HAL Id: hal-02005375
https://hal.archives-ouvertes.fr/hal-02005375
Preprint submitted on 4 Feb 2019
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Green function and Poisson kernel associated to root systems for annular regions
Chaabane Rejeb
To cite this version:
Chaabane Rejeb. Green function and Poisson kernel associated to root systems for annular regions.
2019. �hal-02005375�
Green function and Poisson kernel associated to root systems for annular regions
Chaabane REJEB∗
Abstract
Let ∆k be the Dunkl Laplacian relative to a fixed root system Rin Rd, d ≥2, and to a nonnegative multiplicity function k on R. Our first purpose in this paper is to solve the ∆k-Dirichlet problem for annular regions. Secondly, we introduce and study the ∆k-Green function of the annulus and we prove that it can be expressed by means of ∆k-spherical harmonics. As applications, we obtain a Poisson-Jensen formula for ∆k-subharmonic functions and we study positive continuous solutions for a ∆k-semilinear problem.
MSC (2010) primary: 31B05, 35J08, 35J65; secondary: 31C45, 46F10, 47B39.
Key words: Dunkl-Laplace operator, Poisson kernel, Green function, Dirichlet problem, spherical harmonics, Newton kernel.
1 Introduction
Since the 90’s, extensive studies have been carried out on analysis associated with Dunkl operators. These are commuting differential-difference operators onRd introduced by C.
F. Dunkl (see [6]). The Dunkl analysis includes especially a generalization of the Fourier transform (called the Dunkl transform) and the Laplace operator known as the Dunkl Laplacian (and denoted by ∆k).
The Dunkl theory has many applications as well in mathematical physics and probability theory. In particular, it has been used in the study of the Calogero-Moser-Sutherland and other integrable systems (see [4, 10]) and in the study of Markov processes generalizing Brownian motion (see [22]).
Recently, a special interest has been devoted to potential theory associated with the Dunkl Laplacian. The study focused on ∆k-harmonic functions (see [2, 11, 12, 17, 19, 20]), on
∆k-Newton potential theory (including ∆k-subharmonic functions) (see [13]) and on ∆k- Riesz potentials of Radon measures (see [14]). More recently, by means of the ∆k-Newton kernel, the Green function of the open unit ball has been studied in [15]. Note that finding
∗Universit´e de Tunis El Manar, Facult´e des Sciences de Tunis, Laboratoire d’Analyse Math´ematques et Applications LR11ES11, 2092 El Manar I, Tunis, TUNISIA and Laboratoire de Math´ematiques et Physique Th´eorique CNRS-UMR 7350, Universit´e de Tours, Campus de Grandmont, 37200 Tours, FRANCE; Email:
chaabane.rejeb@gmail.com
∆k-Green functions for other open sets is a rather difficult problem already in the case of the classical Laplace operator. The aim of this paper is to show that we can determine the
∆k-Green function for annular regions inRdby using ∆k-spherical harmonics as a crucial tool.
Let us assume throughout the paper that d≥2. Let Abe the annulus A:={x∈Rd, ρ <∥x∥<1} with ρ∈]0,1[.
After giving some properties of the ∆k-Green functionGk,A of A, we will use it to study the semilinear problem
{ ∆k(uωk) =ϕ(., u)ωk, in the sense of distributions
u=f, on ∂A,
whereωk is a precise weight function (see (2.6) for its expression).
More precisely, under some assumptions on the functionϕ, we will show that iff ∈ C(∂A) is nonnegative, this boundary problem has only and only one positive continuous solution on Awhich satisfies (see Theorem 5.2)
∀ x∈A, u(x) +
∫
A
Gk,A(x, y)ϕ(y, u(y))ωk(y)dy=Pk,A[f](x).
Here Pk,A[f] is the unique solution in C2(A)∩ C(A) of the boundary Dirichlet problem { ∆ku= 0, onA,
u=f, on∂A, that will be given explicitly in Section 3.
This paper is organized as follows. In Section 2, we recall some basics from Dunkl theory that will be used throughout the paper. In Section 3, we give an explicit solution of the boundary Dirichlet problem for the annulus. The Green function Gk,A will be introduced and studied in Section 4. Some applications will be given in the last Section.
Precisely, we will obtain a Poisson-Jensen formula for ∆k-subharmonic functions in the annulus and we will study positive solutions of the above semilinear problem.
2 Basics from Dunkl theory
We start by recalling some useful facts in the Dunkl theory. Let R be a root system in the Euclidian spaceRd, in the sense thatR is a finite set inRd\ {0} such that for every α∈ R,R ∩Rα={±α}and σα(R) =R(where σα is the reflection w.r.t. the hyperplane Hα orthogonal to α). The subgroup W ⊂O(Rd) generated by the reflections σα,α ∈ R, is called the Coxeter-Weyl group associated to R. We refer to ([18]) for more details on root systems and their Coxeter-Weyl groups.
Let k be a fixed nonnegative multiplicity function on R (i.e. k is W-invariant). For ξ∈Rd, theξ-directional Dunkl operator associated to (W, k) is defined by
Dξf(x) :=∂ξf(x) + ∑
α∈R+
k(α)⟨α, ξ⟩f(x)−f(σα.x)
⟨α, x⟩ , f ∈ C1(Rd),
where∂ξ is the usual ξ-directional partial derivative andR+ is a positive subsystem.
Let us denote by P(Rd) (resp. Pn(Rd)) the space of polynomial functions on Rd (resp.
the space of homogeneous polynomials of degreen∈N).
It was shown that there is a unique linear isomorphism Vk from P(Rd) onto itself such thatVk(Pn(Rd)) =Pn(Rd) for everyn∈N,Vk(1) = 1 and
∀ ξ∈Rd, DξVk=∂ξVk. (2.1) The operator Vk is known as the Dunkl intertwining operator (see [7, 8]). It has been extended to a topological isomorphism fromC∞(Rd) onto itself satisfying (2.1) (see [26]).
Furthermore, according to [23], for eachx∈Rd, there is a compactly supported probability measure µx on Rdsuch that
∀ f ∈ C∞(Rd), Vk(f)(x) =
∫
Rdf(y)dµx(y). (2.2) IfW.x denotes the orbit ofx under theW-action andCo(x) its convex hull, then
supp µx ⊂Co(x)⊂B(0,∥x∥). (2.3) The Dunkl-Laplacian is defined as ∆k =∑d
j=1D2ej, where (ej)1≤j≤d is the canonical basis of Rd. It can be expressed as follows
∆kf(x) = ∆f(x)+ ∑
α∈R+
k(α) (
2⟨∇f(x), α⟩
⟨α, x⟩ −∥α∥2f(x)−f(σα(x))
⟨α, x⟩2 )
, f ∈ C2(Rd), (2.4) where ∆ (resp. ∇) is the usual Laplace (resp. gradient) operator (see [6, 8]). Note that if kis the zero function, the Dunkl Laplacian reduces to the classical one which is commutes with the action of O(Rd). For general k≥0, ∆k commutes with theW-action (see [24]) i.e.
∀g∈W, g◦∆k= ∆k◦g. (2.5)
Let L2k(Sd−1), d≥2, be the Hilbert space endowed with the inner product
⟨p, q⟩k:= 1 dk
∫
Sd−1
p(ξ)q(ξ)ωk(ξ)dσ(ξ).
We denote by∥.∥L2
k(Sd−1) the associated Euclidean norm. Here,dσ is the surface measure on the unit sphere Sd−1,ωk is the weight function given by
ωk(x) =∏
α∈R+| ⟨α, x⟩ |2k(α) (2.6)
and dk is the constant
dk=∫
Sd−1ωk(ξ)dσ(ξ). (2.7)
The function ωk is W-invariant and homogeneous of degree 2γ := 2∑
α∈R+k(α).
To simplify, we introduce the constant λk:= d
2+γ−1≥0. (2.8)
Let H∆k,n(Rd) := Pn(Rd)∩Ker∆k be the space of ∆k-harmonic polynomials, homoge- neous of degree nonRd. From [8], we know that ifn̸=m, thenH∆k,n(Rd)⊥ H∆k,m(Rd) inL2k(Sd−1). Moreover, for every n∈N, we have
Pn(Rd) =⊕⌊n/2⌋
j=0 ∥x∥2jH∆k,n−2j(Rd). (2.9) The restriction to the sphereSd−1of an element ofH∆k,n(Rd) is called a ∆k-spherical har- monic of degreen. The space of ∆k-spherical harmonics will be denoted byH∆k,n(Sd−1).
This space has a reproducing kernelZk,n uniquely determined by the properties (see [5, 8]) i) for each x∈Sd−1,Zk,n(x, .)∈ H∆k,n(Sd−1),
ii) for everyf ∈ H∆k,n(Sd−1), we have f(x) =⟨f, Zk,n(x, .)⟩k= 1
dk
∫
Sd−1
f(ξ)Zk,n(x, ξ)ωk(ξ)dσ(ξ), x∈Sd−1. (2.10) From this formula, we can see that
∀g∈W, ∀ x, y∈Sd−1, Zk,n(gx, gy) =Zk,n(x, y). (2.11) In classical case (i.e. k = 0), Z0,n(x, .) is known as the zonal harmonic of degree n (see [1, 5]). Note that if{Yj,n, j = 1, . . . , h(n, d) :=dimH∆k,n(Rd)}is a real-orthonormal basis of H∆k,n(Sd−1) in L2k(Sd−1), then
Zk,n(x, y) =
h(n,d)∑
j=1
Yj,n(x)Yj,n(y). (2.12)
By means of the Dunkl intertwining operator and Gegenbauer polynomials,Zk,n is given explicitly by (see [5], Theorem 7.2.6. or [27])
∀x, y∈Sd−1, Zk,n(x, y) = (n+λk)(2λk)n λk.n! Vk
( Pnλk(
⟨., y⟩))
(x), (2.13) where λk is the constant given by (2.8), Pnµ, µ > −1/2, is the normalized Gegenbauer polynomial (see [8] p. 17) defined by
Pnµ(x) := (−1)n 2n(µ+ 1/2)n
(1−x2)1/2−µ dn
dxn(1−x2)n+µ−1/2, and (x)n:=x(x+ 1). . .(x+n−1) is the Pochhammer symbol.
At the end of this section, to simplify, when k = 0 we will use the usual nota- tions L2(Sd−1) for L20(Sd−1), H∆,n for H∆0,n, ωd−1 := d0 the surface area of Sd−1 and Zn:=Z0,n.
3 ∆
k-Dirichlet problem for the annulus
In this section, by introducing a Poisson type kernel, we will solve the Dirichlet problem for the Dunkl Laplacian in annular regions
AR1,R2 :={x∈Rd: R1 <∥x∥< R2}. Note that from the homogeneity property of ∆k:
δr◦∆k =r−2∆k◦δr, with δr(f)(x) :=f(rx),
it suffices to do this for the annular region Aρ,1 with ρ ∈]0,1[ fixed. In the sequel, to simplify, we will use the notationA instead ofAρ,1.
Recall that the ∆k-Poisson kernel of the unit ball (see [8]) is given by Pk(x, y) =
+∞
∑
n=0
Zk,n(x, y) =
∫
Rd
1− ∥x∥2 (1− ⟨x, z⟩+∥x∥2)d
2+γdµy(z), (x, y)∈B×Sd−1. (3.1) From [8], we know that
1 dk
∫
Sd−1
Pk(x, ξ)ωk(ξ)dσ(ξ) = 1. (3.2)
We start by two preliminary useful results. For each n∈N, we see that the restriction of the Dunkl intertwining operator
Vk:H∆,n(Rd)−→ H∆k,n(Rd) is a linear isomorphism.
In the first result, we will estimate the matrix-norms of this operator and of its inverse where the space H∆,n(Rd) (resp. H∆k,n(Rd)) is endowed with theL2(Sd−1)-norm (resp.
theL2k(Sd−1)-norm). More precisely,
Proposition 3.1 Let n be a nonnegative integer.
1. For every f ∈ H∆,n(Rd), we have
∥Vk(f)∥L2k(Sd−1) ≤dimH∆,n(Rd)∥f∥L2(Sd−1). (3.3) 2. For every f ∈ H∆k,n(Rd), we have
∥Vk−1(f)∥L2(Sd−1)≤ (γ+d2)nωd−1
(d2)n
dimH∆,n(Rd)∥f∥L2k(Sd−1). (3.4)
Proof:1)Letf ∈ H∆,n(Rd). After rewriting the reproducing formula (2.10) in the classical case (i.e. k= 0), applying it tof and using Fubini’s theorem, we get
Vk(f)(x) = 1 ωd−1
∫
Sd−1
f(ξ)Vk[Zn(., ξ)](x)dσ(ξ), x∈Rd. (3.5) But, from [1], Proposition 5.27, we have
∀ z, ξ∈Sd−1, |Zn(z, ξ)| ≤dimH∆,n(Rd) which implies that
∀(z, ξ)∈Rd×Sd−1, |Zn(z, ξ)| ≤(
dimH∆,n(Rd)
)∥z∥n. (3.6)
Thus, using the relations (2.3), (3.5) and (3.6) and Cauchy-Schwarz inequality, we obtain
∀ x∈Rd, |Vk(f)(x)| ≤dimH∆,n(Rd)∥f∥L2(Sd−1)∥x∥n. (3.7) This implies that
∥Vk(f)∥L2
k(Sd−1)≤dimH∆,n(Rd)∥f∥L2(Sd−1).
2)Letf ∈ H∆k,n(Rd). Applying the classical case of the formula (2.10) withf ←Vk−1(f) and using (3.6) and Cauchy-Schwarz inequality, we deduce that
∀ x∈Rd, |Vk−1(f)(x)| ≤dimH∆,n(Rd)∥Vk−1(f)∥L2(Sd−1)∥x∥n.
Now, using the following result (see [8], Proposition 5.2.8): for p ∈ Pn(Rd) and q ∈ H∆,n(Rd), then
1 ωd−1
∫
Sd−1
p(ξ)q(ξ)dσ(ξ) = (γ+ d2)nωd−1
(d2)ndk
∫
Sd−1
p(ξ)Vk(q)(ξ)ωk(ξ)dσ(ξ) withp=q =Vk−1(f), we obtain
∥Vk−1(f)∥2L2(Sd−1)≤ (γ+d2)nωd−1
(d2)ndk
∫
Sd−1
|Vk−1(f)(ξ)f(ξ)|ωk(ξ)dσ(ξ)
≤ (γ+d2)nωd−1
(d2)ndk dimH∆,n(Rd)∥Vk−1(f)∥L2(Sd−1)
∫
Sd−1
|f(ξ)|ωk(ξ)dσ(ξ)
≤ (γ+d2)nωd−1
(d2)n dimH∆,n(Rd)∥Vk−1(f)∥L2(Sd−1)∥f∥L2
k(Sd−1).
This proves the desired relation.
Corollary 3.1 The following inequality holds:
∀ x, y∈Sd−1, |Zk,n(x, y)| ≤((γ+ d2)nωd−1 (d2)n
)2(
dimH∆,n(Rd) )5
. (3.8)
Proof: Let {Yj,n, j = 1, . . . , h(n, d) = dimH∆k,n(Rd)}, is a real-orthonormal basis of H∆k,n(Rd) in L2k(Sd−1). Using (3.7) with f =Vk−1(Yj,n) and (3.4), we deduce that
∀x∈Sd−1,|Yj,n(x)| ≤dimH∆,n(Rd)∥Vk−1(Yj,n)∥L2(Sd−1)
≤ (γ+d2)nωd−1
(d2)n
(dimH∆,n(Rd) )2
.
Consequently, we obtain the result from (2.12).
Following the classical casek= 0 (see [1]), we define the kernelPk,1(., .) on A×Sd−1 by Pk,1(x, ξ) :=
+∞
∑
n=0
ak,n(x)Zk,n(x, ξ), with ak,n(x) = 1−(∥x∥
ρ
)−2λk−2n
1−ρ2λk+2n . (3.9) Proposition 3.2 The kernel Pk,1 satisfies the following properties
i) For each ξ ∈ Sd−1, Pk,1(., ξ) is a ∆k-harmonic function on A and Pk,1(., ξ) = 0 on S(0, ρ).
ii) For every x∈A and ξ ∈Sd−1,
0≤Pk,1(x, ξ)≤Pk(x, ξ). (3.10) iii) Let x∈A and ξ∈Sd−1 fixed. Then
∀ g∈W, Pk,1(gx, gξ) =Pk,1(x, ξ). (3.11) Proof: i)Clearly Pk,1(., ξ) = 0 on S(0, ρ). On the other hand, for any (x, ξ) ∈A×Sd−1 we can write
ak,n(x)Zk,n(x, ξ) =c1,nZk,n(x, ξ)−c2,nKk[Zk,n(., ξ)](x),
wherec1,n, c2,n are two nonnegative constants andKk is the ∆k-Kelvin transform (see [9]) given by
Kk[f](x) =∥x∥−2λkf(x/∥x∥2) =∥x∥2−2γ−df(x/∥x∥2) (3.12) and f is a function defined on Rd\ {0}. As the ∆k-Kelvin transform preserves the ∆k- harmonic functions onRd\{0}(see [9]), we deduce that the functionx7→ak,n(x)Zk,n(x, ξ) is ∆k-harmonic onA.
According to [8] (see also [1] and [5]), we know that dimH∆,n(Rd) =dimH∆k,n(Rd) =
(n+d−1 n
)
−
(n+d−3 n−2
) . Hence, we have
n→+∞lim n2−ddimH∆,n(Rd) = 2 (d−2)!.
Moreover, we have
n→lim+∞n−γ(γ+d2)n
(d2)n
= lim
n→+∞n−γ Γ(d/2) Γ(γ+d/2)
Γ(d/2 +γ+n)
Γ(d/2 +n) = Γ(d/2) Γ(γ+d/2). Consequently, from (3.8), there existsC=C(d, γ)>0 such that
∀x∈Rd, ∀y∈Sd−1, |Zk,n(x, y)| ≤Cn5d+2γ−10∥x∥n. (3.13) This inequality as well as the fact that 0≤ak,n(x)<1 imply that the series
∑
n≥0ak,n(x)Zk,n(x, ξ)
converges uniformly on Aρ,R×Sd−1 for every R ∈]ρ,1[. Then , by Corollary 3.3 in [11], the functionPk,1(., ξ) is ∆k-harmonic onA.
ii)For ε >0 small enough and ξ∈Sd−1, consider the function hε(x) :=∑
n≥0ak,n(x)Zk,n((1−ε)x, ξ).
As above, from the inequality (3.13) and the homogeneity ofZk,n(., ξ), we see that hε de- fines a ∆k-harmonic function in the annular regionAρ,RwithR= (1−ε)−1. Furthermore, hε = 0 onS(0, ρ) and if x∈Sd−1, then
hε(x) =∑
n≥0Zk,n((1−ε)x, ξ) =Pk((1−ε)x, ξ). (3.14) wherePk is the ∆k-Poisson kernel of the unit ball (see [8]). In particular,hε≥0 on Sd−1. Consequently, by the weak minimum principle for ∆k-harmonic functions (see [11] or [21]), we deduce that
∀x∈A, hε(x)≥0.
On the other hand, for each fixed (x, ξ) in A×Sd−1, we have
|Pk,1(x, ξ)−hε(x)| ≤∑
n≥1
(1−(1−ε)n)ak,n(x)|Zk,n(x, ξ)|
≤C∑
n≥1
(1−(1−ε)n)n5d+2γ−10∥x∥n.
Hence, by the monotone convergence theorem we havePk,1(x, ξ) = limε→0hε(x). Finally, we obtainPk,1 ≥0 on A×Sd−1.
• For ξ ∈ Sd−1 fixed, the function x 7→ Pk((1−ε)x, ξ)−hε(x) is ∆k-harmonic on A.
Moreover, since Pk is a nonnegative kernel, we have
∀ x∈S(0, ρ), Pk((1−ε)x, ξ)−hε(x) =Pk((1−ε)x, ξ)≥0.
By (3.14),x7→Pk((1−ε)x, ξ)−hε(x) is the zero function onSd−1. So, the weak maximum principle implies that
∀x∈A, Pk((1−ε)x, ξ)≥hε(x).
Lettingε−→0, we obtainPk(., ξ)≥Pk,1(., ξ) onA.
iii)The result follows immediately from (2.11).
Proposition 3.3 Let f be a continuous function onSd−1. Then the function
Pk,1[f](x) = 1 dk
∫
Sd−1
Pk,1(x, ξ)f(ξ)ωk(ξ)dσ(ξ) (3.15) is the unique solution inC2(A)∩ C(A) of the boundary Dirichlet problem
∆ku= 0, onA;
u=f, onSd−1 u= 0, onS(0, ρ).
Proof: The uniqueness follows from the weak maximum principle for ∆k-harmonic func- tions (see [11] or [21]). The inequality (3.13) allowed us to write for anyx∈Athat
Pk,1[f](x) =
+∞
∑
n=0
un(x), with un(x) = ak,n(x) dk
∫
Sd−1
Zk,n(x, ξ)f(ξ)ωk(ξ)dσ(ξ).
By differentiation theorem under integral sign, the functions un are ∆k-harmonic on A.
Moreover, by (3.13) we have
∀ n, |un(x)| ≤C∥f∥∞n5d+2γ−10∥x∥n. This proves that the series ∑
n≥0un converges uniformly on each closed annular region Aρ,R whenever R∈]ρ,1[. Then, we conclude that Pk,1[f] is ∆k-harmonic onA.
On the other hand, it is easy to see thatPk,1[f] = 0 on S(0, ρ).
It remains to prove that for every ξ∈Sd−1, limx→ξPk,1[f](x) =f(ξ).
- If f ∈ H∆k,m(Rd), then un = 0 if n ̸= m and um(x) = ak,m(x)f(x) = Pk,1[f](x).
Therefore, Pk,1[f] =f on Sd−1.
- Iff ∈ Pm(Rd), then by (2.9), there existf1, . . . , fm, withfj ∈ H∆k,n−2j(Rd) such that f(x) =∑[m/2]
j=0 ∥x∥2jfj(x).
This implies thatPk,1[f] =f onSd−1.
- Iff is an arbitrary polynomial function, the result also holds.
- Suppose that f is a continuous function on Sd−1 and letp be a polynomial function.
By (3.10) and (3.2) we have
|Pk,1[f](x)−f(x)| ≤ |Pk,1[f](x)−Pk,1[p](x)|+|Pk,1[p](x)−p(x)|+|p(x)−f(x)|
≤2∥f−p∥∞+|Pk,1[p](x)−p(x)|.
This inequality as well as the Stone-Weierstrass theorem show that limx→ξPk,1[f](x) =
f(ξ) for everyξ ∈Sd−1. This completes the proof.
Now, for x∈Aand ξ ∈Sd−1, consider the functions bk,n(x) =∥x∥−n(∥x∥
ρ
)−2λk−n1− ∥x∥2λk+2n
1−ρ2λk+2n =ρ−n(1−ak,n(x))
and
Pk,2(x, ξ) :=
+∞
∑
n=0
bk,n(x)Zk,n(x, ξ). (3.16) By means of the Poisson kernel of the unit ball, we can write
Pk,2(x, ρξ) =Pk(x, ξ)−Pk,1(x, ξ). (3.17) This relation as well as the properties ofPkandPk,1prove thatPk,2(., ρξ) is a nonnegative
∆k-harmonic function onA withPk,2(., ρξ) = 0 on Sd−1.
Let f be a continuous function on S(0, ρ) and define the function Pk,2[f](x) = 1
dk
∫
Sd−1
Pk,2(x, ρξ)f(ρξ)ωk(ξ)dσ(ξ), x∈A. (3.18) Using (3.17), we can write
Pk,2[f](x) = 1 dk
∫
Sd−1
(Pk(x, ξ)−Pk,1(x, ξ))
f(ρξ)ωk(ξ)dσ(ξ)
=Pk[δρ.f](x)−Pk,1[δρ.f](x).
Here,Pk[ϕ] denotes the Poisson integral ofϕ andδρ.f(x) =f(ρx).
Then, using Proposition 3.3 and theorem A in [19], we obtain immediately the following result:
Proposition 3.4 Let f be a continuous function on S(0, ρ). Then Pk,2[f]is the unique solution in C2(A)∩ C(A) of the boundary Dirichlet problem
∆ku= 0, onA;
u= 0, onSd−1 u=f, onS(0, ρ).
Definition 3.1 Letf be a continuous function on∂A. We define the∆k-Poisson integral of f for the annulusA by
Pk,A[f](x) :=Pk,1[f](x) +Pk,2[f](x) (3.19) Remark 3.1 1. We can see that Pk,A[1] = 1.
2. Using (3.11) and a similar relation for the kernel Pk,2, we obtain
g.Pk,A[f] =Pk,A[g.f], with g.f(x) :=f(g−1x). (3.20) From Propositions 3.3 and 3.4, we deduce the following main result:
Theorem 3.1 Letf ∈ C(∂A). Then the functionPk,A[f]is the unique solution inC2(A)∩ C(A) of the boundary Dirichlet problem
∆ku= 0, on A;
u=f on ∂A.
From this theorem and the weak maximum principle for ∆k-harmonic function (see [11]), we obtain the following result:
Corollary 3.2 Let h be a ∆k-harmonic function on A and continuous onA. Then,
∀ x∈A, h(x) =Pk,A[h](x).
4 ∆
k-Green function of the annulus
Our aim in this section is to introduce and study the Green function of the annular region A={x∈Rd, ρ <∥x∥<1}for the Dunkl-Laplace operator. In the sequel, we will assume thatd+ 2γ >2 i.e. λk>0 with λk the constant (2.8).
Let us first recall that the ∆k-Newton kernel, introduced in [13], is given by Nk(x, y) :=
∫ +∞
0
pk(t, x, y)dt, (4.1)
withpk the Dunkl heat kernel (see [21, 24]) pk(t, x, y) = 1
(2t)d/2+γck
∫
Rde−(∥x∥2+∥y∥2−2⟨x,z⟩)/4tdµy(z) (4.2) and ck the Macdonald-Mehta constant given by
ck:=
∫
Rdexp(−∥x∥2
2 )ωk(x)dx.
According to ([13]), the positive and symmetric kernelNk takes the following form Nk(x, y) = 1
2dkλk
∫
Rd
(∥x∥2+∥y∥2−2⟨x, z⟩)−λk
dµy(z). (4.3) Note that ify= 0, µy =δ0 (withδx0 the Dirac measure atx0∈Rd) and then
Nk(x,0) = 1
2dkλk∥x∥−2λk.
In addition, for each fixed x∈Rd, the functionNk(x, .) is ∆k-harmonic and of classC∞ on Rd\W.x (where W.x is the W-orbit of x), ∆k-superharmonic (see below for precise definition) on wholeRd and satisfies
−∆k[Nk(x, .)ωk] =δx, in D′(Rd), where
-for Ω ⊂ Rd a W-invariant open set, D(Ω) and D′(Ω) denote respectively the space of C∞-functions on Ω with compact support and the space of Schwartz distributions on Ω.
-forf ∈L1loc(Ω, ωk(x)dx), ∆k(f ωk) is the Schwartz distribution on Ω defined by
⟨∆k(f ωk), φ⟩=⟨f ωk,∆kφ⟩, φ∈ D(Ω).
Moreover, for any x ∈ Rd, Nk(x, x) = +∞. For more details on the properties of the
∆k-Newton kernel one can see Section 6 in [13].
Let Ω be a W-invariant open subset of Rd. Recall that a functionu: Ω−→[−∞,+∞[ is
∆k-subharmonic if (see [13])
1. u is upper semi-continuous on Ω,
2. u is not identically−∞ on each connected component of Ω,
3. usatisfies the volume sub-mean property i.e. for each closed ballB(x, r)⊂Ω, we have u(x)≤MBr(u)(x) := 1
mk[B(0, r)]
∫
Rdu(y)hk(r, x, y)ωk(y)dy. (4.4) Here mk is the measure dmk(x) :=ωk(x)dx and y 7→ hk(r, x, y) is the nonnegative com- pactly supported measurable function given by
hk(r, x, y) :=
∫
Rd1[0,r](√
∥x∥2+∥y∥2−2⟨x, z⟩)dµy(z). (4.5) We refer to [11] for more details on the kernelhk.
The following result gives some useful facts about the Poisson integral of the ∆k- Newton kernel:
Proposition 4.1
i) For each x∈A, the functionPk,A[Nk(x, .)] is the solution of the Dirichlet problem { ∆ku= 0, on A;
u=Nk(x, .) on ∂A.
ii) The function (x, y)7→Pk,A[Nk(x, .)](y) is continuous on A×A.
iii) For each fixed y∈A, the function x7→Pk,A[Nk(x, .)](y) is∆k-harmonic in A.
Proof: i)If x ∈ A, then the function Nk(x, .) is continuous on ∂A and by Theorem 3.1, we obtain the first assertion.
ii)At first, we shall prove the following result
Lemma 4.1 The function (x, y) 7→ Nk(x, y) is continuous on Rd×Rd\ {(x, gx), x ∈ Rd, g∈W}.
Proof: Using the following inequality (see [24] Lemma 4.2)
∀t >0, pk(t, x, y)≤ 1
ck(2t)d/2+γ max
g∈We−∥x−gy∥2/4t,
we can apply the dominated convergence theorem in formula (4.1) to obtain the result of
the lemma.
From the first assertion, for each x ∈ A, Pk,A[Nk(x, .)] is extendable to a continuous function on A withPk,A[Nk(x, .)] =Nk(x, .) on ∂A.
Let (x0, y0)∈A×A. For every (x, y)∈A×A we have
Pk,A[Nk(x, .)](y)−Pk,A[Nk(x0, .)](y0)≤Pk,A[Nk(x, .)](y)−Pk,A[Nk(x0, .)](y) +Pk,A[Nk(x0, .)](y)−Pk,A[Nk(x0, .)](y0)
≤Pk,1[Kx0(x, .)](y) +Pk,2[Kx0(x, .)](y) +Pk,A[Nk(x0, .)](y)−Pk,A[Nk(x0, .)](y0), whereKx0(x, y) :=Nk(x, y)−Nk(x0, y).
We already know that
ylim→y0
Pk,A[Nk(x0, .)](y) =Pk,A[Nk(x0, .)](y0).
Now, let ε > 0 and R > 0 be such that B(x0, R) ⊂ A. Since (x, ξ) 7−→ Nk(x, ξ) is uniformly continuous onB(x0, R)×Sd−1, we deduce that there existsη >0 such that
∀(x, ξ)∈B(x0, η)×Sd−1, |Kx0(x, ξ)|=Nk(x, ξ)−Nk(x0, ξ)< ε.
Then, using (3.15) as well as the inequalities (3.2) and (3.10), we get for everyx∈B(x0, η) and everyy∈A
Pk,1[Kx0(x, .)](y)≤ 1 dk
∫
Sd−1
Pk,1(y, ξ)|Kx0(x, ξ)|ωk(ξ)dσ(ξ)≤ε.
The same idea works if we replace the kernel Pk,1 byPk,2. Finally, we obtain lim
(x,y)→(x0,y0)Pk,A[Nk(x, .)](y) =Pk,A[Nk(x0, .)](y0).
That is the function (x, y)7→Pk,A[Nk(x, .)](y) is continuous on A×A as desired.
iii)According to Corollary 4.6 in [12], it is enough to show that the functionsx7→uy(x) :=
Pk,1[Nk(x, .)](y) and x7→vy(x) :=Pk,2[Nk(x, .)](y) satisfy the volume-mean property.
Let then x0 ∈A and R > 0 such thatB(x0, R)⊂A. As the kernels Nk,hk and Pk,1 are nonnegative, we can use Fubini’s theorem to obtain
MBR(uy)(x0) = 1 dk
∫
Sd−1
Pk,1(y, ξ)MBR[Nk(., ξ)](x0)ωk(ξ)dσ(ξ)
But for any ξ ∈Sd−1, the function Nk(., ξ) is ∆k-harmonic on A. Hence, it satisfies the volume-mean property i.e. MBR[Nk(., ξ)](x0) =Nk(x0, ξ). Therefore, we obtain
MBR(uy)(x0) = 1 dk
∫
Sd−1
Pk,1(y, ξ)Nk(x0, ξ)(x0)ωk(ξ)dσ(ξ)
=Pk,1[Nk(x0, .)](y) =uy(x0).
By the same way, we get that x 7→ vy(x) := Pk,2[Nk(x, .)](y) is also a ∆k-harmonic function inA. This proves the desired result.
Definition 4.1 For x∈A, the function Gk,A(x, .) defined by
Gk,A(x, y) :=Nk(x, y)−Pk,A[Nk(x, .)](y), y∈A, (4.6) is called the ∆k-Green function of A with pole x.
The ∆k-Green function Gk,A has the following properties:
Proposition 4.2 Let x∈A. Then
1. The function Gk,A(x, .) is ∆k-harmonic on A\W.x, is ∆k-superharmonic on A and satisfies
−∆k[Gk,A(x, .)ωk] =δx in D′(A). (4.7) 2. Gk,A(x, x) = +∞ and Gk,A(x, y)<+∞ whenever y /∈W.x .
3. For every ξ∈∂A, limy→ξGk,A(x, y) = 0.
4. For every y∈A,Gk,A(x, y)>0.
5. For every x, y∈A, Gk,A(x, y) =Gk,A(y, x).
6. For every x, y∈A and g∈W, Gk,A(gx, gy) =Gk,A(x, y).
7. The zero function is the greatest ∆k-subharmonic minorant of Gk,A(x, .) on A.
8. The function (x, y)7→Gk,A(x, y) is continuous on A×A\ {(x, gx) : x∈A, g∈W}. Proof: The first and the second assertions follow from the properties of the ∆k-Newton kernel previously mentioned. In addition, by Proposition 4.1, we easily obtain the third statement.
4)As limy→ξ∈∂AGk,A(x, y) = 0, the weak minimum principle for ∆k-superharmonic func- tions (see [13], Theorem 3.1) implies thatGk,A(x, .)≥0 onA.
IfGk,A(x, y0) = 0 for somey0∈A, from the strong maximum principle (see [13]), we must have Gk,A(x, .) is the zero function on A which is impossible because Gk,A(x, x) = +∞. Thus,Gk,A is a positive kernel on A×A.
5)Since Nk is a symmetric kernel, we have to prove that
∀ x, y∈A, Pk,A[Nk(x, .)](y) =Pk,A[Nk(y, .)](x).
Lety∈A and consider the function
Hy(x) :=Pk,A[Nk(x, .)](y)−Pk,A[Nk(y, .)](x).