ANNALES DE
L’INSTITUT FOURIER
LesAnnales de l’institut Fouriersont membres du
Wolfgang Arendt & A. F. M. ter Elst
The Dirichlet problem without the maximum principle Tome 69, no2 (2019), p. 763-782.
<http://aif.centre-mersenne.org/item/AIF_2019__69_2_763_0>
© Association des Annales de l’institut Fourier, 2019, Certains droits réservés.
Cet article est mis à disposition selon les termes de la licence Creative Commons attribution – pas de modification 3.0 France.
http://creativecommons.org/licenses/by-nd/3.0/fr/
THE DIRICHLET PROBLEM
WITHOUT THE MAXIMUM PRINCIPLE
by Wolfgang ARENDT & A. F. M. TER ELST (*)
Abstract. — Consider the Dirichlet problem with respect to an elliptic oper- ator
A=−
d
X
k,l=1
∂kakl∂l−
d
X
k=1
∂kbk+
d
X
k=1
ck∂k+c0
on a bounded Wiener regular open set Ω ⊂ Rd, whereakl, ck ∈ L∞(Ω,R) and bk, c0∈L∞(Ω,C). Suppose that the associated operator onL2(Ω) with Dirichlet boundary conditions is invertible. Then we show that for allϕ∈C(∂Ω) there exists a uniqueu∈C(Ω)∩Hloc1 (Ω) such thatu|∂Ω=ϕandAu= 0.
In the case when Ω has a Lipschitz boundary andϕ∈C(Ω)∩H1/2(Ω), then we show thatucoincides with the variational solution inH1(Ω).
Résumé. — Considérons le problème de Dirichlet par rapport à un opérateur elliptique
A=−
d
X
k,l=1
∂kakl∂l−
d
X
k=1
∂kbk+
d
X
k=1
ck∂k+c0
sur un ensemble ouvert régulier de Wiener borné Ω⊂Rd, oùakl, ck∈L∞(Ω,R) et bk, c0 ∈ L∞(Ω,C). Supposons que 0 n’est pas une valeur propre de Aavec conditions aux limites Dirichlet. Alors nous montrons que pour toutϕ∈C(∂Ω) il existe un uniqueu∈C(Ω)∩Hloc1 (Ω) tel queu|∂Ω=ϕetAu= 0.
Dans le cas où Ω a une frontière Lipschitz etϕ∈C(Ω)∩H1/2(Ω), nous montrons queucoïncide avec la solution variationnelle dansH1(Ω).
Keywords:Dirichlet problem, Wiener regular, holomorphic semigroup.
2010Mathematics Subject Classification:31C25, 35J05, 31B05.
(*) The second-named author is most grateful for the hospitality extended to him during a fruitful stay at the University of Ulm. He wishes to thank the University of Ulm for financial support. Part of this work is supported by an NZ-EU IRSES counterpart fund and the Marsden Fund Council from Government funding, administered by the Royal Society of New Zealand. Part of this work is supported by the EU Marie Curie IRSES program, project “AOS”, No. 318910.
1. Introduction
Let Ω ⊂ Rd be an open bounded set with boundary Γ. Throughout this paper we assume that d > 2. The classical Dirichlet problem is to find for each ϕ ∈ C(Γ) a function u ∈ C(Ω) such that u|Γ = ϕ and
∆u= 0 as distribution on Ω. The set Ω is calledWiener regularif for every ϕ∈C(Γ) there exists a uniqueu∈C(Ω) such thatu|Γ =ϕand ∆u= 0 as distribution on Ω.
The Dirichlet problem has been extended naturally to more general second-order operators. For all k, l ∈ {1, . . . , d} let akl: Ω → R be a bounded measurable function and suppose that there exists aµ >0 such that
(1.1) Re
d
X
k,l=1
akl(x)ξkξl>µ|ξ|2
for allx∈Ω andξ∈Cd. Further, for allk∈ {1, . . . , d}letbk, ck, c0: Ω→C be bounded and measurable. Define the mapA:Hloc1 (Ω)→ D0(Ω) by
hAu, viD0(Ω)×D(Ω)=
d
X
k,l=1
Z
Ω
akl(∂ku)∂lv+
d
X
k=1
Z
Ω
bku ∂kv
+
d
X
k=1
Z
Ω
ck(∂ku)v+ Z
Ω
c0u v for allu∈Hloc1 (Ω) andv∈Cc∞(Ω). Givenϕ∈C(Γ), by aclassical solution of the Dirichlet problem we understand a function u ∈ C(Ω)∩Hloc1 (Ω) satisfyingAu= 0 andu|Γ =ϕ. For the pure second-order case (that isbk= ck = c0 = 0) Littman–Stampacchia–Weinberger [11] proved that for all ϕ∈C(Γ) there exists a unique classical solutionu. Then Stampacchia [13, Théorème 10.2] added real valued lower order terms, under the condition (see [13], (9.2’)) that there exists aµ0>0 such that
(1.2)
Z
Ω
c0v+
d
X
k=1
Z
Ω
bk∂kv>µ0 Z
Ω
v
for allv∈Cc∞(Ω)+. Gilbarg–Trudinger [10, Theorem 8.31] merely assume that
(1.3)
Z
Ω
c0v+
d
X
k=1
Z
Ω
bk∂kv>0
for allv∈Cc∞(Ω)+ in order to obtain the same conclusion. A consequence of these assumptions is a weak maximum principle, which implies that
kukC(Ω) 6 kϕkC(Γ) for all u ∈ Hloc1 (Ω)∩C(Ω) satisfying Au = 0 and u|Γ =ϕ. We may consider (1.3) as a kind of submarkov condition since it is equivalent to−A1Ω60 inD0(Ω).
The aim of this paper is to show that the positivity condition (1.3) and the maximum principle are not needed for the well-posedness of the Dirichlet problem. In addition we allow thebkandc0to be complex valued.
In order to state the main results of this paper in a more precise way we need a few definitions. Define the forma:H1(Ω)×H1(Ω)→Cby
(1.4) a(u, v) =
d
X
k,l=1
Z
Ω
akl(∂ku)∂lv+
d
X
k=1
Z
Ω
bku ∂kv
+
d
X
k=1
Z
Ω
ck(∂ku)v+ Z
Ω
c0u v.
LetAD be the operator inL2(Ω) associated with the forma|H1
0(Ω)×H01(Ω). In other words,AD is the realisation of the elliptic operator A in L2(Ω) with Dirichlet boundary conditions. This operator has a compact resolvent.
Moreover, if (1.3) is valid, then kerAD={0}by [10, Corollary 8.2]. Instead of (1.3) we assume the condition kerAD={0}, which is equivalent to the uniqueness of the Dirichlet problem (cf. Proposition 2.3 below).
The main result of this paper is the following well-posedness result for the Dirichlet problem.
Theorem 1.1. — Let Ω ⊂ Rd be an open bounded Wiener regular set with d > 2. For all k, l ∈ {1, . . . , d} let akl: Ω → R be a bounded measurable function and suppose that there exists aµ >0such that
Re
d
X
k,l=1
akl(x)ξkξl>µ|ξ|2
for allx∈Ωand ξ∈Cd. Further, for all k∈ {1, . . . , d} letbk, c0: Ω→C andck: Ω→Rbe bounded and measurable. LetADbe as above. Suppose 06∈σ(AD). Then for allϕ∈C(Γ)there exists a uniqueu∈C(Ω)∩Hloc1 (Ω) such thatu|Γ =ϕandAu= 0.
Moreover, there exists a constant c >0such that kukC(Ω)6ckϕkC(Γ)
for all ϕ ∈ C(Γ), where u ∈ C(Ω)∩Hloc1 (Ω) is such that u|Γ = ϕ and Au= 0.
Instead of the homogeneous equation Au = 0 one can also consider the inhomogeneous equation Au= f0+Pd
k=1∂kfk. We shall do that in Theorem 2.13.
Adopt the notation and assumptions of Theorem 1.1. DefineP:C(Γ)→ C(Ω) by P ϕ = u, where u ∈ C(Ω)∩Hloc1 (Ω) is such that u|Γ = ϕ and Au= 0. Note thatP ϕis theclassical solutionof the Dirichlet problem.
If Ω has even a Lipschitz boundary (which implies Wiener regularity), then there is also a variational solution of the Dirichlet problem that we describe next. Denote by Tr :H1(Ω) → L2(Γ) the trace operator. Again letakl, bk, ck, c0 ∈L∞(Ω) and suppose that the ellipticity condition (1.1) is satisfied. Further suppose that 06∈σ(AD). Then for eachϕ∈TrH1(Ω) there exists a uniqueu∈H1(Ω), called thevariational solution, such that Au= 0 and Tru=ϕ(cf. Lemma 2.1). Defineγ: TrH1(Ω)→ H1(Ω) by settingγϕ=u.
The second result of this paper says that the variational solution and the classical solution coincide, if both are defined.
Theorem 1.2. — Adopt the notation and assumptions of Theorem 1.1.
Suppose thatΩhas a Lipschitz boundary. Letϕ∈C(Γ)∩TrH1(Ω). Then P ϕ=γϕalmost everywhere onΩ.
The last main result of this paper concerns a parabolic equation. LetAc denote the part of the operatorAD inC0(Ω). So
D(Ac) ={u∈D(AD)∩C0(Ω) :ADu∈C0(Ω)}
andAc=AD|D(Ac).
Theorem 1.3. — Adopt the notation and assumptions of Theorem 1.1.
Then −Ac generates a holomorphic C0-semigroup on C0(Ω). Moreover, e−tAcu= e−tADufor allu∈C0(Ω)andt >0.
In Section 2 we prove Theorem 1.1 via an iteration argument. Section 3 is devoted to the comparison of the classical and the variational solutions of the Dirichlet problem. Theorem 1.2 is proved there with the help of a deep result of Dahlberg [7]. We consider the semigroup on C0(Ω) in Section 4 and prove Theorem 1.3.
2. The Dirichlet problem
In this section we prove Theorem 1.1 on the well-posedness of the Dirich- let problem. The technique is a reduction to the Stampacchia result men- tioned in the introduction. For this reason we introduce the following two forms and operators.
Adopt the notation and assumptions of Theorem 1.1. For allλ∈Rdefine the formsaλ,bλ:H1(Ω)×H1(Ω)→Cby
aλ(u, v) =a(u, v) +λ(u, v)L2(Ω) and bλ(u, v) =
d
X
k,l=1
Z
Ω
akl(∂ku)∂lv+
d
X
k=1
Z
Ω
ck(∂ku)v+λ Z
Ω
u v,
where a is as in (1.4). Define similarly Aλ,Bλ: Hloc1 (Ω) → D0(Ω) and let BD be the operator associated with the sesquilinear formb0|H1
0(Ω)×H01(Ω). It follows from ellipticity that there exists aλ0>0 such that
µ
2kvk2H1(Ω)6Reaλ0(v) and µ
2 kvk2H1(Ω)6Rebλ0(v)
for all v ∈ H1(Ω). Note that Bλ satisfies the submarkovian condition
−Bλ1Ω 60, that is (1.3), and even Stampacchia’s condition (1.2) for all λ > 0. So we can and will apply Stampacchia’s result (in the proof of Lemma 2.8).
We first investigate the operatorADinL2(Ω). Note thatf0+Pd
k=1∂kfk∈ D0(Ω) for allf0, f1, . . . , fd∈L1(Ω). The next lemma is also valid if theakl andck are complex valued.
Lemma 2.1. — Let f1, . . . , fd ∈ L2(Ω). Let p˜ ∈ (1,∞) be such that p˜> d+22d . Further let f0 ∈ Lp˜(Ω). Then there exists a unique u∈ H01(Ω) such thatAu=f0+Pd
k=1∂kfk.
Proof. — There exists a uniqueT ∈ L(H01(Ω)) such that (T u, v)H1 0(Ω)= a(u, v) for all u, v ∈ H01(Ω). Then T is injective because kerAD = {0}.
Moreover, the inclusion H01(Ω) ,→ L2(Ω) is compact. Hence the opera- torT is invertible by the Fredholm–Lax–Milgram lemma, [5, Lemma 4.1].
Clearlyv 7→Pd
k=1(fk, ∂kv)L2(Ω) is continuous fromH01(Ω) into C. Define F:Cc∞(Ω)→CbyF(v) =hf0, viD0(Ω)×D(Ω). We claim thatF extends to a continuous function fromH01(Ω) into C. Ifd>3, then H01(Ω)⊂Lr(Ω), wherer=d−22d . SoH01(Ω)⊂Lq(Ω), whereqis the dual exponent of ˜p. The last inclusion is also valid ifd= 2. So in any case the mapF extends to a continuous function fromH01(Ω) intoC. Then the lemma follows.
The next lemma is valid for a general bounded open set Ω and does not use the condition 06∈σ(AD). It is an extension of [1, Lemma 4.2].
Lemma 2.2. — Letu ∈ C0(Ω)∩Hloc1 (Ω) and f1, . . . , fd ∈ L2(Ω). Let
˜
p∈ (1,∞) be such that p˜ > d+22d . Further let f0 ∈ Lp˜(Ω). Suppose that Au=f0+Pd
k=1∂kfk. Thenu∈H01(Ω).
Proof. — As at the end of the previous proof there exists anM0>0 such that|R
Ωf0v|6M0kvkH1(Ω)for allv∈H01(Ω). SetM =M0+Pd
k=1kfkk2. Let ε > 0. Setvε = (Reu−ε)+. Then suppvε ⊂Ω is compact. Hence there exists an open Ω1 ⊂ Rd such that suppvε ⊂ Ω1 ⊂ Ω1 ⊂ Ω. Then vε∈H01(Ω1). Moreover,
d
X
k,l=1
Z
Ω1
akl(∂ku)∂lv+
d
X
k=1
Z
Ω1
bku ∂kv+
d
X
k=1
Z
Ω1
ck(∂ku)v+ Z
Ω1
c0u v
= Z
Ω1
f0v+
d
X
k=1
Z
Ω1
fk∂kv (2.1)
for allv ∈ Cc∞(Ω1). Since u|Ω1 ∈H1(Ω1) it follows that (2.1) is valid for allv∈H01(Ω1). Choosingv=vε gives
d
X
k,l=1
Z
Ω
akl(∂ku)∂lvε+
d
X
k=1
Z
Ω
bku ∂kvε+
d
X
k=1
Z
Ω
ck(∂ku)vε+ Z
Ω
c0u vε
6M0kvεkH1(Ω)+
d
X
k=1
kfkk2k∂kvεk26MkvεkH1(Ω). On the other hand, ∂kvε = ∂k((Reu−ε)+) = 1[Reu>ε]∂kReu for all k∈ {1, . . . , d}by [10, Lemma 7.6]. Therefore
Re
d
X
k,l=1
Z
Ω
akl(∂ku)∂lvε+ Re
d
X
k=1
Z
Ω
bku ∂kvε
+ Re
d
X
k=1
Z
Ω
ck(∂ku)vε+ Re Z
Ω
c0u vε
=
d
X
k,l=1
Z
Ω
akl(∂kvε)∂lvε+ Re
d
X
k=1
Z
Ω
bku ∂kvε
+
d
X
k=1
Z
Ω
ck(∂kReu)vε+ Re Z
Ω
c0u vε
= Rea(vε) +ε
d
X
k=1
Z
Ω
(Rebk)∂kvε−
d
X
k=1
Z
Ω
(Imbk) (Imu)∂kvε
+ε Z
Ω
(Rec0)vε− Z
Ω
(Imc0) (Imu)vε
>µ
2 kvεk2H1(Ω)−λ0kvεk22−ε M0|Ω|1/2kvεkH1(Ω)−M0kuk2kvεkH1(Ω),
whereM0 =kc0k∞+Pd
k=1kbkk∞. Sincekvεk2=k(Reu−ε)+k26kuk26
|Ω|1/2kukC0(Ω), it follows that µ
2k(Reu−ε)+k2H1(Ω)6M00k(Reu−ε)+kH1(Ω)+λ0|Ω| kuk2C
0(Ω)
for allε∈(0,1], whereM00=M+M0|Ω|1/2(kukC0(Ω)+ 1).
Therefore the sequence ((Reu−2−n)+)n∈N0 is bounded inH01(Ω). Pass- ing to a subsequence if necessary, we may assume without loss of generality that there exists aw∈H01(Ω) such that lim(Reu−2−n)+=w weakly in H01(Ω). Then lim(Reu−2−n)+ = w in L2(Ω). But lim(Reu−2−n)+ = (Reu)+ in L2(Ω). So (Reu)+ = w ∈ H01(Ω). Similarly one proves that (Reu)−,(Imu)+,(Imu)− ∈H01(Ω). Sou∈H01(Ω).
Lemma 2.2 together with the condition 06∈σ(AD) gives the uniqueness in Theorem 1.1.
Proposition 2.3. — For all ϕ ∈ C(Γ) there exists at most one u ∈ C(Ω)∩Hloc1 (Ω) such thatu|Γ=ϕand Au= 0.
Proof. — Letu∈C(Ω)∩Hloc1 (Ω) and suppose thatu|Γ= 0 andAu= 0.
Thenu∈C0(Ω). Henceu∈H01(Ω) by Lemma 2.2. AlsoAu= 0. Therefore u∈D(AD) andADu= 0. But 06∈σ(AD). Sou= 0.
In the next proposition we use that Ω is Wiener regular.
Proposition 2.4. — Letλ > λ0 andp∈(d,∞]. Letf0∈Lp/2(Ω) and f1, . . . , fd ∈ Lp(Ω). Then there exists a unique u∈ H01(Ω)∩C0(Ω) such thatBλu=f0+Pd
k=1∂kfk.
Proof. — Sinceakl andckare real valued for allk, l∈ {1, . . . , d}we may assume thatf0, . . . , fdare real valued. By [10, Theorem 8.31] there exists a uniqueu∈C(Ω)∩Hloc1 (Ω) such thatBλu=f0+Pd
k=1∂kfk andu|Γ= 0.
Thenu∈C0(Ω) and the existence follows from Lemma 2.2. The uniqueness
follows from Proposition 2.3.
Corollary 2.5. — Letλ > λ0 andp∈(d,∞]. Let f0∈ Lp/2(Ω) and f1, . . . , fd∈Lp(Ω). Letu∈H01(Ω)and suppose thatBλu=f0+Pd
k=1∂kfk. Thenu∈C0(Ω).
Proof. — By Proposition 2.4 there exists a ˜u∈H01(Ω)∩C0(Ω) such that Bλu˜=f0+Pd
k=1∂kfk. ThenBλ(u−u) = 0. So˜ bλ(u−u, v) = 0 first for˜ allv∈Cc∞(Ω) and then by density for allv ∈H01(Ω). Choosev =u−u.˜ Then µ2ku−uk˜ 2H1(Ω)6Rebλ(u−u) = 0. So˜ u= ˜u∈C0(Ω).
We next wish to add the other lower order terms.
Proposition 2.6. — There exists ac >0such that for allΦ∈C1(Rd) there exists a uniqueu∈H1(Ω)∩C(Ω)such that u|Γ= Φ|Γ andAu= 0.
Moreover,
kukC(Ω)6ckΦ|ΓkC(Γ).
For the proof we need some lemmas. In the next lemma we introduce a parameterδin order to avoid duplication of the proof.
Lemma 2.7. — Fixδ∈[0, λ0+ 1].
(1) For allf ∈L2(Ω)andλ > λ0there exists a uniqueu∈H01(Ω)such that
(2.2) bλ(u, v) =
d
X
k=1
(bkf, ∂kv)L2(Ω)+ ((c0−δ1Ω)f, v)L2(Ω) for allv∈H01(Ω).
For allλ > λ0 defineRλ:L2(Ω)→L2(Ω) byRλf =u, whereu∈H01(Ω) is as in(2.2).
(2) There exists ac1>0such that
kRλfkLq(Ω)6c1(λ−λ0)−1/4kfkL2(Ω)
for allλ > λ0andf ∈L2(Ω), where 1q =12−4d1. (3) There exists ac2>1such that
kRλfkLq(Ω)6c2kfkLp(Ω)
for allλ∈[λ0+ 1,∞),p, q∈[2,∞]andf ∈Lp(Ω)with 1q = 1p−4d1. (4) Ifλ > λ0,p∈(d,∞]andf ∈Lp(Ω), then Rλf ∈C0(Ω).
Proof.
(1). This follows from the Lax–Milgram theorem.
(2). Define M =kc0−δ1ΩkL∞(Ω)+Pd
k=1kbkkL∞(Ω). Let λ > λ0, f ∈ L2(Ω) and setu=Rλf. Then
µ
2 kuk2H1(Ω)+ (λ−λ0)kuk2L2(Ω)
6Rebλ0(u) + (λ−λ0)kuk2L
2(Ω)
= Rebλ(u)
= Re
d
X
k=1
(bkf, ∂ku)L2(Ω)+ Re((c0−δ1Ω)f, u)L2(Ω) 6MkfkL2(Ω)kukH1(Ω).
SokukH1(Ω)62µ−1MkfkL2(Ω) and
kRλfkL2(Ω)=kukL2(Ω)6
s 2
µ(λ−λ0)MkfkL2(Ω).
The Sobolev embedding theorem implies that there exists a c1 >0 such thatkvkLq
1(Ω)6c1kvkH1(Ω) for allv ∈H01(Ω), where q1
1 = 12 −2d1. (The extra factor 2 is to avoid a separate case ford= 2.) ThenkRλfkLq
1(Ω)6 2µ−1c1MkfkL2(Ω). Hence
kRλfkLq(Ω)6kRλfk1/2L
2(Ω)kRλfk1/2L
q1(Ω)6c2(λ−λ0)−1/4kfkL2(Ω), wherec2= (2/µ)3/4c1/21 M.
(3). Apply Corollary 2.5 with p = 4d and λ = λ0+ 1. It follows that Rλ0+1f ∈C0(Ω) for allf ∈Lp(Ω). Clearly the mapRλ0+1|Lp(Ω):Lp(Ω)→ C0(Ω) has a closed graph. Hence it is continuous. In particular, there exists ac3>0 such thatkRλ0+1fkL∞(Ω)=kRλ0+1fkC0(Ω) 6c3kfkLp(Ω) for all f ∈Lp(Ω).
Letλ>λ0+ 1 andf ∈L2(Ω). Writeu=Rλf and u0=Rλ0+1f. Then bλ(u, v) =bλ0+1(u0, v) andbλ(u−u0, v) =−(λ−λ0−1) (u, v)L2(Ω)for all v∈H01(Ω). Henceu−u0∈D(BD) and (BD+λ I)(u−u0) =−(λ−λ0−1)u0. Consequently
Rλ= I−(λ−λ0−1) (BD+λ I)−1 Rλ0+1
for all λ>λ0+ 1. Since the semigroup generated by −BD has Gaussian bounds, there exists ac4>1 such thatk(BD+λ I)−1k∞→∞6c4λ−1 for allλ>λ0+ 1. ThenkRλfkL∞(Ω)62c3c4kfkLp(Ω) for allλ>λ0+ 1 and f ∈Lp(Ω).
Finally let p0 ∈ (2,4d) and let q0 ∈ (2,∞) be such that q10 = p10 − 4d1. There exists aθ ∈(0,1) such that p10 = 1−θ2 + θp. Then q10 = 1−θq , where
1
q = 12−4d1. Letc1>0 be as in Statement (2). The operatorRλis bounded from L2(Ω) into Lq(Ω) with norm at most c1 by Statement (2), and we just proved that the operatorRλis bounded fromLp(Ω) intoL∞(Ω) with norm at most 2c3c4. Hence by interpolation the operator Rλ is bounded fromLp0(Ω) intoLq0(Ω) with norm bounded byc1−θ1 (2c3c4)θ6c1+ 2c3c4, which gives Statement (3).
(4). This is a special case of Corollary 2.5.
The main step in the proof of Proposition 2.6 is the next lemma.
Lemma 2.8. — There exist λ > λ0 and c > 0 such that for all Φ ∈ C1(Ω)∩H1(Ω)there exists a uniqueu∈H1(Ω)∩C(Ω)such thatu|Γ= Φ|Γ
andAλu= 0. Moreover,
kukC(Ω)6ckΦ|ΓkC(Γ).
Proof. — Chooseδ= 0 in Lemma 2.7. Letc1andc2be as in Lemma 2.7.
Letλ∈(λ0+ 1,∞) be such that c1c2d−12 (λ−λ0)−1/4(1 +|Ω|)6 12. Let Rλ be as in Lemma 2.7. Setϕ= Φ|Γ.
There exist unique w,w˜ ∈ H01(Ω) such that aλ(w, v) = aλ(Φ, v) and bλ( ˜w, v) = bλ(Φ, v) for all v ∈H01(Ω). Then ˜w∈C0(Ω) by Corollary 2.5.
Defineu= Φ−wand ˜u= Φ−w. Then ˜˜ u∈H1(Ω)∩C(Ω) and ˜u|Γ =ϕ.
Moreover,aλ(u, v) = 0 andbλ(˜u, v) = 0 for allv∈H01(Ω), andkuk˜ C(Ω)6 kϕkC(Γ)by the result of Stampacchia mentioned in the introduction ([13, Théorème 3.8]).
Letv∈H01(Ω). Then
bλ(˜u−u, v) =
d
X
k=1
(bku, ∂kv)L2(Ω)+ (c0u, v)L2(Ω)
and ˜u−u=Rλuby the definition ofRλ.
For all n ∈ {0, . . . ,2d} define pn = 2d−n4d . Then p0 = 2, p2d−1 = 4d, p2d =∞and p1
n = p1
n−1 −4d1 for alln∈ {1, . . . ,2d}. So k˜u−ukLpn(Ω)6 c2kukLpn−1(Ω)for alln∈ {2, . . . ,2d}and
k˜u−ukLp
1(Ω)6c1(λ−λ0)−1/4kukL2(Ω) by Lemma 2.7 (3) and (2). Then
kukLp
1(Ω)6c1(λ−λ0)−1/4kukL2(Ω)+ (1 +|Ω|)k˜ukL∞(Ω) and
kukLpn(Ω)6c2kukLpn−1(Ω)+ (1 +|Ω|)k˜ukL∞(Ω) for alln∈ {2, . . . ,2d}. It follows by induction tonthat
kukLpn(Ω)6c1cn−12 (λ−λ0)−1/4kukL2(Ω)+ (1 +|Ω|)
n−1
X
k=0
ck2k˜ukL∞(Ω)
for alln ∈ {2, . . . ,2d}. So u ∈ Lp2d−1(Ω) = L4d(Ω) and ˜u−u= Rλu ∈ C0(Ω) by Lemma 2.7 (4). In particularu∈C(Ω). Moreover,
kukL∞(Ω)
=kukLp
2d(Ω)
6c1c2d−12 (λ−λ0)−1/4kukL2(Ω)+ 2d(1 +|Ω|)c2d−12 k˜ukL∞(Ω)
6c1c2d−12 (λ−λ0)−1/4(1 +|Ω|)kukL∞(Ω)+ 2d(1 +|Ω|)c2d−12 k˜ukL∞(Ω) 6 1
2kukL∞(Ω)+ 2d(1 +|Ω|)c2d−12 kuk˜ L∞(Ω) by the choice ofλ. So
kukL∞(Ω)64d(1 +|Ω|)c2d−12 k˜ukL∞(Ω)64d(1 +|Ω|)c2d−12 kϕkC(Γ)
and the proof of the lemma is complete.
We next wish to remove the λ in Lemma 2.8. For future purposes, we consider the full inhomogeneous problem.
Proposition 2.9. — Letp∈(d,∞],f0 ∈Lp/2(Ω) and letf1, . . . , fd ∈ Lp(Ω). Letu∈H01(Ω)be such thatAu=f0+Pd
k=1∂kfk. Thenu∈C0(Ω).
Proof. — Without loss of generality we may assume that p ∈ (d,4d).
Chooseλ=δ =λ0+ 1 in Lemma 2.7 and in Proposition 2.4. By Propo- sition 2.4 there exists a unique ˜u ∈ H01(Ω) ∩C0(Ω) such that Bλu˜ = f0+Pd
k=1∂kfk. Ifv∈Cc∞(Ω), then bλ(˜u, v) =hf0+
d
X
k=1
∂kfk, viD0(Ω)×D(Ω)
=a(u, v)
=bλ(u, v) +
d
X
k=1
(bku, ∂kv)L2(Ω)+ ((c0−δ1Ω)u, v)L2(Ω). So
bλ(˜u−u, v) =
d
X
k=1
(bku, ∂kv)L2(Ω)+ ((c0−δ1Ω)u, v)L2(Ω)
and by density for allv ∈H01(Ω). Henceu−u˜ =Rλu, whereRλ is as in Lemma 2.7. For alln ∈ {0, . . . ,2d−1} define pn = 2d−n4d . Then u−˜u∈ L2(Ω) = Lp0(Ω). It follows by induction to n that u ∈ Lpn−1(Ω) and u−u˜∈Lpn(Ω) for alln∈ {1, . . . ,2d−1}, where the last part follows from Lemma 2.7 (3). Henceu−u˜ ∈Lp2d−1(Ω) = L4d(Ω) and u∈Lp(Ω). Then Lemma 2.7 (4) givesu−u˜=Rλu∈C0(Ω) and thereforeu∈C0(Ω).
Corollary 2.10. — Letp∈(d,∞]. Then(AD)−1(Lp(Ω))⊂C0(Ω).
Corollary 2.11. — There exists ac0 >0such thatk(AD)−1fkL∞(Ω)6 c0kfkL∞(Ω)for allf ∈L∞(Ω).
Proof. — Closed graph theorem.
Proof of Proposition 2.6. — Let c, λ > 0 be as in Lemma 2.8 and let c0 > 0 be as in Corollary 2.11. By Lemma 2.8 there exists a unique ˜u ∈ H1(Ω)∩C(Ω) such that ˜u|Γ = Φ|Γ and Aλu˜ = 0. By Lemma 2.1 there exists a uniquew∈H01(Ω) such thata(w, v) =a(Φ|Ω, v) for allv∈H01(Ω).
Setu= Φ|Ω−wand ˜w= Φ|Ω−u. Then˜
a(w, v) =a(Φ|Ω, v) =aλ(Φ|Ω, v)−λ(Φ, v)L2(Ω)=aλ( ˜w, v)−λ(Φ, v)L2(Ω)
=a( ˜w, v) +λ( ˜w, v)L2(Ω)−λ(Φ, v)L2(Ω)=a( ˜w, v)−λ(˜u, v)L2(Ω) for allv∈H01(Ω). So
a(˜u−u, v) =a(w−w, v) =˜ −λ(˜u, v)L2(Ω).
Since ˜u−u ∈ H01(Ω) it follows that AD(˜u−u) = −λu. Consequently,˜ u= ˜u+λ(AD)−1u˜∈C0(Ω) by Corollary 2.10. Moreover,
kukC(Ω)=kukL∞(Ω)6k˜ukL∞(Ω)+λk(AD)−1uk˜ L∞(Ω) 6(1 +c0λ)k˜ukL∞(Ω)6(1 +c0λ)ckΦ|ΓkC(Γ)
and the proof of Proposition 2.6 is complete.
Define ||| · |||: Hloc1 (Ω)→[0,∞] by
|||u|||= sup
δ>0
sup
Ω0⊂Ω open d(Ω0,Γ)=δ
δ Z
Ω0
|∇u|2 1/2
.
Finally we need the following Caccioppoli inequality.
Proposition 2.12. — There exists a c0 > 1 such that |||u||| 6 c0kukL2(Ω) for allu∈H1(Ω)such thatAu= 0.
Proof. — See [9, Theorem 4.4].
Now we are able to prove Theorem 1.1.
Proof of Theorem 1.1. — The uniqueness is already proved in Proposi- tion 2.3.
Letc >0 andc0>1 be as in Propositions 2.6 and 2.12. Let Φ∈C1(Rd)∩
H1(Rd). By Proposition 2.6 there exists a unique u∈H1(Ω)∩C(Ω) such
thatu|Γ= Φ|Γ andAu= 0. Moreover,
kukC(Ω)+|||u|||6kukC(Ω)+c0kukL2(Ω)
6(2 +|Ω|)c0kukC(Ω) 6(2 +|Ω|)c c0kΦ|ΓkC(Γ). (2.3)
It follows from (2.3) that we can define a linear mapF: {Φ|Γ: Φ∈C1(Rd)∩
H1(Rd)} → H1(Ω)∩C(Ω) by F(Φ|Γ) = u, whereu ∈ H1(Ω)∩C(Ω) is such that u|Γ = Φ|Γ and Au = 0. Now let ϕ ∈ C(Γ). By the Stone–
Weierstraß theorem there are Φ1,Φ2, . . . ∈ C1(Rd)∩H1(Rd) such that lim Φn|Γ = ϕ in C(Γ). Set un = F(Φn|Γ) for all n ∈ N. Then it follows from (2.3) that (un)n∈N is a Cauchy sequence in C(Ω). Let u = limun in C(Ω). Also (un)n∈N is a Cauchy sequence in Hloc1 (Ω) by (2.3). So u∈ Hloc1 (Ω). SinceAun= 0 for alln∈N, one deduces thatAu= 0. Moreover, u|Γ = limun|Γ= lim Φn|Γ=ϕ. This proves existence. Finally,
kukC(Ω)= limkunkC(Ω)
6lim(2 +|Ω|)c c0kΦn|ΓkC(Γ)= (2 +|Ω|)c c0kϕkC(Γ).
This completes the proof of Theorem 1.1.
Theorem 1.1 has the following extension.
Theorem 2.13. — Adopt the notation and assumptions of Theorem 1.1.
Let ϕ ∈ C(Γ), p ∈ (d,∞], f0 ∈ Lp/2(Ω) and let f1, . . . , fd ∈ Lp(Ω).
Then there exists a unique u ∈ C(Ω)∩Hloc1 (Ω) such that u|Γ = ϕ and Au=f0+Pd
k=1∂kfk.
Proof. — The uniqueness follows as in the proof of Proposition 2.3.
By Lemma 2.1 there exists au0∈H01(Ω) such thatAu0=f0+Pd
k=1∂kfk. Thenu0 ∈C0(Ω) by Proposition 2.9. By Theorem 1.1 there exists au1 ∈ C(Ω)∩Hloc1 (Ω) such thatu1|Γ=ϕandAu1= 0. Defineu=u0+u1. Then u∈C(Ω)∩Hloc1 (Ω). Moreover, u|Γ =ϕandAu=f0+Pd
k=1∂kfk. We conclude this section with some results for the classical solution.
They will be used in Section 3 and are of independent interest. Recall that P:C(Γ) → C(Ω) is given by P ϕ = u, where u∈ C(Ω)∩Hloc1 (Ω) is the classical solution, sou|Γ=ϕandAu= 0.
Proposition 2.14. — LetΦ∈C(Ω)∩Hloc1 (Ω). Suppose there exists a w∈H01(Ω) such thatAΦ =Aw. Then w∈C(Ω)andP(Φ|Γ) = Φ−w.
Proof. — Write ˜w= Φ−P(Φ|Γ). Then ˜w∈C0(Ω)∩Hloc1 (Ω) andAw˜= AΦ =Aw=f0+Pd
k=1∂kfk, where f0=c0w+Pd
l=1cl∂lw∈L2(Ω) and fk =−Pd
l=1alk∂lw−bkw∈L2(Ω) for allk∈ {1, . . . , d}. So ˜w∈H01(Ω)