Polynomials Associated with Dihedral Groups

There is a commutative algebra of differential-difference operators, with two parameters, associated to any dihedral group with an even number of reflections. The intertwining operator relates this algebra to the algebra of partial derivatives. This paper presents an explicit form of the action of t...

Повний опис

Збережено в:
Бібліографічні деталі
Опубліковано в: :Symmetry, Integrability and Geometry: Methods and Applications
Дата:2007
Автор: Dunkl, C.F.
Формат: Стаття
Мова:English
Опубліковано: Інститут математики НАН України 2007
Онлайн доступ:https://nasplib.isofts.kiev.ua/handle/123456789/147808
Теги: Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Цитувати:Polynomials Associated with Dihedral Groups / C.F. Dunkl // Symmetry, Integrability and Geometry: Methods and Applications. — 2007. — Т. 3. — Бібліогр.: 10 назв. — англ.

Репозитарії

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-147808
record_format dspace
spelling Dunkl, C.F.
2019-02-16T08:35:28Z
2019-02-16T08:35:28Z
2007
Polynomials Associated with Dihedral Groups / C.F. Dunkl // Symmetry, Integrability and Geometry: Methods and Applications. — 2007. — Т. 3. — Бібліогр.: 10 назв. — англ.
1815-0659
2000 Mathematics Subject Classification: 33C45; 33C80; 20F55
https://nasplib.isofts.kiev.ua/handle/123456789/147808
There is a commutative algebra of differential-difference operators, with two parameters, associated to any dihedral group with an even number of reflections. The intertwining operator relates this algebra to the algebra of partial derivatives. This paper presents an explicit form of the action of the intertwining operator on polynomials by use of harmonic and Jacobi polynomials. The last section of the paper deals with parameter values for which the formulae have singularities.
en
Інститут математики НАН України
Symmetry, Integrability and Geometry: Methods and Applications
Polynomials Associated with Dihedral Groups
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title Polynomials Associated with Dihedral Groups
spellingShingle Polynomials Associated with Dihedral Groups
Dunkl, C.F.
title_short Polynomials Associated with Dihedral Groups
title_full Polynomials Associated with Dihedral Groups
title_fullStr Polynomials Associated with Dihedral Groups
title_full_unstemmed Polynomials Associated with Dihedral Groups
title_sort polynomials associated with dihedral groups
author Dunkl, C.F.
author_facet Dunkl, C.F.
publishDate 2007
language English
container_title Symmetry, Integrability and Geometry: Methods and Applications
publisher Інститут математики НАН України
format Article
description There is a commutative algebra of differential-difference operators, with two parameters, associated to any dihedral group with an even number of reflections. The intertwining operator relates this algebra to the algebra of partial derivatives. This paper presents an explicit form of the action of the intertwining operator on polynomials by use of harmonic and Jacobi polynomials. The last section of the paper deals with parameter values for which the formulae have singularities.
issn 1815-0659
url https://nasplib.isofts.kiev.ua/handle/123456789/147808
citation_txt Polynomials Associated with Dihedral Groups / C.F. Dunkl // Symmetry, Integrability and Geometry: Methods and Applications. — 2007. — Т. 3. — Бібліогр.: 10 назв. — англ.
work_keys_str_mv AT dunklcf polynomialsassociatedwithdihedralgroups
first_indexed 2025-11-24T16:26:16Z
last_indexed 2025-11-24T16:26:16Z
_version_ 1850482758082625536
fulltext Symmetry, Integrability and Geometry: Methods and Applications SIGMA 3 (2007), 052, 19 pages Polynomials Associated with Dihedral Groups Charles F. DUNKL Department of Mathematics, University of Virginia, Charlottesville, VA 22904-4137, USA E-mail: cfd5z@virginia.edu URL: http://www.people.virginia.edu/∼cfd5z Received February 06, 2007; Published online March 23, 2007 Original article is available at http://www.emis.de/journals/SIGMA/2007/052/ Abstract. There is a commutative algebra of differential-difference operators, with two parameters, associated to any dihedral group with an even number of reflections. The intertwining operator relates this algebra to the algebra of partial derivatives. This paper presents an explicit form of the action of the intertwining operator on polynomials by use of harmonic and Jacobi polynomials. The last section of the paper deals with parameter values for which the formulae have singularities. Key words: intertwining operator; Jacobi polynomials 2000 Mathematics Subject Classification: 33C45; 33C80; 20F55 1 Introduction The dihedral group of type I2 (2s) acts on R2, contains 2s reflections and the rotations through angles of mπ s for 1 ≤ m ≤ 2s − 1, and is of order 4s, where s is a positive integer. It is the symmetry group of the regular 2s-gon and has two conjugacy classes of reflections (the mirrors passing through midpoints of pairs of opposite edges and those joining opposite vertices). There is an associated commutative algebra of differential-difference (“Dunkl”) operators with two parameters, denoted by κ0, κ1. It is convenient to use complex coordinates for R2, that is, z = x1 + ix2, z = x1 − ix2. Notations like f (z) will be understood as functions of z, z; except that f (z, z) will be used to indicate the result of interchanging z and z. Let N, N0, Q denote the sets of positive integers, nonnegative integers and rational numbers, respectively. Let ω = eiπ/s, then the reflections in the group are (z, z) 7→ (zωm, zω−m), 0 ≤ m < 2s and the rotations are (z, z) 7→ (zωm, zω−m, ), 1 ≤ m < 2s. Note that f (zωm) is the abbreviated form of f (zωm, zω−m). The differential-difference operators are defined by Tf (z) := ∂ ∂z f (z) + κ0 s−1∑ j=0 f (z)− f ( zω2j ) z − zω2j + κ1 s−1∑ j=0 f (z)− f ( zω2j+1 ) z − zω2j+1 , T f (z) := ∂ ∂z f (z)− κ0 s−1∑ j=0 f (z)− f ( zω2j ) z − zω2j ω2j − κ1 s−1∑ j=0 f (z)− f ( zω2j+1 ) z − zω2j+1 ω2j+1, for polynomials f (z). (The second formula implicitly uses the relation − ωm z−zωm = 1 z−zω−m .) The key fact is that T and T commute. The explicit action of T and T on monomials is given by Tzazb = aza−1zb + s b(a−b−1)/sc∑ j=0 ( κ0 + (−1)j κ1 ) za−1−jszb+js, (1.1) Tzazb = bzazb−1 − s b(a−b)/sc∑ j=1 ( κ0 + (−1)j κ1 ) za−jszb−1+js, (1.2) mailto:cfd5z@virginia.edu http://www.people.virginia.edu/~cfd5z http://www.emis.de/journals/SIGMA/2007/052/ 2 C.F. Dunkl for a ≥ b; the relations remain valid when both (z, z) and ( T, T ) are interchanged. The Laplacian is 4TT . These results are from [2, Section 3]. The harmonic polynomials and formulae (1.1) and (1.2) also appear in Berenstein and Burman [1, Section 2]. The aim of this paper is to find an explicit form of the intertwining operator V . This is the unique linear transformation that maps homogeneous polynomials to homogeneous polynomials of the same degree and satisfies TV f (z) = V ∂ ∂z f (z) , TV f (z) = V ∂ ∂z f (z) , V 1 = 1. The operator was defined for general finite reflection groups in [4]. Rösler [8] proved that V is a positive operator when κ0, κ1 > 0; this roughly means that if a polynomial f satisfies f (y) ≥ 0 for all y with ‖y‖ < R (for some R) then V f (y) ≥ 0 on the same set. The present paper does not shed light on the positivity question since the formulae are purely algebraic. In Section 5 the special case − (κ0 + κ1) ∈ N is considered in more detail. These values of (κ0, κ1) are apparent singularities in the expressions for V zazb which are found in Section 4. The book by Y. Xu and the author [7] is a convenient reference for the background of this paper. In a way, to find V zazb only requires to solve a set of equations involving V zjzk for 0 ≤ j ≤ a, 0 ≤ k ≤ b. This can be implemented in computer algebra for small a, b but it is not really an explicit description. For example, by direct computation we find that V z2 = (κ0 + κ1 + 1) z2 + (κ0 − κ1) z2 (2κ0 + 1) (2κ1 + 1) (2κ0 + 2κ1 + 1) , s = 2, V z2 = 2z2 (sκ0 + sκ1 + 1) (sκ0 + sκ1 + 2) , s > 2. The idea is to find the harmonic expansion of V zazb; suppose f (z) is (real-) homogeneous of degree n then there is a unique expansion f (z) = bn/2c∑ j=0 (zz)j fn−2j (z) where fn−2j is homo- geneous of degree n − 2j and is harmonic, that is, TTfn−2j = 0, for 0 ≤ j ≤ n/2. There is some more information easily available for the expansion of V zazb. Let n = a + b and suppose V zazb = n∑ j=0 cjz n−jzj for certain coefficients cj . Because V commutes with the action of the group we deduce that V ( (ωz)a (ωz)b ) = ωa−b n∑ j=0 cjz n−jzj = n∑ j=0 ωn−2jcjz n−jzj ; thus cj 6= 0 implies n− 2j ≡ a− b mod (2s) or j ≡ b mod s. Further V ( zazb ) = n∑ j=0 cjz n−jzj , so it will suffice to determine V zazb for a ≥ b. We will use the Poisson kernel to calculate the polynomials denoted Kn (x, y) := V x ( 1 n! (x1y1 + x2y2) n) (see [5, p. 1219]), where y ∈ R2 and V x acts on the variable x. Thus V xn−j 1 xj 2 is j! (n− j)! times the coefficient of yn−j 1 yj 2 in Kn (x, y). This is adapted to complex coordinates by setting w = y1 + iy2, in which case x1y1 + x2y2 = 1 2 (zw + zw). 2 The Poisson kernel Actually it is only the series expansion of this kernel that is used. For now we assume κ0, κ1 ≥ 0. The measure on the circle T := { eiθ : −π < θ ≤ π } associated to the group I2 (2s) and the Polynomials Associated with Dihedral Groups 3 operators T , T is dµ ( eiθ ) := 1 2B ( κ0 + 1 2 , κ1 + 1 2 ) (sin2 sθ )κ0 ( cos2 sθ )κ1 dθ. Suppose g is a function of t = cos 2sθ then∫ T g (t (θ)) dµ ( eiθ ) = 2−κ0−κ1 B ( κ0 + 1 2 , κ1 + 1 2 ) ∫ 1 −1 g (t) (1− t)κ0−1/2 (1 + t)κ1−1/2 dt. The inner product in L2 (T, µ) is 〈f, g〉 := ∫ T f (z) g (z)dµ (z) and ‖f‖ := 〈f, f〉1/2. Throughout the polynomials under consideration have real coefficients so that g (z, z) = g (z, z). By the group invariance of µ the integral ∫ T zazbdµ (z)is real-valued when a ≡ b mod (2s) and vanishes otherwise. There is an orthogonal decomposition L2 (T, µ) = ∞∑ n=0 ⊕Hn; for n > 0 each Hn is of dimension two and consists of the polynomials in z, z (real-) homogeneous of degree n and annihilated by TT (the harmonic property), while H0 consists of the constant functions. The Poisson kernel is the reproducing kernel for harmonic polynomials (for more details see [3, 5]). Xu [10] investigated relationships between harmonic polynomials, the intertwining operator and the Poisson kernel for the general reflection group. The paper of Scalas [9] concerns boundary value problems for the dihedral groups. The projection of the kernel onto Hn is denoted by Pn (z, w) and satisfies∫ T Pn (z, w) g (w) dµ (w) = g (z) for each polynomial g ∈ Hn. There is a formula for Pn in terms of { Kn−2j : 0 ≤ j ≤ n 2 } (see [5, p. 1224]) which can be inverted. In the present case Pn (z, w) = bn/2c∑ j=0 (γ0)n j! (2− n− γ0)j 2n−2j (zzww)j Kn−2j (z, w) , where γ0 = sκ0 + sκ1 + 1. The inverse relation is Kn (z, w) = 2−n bn/2c∑ j=0 1 j! (γ0)n−j (zzww)j Pn−2j (z, w) . (2.1) This is a consequence of the following: Proposition 1. Suppose there are two sequences {ξn : n ∈ N0} and {ηn : n ∈ N0} in a vector space over Q (γ0) where γ0 is transcendental, then ξn = bn/2c∑ j=0 (γ0)n j! (2− n− γ0)j ηn−2j , n ∈ N0, if and only if ηn = bn/2c∑ j=0 1 j! (γ0)n−j ξn−2j , n ∈ N0. 4 C.F. Dunkl Proof. Consider the matrices A and B defined by ξn = ∑ j Ajnηj , ηn = ∑ j Bjnξj ; these matrices are triangular and the diagonal entries are nonzero, hence they are nonsingular. It suffices to show B is a one-sided inverse of A; this is actually finite-dimensional linear algebra, since one can truncate to the range 0 ≤ n, j ≤ M for any M ∈ N. Indeed bn/2c∑ j=0 (γ0)n j! (2− n− γ0)j bn/2−jc∑ i=0 1 i! (γ0)n−2j−i ξn−2j−2i = bn/2c∑ k=0 ξn−2k (γ0)n k! (γ0)n−k k∑ j=0 (−k)j (1− n− γ0 + k)j j! (2− n− γ0)j = bn/2c∑ k=0 ξn−2k (γ0)n (1− k)k k! (γ0)n−k (2− n− γ0)k = ξn; using the substitution i = k − j we obtain (γ0)n−2j−i = (γ0)n−k−j = (−1)j(γ0)n−k (1−n−γ0+k)j and 1 i! = (−1)j (−k)j k! ; the sum over j is found by the Chu–Vandermonde formula. � Set ξn = Pn(z,w) (zzww)n/2 and ηn = 2nKn(z,w) (zzww)n/2 for n ∈ N0 to prove equation (2.1). Suppose for each n ∈ N there exist a basis {hn1, hn2} and a biorthogonal basis {gn1, gn2} for Hn with real coefficients in z, z (so hn1 (z, z) = hn1 (z, z), for example). Thus 〈hni, gnj〉 = δij/λni,with structural constants λni. Then Pn (z, w) = 2∑ i=1 λnihni (z, z) gni (w,w) . (2.2) Once this is made sufficiently explicit we can compute Kn (z, w) and V zn−jzj . The description of harmonic polynomials is in terms of the case s = 1 (corresponding to the group I2 (2) = Z2×Z2). In terms of Jacobi polynomials the polynomials annihilated by T are: f2n ( reiθ ) := r2nP (κ0− 1 2 ,κ1− 1 2) n (cos 2θ) + i 2 ( r2 sin 2θ ) r2n−2P (κ0+ 1 2 ,κ1+ 1 2) n−1 (cos 2θ) , (2.3) f2n+1 ( reiθ ) := ( n + κ0 + 1 2 ) r cos θ r2nP (κ0− 1 2 ,κ1+ 1 2) n (cos 2θ) (2.4) + i ( n + κ1 + 1 2 ) r sin θ r2nP (κ0+ 1 2 ,κ1− 1 2) n (cos 2θ) ; where the subscript indicates the degree of homogeneity, (clearly fn is a polynomial with real coefficients in z, z; cos 2θ = ( z2 + z2 ) / (2zz) and i 2 ( r2 sin 2θ ) = 1 4 ( z2 − z2 ) ). The real and imaginary parts form a basis for the harmonic polynomials. Specifically let f0 n (z) := Re fn (z) , f1 n (z) := i Im fn (z) . This implies that both f0 n and f1 n have real coefficients in z, z and f0 n (z, z) = f0 n (z, z) , f1 n (z, z) = −f1 n (z, z). When s > 1 and 1 ≤ t < s it is known [2, p. 182] that { ztfn (zs) , ztfn (zs) } is an orthogonal basis for Hns+t for n ≥ 0. Henceforth we denote hns+t,1 (z) = gns+t,1 (z) = ztfn (zs) = hns+t,2 = gns+t,2 and λns+t,1 = λns+t,2 = ‖fn‖−2. The integral 〈 ztfn (zs) , ztfn (zs) 〉 reduces to the case s = 1 and t = 0. When s ≥ 1 { f0 n (zs) , f1 n (zs) } is an orthogonal basis for Hns and zsfn−1 (zs) is orthogonal to fn (zs). By orthogonality ‖fn‖2 = ∥∥f0 n ∥∥2 + ∥∥f1 n ∥∥2 and Polynomials Associated with Dihedral Groups 5 the latter two norms are standard Jacobi polynomial facts. The associated structural constants are denoted by labeled λ’s. Thus λ0 2n := ∥∥f0 2n ∥∥−2 = n! (κ0 + κ1 + 1)n (κ0 + κ1 + 2n)( κ0 + 1 2 ) n ( κ1 + 1 2 ) n (κ0 + κ1 + n) , (2.5) λ1 2n := ∥∥f1 2n ∥∥−2 = (n− 1)! (κ0 + κ1 + 1)n (κ0 + κ1 + 2n)( κ0 + 1 2 ) n ( κ1 + 1 2 ) n , (2.6) λ2n := ‖f2n‖−2 = n! (κ0 + κ1 + 1)n( κ0 + 1 2 ) n ( κ1 + 1 2 ) n ; (2.7) and λ0 2n+1 := ∥∥f0 2n+1 ∥∥−2 = n! (κ0 + κ1 + 1)n (κ0 + κ1 + 2n + 1)( n + κ0 + 1 2 ) ( κ0 + 1 2 ) n+1 ( κ1 + 1 2 ) n+1 , (2.8) λ1 2n+1 := ∥∥f1 2n+1 ∥∥−2 = n! (κ0 + κ1 + 1)n (κ0 + κ1 + 2n + 1)( n + κ1 + 1 2 ) ( κ0 + 1 2 ) n+1 ( κ1 + 1 2 ) n+1 , (2.9) λ2n+1 := ‖f2n+1‖−2 = n! (κ0 + κ1 + 1)n( κ0 + 1 2 ) n+1 ( κ1 + 1 2 ) n+1 . (2.10) From this point on we no longer need the measure µ on the circle. Only the algebraic expressions are used. The condition κ0, κ1 ≥ 0 is replaced by the requirement that none of −κ0+ 1 2 , −κ1+ 1 2 , −s (κ0 + κ1) equal a positive integer. The exceptional case − (κ0 + κ1) ∈ N is taken up in the last section. In the next section we compute the structural constants for the biorthogonal bases {fn (zs) , fn (zs)} and {zsfn−1 (zs) , zsfn−1 (zs)} for Hns (see [3, p. 461]. It is easier to carry this out with material developed in the next section. 3 Expressions for coefficients This is a detailed study of the coefficients of fn (z) in terms of powers of z, z. The expressions are in the form of a single sum of hypergeometric 3F2-type, and can not be simplified any further. For a polynomial f in z, z define c (f ; a, b) to be the coefficient of zazb in f , that is, f (z, z) = ∑ a,b≥0 c (f ; a, b) zazb. Since we restrict to polynomials with real coefficients the equation c ( f (z); a, b ) = c (f ; b, a) is valid. Further c (f (zs) ; as, bs) = c (f ; a, b). Recall Kn (z, w) := 1 2nn! V z ((zw + zw)n) , thus V zn−jzj is 2nj! (n− j)! times the coefficient of wjwn−j in Kn (z, w). To adapt the notation from equation (2.2) for P0 set h01 = g01 = λ01 = 1 and h02 = g02 = λ02 = 0. Then Kn (z, w) = 2−n bn/2c∑ j=0 1 j! (sκ0 + sκ1 + 1)n−j (3.1) × (zzww)j 2∑ i=1 λn−2j,ihn−2j,i (z, z) gn−2j,i (w,w) . 6 C.F. Dunkl Proposition 2. For 0 ≤ m ≤ n, V ( zn−mzm ) = m! (n−m)! bn/2c∑ j=0 1 j! (sκ0 + sκ1 + 1)n−j (3.2) × (zz)j 2∑ i=1 λn−2j,ic (gn−2j,i;n−m− j, m− j) hn−2j,i (z, z) . The nonzero terms appear at increments (in j) of 2s. We start by finding c ( f0 n;n− j, j ) and c ( f1 n;n− j, j ) . This is straightforward and will serve as motivation for introducing a specific useful 3F2-series. Consider f0 2n (z) and recall that P (α,β) n (t) = (α + 1)n n! 2F1 ( −n, n + α + β + 1 α + 1 ; 1− t 2 ) . When z = reiθ we have 1 2 (1− cos 2θ) = − (z − z)2 / (4zz) so r2nP (κ0− 1 2 ,κ1− 1 2) n (cos 2θ) = ( κ0 + 1 2 ) n n! n∑ l=0 2l∑ i=0 (−n)l (n + κ0 + κ1)l (2l)! l! ( κ0 + 1 2 ) l i! (2l − i)! 2−2lzn+l−izn−l+i (−1)l+i = ( κ0 + 1 2 ) n n! n∑ j=−n (−1)j zn+jzn−j n−j∑ i=max(−2j,0) (−n)j+i (n + κ0 + κ1)j+i ( 1 2 ) j+i i! ( κ0 + 1 2 ) j+i (2j + i)! ; (substituting l = i + j, so 0 ≤ i + j ≤ n and 0 ≤ i ≤ 2i + 2j are the ranges of the summation) by the (z, z)-symmetry it suffices to consider j ≥ 0. Thus c ( f0 2n;n + j, n− j ) = ( κ0 + 1 2 ) n (−n)j (n + κ0 + κ1)j ( 1 2 ) j( κ0 + 1 2 ) j (2j)!n! (−1)j × n−j∑ i=0 (j − n)i (n + κ0 + κ1 + j)i ( 1 2 + j ) i i! ( κ0 + 1 2 + j ) i (2j + 1)i = (n + κ0 + κ1)j ( κ0 + 1 2 + j ) n−j 22jj! (n− j)! 3F2 ( j − n, n + κ0 + κ1 + j, j + 1 2 κ0 + 1 2 + j, 2j + 1 ; 1 ) ; this used (2j)! = 22jj! ( 1 2 ) j and ( κ0 + 1 2 ) n / ( κ0 + 1 2 ) j = ( κ0 + 1 2 + j ) n−j . The sum, which appears to be a mysterious combination of the parameters, actually has a nice form revealing more useful information. Definition 1. For n ∈ N0 and parameters a, b, c1, c2 let En (a, b; c1, c2) := (a)n (c2)n n! (c1 + c2)n 3F2 ( −n, b, c1 1− n− a, 1− c2 − n ; 1 ) = 1 n! (c1 + c2)n n∑ j=0 (−n)j j! (a)n−j (b)j (c1)j (c2)n−j . Observe the symmetry En (a, b; c1, c2) = (−1)n En (b, a; c2, c1). This follows from manipula- tions such as (a)n−j = (−1)j (a)n / (1− n− a)j . The following transformation is relevant to the calculation of coefficients. Polynomials Associated with Dihedral Groups 7 Proposition 3. For n ∈ N0 and parameters a, b, c1, c2 En (a, b; c1, c2) = (a + c1)n n! 3F2 ( −n, n + a + b + c1 + c2 − 1, c1 a + c1, c1 + c2 ; 1 ) . Proof. Use the transformation 3F2 ( −n, A,B C,D ; 1 ) = (D −B)n (D)n 3F2 ( −n, C −A,B C, 1 + B −D − n ; 1 ) . First set A = b, B = c1, C = 1− n− a,D = 1− n− c2 then (a)n (c2)n n! (c1 + c2)n 3F2 ( −n, b, c1 1− n− a, 1− c2 − n ; 1 ) = (a)n n! 3F2 ( −n, 1− n− a− b, c1 1− n− a, c1 + c2 ; 1 ) . Set A = 1− n− a− b, B = c1, C = c1 + c2, D = 1− n− a to obtain the stated formula. In the calculation the reversal such as (1− n− a)n = (−1)n (a)n is used several times. � We arrive at a pleasing formula: c ( f0 2n;n + j, n− j ) = (n + κ0 + κ1)j 22jj! En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) . It is useful because it clearly displays the result of setting one or both parameters equal to zero (or a negative integer). That is En (κ0, 0; c1, c2) = (κ0)n(c2)n n!(c1+c2)n and n ≥ 1 implies En (0, 0; c1, c2) = 0. (When κ0 = κ1 = 0 the polynomial f0 2n is a multiple of the Chebyshev polynomial of the first kind, that is f0 2n (z) = (n)n n!22n ( z2n + z2n ) , a fact obvious from the definition of f0 2n.) The remaining basis polynomials can all be expressed in terms of the function E. Proposition 4. For n ∈ N f0 2n (z) = n∑ j=1 ( zn+jzn−j + zn−jzn+j ) 1 22jj! × (n + κ0 + κ1)j En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) + znznEn ( κ0, κ1; 1 2 , 1 2 ) , f1 2n (z) = n∑ j=1 ( zn+jzn−j − zn−jzn+j ) 1 22j (j − 1)! × (n + κ0 + κ1 + 1)j−1 En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) . Proof. The expansion for f0 2n has already been determined. Next i 2r2 sin 2θ = 1 4 ( z2 − z2 ) so( i 2 r2 sin 2θ ) r2n−2P (κ0+ 1 2 ,κ1+ 1 2) n−1 (cos 2θ) = ( κ0 + 3 2 ) n−1 (n− 1)! × n−1∑ l=0 (1− n)l (n + κ0 + κ1 + 1)l l! ( κ0 + 3 2 ) l 2−2l−2 (−1)l (z + z) (z − z)2l+1 (zz)n−1−l , and 2−2l−2 (z + z) (z − z)2l+1 (zz)n−1−l = 2−2l−2 2l+2∑ i=0 (2l + 1)! (2l + 2− 2i) i! (2l + 2− i)! (−1)i zn+l+1−izn−l−1+i 8 C.F. Dunkl = 2l+2∑ i=0 l! ( 1 2 ) l+1 (l + 1− i) i! (2l + 2− i)! (−1)i zn+l+1−izn−l−1+i; substitute l = j+i−1. By the symmetry f1 2n (z) = −f1 2n (z) it suffices to find c ( f1 2n;n + j, n− j ) for 1 ≤ j ≤ n. Indeed c ( f1 2n;n + j, n− j ) = j ( κ0 + 3 2 ) n−1 (1− n)j−1 (n + κ0 + κ1 + 1)j−1 ( 1 2 ) j( κ0 + 3 2 ) j−1 (2j)! (n− 1)! (−1)j−1 × n−j∑ i=0 (j − n)i (n + κ0 + κ1 + j)i ( 1 2 + j ) i i! ( κ0 + 1 2 + j ) i (2j + 1)i = (n + κ0 + κ1 + 1)j−1 22j (j − 1)! En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) . This completes the proof. � Proposition 5. For n ∈ N0 f0 2n+1 (z) = ( n + κ0 + 1 2 ) n∑ j=0 ( zn+1+jzn−j + zn−jzn+1+j ) 1 22j+1j! × (n + κ0 + κ1 + 1)j En−j ( κ0, κ1; j + 1 2 , j + 3 2 ) , f1 2n+1 (z) = ( n + κ1 + 1 2 ) n∑ j=0 ( zn+1+jzn−j − zn−jzn+1+j ) 1 22j+1j! × (n + κ0 + κ1 + 1)j En−j ( κ0, κ1; j + 3 2 , j + 1 2 ) . Proof. The second equation is straightforward: f1 2n+1 (z) = 1 2 ( n + κ1 + 1 2 ) (z − z) r2nP (κ0+ 1 2 ,κ1− 1 2) n (cos 2θ) = ( n + κ1 + 1 2 ) ( κ0 + 3 2 ) n n! × n∑ l=0 2l+1∑ i=0 (−n)l (n + κ0 + κ1 + 1)l (2l + 1)! l! ( κ0 + 3 2 ) l i! (2l + 1− i)! 2−2l−1zn+1+l−izn−l+i (−1)l+i = ( n + κ1 + 1 2 ) ( κ0 + 3 2 ) n n! n∑ j=0 ( zn+1+jzn−j − zn−jzn+1−j ) × (−1)j (−n)j (n + κ0 + κ1 + 1)j ( 1 2 ) j+1( κ0 + 3 2 ) j (2j + 1)! n−j∑ i=0 (j − n)i (n + κ0 + κ1 + j + 1)i ( 3 2 + j ) i i! ( κ0 + 3 2 + j ) i (2j + 2)i , (substituting l = j + i for 0 ≤ j ≤ n) thus c ( f1 2n+1;n + 1 + j, n− j ) = −c ( f1 2n+1;n− j, n + 1 + j ) = ( n + κ1 + 1 2 ) (n + κ0 + κ1 + 1)j j!22j+1 En−j ( κ0, κ1; j + 3 2 , j + 1 2 ) . Note that (2j + 1)! = 22j+1j! ( 1 2 ) j+1 and ( κ0 + 3 2 ) n / ( κ0 + 3 2 ) j = ( κ0 + 3 2 + j ) n−j . For f0 2n+1 reverse the parameters, that is, P (κ0− 1 2 ,κ1+ 1 2) n (cos 2θ) = (−1)n P (κ1+ 1 2 ,κ0− 1 2) n (− cos 2θ) , Polynomials Associated with Dihedral Groups 9 and note 1 2 (1 + cos 2θ) = (z + z)2 / (4zz) and r cos θ = 1 2 (z + z). Thus f0 2n+1 (z) = (−1)n ( n + κ0 + 1 2 ) ( κ1 + 3 2 ) n n! × n∑ l=0 2l+1∑ i=0 (−n)l (n + κ0 + κ1 + 1)l (2l + 1)! l! ( κ1 + 3 2 ) l i! (2l + 1− i)! 2−2l−1zn+1+l−izn−l+i = (−1)n ( n + κ0 + 1 2 ) ( κ1 + 3 2 ) n n! n∑ j=0 ( zn+1+jzn−j + zn−jzn+1−j ) × (−n)j (n + κ0 + κ1 + 1)j ( 1 2 ) j+1( κ1 + 3 2 ) j (2j + 1)! n−j∑ i=0 (j − n)i (n + κ0 + κ1 + j + 1)i ( 3 2 + j ) i i! ( κ1 + 3 2 + j ) i (2j + 2)i , thus c ( f0 2n+1;n + 1 + j, n− j ) = c ( f0 2n+1;n− j, n + 1 + j ) = ( n + κ0 + 1 2 ) (n + κ0 + κ1 + 1)j j!22j+1 (−1)n−j En−j ( κ1, κ0; j + 3 2 , j + 1 2 ) . The symmetry relation Em (b, a; c2, c1) = (−1)m Em (a, b; c1, c2) finishes the computation. � To find the coefficients of fn we use contiguity relations satisfied by Em. Lemma 1. For m ∈ N0 and parameters a, b, c (m + a + c) Em (a, b; c, c + 1)− (m + b + c) Em (a, b; c + 1, c) (3.3) = 2 (m + 1) Em+1 (a, b; c, c) , (m + a + c) Em (a, b; c, c + 1) + (m + b + c) Em (a, b; c + 1, c) (3.4) = m + 2c + 1 2c + 1 (m + a + b + 2c) Em (a, b; c + 1, c + 1) . Proof. We compute the coefficient of (b)j for 0 ≤ j ≤ m + 1 in the two identities. Note that (m + b + c) (b)j = (b)j+1 + (m + c− j) (b)j , then replace j by j − 1 for the first term. The coefficient of (b)j in (m + b + c) Em (a, b; c + 1, c) is 1 m! (2c + 1)m j! × { (−m)j (m + c− j) (a)m−j (c)m−j (c + 1)j + j (−m)j−1 (a)m+1−j (c)m+1−j (c + 1)j−1 } = (−m)j−1 m! (2c + 1)m j! (a)m−j (c)m+1−j (c + 1)j−1 {(−m + j − 1) (c + j) + j (a + m− j)} . The coefficient of (b)j in the left side of (3.3) is (−m)j−1 m! (2c + 1)m j! (a)m−j (c)j (c + 1)m−j × {(m + a + c) (−m + j − 1)− (−m + j − 1) (c + j)− j (a + m− j)} = (−m)j−1 m! (2c + 1)m j! (a)m−j (c)j (c + 1)m−j (a + m− j) (−m− 1) = 2 (−1−m)j m! (2c)m+1 j! (a)m+1−j (c)j (c)m+1−j . 10 C.F. Dunkl This proves equation (3.3). For the right side of (3.4) the coefficient of (b)j is found similarly as before ((m + a + b + 2c) (b)j = (m + a + 2c− j) (b)j +(b)j+1, and so on). The coefficient of (b)j in the left side is (−m)j−1 m! (2c + 1)m j! (a)m−j (c)j (c + 1)m−j {(m + a + 2c) (−m + j − 1) + j (a− 1)} , and in the right side m + 2c + 1 m! (2c + 1) (2c + 2)m { (−m)j j! (m + a + 2c− j) (a)m−j (c + 1)j (c + 1)m−j + (−m)j−1 j! j (a)m+1−j (c + 1)j−1 (c + 1)m+1−j } = (−m)j−1 m! (2c + 1)m j! (a)m−j (c + 1)j−1 (c + 1)m−j × {(−m + j − 1) (m + a + 2c− j) (c + j) + j (a + m− j) (c + m + 1− j)} ; the expression in {·} equals c (m + a + 2c) (−m + j − 1) + cj (a− 1) which proves (3.4). � Proposition 6. For n ∈ N0 f2n (z) = n∑ j=1 ( (n + κ0 + κ1 + j) zn+jzn−j + (n + κ0 + κ1 − j) zn−jzn+j ) × 1 22jj! (n + κ0 + κ1 + 1)j−1 En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) + En ( κ0, κ1; 1 2 , 1 2 ) znzn, f2n+1 (z) = n+1∑ j=1 ( (n + 1 + j) zn+jzn+1−j + (n + 1− j) zn−jzn+1+j ) × 1 22jj! (n + κ0 + κ1 + 1)j En+1−j ( κ0, κ1; j + 1 2 , j + 1 2 ) + (n + 1) En+1 ( κ0, κ1; 1 2 , 1 2 ) znzn+1. Proof. Recall fn = f0 n + f1 n. For 0 ≤ j ≤ n from Proposition 4 we find c (f2n;n + j, n− j) = c ( f0 2n;n + j, n− j ) + c ( f1 2n;n + j, n− j ) = (n + κ0 + κ1 + 1)j−1 22jj! ((n + κ0 + κ1) + j) En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) , c (f2n;n− j, n + j) = c ( f0 2n;n + j, n− j ) − c ( f1 2n;n + j, n− j ) = (n + κ0 + κ1 + 1)j−1 22jj! ((n + κ0 + κ1)− j) En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) . It remains to compute c (f2n+1;n + 1 + j, n− j) and c (f2n+1;n− j, n + 1− j) for 0 ≤ j ≤ n. Write the arguments as ( n + 1 2 + ε ( j + 1 2 ) , n + 1 2 − ε ( j + 1 2 )) with ε = ±1. Then, by Proposi- tion 5, c ( f2n+1;n + 1 2 + ε ( j + 1 2 ) , n + 1 2 − ε ( j + 1 2 )) Polynomials Associated with Dihedral Groups 11 = c ( f0 2n+1;n + 1 + j, n− j ) + εc ( f1 2n+1;n + 1 + j, n− j ) = (n + κ0 + κ1 + 1)j j!22j+1 {( n + κ0 + 1 2 ) En−j ( κ0, κ1; j + 1 2 , j + 3 2 ) + ε ( n + κ1 + 1 2 ) En−j ( κ0, κ1; j + 3 2 , j + 1 2 )} . When ε = 1 by (3.4) we obtain c (f2n+1;n + 1 + j, n− j) = (n + κ0 + κ1 + 1)j+1 (j + 1)!22j+2 (n + j + 2) En−j ( κ0, κ1; j + 3 2 , j + 3 2 ) , and when ε = −1 by (3.3) we obtain c (f2n+1;n− j, n + 1− j) = (n + κ0 + κ1 + 1)j j!22j (n− j + 1) En−j+1 ( κ0, κ1; j + 1 2 , j + 1 2 ) . The stated formula for f2n+1 uses c (f2n+1;n + j, n + 1− j) explicitly (j is shifted by 1). � For Hns with n > 0 we intend to use both the orthogonal basis { f0 n (zs) , f1 n (zs) } as well as the biorthogonal bases { fn (zs) , fn (zs) } and { zsfn−1 (zs) , zsfn−1 (zs) } . For the latter we need the value of νn := 〈fn (zs) , zsfn−1 (zs)〉. Instead of doing the integral directly we use the two formulae for Pns (z, w), that is, Pns (z, w) = λ0 nf0 n (zs) f0 n (ws) + λ1 nf1 n (zs) f1 n (ws) = ν−1 n (fn (zs) wsfn−1 (ws) + fn (zs) wsfn−1 (ws)) . From the coefficients of wns in the equation we obtain λ0 nc ( f0 n; 0, n ) f0 n (zs) + λ1 nc ( f1 n;n, 0 ) f1 n (zs) = ν−1 n c (fn−1;n− 1, 0) fn (zs) . But fn = f0 n + f1 n so by the linear independence of { f0 n, f1 n } there are two equations for cn (one is redundant). Thus νn = c (fn−1;n− 1, 0) λ0 nc (f0 n; 0, n) = c (fn−1;n− 1, 0) λ1 nc (f1 n;n, 0) . The calculation has two cases depending on n being even or odd: ν2n = 2 ( κ0 + 1 2 ) n ( κ1 + 1 2 ) n (n− 1)! (κ0 + κ1 + 1)n−1 (κ0 + κ1 + 2n) , n ≥ 1, (3.5) ν2n+1 = 2 ( κ0 + 1 2 ) n+1 ( κ1 + 1 2 ) n+1 n! (κ0 + κ1 + 1)n (κ0 + κ1 + 2n + 1) , n ≥ 0. (3.6) 4 The intertwining operator We describe V zazb for a ≥ b. It is helpful to consider the representations of I2 (2s) since V commutes with the group action on polynomials. Since zz is invariant it suffices to consider (zz)b za−b, or zm. The residue of m mod2s is the determining factor. Suppose m ≡ j mod2s and j 6= 0, s. The representation of I2 (2s) on span {zm, zm} is irreducible and isomorphic to the one on span { zj , zj } if 1 ≤ j < s, and to the one on span { z2s−j , z2s−j } if s < j < 2s. If m ≡ 0 mod 2s then span {zm, zm} is the direct sum of the identity and determinant representations (on C1 and 12 C.F. Dunkl C ( z2s − z2s ) respectively). If m ≡ smod2s then span {zm, zm} is the direct sum of the two representations realized on C (zs − zs) and C (zs + zs) (these are relative invariants). Recall Pm (z, w) = 2∑ i=1 λmihmi (z, z) gmi (w,w) and equation (3.2) shows that the nonzero terms in the expansion of V zazb occur only when the condition c (ga+b−2j,i; a− j, b− j) 6= 0 (4.1) is satisfied. If m ≡ 0 mod s then gmi (w,w) is a polynomial in ws, ws thus (4.1) is equivalent to a− j ≡ b− j ≡ 0 mod s, in particular a ≡ b mod s. In this case suppose a = us + r ≥ b = vs + r with 0 ≤ r < s. Set b−j = (v − k) s then j = ks+r, a−j = (u− k) s, a+b−2j = (u + v − 2k) s, and 0 ≤ k ≤ v ≤ u. We see that the nonzero terms occur for P(u+v−2k)s with 0 ≤ k ≤ v. If m ≡ t mod s and 1 ≤ t < s then gm1 (w,w) = wtf(m−t)/s (ws, ws) and (4.1) implies a−j ≡ t mod s, b−j ≡ 0 mod s; further gm2 (w,w) = gm1 (w,w) and (4.1) implies a−j ≡ 0 mod s, b− j ≡ t mod s. Theorem 1. Suppose a − b ≡ t mod s, 1 ≤ t < s and a > b. Let b = vs + r with v ≥ 0 and 0 ≤ r < s and a = us + r + t, then V ( zazb ) = a!b! v∑ k=0 1 (ks + r)! (sκ0 + sκ1 + 1)a+(v−k)s λu+v−2k × c (fu+v−2k;u− k, v − k) (zz)ks+r ztfu+v−2k (zs) + a!b! v∑ k=1−b(r+t)/sc 1 ((k − 1) s + r + t)! (sκ0 + sκ1 + 1)b+(u−k+1)s λu+v+1−2k × c (fu+v+1−2k; v − k, u− k + 1) (zz)(k−1)s+r+t zs−tfu+v+1−2k (zs) . Proof. Since 0 < a − b = (u− v) s + t we have u ≥ v. For the first part of the series, corresponding to i = 1 in Pns+t let b− j = (v − k) s with k ≤ v; then j = b− (v − k) s = ks + r, implying k ≥ 0. Further a − j = (u− k) s + t, a + b − 2j = (u + v − 2k) s + t = a + b − 2r − 2ks (and a + b − j = a + (b− j) = a + (v − k) s). Also c ( ztfu+v−2k (zs) ; a− j, b− j ) = c (fu+v−2k;u− k, v − k) . This proves the first part. For the second part, with i = 2 in Pns−t let b − j = (v − k) s + (s− t), thus k ≤ v. Then j = (k − 1) s + r + t. The requirement j ≥ 0 implies 1− k ≤ r+t s , that is k ≥ 1− ⌊ r+t s ⌋ (if 0 ≤ r + t < s then k ≥ 1, otherwise s ≤ r + t < 2s and k ≥ 0). Also a − j = (u− k + 1) s and a + b − 2j = (u + v + 1− 2k) s + (s− t) (and a+b−j = b+(a− j) = b+(u− k + 1) s). In this case we use c ( zs−tfu+v+1−2k (zs) ; a− j, b− j ) = c (fu+v+1−2k; v − k, u− k + 1). � Note that the degrees of fm have the same parity as u + v in the first sum, and the opposite in the second sum. By Proposition 6 we can find the coefficients explicitly. If u+v is even then c (fu+v−2k;u− k, v − k) = 1 2u−v ( u−v 2 ) ! ( u + v 2 − k + κ0 + κ1 ) u−v 2 × Ev−k ( κ0, κ1; u− v + 1 2 , u− v + 1 2 ) , and c (fu+v+1−2k; v − k, u− k + 1) = (v − k + 1) 2u−v ( u−v 2 ) ! ( u + v 2 − k + κ0 + κ1 + 1 ) u−v 2 × Ev−k+1 ( κ0, κ1; u− v + 1 2 , u− v + 1 2 ) . Polynomials Associated with Dihedral Groups 13 If u + v is odd then c (fu+v−2k;u− k, v − k) = (u− k + 1) 2u−v+1 ( u−v+1 2 ) ! ( u + v + 1 2 − k + κ0 + κ1 ) u−v+1 2 × Ev−k ( κ0, κ1; u− v 2 + 1, u− v 2 + 1 ) , and c (fu+v+1−2k; v − k, u− k + 1) = (v − k + 1) 2u−v+1 ( u−v+1 2 ) ! ( u + v + 3 2 − k + κ0 + κ1 ) u−v−1 2 × Ev−k ( κ0, κ1; u− v 2 + 1, u− v 2 + 1 ) . Theorem 2. Suppose a ≡ b mod s, and a ≥ b. Let a = us + r ≥ b = vs + r with 0 ≤ r < s and v ≥ 0. If a > b then V ( zazb ) = a!b! v∑ k=0 1 (ks + r)! (sκ0 + sκ1 + 1)b+(u−k)s ν−1 u+v−2k × (zz)ks+r {c (fu+v−2k−1;u− k − 1, v − k) fu+v−2k (zs) + c (fu+v−2k−1; v − k − 1, u− k) fu+v−2k (zs)}, V ( 1 2 ( zazb − zbza )) = a!b! v∑ k=0 1 (ks + r)! (sκ0 + sκ1 + 1)b+(u−k)s λ1 u+v−2k × c ( f1 u+v−2k;u− k, v − k ) (zz)ks+r f1 u+v−2k (zs) . If a ≥ b then V ( 1 2 ( zazb + zbza )) = a!b! v∑ k=0 1 (ks + r)! (sκ0 + sκ1 + 1)b+(u−k)s λ0 u+v−2k × c ( f0 u+v−2k;u− k, v − k ) (zz)ks+r f0 u+v−2k (zs) . Proof. The three different expansions for zazb, 1 2 ( zazb − zbza ) and 1 2 ( zazb + zbza ) use the bases { fj , fj } , { f1 j } and { f0 j } respectively. Suppose a = us + r ≥ b = vs + r with 0 ≤ r < s. Set b − j = (v − k) s then j = ks + r, a − j = (u− k) s, a + b − 2j = (u + v − 2k) s, and 0 ≤ k ≤ v ≤ u. Consider the case a > b, that is, u > v. For arbitrary m ≥ 1 the basis {fm (zs, zs) , fm (zs, zs)} for Hsm has the biorthogonal set {zsfm−1 (zs, zs) , zsfm−1 (zs, zs)} and c (zsfn1+n2−1 (zs, zs) ;n1s, n2s) = c (fn1+n2−1;n1 − 1, n2) , c (zsfn1+n2−1 (zs, zs) ;n1s, n2s) = c (fn1+n2−1;n2 − 1, n1) . The constants νm are given in equations (3.5) and (3.6). This demonstrates the first series. The remaining two follow from Proposition 3.2. � Observe that in the series for V ( zazb ) the lowest-degree term with k = v < u reduces to one summand since c (fu−v−1;−1, u− v) = 0. Each term in V ( zazb − zbza ) is of the same representation type, C (zs − zs) when a− b ≡ smod2s or C ( z2s − z2s ) when a− b ≡ 0 mod 2s. Similarly each term in V ( zazb + zbza ) is of the representation type C (zs + zs) or C1 (depending on the parity of a−b s ). The coefficients can be found from Propositions 4 and 5. 14 C.F. Dunkl For a > b consider zazb as (zz)b times the (ordinary) harmonic polynomial za−b. The fact that V ( zazb ) is L2 (T, µ)-orthogonal to Hn for n < a− b, equivalently, that the above series for V ( zazb ) contain no terms involving Hn with n < a−b (that is, a term like cn (zz)(a+b−n)/2 pn (z) with pn ∈ Hn), is a special case of a result of Xu [10]. This paper also has formulae for V z2m when s = 2, that is, the group I2 (4). 5 Singular values The term “singular values” refers to the set Ks of pairs (κ0, κ1) ∈ C2 for which V is not defined on all polynomials in z, z. Let K0 := { (κ0, κ1) ∈ C2 : {κ0, κ1} ∩ ( −1 2 − N0 ) 6= ∅ } , (at least one of κ0, κ1 is in { −1 2 ,−3 2 , . . . } ). It was shown by de Jeu, Opdam and the author [6, p. 248] that Ks = K0 ∪ { (κ0, κ1) : κ0 + κ1 = − j s , j ∈ N, j s /∈ N } . To illustrate how the singular values appear in the formulae for V consider V z2ns+1 (for s > 1, n ≥ 1) which has only one term in the formula from Theorem 1. In particular c ( V z2ns+1; 2ns + 1, 0 ) = (2ns + 1)! (κ0 + κ1 + 1)2n (n + κ0 + κ1 + 1)n 24nn! ( κ0 + 1 2 ) n ( κ1 + 1 2 ) n (sκ0 + sκ1 + 1)2ns+1 . The denominator vanishes for κ0, κ1 = −1 2 ,−3 2 , . . . ,−2n−1 2 and κ0+κ1 = −k s for 1 ≤ k ≤ 2ns+1. There appear to be singularities at κ0 + κ1 = −k for 1 ≤ k ≤ 2n but the term (κ0 + κ1 + 1)2n in the numerator cancels these zeros. The same cancellation occurs for arbitrary V ( zazb ) in a more complicated way. The formula for Kn (z, w) has the factors (s (κ0 + κ1) + 1)n−j in the denominators thus the individual terms can have simple poles at κ0 + κ1 = −k s for k ∈ N. We will show directly that the singularities at κ0 + κ1 = −m are removable when Kn (z, w) is expressed as a quotient of polynomials in κ0, κ1. It turns out that the terms with poles can be paired in such a way that the sum of each pair has a removable singularity. The pairs correspond to {Pk, P2sm−k} for certain values of k. Throughout we assume that (κ0, κ1) /∈ K0. We use an elementary algebraic result: suppose a rational function F (α, β) (with coefficients in the ring Q [z, z, w, w]) vanishes for a countable set of values {α = 0, β = rn : n ∈ N0} (which are not poles) then F (α, β) is divisible by α; indeed the numerator of F (0, β) is a polynomial in β vanishing at all β = rn hence is zero. This result will be applied with α = κ0 + κ1 + m, β = κ0 − κ1. Most of the section concerns the proof of the following result: let κ0 + κ1 = −m then PN (z, w) = 0 for N > 2sm and PN (z, w) + (zzww)N−sm P2sm−N (z, w) = 0 for 0 ≤ N ≤ 2sm. The Poisson kernels Pn were described in equation (2.2). There are a number of cases, roughly corresponding to the representations of I2 (2s). Proposition 7. Suppose − (κ0 + κ1) = m ∈ N then f0 2n (z) = ( κ0 + 1 2 ) n (m− n)!( κ0 + 1 2 ) m−n n! (zz)2n−m f0 2m−2n (z) , 0 ≤ n ≤ m, f1 2n (z) = ( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n (n− 1)! (zz)2n−m f1 2m−2n (z) , 1 ≤ n ≤ m− 1. Proof. The argument uses the Jacobi polynomials directly. Recall z=reiθ. Then for 0 ≤ n ≤ m f0 2n (z) = r2n ( κ0 + 1 2 ) n n! 2F1 ( −n, n−m κ0 + 1 2 ; 1− cos 2θ 2 ) , Polynomials Associated with Dihedral Groups 15 f0 2m−2n (z) = r2m−2n ( κ0 + 1 2 ) m−n (m− n)! 2F1 ( − (m− n) , (m− n)−m κ0 + 1 2 ; 1− cos 2θ 2 ) , while for 1 ≤ n ≤ m− 1 f1 2n (z) = ir2n sin 2θ ( κ0 + 3 2 ) n−1 (n− 1)! 2F1 ( 1− n, n−m + 1 κ0 + 3 2 ; 1− cos 2θ 2 ) , f1 2m−2n (z) = ir2m−2n sin 2θ ( κ0 + 3 2 ) m−n−1 (m− n− 1)! 2F1 ( − (m− n− 1) , 1− n κ0 + 1 2 ; 1− cos 2θ 2 ) . This proves the formulae. � Proposition 8. Suppose − (κ0 + κ1) = m ∈ N and 0 ≤ n < m then f0 2n+1 (z) = ( κ0 + 1 2 ) n+1 (m− n− 1)!( κ0 + 1 2 ) m−n n! (zz)2n−m+1 f0 2m−2n−1 (z) , f1 2n+1 (z) = ( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n−1 n! (zz)2n−m+1 f1 2m−2n−1 (z) . Proof. Similarly to the even case we have f0 2n+1 (z) = r2n+1 cos θ ( κ0 + 1 2 ) n+1 n! 2F1 ( −n, n−m + 1 κ0 + 1 2 ; 1− cos 2θ 2 ) , f0 2m−2n−1 (z) = r2m−2n−1 cos θ ( κ0 + 1 2 ) m−n (m− n− 1)! 2F1 ( − (m− n− 1) ,−n κ0 + 1 2 ; 1− cos 2θ 2 ) , and f1 2n+1 (z) = ir2n+1 sin θ ( κ1 + n + 1 2 ) ( κ0 + 3 2 ) n n! 2F1 ( −n, n−m + 1 κ0 + 3 2 ; 1− cos 2θ 2 ) , f1 2m−2n−1 (z) = ir2m−2n−1 sin θ ( κ1 + m− n− 1 2 ) ( κ0 + 3 2 ) m−n−1 (m− n− 1)! × 2F1 ( − (m− n− 1) ,−n κ0 + 3 2 ; 1− cos 2θ 2 ) . Thus f1 2n+1 (z) f1 2m−2n−1 (z) = r4n−2m+2 (m− n− 1)! ( κ0 + 1 2 ) n+1 ( −m− κ0 + n + 1 2 ) n! ( κ0 + 1 2 ) m−n ( −κ0 − n− 1 2 ) = r4n−2m+2 (m− n− 1)! ( κ0 + 1 2 ) n n! ( κ0 + 1 2 ) m−n−1 . � Proposition 9. Suppose − (κ0 + κ1) = m ∈ N and 0 ≤ n < m then f2n (z) = ( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n n! (zz)2n−m zf2m−2n−1 (z) . Proof. We use the expressions from Proposition 6. First we show for 0 ≤ j ≤ min (n, m− n) that En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) = ( κ0 + 1 2 ) m−n (n− j)!( κ0 + 1 2 ) n (m− n− j)! Em−n−j ( κ0, κ1; j + 1 2 , j + 1 2 ) . 16 C.F. Dunkl Indeed by Proposition 3( κ0 + 1 2 ) j En−j ( κ0, κ1; j + 1 2 , j + 1 2 ) = ( κ0 + 1 2 ) n (n− j)! 3F2 ( j − n, n−m + j, j + 1 2 κ0 + j + 1 2 , 2j + 1 ; 1 ) , and ( κ0 + 1 2 ) j Em−n−j ( κ0, κ1; j + 1 2 , j + 1 2 ) = ( κ0 + 1 2 ) m−n (m− n− j)! 3F2 ( n−m + j, j − n, j + 1 2 κ0 + j + 1 2 , 2j + 1 ; 1 ) . Let g1 (z) := n!( κ0 + 1 2 ) n f2n (z) , g2 (z) := (m− n− 1)!( κ0 + 1 2 ) m−n (zz)2n−m zf2m−2n−1 (z) , and for j ≥ 0 let bj := 1 22jj! 3F2 ( n−m + j, j − n, j + 1 2 κ0 + j + 1 2 , 2j + 1 ; 1 ) . Then g1 (z) = n∑ j=−n (n−m + 1)|j|−1 n! (n− |j|)! b|j| (n−m + j) zn+jzn−j , g2 (z) = m−n∑ j=n−m (−n)|j| (m− n− 1)! (m− n− |j|)! b|j| (m− n + j) zn+jzn−j . Thus c (g1;n, n) = b0 = c (g2;n, n). Suppose |j| ≥ 1, then c (g1;n + j, n− j) = (n−m + 1)|j|−1 (−n)|j| (−1)j (n−m + j) b|j| for |j| ≤ n, and the equation remains valid if n < |j| ≤ m−n because (−n)|j| = 0 for |j| > n. Also c (g2;n + j, n− j) = (−n)|j| (n + 1−m)|j|−1 (−1)j−1 (m− n− j) b|j|, and the equation remains valid if m−n < |j| ≤ n (that is n−m+ |j|−1 ≥ 0). Thus c (g1;n + j, n− j) = c (g2;n + j, n− j) for |j| ≤ max (n, m− n) and g1 = g2. � Note that if k = 0, 1, 2, . . . then (−k)j = 0 for j > k. Recall the structural constants for the Poisson kernel Pn from equations (2.5)–(2.10). These are rational functions of κ0, κ1 defined for all (κ0, κ1) /∈ K0. Proposition 10. Suppose − (κ0 + κ1) = m ∈ N, 1 ≤ n ≤ m− 1, and n 6= m 2 then λ0 2m−2n λ0 2n = − (( κ0 + 1 2 ) n (m− n)!( κ0 + 1 2 ) m−n n! )2 , λ1 2m−2n λ1 2n = − (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n (n− 1)! )2 , λ0 2m = − ( m!( κ0 + 1 2 ) m )2 , λ1 2m = 0. Proof. Recall λ0 2n = n!(κ0+κ1+1)n−1(κ0+κ1+2n) (κ0+ 1 2)n (κ1+ 1 2)n (for n ∈ N). Also ( κ1 + 1 2 ) n = ( −m− κ0 + 1 2 ) n = (−1)n (κ0 + 1 2 + m− n ) n = (−1)n (κ0+ 1 2)m (κ0+ 1 2)m−n , and similarly ( κ1 + 1 2 ) m−n = (−1)m−n (κ0+ 1 2)m (κ0+ 1 2)n . Thus λ0 2m−2n λ0 2n = (−1)m ( ( κ0 + 1 2 ) n( κ0 + 1 2 ) m−n )2 (m− n)! (1−m)m−n−1 (m− 2n) n! (1−m)n−1 (−m + 2n) . Polynomials Associated with Dihedral Groups 17 But (1−m)m−n−1 (1−m)n−1 = (−1)m (m−1)!(m−n)! (m−1)!n! (note (−k)j = (−1)j k! (k−j)! for k ∈ N0). Next λ1 2n = (n−1)!(κ0+κ1+1)n(κ0+κ1+2n) (κ0+ 1 2)n (κ1+ 1 2)n . Similarly we find λ1 2m−2n λ1 2n = (−1)m ( ( κ0 + 1 2 ) n( κ0 + 1 2 ) m−n )2 (m− n− 1)! (1−m)m−n (m− 2n) (n− 1)! (1−m)n (−m + 2n) , and (1−m)m−n (1−m)n = (−1)m (m−1)!(m−n−1)! (m−1)!(n−1)! . The special case λ0 2m follows from setting n = 0 in the first formula. The term (κ0 + κ1 + 1)m shows λ1 2m = 0. � The following two propositions are proven by similar calculations. Proposition 11. Suppose − (κ0 + κ1) = m ∈ N, 0 ≤ n ≤ m− 1, and n 6= m−1 2 then λ0 2m−2n−1 λ0 2n+1 = − (( κ0 + 1 2 ) n+1 (m− n− 1)!( κ0 + 1 2 ) m−n n! )2 , λ1 2m−2n−1 λ1 2n+1 = − (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n−1 n! )2 . Proposition 12. Suppose − (κ0 + κ1) = m ∈ N, 0 ≤ n ≤ m− 1 then λ2m−2n−1 λ2n = − (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n n! )2 . Proposition 13. Suppose − (κ0 + κ1) = m ∈ N then λ0 m = 0 = λ1 m. If n > 2m then λ0 n = 0 = λ1 n. If n ≥ 2m then λn = 0. Proof. Since both λ0 n and λ1 n contain the factor (κ0 + κ1 + n) for n even or odd, it follows that λ0 m = 0 = λ1 m. The term (κ0 + κ1 + 1)j vanishes for j > m, and j = k − 1 for λ0 2k, j = k for each of λ2k, λ2k+1, λ1 2k, λ0 2k+1, λ1 2k+1. � Theorem 3. Suppose − (κ0 + κ1) = m ∈ N then PN (z, w) = 0 for N > 2sm and PN (z, w) + (zzww)N−sm P2sm−N (z, w) = 0 for 0 ≤ N ≤ 2sm. Proof. If N = sk > 2sm then Psk (z, w) = λ0 kf 0 k (zs) f0 k (ws) + λ1 kf 1 k (zs) f1 k (ws) and λ0 k = λ1 k = 0 by Proposition 13. If N = sk + t with 1 ≤ t < s and N > 2sm then Psk+t (z, w) = λk ( ztwtfk (zs) fk (ws) + ztwtfk (zs) fk (ws) ) , k ≥ 2sm and λk = 0. If N = sm then λ0 m = λ1 m = 0. Suppose N = 2ns and 0 < N < 2sm (so 0 < n < m), then P2ns (z, w) = λ0 2n (( κ0 + 1 2 ) n (m− n)!( κ0 + 1 2 ) m−n n! )2 (zzww)(2n−m)s f0 2m−2n (zs) f0 2m−2n (ws) + λ1 2n (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n (n− 1)! )2 (zzww)(2n−m)s f1 2m−2n (zs) f1 2m−2n (ws) , thus by Propositions 7 and 10 (zzww)(m−2n)s P2ns (z, w) + P2ms−2ns (z, w) = λ0 2n  (( κ0 + 1 2 ) n (m− n)!( κ0 + 1 2 ) m−n n! )2 + λ0 2m−2n λ0 2n  f0 2m−2n (zs) f0 2m−2n (ws) 18 C.F. Dunkl + λ1 2n  (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n (n− 1)! )2 + λ1 2m−2n λ1 2n  f1 2m−2n (zs) f1 2m−2n (ws) = 0. For the special case N = 2sm we have P2sm (z, w) = λ0 2m ((κ0+ 1 2)m m! )2 (zzww)ms = − (zzww)ms P0 because λ1 2m = 0, P0 = 1 and λ0 2m = − ( m! (κ0+ 1 2)m )2 . Similarly by use of Propositions 8 and 11 we show the result holds for N = (2n + 1) s for 0 ≤ n < m and 2n+1 6= m. Suppose N = sk + t with 1 ≤ t < s and 0 < N < 2sm, then 2sm−N = s (2m− k − 1)+(s− t). One of k, 2m−k−1 is even so assume k = 2n with 0 ≤ n < m (otherwise replace N by 2sm−N and t by s− t). By Propositions 9 and 12 λ2nztwtf2n (zs) f2n (ws) +λ2m−2n−1 (zzww)(2n−m)s+t zs−tws−tf2m−2n−1 (zs) f2m−2n−1 (ws) = λ2n (zw)(2n−m)s+t (zw)(2n−m+1)s f2m−2n−1 (zs) f2m−2n−1 (ws) ×  (( κ0 + 1 2 ) n (m− n− 1)!( κ0 + 1 2 ) m−n n! )2 + λ2m−2n−1 λ2n  = 0. Add this equation to its complex conjugate to show P2ns+t (z, w) + (zzww)(2n−m)s+t P(2m−2n)s−t (z, w) = 0. � Theorem 4. For n, m∈N equation (3.1) for Kn(z, w) has a removable singularity at κ0+κ1=−m. Proof. Consider the series Kn (z, w) = 2−n bn/2c∑ j=0 1 j! (sκ0 + sκ1 + 1)n−j (zzww)j Pn−2j (z, w) . The possible poles occur at n− j ≥ sm (that is, (1− sm)n−j = 0) and the multiplicities do not exceed 1. Thus there are no poles if n < sm. If n − 2j > 2sm then Pn−2j (z, w) is divisible by (κ0 + κ1 + m), by Theorem 3, and the singularity is removable. It remains to consider the case n − 2j ≤ 2sm and n − j ≥ sm. Suppose j = j0 satisfies these inequalities and let j1 = n − j0 − sm. Then j1 ≥ 0 and n − 2j1 = 2sm − n + 2j0 ≥ 0, hence j = j1 appears in the sum. But 2sm − (n− 2j0) = n − 2j1 so Theorem 3 applies. We can assume j1 ≤ j0. Consider the following subset of the sum for Kn (z, w): (zzww)j0 Pn−2j0 (z, w) j0! (sκ0 + sκ1 + 1)n−j0 + (zzww)j1 Pn−2j1 (z, w) j1! (sκ0 + sκ1 + 1)n−j1 = (zzww)j0 j0! (sκ0 + sκ1 + 1)n−j0 Cn,j0 . with Cn,j0 = Pn−2j0 (z, w) + j0! (zzww)j1−j0 Pn−2j1 (z, w) j1! (sκ0 + sκ1 + 1 + n− j0)j0−j1 . The expression Cn,j0 has no pole at κ0 + κ1 = −m since 1 − sm + n − j0 ≥ 1. Indeed (1− sm + n− j0)j0−j1 = (j1 + 1)j0−j1 = j0!/j1!. In the special case n − 2j0 = ms, and j0 = j1 = (n− sm) /2 we replace Cn,j0 by Pn−2j0 (z, w). By Theorem 3 Cn,j0 = 0 when κ0 + κ1 = −m, thus Cn,j0 is divisible by (κ0 + κ1 + m) s. The sum of the two terms (j = j0 and j = j1) has a removable singularity there. � Polynomials Associated with Dihedral Groups 19 The expressions for V ( zazb ) are derived from the series (3.1) for Kn (z, w) thus the result about singularities at κ0 + κ1 = −m being removable by grouping the expansion into certain pairs applies. Note that in the above proof the paired terms are Pn−2j0 and Pn−2j1 with j0+j1 = n− sm. To analyze V ( zazb ) it suffices to identify the pairs. For the case a ≡ b mod s and a ≥ b let a = us + r, b = vs + r and 0 ≤ r < s. The paired indices in the sum from Theorem 2 consist of {(k, k′) : 0 ≤ k < k′ ≤ v, k + k′ = u + v −m} . Indeed for k, k′ with 0 ≤ k, k′ ≤ v define j by a + b − 2j = (u + v − 2k) s so that j = ks + r and similarly set j′ := k′s + r. The pairing condition j + j′ = a + b− sm is equivalent to k + k′ = u + v −m. Thus k, k′ are paired exactly when k + k′ = u + v −m and 0 ≤ k, k′ ≤ v. For the case a− b ≡ t mod s, and with a = us + r + t > b = vs + r, 0 ≤ r < s, 1 ≤ t < s the pairing in the formula from Theorem 1 combines terms from the first sum with corresponding terms in the second. For the first sum suppose 0 ≤ k ≤ v and j := ks + r so that a + b− 2j = t + (u + v − 2k) s. For the second sum let 1 − ⌊ r+t s ⌋ ≤ k′ ≤ v and let j′ := (k′ − 1) s + r + t so that a + b− 2j′ = (s− t) + (u + v − 2k′ + 1) s. The pairing condition j + j′ = a + b− sm is equivalent to k + k′ = u + v + 1 − m. To remove the singularities at κ0 + κ1 = −m combine the term in the first sum of index k with the term in the second of index k′ for all pairs (k, k′) satisfying k + k′ = u + v + 1−m, 0 ≤ k ≤ v, 1− ⌊ r+t s ⌋ ≤ k′ ≤ v. References [1] Berenstein A., Burman Y., Quasiharmonic polynomials for Coxeter groups and representations of Cherednik algebras, math.RT/0505173. [2] Dunkl C., Differential-difference operators associated to reflection groups, Trans. Amer. Math. Soc. 311 (1989), 167–183. [3] Dunkl C., Poisson and Cauchy kernels for orthogonal polynomials with dihedral symmetry, J. Math. Anal. Appl. 143 (1989), 459–470. [4] Dunkl C., Operators commuting with Coxeter group actions on polynomials, in Invariant Theory and Tableaux, Editor D. Stanton, Springer, Berlin – Heidelberg – New York, 1990, 107–117. [5] Dunkl C., Integral kernels with reflection group invariance, Can. J. Math. 43 (1991), 1213–1227. [6] Dunkl C., de Jeu M., Opdam E., Singular polynomials for finite reflection groups, Trans. Amer. Math. Soc. 346 (1994), 237–256. [7] Dunkl C., Xu Y., Orthogonal polynomials of several variables, Encycl. of Math. and its Applications, Vol. 81, Cambridge University Press, Cambridge, 2001. [8] Rösler M., Positivity of Dunkl’s intertwining operator, Duke Math. J. 98 (1999), 445–463, q-alg/9710029. [9] Scalas F., Poisson integrals associated to Dunkl operators for dihedral groups, Proc. Amer. Math. Soc., 133 (2005), 1713–1720. [10] Xu Y., Intertwining operator and h-harmonics associated with reflection groups, Can. J. Math. 50 (1998), 193–208. http://arxiv.org/abs/math.RT/0505173 http://arxiv.org/abs/q-alg/9710029 1 Introduction 2 The Poisson kernel 3 Expressions for coefficients 4 The intertwining operator 5 Singular values References