B 1
∂w/∂ν = 0 on ∂B 1 .
Since (1 − χ)f + v∆χ + 2Dχ · Dv + λχv ∈ L q (B 1 ) for any q ≥ 1, choosing λ different from an eigenvalue implies that w λ ∈ C( ¯ B 1 ). Hence
x→0 lim w λ (x)
v λ (x) = 0,
so that near 0 we have u λ (x) = − log(2 − log |x|) + w λ < 0.
Our final results, together with Theorem 1.6, give an almost complete char- acterization of the set of λ’s for which L + λI satisfies a k–(UMP), in the spirit of Theorem 1.2. Of course, due to the validity of a classical maximum principle for λ < 0, in the following we concentrate on λ > 0. As usual, we denote by
R λ = (L + λI) −1 : L p (Ω) → C 0 (Ω) the resolvent of L, whenever it exists. Finally, let us set
Λ + (k) := n
λ > 0 : L + λI satisfies a k–(UMP) o
and λ + = λ + (k) := sup Λ + (k). In the next result we show that the set of λ’s for which a k–(UMP) holds is an interval, in whose interior a strong k–(UMP) is satisfied; moreover, such a result characterizes completely such an interval.
We focus on λ + , an analogous result being valid for λ − .
Theorem 1.10. Assume (1.5) and (1.6), and suppose that for all λ < λ + the following condition holds:
if L + λI satisfies a k–(UMP), then R λ (P k p ) ⊆ P k p . (1.8) Then L + λI satisfies a k–(UMP) if and only if λ ∈ (0, λ + ]. Moreover, the k–(UMP) is strong if λ ∈ (0, λ + ).
In general, it is not clear if condition (1.8) can be verified, and without it, we cannot prove that L + λ + I satisfies a k–(UMP). For this, a useful tool is the following result on a priori estimates for R λ (P k p ):
Proposition 1.11 (A priori estimates). Take λ > 0 and suppose that L + λI satisfies a k–(UMP). Then, for any nonconstant f ∈ P k p , setting u = R λ f , we have
if k ˜ f k L
p(Ω) < λk˜uk L
p(Ω) , then 2M µ(Ω) 1/p λ > 1. (1.9) Moreover, if λk˜uk L
p(Ω) ≤ k ˜ f k L
p(Ω) , then u ∈ P h p , where
h = min n
2M µ(Ω) 1/p λk, k o
≤ k. (1.10)
If M µ(Ω) 1/p λ < 1, then u ∈ P h p , where
h = max
min n
2M µ(Ω) 1/p λk, k o
, M µ(Ω) 1/p λ 1 − Mµ(Ω) 1/p λ k
. (1.11) As a consequence, we have the
Corollary 1.12. Suppose that L + λI satisfies a k–(UMP) and 2M µ(Ω) 1/p λ ≤ 1. Then, for every f ∈ P k p , also R λ f ∈ P k p .
Proof. It is an immediate consequence of (1.10), since h ≤ k.
Remark 1.13. This corollary implies that condition (1.8) is satisfied if λ is small, in the spirit of bifurcation.
As a straightforward consequence, in addition to (1.10), we have the following sufficient condition for Theorem 1.10:
Proposition 1.14. Assume (1.5) and (1.6); moreover, suppose that λ + ∈ R and that 2M µ(Ω) 1/p λ + ≤ 1. Then the conclusion of Theorem 1.10 holds.
2 Proof of Theorem 1.6
We introduce the operator ˜ R 0 : L p → ˜ C defined by
˜
u = ˜ R 0 f ⇐⇒ L˜u = ˜ ˜ f , which is well defined by assumption (1.6).
The first lemma we prove gives a condition that ensures the existence of R λ .
Lemma 2.1. There exists Λ 1 > 0 such that for all λ ∈ [−Λ 1 , Λ 1 ] \ {0} the resolvent R λ : L p (Ω) → C 0 (Ω) of L is well defined. Moreover, there exists C > 0 such that, if ˜ f ∈ L p and λ ∈ [−Λ 1 , Λ 1 ] \ {0} then
kR λ f k ˜ C
0(Ω) ≤ Ck ˜ f k L
p(Ω) . More precisely, C := M
1−Λ
1k ˜ R
0k
C→ ˜˜ C, where M is the constant appearing in (1.6).
Here k ˜ R 0 k C→ ˜ ˜ C denotes the norm of the restriction of the operator ˜ R 0 from C to ˜ ˜ C, which is well defined, since ˜ C ⊂ L p .
Proof. Rewrite equation (1.2) as the system ( L˜ u + λ˜ u = ˜ f ,
λu = f . (2.1)
Applying ˜ R 0 , the first equation in (2.1) can be rewritten as
(I + λ ˜ R 0 )˜ u = ˜ R 0 f . ˜ (2.2) Now, observe that ˜ R 0 f ∈ ˜ ˜ C; take |λ| < 1/k ˜ R 0 k C→ ˜ ˜ C , so that I + λ ˜ R 0 is invertible from ˜ C to ˜ C, and equation (2.2) is solved by ˜u = (I + λ ˜ R 0 ) −1 R ˜ 0 f . ˜
Finally, take Λ 1 ∈ (0, 1/k ˜ R 0 k C→ ˜ ˜ C ); therefore, from the triangle inequality, (2.2) and (1.6), for all λ with |λ| ≤ Λ 1 we have that
k˜uk L
∞(Ω) − Λ 1 k ˜ R 0 k C→ ˜ ˜ C k˜uk L
∞(Ω) ≤ k˜uk L
∞(Ω) − |λ|k ˜ R 0 k C→ ˜ ˜ C k˜uk L
∞(Ω)
≤ k(I + λ ˜ R 0 )˜ uk L
∞(Ω) = k ˜ R 0 f k ˜ L
∞(Ω)
= k˜uk L
∞(Ω) ≤ Mk ˜ f k L
p(Ω) . The thesis follows.
Note that the proof above provides the estimate Λ 1 < 1
k ˜ R 0 k C→ ˜ ˜ C
.
After proving that R λ is well defined, we have our second
Lemma 2.2. For any k > 0 there exists Λ 2 := Λ 2 (k) ∈ (0, Λ 1 ] such that for all λ ∈ [−Λ 2 , Λ 2 ] \ {0} the operator L + λI has a k–(UMP). Moreover, a strong k–(UMP) holds if λ ∈ (−Λ 2 , Λ 2 ) \ {0}.
Proof. If f ∈ F k p , f ≥ 0, then f = µ(Ω) 1 kfk L
1(Ω) . Thus, using the second equation in (2.1), one has
λu = λR λ ( ˜ f + f ) = λR λ ( ˜ f ) + f = λR λ ( ˜ f ) + 1
µ(Ω) kfk L
1(Ω)
≥ 1
µ(Ω) kfk L
1(Ω) − |λ|kR λ f k ˜ L
∞(Ω) .
By the previous Lemma it results that if λ ∈ [−Λ 1 , Λ 1 ] \ {0}, then λu ≥ 1
µ(Ω) kfk L
1(Ω) − |λ| M 1 − Λ 1 k ˜ R 0 k C→ ˜ ˜ C
k ˜ f k L
p(Ω)
≥ 1
µ(Ω) − k|λ| M 1 − Λ 1 k ˜ R 0 k C→ ˜ ˜ C
!
kfk L
1(Ω) ,
since f ∈ F k p . The Lemma is thus completely proved, provided that
Λ 2 = min (
Λ 1 , 1 − Λ 1 k ˜ R 0 k C→ ˜ ˜ C
kM µ(Ω) )
.
Proof of Theorem 1.6. Fixed k > 0, the result follows by taking Λ = Λ 2 (k) as given in Lemma 2.2.
3 Proof of Theorem 1.10
In this section we prove Theorem 1.10, so that we assume all the conditions of the previous section. Let us fix k > 0.
Lemma 3.1. Assume that there exists λ 0 > 0 such that L+λ 0 I has a k–(UMP) and R λ
0(P k p ) ⊆ P k p . Then R λ : P k p → C 0 (Ω) is defined for all λ ∈ (0, λ 0 ] and
1. the function λ 7→ R λ f is analytic in (0, λ 0 ] for all f ∈ P k p ; 2. L + λI satisfies a k–(UMP) for all λ ∈ (0, λ 0 ];
3. L + λI satisfies a strong k–(UMP) for all λ ∈ (0, λ 0 ).
Proof. First, by (1.7), we immediately obtain R λ
01 = λ 1
0, and 1 ∈ P k p for any k > 0, since ˜1 = 0.
Now, take f ∈ P k p ∩C 0 ( ¯ Ω); using the k–(UMP), the inclusion R λ
0(P k p ) ⊆ P k p , assumption (1.6) and system (2.1) for λ = λ 0 , we immediately get
0 ≤ R λ
0f = ˜ u + u ≤ Mk ˜ f − λ 0 uk ˜ L
p(Ω) + f
λ 0 ≤ Mk(kfk L
1(Ω) + λ 0 kuk L
1(Ω) ) + f λ 0
= 2M kkfk L
1(Ω) + f λ 0 ≤
2M kµ(Ω) + 1 λ 0
kfk C
0(Ω) .
(3.1) Thus
kR λ
0k = sup
f ∈Pp k∩C0( ¯Ω)
kf k
C0(Ω)≤1
kR λ
0f k C
0(Ω) ≤ η + 1 λ 0
,
where we have set η := 2M kµ(Ω)λ 0 .
Now, for any λ ∈ R we rewrite Lu + λu = f as Lu + λ 0 u = f + (λ 0 − λ)u, and applying R λ
0, we get
u − (λ 0 − λ)R λ
0u = R λ
0f. (3.2) If (λ 0 −λ)kR λ
0k < 1, then I −(λ 0 −λ)R λ
0is invertible from the cone P k p ∩C 0 (Ω) to itself and (3.2) is solvable, since R λ
0f ∈ P k p ∩ C 0 (Ω).
Now, if kR λ
0k = 1/λ 0 , we find
(λ 0 − λ)kR λ
0k < 1 ⇐⇒ λ 0 − λ
λ 0 < 1 ⇐⇒ λ > 0, which implies that (3.2) is solvable for any λ ∈ (0, λ 0 ).
On the other hand, if kR λ
0k > 1/λ 0 , since kR λ
0k ≤ η+1 λ
0, we have λ > η
η + 1 λ 0 ⇐⇒ (λ 0 − λ) η + 1
λ 0 < 1 =⇒ (λ 0 − λ)kR λ
0k < 1.
This implies that if λ ∈
η η+1 λ 0 , λ 0
, then (3.2) is solvable.
If f ∈ P k p then R λ
0f ∈ P k p ∩ C 0 (Ω) by assumption and (3.2) is solvable as well, provided that λ ∈
η η+1 λ 0 , λ 0
, or λ ∈ (0, λ 0 ) if kR λ
0k = 1/λ 0 , the easier case that we don’t discuss.
Now, since u = [I − (λ 0 − λ)R λ
0] −1 R λ
0f by (3.2), from Neumann’s formula for operators we get
R λ f = [I − (λ 0 − λ)R λ
0] −1 R λ
0f =
∞
X
n=0
(λ 0 − λ) n R n λ
0!
R λ
0f, (3.3)
and thus the function λ 7→ R λ f is analytic.
Since f ≥ 0, by assumption we have R λ
0f ≥ 0. Moreover, R λ
0f be- longs to P k p by hypothesis; so we can iterate obtaining R λ f ≥ 0 whenever λ ∈
η η+1 λ 0 , λ 0
i and f ≥ 0.
In order to conclude the proof of claims 1 and 2, let us show that it is possible to extend the results above also for λ ∈
0, η+1 η λ 0
i . Indeed, by assumption (1.6) and system (2.1) for λ 1 = η+ η
12
λ 0 , operating as for (3.1), we find kR λ
1f k C
0(Ω) ≤ k˜uk C
0(Ω) + kuk C
0(Ω) ≤ Mk ˜ f k L
p(Ω) + f
λ 1
≤ Mkkfk L
1(Ω) + f
λ 1 ≤ η λ 0
+ 1 λ 1
kfk C
0(Ω) ≤ η + 1
λ 1 kfk C
0(Ω) . Moreover, since λ 1 ∈
η η+1 λ 0 , λ 0
, by the previous part we know that R λ
1(P k p ) ⊆
P k p , so that we can proceed as above, extending the results for λ belonging to the
interval
η η+1 λ 1 , λ 0
=
η η+1
η η+
12λ 0 , λ 0
. Now, we repeat the procedure above starting from η+1 η η+ η
12