Special Functions Associated with Root Systems:

A First Introduction for Non-Specialists

Tom H. Koornwinder
University of Amsterdam, Department of Mathematics,
Plantage Muidergracht 24, 1018 TV Amsterdam, The Netherlands
e-mail: [email protected]

Abstract A survey is given about recent developments in special functions associated with root
systems. The paper addresses a general mathematical audience and it does not assume earlier
knowledge of either special functions or root systems. The emphasis is on multivariable orthogonal
polynomials: the Heckman-Opdam Jacobi polynomials and Maconald’s q-analogues of these.

1. Introduction
In this paper I present a brief survey of the active area of Special Functions associated
with Root Systems. The article is intended for a general mathematical audience. It
will not suppose prerequisites on either special functions or root systems. It will also
skip many technical details. Some early developments in this area took place during the
seventies ([21], [40]). During the last ten years important break-throughs were made by
Heckman and Opdam [14], [35], [36], [15], I. G. Macdonald [29], [30], C. F. Dunkl [10] and
I. Cherednik [4], [5], [6], [7].
A lot of the motivation for the subject of this paper comes from analysis on semisimple
Lie groups. Spherical functions on Riemannian symmetric spaces of the compact or non-
compact type can be written as special functions depending on parameters which assume
only special discrete values. In the one-variable cases these special functions were classical,
also for parameter values without group theoretic interpretation, but in the more-variable
cases they were new. In the case of group theoretic interpretation, many properties of
these special functions, as well as associated harmonic analysis, immediately follow by
group theoretic arguments. The case of more general parameter values yields the special
functions associated with root systems. Properties derived in the group case can still be
formulated in the general case, but now as conjectures rather than theorems. This paper
describes some of the progress which has been made in proving these conjectures. For
convenience, I will restrict to the polynomial (compact) case, with Bessel functions as a
sole exception. (For an introduction to the non-polynomial case see [37], [16].) Neither will
I discuss the recent work on commuting operators with elliptic functions as coefficients.

July 29, 1999; slightly updated version of my paper in “Special functions and differential equations”,
K. Srinivasa Rao, R. Jagannathan, G. Vanden Berghe & J. Van der Jeugt (eds.), Allied Publishers, Madras,
1998, pp. 10–24.
 –2–

An important aspect of the whole theory, which will not be discussed very much in this
paper, is the connection with completely integrable systems, for instance the generalized
Calogero-Moser system.
Special functions associated with root systems have also been developed in the q-case,
where q is a deformation parameter giving back the earlier cases when q = 1. Motivation
and development of the theory in the q-case has been quite different from the q = 1 case.
Except for the case of Hall polynomials [27], theory was developed [30], [29], [23] without
interpretation in group theory. But afterwards quantum groups looked very promising as
a natural setting for these polynomials. This had already turned out to be true in the
one-variable case [24], and more recently some interpretations of more-variable cases on
quantum groups were found [33], [12].
In any case, a quantum group interpretation for generic values of the parameters can-
not be expected. But, by Cherednik’s work [4], [5], [6] we know already another algebraic
setting for special functions associated with root systems: affine and graded Hecke algebras
[26]. As shown by work of Opdam [37], this new algebraic context also allows harmonic
analysis.
An earlier version of this paper appeared in [25]. It is my intention to make updates
of this paper regularly.

2. The one-variable case

In this section I will introduce three classical families of special functions, each depending
on a real parameter k ≥ 0, and such that the cases k = 0 and k = 1 are elementary. The
three families are connected with each other by limit transitions. Later, for each of the
families I will discuss generalizations which are associated with root systems.

2.1. Bessel functions. Consider Bessel functions in a non-standard notation:

∞
X (− 14 x2 )j
Jk (x) := (x ∈ R). (2.1)
j=0
(k + 12 )j j!

Here we use the notation for shifted factorial:

(a)j := a (a + 1) . . . (a + j − 1) (j = 1, 2, . . .); (a)0 := 1.

The function Jk is related to the Bessel function Jα in standard notation [11, Ch. VII] by
1
2k− 2 Γ(k + 21 )
Jk (x) = 1 Jk− 12 (x).
xk− 2
Note that
Jk (x) = Jk (−x), Jk (0) = 1. (2.2)
The cases k = 0 and k = 1 yield elementary functions:
sin x
J0 (x) = cos x, J1 (x) = . (2.3)
x
 –3–

The function x 7→ Jk (λx) (λ ∈ R) is an eigenfunction of a differential operator:

d2
 
2k d
+ Jk (λ x) = −λ2 Jk (λx).
dx2 x dx

It is the unique C ∞ solution of this differential equation under conditions (2.2).

2.2. Ultraspherical polynomials. Consider ultraspherical or Gegenbauer polynomials

[11, §10.9], i.e. polynomials Cnk of degree n on R such that
Z π
Cnk (cos x) Cm
k
(cos x) (sin x)2k dx = 0 (n, m ∈ Z+ , n 6= m). (2.4)
0

Then the Cnk are determined up to a constant factor (in general, we will not use the
standard normalization for Gegenbauer polynomials). For k = 0, 1 we have:

sin((n + 1)x)
Cn0 (cos x) = const. cos(nx), Cn1 (cos x) = const. . (2.5)
sin x

The function x 7→ Cnk (cos x) is an eigenfunction of a differential operator:

d2
 
d
2
+ 2k cot x Cnk (cos x) = −n(n + 2k) Cnk (cos x).
dx dx

For (nN ) being a sequence of positive integers such that nN /N → λ for some λ ≥ 0 as
N → ∞, we have the limit result

CnkN (cos(x/N ))
lim = Jk (λx).
n→∞ CnkN (1)

2.3. q-Ultraspherical polynomials. Let 0 < q < 1 and define for a ∈ C, n ∈ Z+ :

∞
Y (a; q)∞
(a; q)∞ := (1 − aq j ), (a; q)n := .
j=0
(q n a; q)∞

The infinite product converges because of the condition on q. When we compare with each
other the q-binomial series and the binomial series (see Gasper & Rahman, [13, §1.3])
∞ ∞
X (a; q)n n (az; q)∞ X (a)n n
z = , z = (1 − z)−a ,
n=0
(q; q)n (z; q)∞ n=0
n!

with both power series converging for |z| < 1, we see that, at least formally,

(q a z; q)∞
lim = (1 − z)−a (2.6)
q↑1 (z; q)∞
 –4–

(see Koornwinder [22, Appendix A] for more rigorous statements).

We will consider q-ultraspherical polynomials [1] in a non-standard notation. These
are polynomials Cnk,q of degree n on R such that
π  (e ; q)∞ 2
Z  2ix 
Cnk,q (cos x) Cm
k,q
(cos x)  (q k e2ix ; q)∞  dx = 0
  (n, m ∈ Z+ , n 6= m). (2.7)
0

Then the Cnk,q are determined up to a constant factor. It follows from (2.6) that the weight
function in (2.7) tends to (2 sin x)2k as q ↑ 1, i.e. to the weight function in (2.4). In fact,
we have accordingly with suitable normalization the limit relation

lim Cnk,q (cos x) = Cnk (cos x),

q↑1

which also follows from explicit expressions for the (q-)ultraspherical polynomials. If we
put
Pn (eix ) := Cnk,q (cos x)
then Pn is an eigenfunction of a q-difference operator:

1 − q k e2ix 1 1 − q k e−2ix 1 1 1

2ix
Pn (q 2 eix ) +
−2ix
Pn (q − 2 eix ) = (q − 2 n + q 2 n+k ) Pn (eix ).
1−e 1−e
Note that the Pn on the left hand side have arguments off the unit circle, while orthog-
onality is on the unit circle. The cases k = 0 and k = 1 are elementary as in (2.5) (not
depending on q):

sin((n + 1)x)
Cn0,q (cos x) = const. cos(nx), Cn1,q (cos x) = const. .
sin x
The q-ultraspherical polynomials form a subclass of the Askey-Wilson polynomials [2]: a
family of orthogonal polynomials depending, apart from q, on four non-trivial parameters.

2.4. Dunkl operators in one variable. We will now generalize the elementary for-
mulas
d iλx
eiλx = J0 (λx) + iλx J1 (λx) and e = iλ eiλx (2.8)
dx
(the first formula follows by (2.3)). Dunkl [10] generalized the operator d/dx to a mixture
of a differential and a reflection operator:

f (x) − f (−x)
(D(k) f )(x) := f 0 (x) + k . (2.9)
x
Note that this Dunkl operator sends smooth functions to smooth functions. Let us define
a generalized exponential function in terms of Bessel functions (2.1) by

iλx
Ek (λx) := Jk (λx) + Jk+1 (λx). (2.10)
2k + 1
 –5–

Then it follows immediately from well-known differential recurrence formulas for Bessel
functions that
D(k) Ek (λx) = iλ Ek (λx). (2.11)
Formulas (2.10) and (2.11) generalize the formulas in (2.8). The function x 7→ Ek (λx) is
the unique C ∞ function which equals 1 in 0 and which is an eigenfunction with eigenvalue
iλ of D(k) .
For (D(k) )2 we compute

2k 0 f (x) − f (−x)
(D(k) )2 f (x) = f 00 (x) + f (x) − k .
x x2

Thus, on even functions f the square of the Dunkl operator acts as the differential operator
(d/dx)2 + 2kx−1 d/dx. In particular, its action on

Jk (λx) = 12 (Ek (λx) + Ek (−λx))

yields
(D(k) )2 Jk (λx) = −λ2 Jk (λx).

3. Preliminaries about root systems

3.1. Definition of root system. A reference for this subsection is Humphreys [18].
Let V be a d-dimensional real vector space with inner product h . , . i. For α ∈ V \{0} let
sα denote the orthogonal reflection with respect to the hyperplane orthogonal to α (cf.
Fig. 1):
2hβ, αi
sα (β) := β − α (β ∈ V ).
hα, αi
A root system in V is a finite non-empty subset R of V \{0} which spans V and which
satisfies for all α, β ∈ R the two properties that

2hβ, αi
sα (β) ∈ R and ∈ Z.
hα, αi

The elements of R are called roots. By the rank of R we mean the dimension d of V .
Clearly, if α ∈ R then −α = sα (α) ∈ R. It can be shown that, for a given root α, we have
R ∩ Rα = {±α} or {±α, ±2α} or {±α, ± 12 α}. The root system R is called reduced if, for
all α ∈ R, we have R ∩ Rα = {±α}.
The root system R is called irreducible if R cannot be written as the orthogonal disjoint
union R = R1 ∪ R2 of two root systems R1 and R2 . Any root system can be written as a
unique orthogonal disjoint union of irreducible root systems. The reduced irreducible root
systems can be classified as four infinite families An , Bn , Cn , Dn of classical root systems
and five exceptional root systems G2 , F4 , E6 , E7 , E8 . Here the subscript denotes the rank
 –6–

of the root system. There is one infinite family of non-reduced irreducible root systems:
of type BCn .
An example for d = 1 is the set R := {±2} ⊂ R (root system of type A1 , cf. Fig. 2). An
example for d = 2 is the set R = R+ ∪ (−R+ ), where R+ := {(1, −1), (2, 0), (1, 1), (0, 2)} ⊂
R2 (root system of type C2 , cf. Fig. 3). In general, when we have a root system R in V then
we can write it as a disjoint union R = R+ ∪ (−R+ ), where R+ and −R+ are separated
from each other by a hyperplane in V through the origin. The choice for the set of positive
rootaR+ is not unique. For a given choice of R+ there is a unique subset α1 , . . . , αd of R+
(the so-called simple roota) such that each α ∈ R+ can be written as a linear combination
of α1 , . . . , αd with nonnegative integer coefficients. The simple roots form a basis of V .
Let GL(V ) be the group of invertible linear transformations of V . The Weyl group
W of the root system R is the subgroup of GL(V ) which is generated by the reflections sα
(α ∈ R). The group W is finite and it acts on R. It permutes the possible choices of R+
in a simply transitive way.
If v ∈ V \{0} then write
2v
v̌ := .
hv, vi
The set Rˇ := {α̌ | α ∈ R} is again a root system in V , the so-called dual root system to
R. The root systems R and Rˇ have the same Weyl group. Corresponding to a choice of
R+ there is a choice R+ˇ := {α̌ | α ∈ R+ } of positive roots for R+ˇ.

3.2. Dunkl operators associated with R. Let R be a reduced root system in V . Let
k: α 7→ kα : R → [0, ∞) be a function which is W -invariant, i.e., which satisfies kwα = kα for
all w ∈ W and all α ∈ R. If R is an irreducible (reduced) root system then the Weyl group
is transitive on all roots of equal length and there are at most two different root lengths.
Thus kα then assumes at most two different values. See the above examples: one root
length in A1 and two root lengths in C2 . The function k is called a multiplicity function.
The reason for this name is that root systems naturally arise in the structure theory of real
semisimple Lie algebras, where roots have an interpretation as joint eigenvalues of certain
operators and the kα then are (integer) multiplicities of such eigenvalues.
For ξ ∈ V we will denote by ∂ξ the corresponding directional derivative. The Dunkl
operators [10], [19] associated with the root system R and the multiplicity function k are
(k)
defined as the operators Dξ : C ∞ (V ) → C ∞ (V ) (ξ ∈ V ) given by

(k)
X f (x) − f (sα x)
(Dξ f )(x) := (∂ξ f )(x) + kα hα, ξi . (3.1)
hα, xi
α∈R+

This definition
Pis easily seen to be independent of the choice of R+ , since we may replace
1
P
α∈R+ by 2 α∈R . In case of root system A1 formula (3.1) reduces for ξ := 1 to formula
(2.9). Note that the operator (3.1) consists of a term involving a first order derivative and
terms involving reflection operators, just as we have seen in (2.9). It is an amazing fact,
(k)
which can be proved in a rather straightforward way, that the operators Dξ commute:
(k)
[Dξ , Dη(k) ] = 0 (ξ, η ∈ V ).
 –7–
(k)
Let D(k) be the algebra generated by the operators Dξ . This is a commutative algebra.
It can be shown that each W -invariant operator D in D(k) , when restricted in its action
to the W -invariant C ∞ functions on V , coincides with a partial differential operator (so
its reflection terms vanish when acting on a W -invariant function). The joint W -invariant
smooth eigenfunctions of the W -invariant operators in D(k) are called Bessel functions
associated with R. In the example A1 things reduce to the one-variable considerations of
§2.1 and §2.4. More generally, one may study the joint eigenfunctions of the full algebra
D(k) and one may try to do harmonic analysis for these eigenfunctions. A lot of satisfactory
results have been obtained, see [19] and the references given there.

3.3. Weight lattice asscociated with R. A reference for this subsection is again
Humphreys [18]. We still assume a root system R in V . The weight lattice P of R is
defined by
2hλ, αi
P := {λ ∈ V | ∈ Z for all α ∈ R}.
hα, αi
The subset P+ of dominant weights is then defined by

2hλ, αi
P+ := {λ ∈ P | ≥0 for all α ∈ R+ }.
hα, αi

Note that R ⊂ P , but it is in general not true that R+ ⊂ P+ . Let α1 , . . . , αd be the simple
roots in R+ . These uniquely determine λ1 , . . . , λd ∈ V such that

2hλi , αj i
= δi,j (i, j = 1, . . . , d). (3.2)
hαj , αj i

Then λ1 , . . . , λd are linear independent elements of P+ (the so-called fundamental dominant

weights), P is the Z-span of λ1 , . . . , λd , and P+ is the Z+ -span of λ1 , . . . , λd .
It is easily seen that w(P ) = P for w ∈ W , so the Weyl group acts on P . Moreover,
it can be shown that each Weyl group orbit in P has a one-point intersection with P+ :

∀λ ∈ P Card(Wλ ∩ P+ ) = 1.

Thus the dominant weights can be used as a set of representatives for the W -orbits in P .
We introduce a partial ordering ≺ on P which is induced by the root system: for
λ, µ ∈ P we say that
P
µ≺λ iff λ−µ= α∈R+ mα α for certain nonnegative integers mα . (3.3)

For root system C2 the above concepts are illustrated in Fig. 4.

3.4. Trigonometric polynomials associated with R. Let P be the weight lattice of

a root system R in V . For λ ∈ P define the function eλ on V by

eλ (x) := eihλ,xi (x ∈ V ).
 –8–

Note that eλ eµ = eλ+µ . Thus the space

A := Span{eλ | λ ∈ P }

is an algebra: the algebra of trigonometric polynomials on V (with respect to R). For a

function f on V write (wf )(x) := f (w−1 x) (w ∈ W , x ∈ V ). Then weλ = ewλ (w ∈ W ,
λ ∈ P ). Put X
mλ := eµ (λ ∈ P+ ).
µ∈W λ

Then the functions mλ are W -invariant and they form a basis of the space AW of W -
invariant elements in A.
Let the dual root lattice Qˇbe defined by

Qˇ:= {λ ∈ V | hλ, µi ∈ Z for all µ ∈ P }.

This lattice gives rise to a torus

T := V /(2πQˇ).
Let x 7→ ẋ be the natural mapping of V onto T . Then each function f in A actually lives
on T : f (x) = fe(ẋ) for a suitable function fe on T .
In the example A1 we have P = Z, P+ = {0, 1, 2, . . .}, the algebra A is spanned by the
functions x 7→ einx (n ∈ Z) and the subalgebra AW by the functions 1 and x 7→ 2 cos(nx)
(n = 1, 2, . . .). The torus T equals R/(2πZ).

4. Jacobi polynomials associated with R

4.1. Definition of Jacobi polynomials for R. Let R be a root system in V and let
k: R → [0, ∞) be a W -invariant multiplicity function as before. Define a weight function
δk on T by Y Y
δk (x) := |2 sin(hα, xi)|2kα = |2 sin( 12 hα, xi)|kα . (4.1)
α∈R+ α∈R

Hence δk is nonnegative, W -invariant, and invariant under translations by elements of Qˇ.

Define an inner product on the linear space A by
Z
hf, gik := f (x) g(x) δk (x) dẋ (f, g ∈ A). (4.2)
T

Here dẋ denotes the measure Ron T which is induced from Lebesgue measure on V and
which is normalized such that T dẋ = 1.
(k)
The Jacobi polynomial Pλ (cf. [14]) of “degree” λ ∈ P+ and of “order” k is an
element of AW of the form
(k)
X
Pλ = cλ,µ mµ
µ∈P+
µ≺λ
 –9–

such that cλ,λ = 1 and

(k)
hPλ , mµ ik = 0 if µ ∈ P+ and µ  λ. (4.3)

(k)
Instead of (4.3) we can equivalently require that Pλ satisfies the second order differ-
ential equation
 
(k) (k)
X
kα cot( 21 hα, xi) ∂α
P
∆+ Pλ (x) = −hλ, λ + α∈R+ kα αi Pλ (x). (4.4)
α∈R+

In the example A1 we get k ∈ [0, ∞), δk (x) = |2 sin x|2k (x ∈ R), and
n
X
Pn(k) (x) = cn,m eimx (n = 0, 1, 2, . . .) (4.5)
m=−n
n−m even

with cn,m = cn,−m and cn,n = 1. Equation (4.3) becomes

Z 2π
1
Pn(k) (x) cos((n − 2m)x) | sin x|2k dx = 0 (m = 1, 2, . . . , [ 12 n])
2π 0

and equation (4.4) takes the form

d2
 
d
2
+ 2k cot x Pn(k) (x) = −n(n + 2k) Pn(k) (x).
dx dx

(k)
We see that Pn (x) = const. Cnk (cos x), where Cnk is the ultraspherical polynomial of §2.2.
The case of the (non-reduced) root system BC1 would have given us, more generally, the
classical one-variable Jacobi polynomials.

4.2. Three problems and their solutions. As soon as the above definition of Jacobi
polynomials associated with R is given, three highly nontrivial questions can naturally be
posed:
1. It follows immediately from the definition that the orthogonality

(k)
hPλ , Pµ(k) ik = 0 (4.6)

holds if µ  λ or λ  µ. What about (4.6) if λ and µ are not related in the partial
ordering?
2. Prove the existence of a commutative algebra of differential operators with d alge-
braically independent generators, such that the operators in this algebra have the
(k)
Pλ (λ ∈ P+ ) as joint eigenfunctions. (Note that the operator in (4.4) can be taken
as one of the generators.)
 –10–
(k) (k)
3. Give an explicit expression for hPλ , Pλ ik , or rather for its two factors
(k) (k)
hPλ , Pλ ik
Z
(k) (k)
and δk (x) dẋ. (4.7)
hP0 , P0 ik T

In the past few years all these questions have been answered in the positive sense. Let me
give some indications.
• If problem 2 can be solved then the answer to 1 follows readily, cf. [14]. Indeed, we
(k)
need sufficiently many differential operators having the Pλ as eigenfunctions such
that the joint eigenvalues, in their dependence on λ, separate the points of P+ .
(k) (k)
• For certain special choices of k the functions Pλ , renormalized such that Pλ (0) = 1,
have an interpretation as spherical functions on compact symmetric spaces G/K, cf.
1
m−1
[17]. (For instance, in case A1 the ultraspherical polynomial Cn2 can be inter-
preted as spherical function on the (m − 1)-dimensional sphere SO(m)/SO(m − 1).)
Then problems 1, 2 and the first half of problem 3 can be solved by using the group
theoretic interpretation. The orthogonality (4.6) for general λ, µ follows by Schur’s
orthogonality relations for matrix elements of irreducible unitary representations of
G. The first expression in (4.7) was explicitly computed by Vretare [42] in terms of
Harish-Chandra’s c-function related to the spherical functions on the corresponding
non-compact symmetric space. The algebra of differential operators in problem 2 can
be obtained by taking the radial parts of the G-invariant differential operators on
G/K.
• For the classical root systems question 2 could be answered in a positive way by
giving explicit expressions for generators of the algebra, see [21] for BC2 and A2 , and
Olshanetsky & Perelomov [34], Sekiguchi [39] and Debiard [8] for the higher rank
cases.
• Heckman and Opdam [14] have given positive answers to 2, and hence to 1, by use of
deep transcendental arguments. This also solved part of Problem 3 (the first expres-
sion in (4.7)). In 1982 Macdonald [28] had already given conjectures for the explicit
evaluation of the second expression in (4.7), which could be proved in a number of
special cases.
• Problem 3 for general λ was solved by Opdam [36] by using so-called shift operators
[35]. The most simple example, for case A1 , of such operators is the following pair of
differential recurrence relations for Gegenbauer polynomials:
d k k+1
C (x) = const. Cn−1 (x),
dx n
 
2 −k+ 21 d 2 k+ 12 k+1
(1 − x ) ◦ (1 − x ) Cn−1 (x) = const. Cnk (x).
dx
R1 1
By use of these two formulas we can write −1 (Cnk (x))2 (1 − x2 )k− 2 dx as an explicit
R1 k+1 1
constant times −1 (Cn−1 (x))2 (1−x2 )k+ 2 dx. Opdam’s shift operators in general have
a similar structure of lowering λ and raising k, or conversely. The case of root system
BC2 was already considered in [21], [40].
 –11–

4.3. Dunkl type operators. Some years after Heckman first solved the problems 1
and 2 of the previous subsection he discovered a dramatical simplification [15] for proving
these results. For a given root system R in V and a given multiplicity function k he wrote
down a trigonometric variant of the Dunkl operators (3.1) for ξ ∈ V :

(k)
X
1
(Dξ f )(x) := (∂ξ f )(x) + 2 kα hα, ξi cot( 21 hα, xi) (f (x) − f (sα x))
α∈R+

(x ∈ V, f ∈ C ∞ (V )). (4.8)

(k)
Now the operators Dξ will no longer commute, in general. However, Heckman showed
P (k) j
that the operators η∈W ξ (Dη ) (ξ ∈ V , j = 0, 1, 2, . . .), when restricted to the W -
invariant C ∞ functions on V , coincide with differental operators which commute with
each other and form a commutative algebra. This is the algebra looked for in Problem 2
(k)
of the previous subsection. The Jacobi polynomials Pλ are the joint eigenfunctions of
the operators in this algebra. By this approach, Heckman also obtained a quick existence
proof for Opdam’s shift operators.
Next Cherednik [4] (see also Opdam [37, Definition 2.2]) made a slight but significant
variation in Heckman’s Dunkl type operators (4.8). He put

(k)
X 1
(D
e f )(x) := (∂ξ f )(x) +
ξ kα hα, ξi (f (x) − f (sα x))
1 − e−α (x)
α∈R+
X
− 21 kα hα, ξi f (x).
α∈R+

Cherednik’s operators have the nice property that they mutually commute, without the
need of first restricting to W -invariant functions. On the other hand, they do not share
(k) (k)
the property w Dξ w−1 = Dwξ of Heckman’s operators. This has to be replaced by a
relation coming from graded Hecke algebras. Anyhow, by means of Cherednik’s operators
one can draw the same conclusions as by Heckman’s operators, and in a similar way.

5. Macdonald polynomials associated with R

5.1. Definition of Macdonald polynomials. The reference for this subsection is

Macdonald [29], see also the summary in [31]. Let 0 < q < 1. Assume that R is an
irreducible and reduced root system. We keep the assumptions of §4.1 except that we
+ +
replace the weight function δk in (4.1) by δk,q (x) := δk,q (x) δk,q (−x), where

+
Y (eihα,xi ; q)∞
δk,q (x) := .
α∈R+
(q kα eihα,xi ; q)∞
 –12–

Then 2
(eihα,xi ; q)∞  (eihα,xi ; q)∞
Y Y 

δk,q (x) = =   .
α∈R
(q kα eihα,xi ; q)∞ α∈R
 (q kα eihα,xi ; q)
∞

+

Hence δk,q is nonnegative, W -invariant, and invariant under translations by elements of

(k,q)
Qˇ. Then the Macdonald polynomials Pλ are defined just as the Jacobi polynomials
(k)
Pλ , but with the inner product in (4.2) replaced by
Z
hf, gik,q := f (x) g(x) δk,q (x) dẋ (f, g ∈ A). (5.1)
T

(k,q)
Thus the Macdonald polynomial Pλ (λ ∈ P+ ) is an element of AW of the form

(k,q)
X
Pλ = cλ,µ mµ
µ∈P+
µ≺λ

such that cλ,λ = 1 and

(k,q)
hPλ , mµ ik,q = 0 if µ ∈ P+ and µ  λ.

Note that limq↑1 δk,q (x) = δk (x) by (2.6). Accordingly we have limq↑1 Pλk,q (x) = Pλk (x),
see [29, §11].
Just as in the case q = 1 we immediately see from the definition of Macdonald poly-
nomials that the orthogonality
(k)
hPλ , Pµ(k) ik,q = 0 (5.2)
holds if µ  λ or λ  µ.
In the case of root system A1 we get, for k ∈ [0, ∞) and n = 0, 1, 2, . . . that Pnk,q is
of the form as given in the right-hand side of (4.5) and satisfying

2π  (e ; q)∞ 2
Z  2ix 
1
Pn(k) (x) cos((n − 2m)x)  k 2ix  dx = 0 (m = 1, 2, . . . , [ 21 n]).
2π 0 (q e ; q)∞ 

Hence Pnk,q (x) = const. Cnk,q (cos x), where Cnk,q is the q-ultraspherical polynomial of §2.3.

5.2. Minuscule and quasi-minuscule elements. Relative to the partial ordering on

R (see (3.3)) there is a unique maximal element γ, which is of course a positive root (the
so-called maximal root). Its decomposition in simple roots

d
X
γ= ki αi (5.3)
i=1

has the property that ki > 0 for all i (see [18, §10.4, Lemma A]).
 –13–

Let the dual weight lattice Pˇ be the weight lattice corresponding to the dual root sys-
tem Rˇ. It has fundamental dominant weights 2λi /hαi , αi i (i = 1, . . . , d), where λ1 , . . . , λd
are the fundamental dominant weights for P .
An element σ ∈ V \{0} is called minuscule (more precisely a minuscule fundamental
weight for the root system Rˇ) if hσ, αi = 0 or 1 for all α ∈ R+ . It can be shown (see [3,
Chap. 6, Exercises §1, no. 24])) that the minuscule elements are the fundamental dominant
weights 2λi /hαi , αi i for Pˇ precisely for those i ∈ {1, . . . , d} for which ki = 1 in (5.3). The
coefficients ki of (5.3) are explicitly given in [18, §12.2, Table 2]. From this Table it can
be concluded that the number of minuscule elements is n for root system An , 2 for root
systems Dn , E6 , 1 for root systems Bn , Cn , E7 , and 0 for root systems E8 , F4 , G2 .
In these last three cases there still exists a quasi-minuscule element, i.e.. σ ∈ V such
that hσ, αi = 0, 1 or 2 for all α ∈ R+ , with the values 1 and 2 both being attained. A
very special choice of such a quasi-minuscule element can then be made such that the
statements below will be true.

5.3. Macdonald’s q-difference operators. Macdonald gives some explicit q-difference

(k,q)
operators of which the Pλ are eigenfunctions. In order to state his results we need a
few preparations.
Let v ∈ V . For f a holomorphic function on a suitable neighbourhood of V in V + iV
define a function Tv f by

(Tv f )(x) := f (x − i(log q)v) (x ∈ V ).

Hence
Tv eλ = q hv,λi eλ (λ ∈ P ).
Let σ be a (quasi-)minuscule element as above and let Wσ be the stabilizer of σ in
W . Put !
+
X T σ δ k,q
Eσ(k,q) f := |Wσ |−1 w + Tσ f .
w∈W
δ k,q

Then it can be shown in the minuscule case that Es i maps AW into itself and that the
operator Eσ is self-adjoint with respect to the inner product (5.1). In order to handle the
quasi-minuscule case as well put
+
!
X T σ δ k,q
Dσ(k,q) f := |Wσ |−1 w + (Tσ f − f ) .
w∈W
δk,q

If σ is minuscule this is a trivial change, since then

Dσ(k,q) = Eσ(k,q) − Eσ(k,q) (1).

This will not be true in the quasi-minuscule case, but then it can still be proved that Ds i
maps AW into itself and that the operator Dσ is self-adjoint with respect to the inner
product (5.1).
 –14–

Although these operators, except for root system An (where they were independently
found by Ruijsenaars [38]) do not yet give a full commutative algebra of operators having
(k,q)
the Pλ as joint eigenfunctions, the additional parameter q gives enough freedom such
that already the eigenvalue of one such operator separates the elements of P+ for generic
q, by which a positive answer to question 1 in §4.2 can be given for the case of Macdonald
polynomials. Taking limits for q ↑ 1 then yields the same positive answer for the case of Ja-
cobi polynomials. This is an alternative to Heckman’s approach via Problem 2. Macdonald
(k,q) (k,q)
also gives conjectured explicit expressions for the squared norms hPλ , Pλ ik,q .

5.4. Askey-Wilson polynomials for root system BCn . The author [23] introduced
for the non-reduced root system BCn a class of polynomials having two more parameters
than Macdonald’s class for BCn . This extended class reduces for n = 1 to the Askey-Wilson
polynomials [2]. In [23] only one explicit q-difference operator was given having the BCn -
polynomials as eigenfunctions, but this was sufficient for establishing orthogonality. Later,
van Diejen [9] gave explicit expressions for the generators af a full commutative algebra of
operators having the BCn polynomials as joint eigenfunctions.

5.5. Cherednik’s approach to Macdonald polynomials. Cherednik [5], [6], [7]

succeeded to give positive answers to questions 2 and 3 in §4.2. In the context of certain
representations of affine Hecke algebras he could realize a commutative algebra of operators
which have the Macdonald polynomials as joint eigenfunctions. In the same context he
could realize q-analogues of Opdam’s shift operators and next, by the same technique as
in Opdam, prove Macdonald’s conjectures in the q-case.
It is beyond the scope of this short survey to explain Cherednik’s approach in any
detail. See the expository papers by I. G. Macdonald [32] and by Kirillov [20]. Let me here
only give a few indications. Just as a Hecke algebra is a deformation of the group algebra of
a Weyl group, an affine Hecke algebra (cf. [26]) deforms the group algebra of an affine Weyl
group. If R is an irreducible root system in V with Weyl group W then the (extended)
affine Weyl group is the semidirect product W f := W n P ∨ , where the dual weight lattice
P ∨ is defined as P ∨ := {λ ∈ V | hλ, αi ∈ Z}, an abelian group under addition. Then W f
∨
acts as a group of motions on V , with P acting as a group of translations. The group
f also acts on A, with W acting as before and with the action of P ∨ still depending on
W
a parameter q.
The affine Hecke algebra H can be defined in terms of generators and relations which
still depend on the values of a W -invariant function α 7→ tα on R. Corresponding to a
choice of R+ we can define P+∨ . The embedding of P+∨ in H then generates a commutative
subalgebra Y of H.
f on A in order to define an action of H on A,
For given q we can use the action of W
by specifying the action for a set of generators of H (Demazure operators). This action
depends on q and the tα . Put tα = q −kα /2 , where k: R → [0, ∞) is a multiplicity function.
(k,q)
Then Cherednik proves that the Macdonald polynomials Pλ are the joint eigenfunctions
of the W -invariant elements in the commutative algebra Y. This answers question 2 in
§4.2.
 –15–

A next problem concerns the extension of Cherednik’s approach to the BCn polyno-
mials. As van Diejen [9] already answered question 2 in §4.2 in a positive way for this
case, it would be nice to complement van Diejen’s constructive approach with the deep
conceptual approach via affine Hecke algebras. In fact, Macdonald [32], at the end of his
Bourbaki lecture, claims that he is able to do this.

References
[1] R. Askey & M. E. H. Ismail, A generalization of ultraspherical polynomials, in Studies
in Pure Mathematics, P. Erdös (ed.), Birkhäuser, 1983, pp. 55-78.
[2] R. Askey & J. Wilson, Some basic hypergeometric orthogonal polynomials that gen-
eralize Jacobi polynomials, Memoirs Amer. Math. Soc. (1985), no. 319.
[3] N. Bourbaki, Groupes et algèbres de Lie, Chapitres 4, 5 et 6, Hermann, 1968.
[4] I. Cherednik, A unification of Knizhnik-Zamolodchikov and Dunkl operators via affine
Hecke algebras, Invent. Math. 106 (1991), 411–431.
[5] I. Cherednik, Double affine Hecke algebras, Knizhnik-Zamolodchikov equations, and
Macdonald’s operators, Duke Math. J., Internat. Math. Res. Notices (1992), no. 9,
171–180.
[6] I. Cherednik, The Macdonald constant-term conjecture, Duke Math. J., Internat.
Math. Res. Notices (1993), no. 6, 165–177.
[7] I. Cherednik, Double affine Hecke algebras and Macdonald’s conjectures, Ann. of
Math. (2) 141 (1995), 191–216.
[8] A. Debiard, Système différentiel hypergéométrique et parties radiales des opérateurs
invariants des espaces symétriques de type BCp , in Séminaire d’Algèbre Paul Dubreil
et Marie-Paule Malliavin, M.-P. Malliavin (ed.), Lecture Notes in Math. 1296, Springer,
1988, pp. 42–124.
[9] J. F. van Diejen, Commuting difference operators with polynomial eigenfunctions,
Compositio Math. 95 (1995), 183–233.
[10] C. F. Dunkl, Differential-difference operators associated to reflection groups, Trans.
Amer. Math. Soc. 311 (1989), 167–183.
[11] A. Erdélyi, W. Magnus, F. Oberhettinger & F. G. Tricomi, Higher transcendental
functions, Vol. II, McGraw-Hill, 1953.
[12] P. I. Etingof & A. A. Kirillov. Jr., Macdonald’s polynomials and representations of
quantum groups, Math. Res. Lett. 1 (1994), no. 3, 279–296.
[13] G. Gasper & M. Rahman, Basic hypergeometric series, Cambridge University Press,
1990.
[14] G. J. Heckman (part I with E. M. Opdam), Root systems and hypergeometric func-
tions I, II, Compositio Math. 64 (1987), 329–352, 353–373.
[15] G. J. Heckman, An elementary approach to the hypergeometric shift operators of
Opdam, Invent. Math. 103 (1991), 341–350.
 –16–

[16] G. J. Heckman (part I, II) and E. M. Opdam (part III), Hypergeometric functions
and representation theory, in “Seminar Hypergeometric Functions”, Communications
of the Mathematical Institute 18, Utrecht University, 1994, pp. 72–93.
[17] S. Helgason, Groups and geometric analysis, Academic Press, 1984.
[18] J. E. Humphreys, Introduction to Lie algebras and representation theory, Springer,
1972.
[19] M. F. E. de Jeu, Dunkl operators, Dissertation, University of Leiden, 1994.
[20] A. A. Kirillov, Jr., Lectures on the affine Hecke algebras and Macdonald conjectures,
Bull. Amer. Math. Soc. (N. S.) 34 (1997), 251–292.
[21] T. H. Koornwinder, Orthogonal polynomials in two variables which are eigenfunctions
of two algebraically independent partial differential operators I, II, III, IV, Nederl.
Akad. Wetensch. Proc. Ser. A 77 (1974), 48–58, 59–66, 357–369, 370–381.
[22] T. H. Koornwinder, Jacobi functions as limit cases of q-ultraspherical polynomials, J.
Math. Anal. Appl. 148 (1990), 44–54.
[23] T. H. Koornwinder, Askey-Wilson polynomials for root systems of type BC, in Hy-
pergeometric functions on domains of positivity, Jack polynomials, and applications,
D. St. P. Richards (ed.), Contemp. Math. 138, Amer. Math. Soc., 1992, pp. 189–204.
[24] T. H. Koornwinder, Askey-Wilson polynomials as zonal spherical functions on the
SU (2) quantum group, SIAM J. Math. Anal. 24 (1993), 795–813.
[25] T. H. Koornwinder, Special functions associated with root systems: recent progress,
in From Universal Morphisms to Megabytes — a Baayen Space Odyssey, K. R. Apt,
A. Schrijver & N. M. Temme (eds.), CWI, Amsterdam, 1994, pp. 391–404.
[26] G. Lusztig, Affine Hecke algebras and their graded version, J. Amer. Math. Soc. 2
(1989), 599–635.
[27] I. G. Macdonald, Symmetric functions and Hall polynomials, Oxford University Press,
1979.
[28] I. G. Macdonald, Some conjectures for root systems, SIAM J. Math. Anal. 13 (1982),
988–1007.
[29] I. G. Macdonald, Orthogonal polynomials associated with root systems, preprint, 1987.
[30] I. G. Macdonald, A new class of symmetric functions, in Séminaire Lotharingien de
Combinatoire, 20e Session, L. Cerlienco & D. Foata (eds.), Publication de l’Institut
de Recherche Mathématique Avancée 372/S-20, Strasbourg, 1988, pp. 131–171.
[31] I. G. Macdonald, Orthogonal polynomials associated with root systems, in Orthogonal
polynomials: theory and practice, P. Nevai (ed.), NATO ASI Series C 294, Kluwer,
1990, pp. 311–318.
[32] I. G. Macdonald, Affine Hecke algebras and orthogonal polynomials, Séminaire Bour-
baki 1994–95, no. 797.
[33] M. Noumi, Macdonald’s symmetric polynomials as zonal spherical functions on some
quantum homogeneous spaces, Adv. in Math. 123 (1996), 16–77.
 –17–

[34] M. A. Olshanetsky & A. M. Perelomov, Quantum integrable systems related to Lie

algebras, Phys. Rep. 94 (1983), 313–404.
[35] E. M. Opdam, Root systems and hypergeometric functions III, IV, Compositio Math.
67 (1988), 21–49, 191–209.
[36] E. M. Opdam, Some applications of hypergeometric shift operators, Invent. Math. 98
(1989), 1–18.
[37] E. M. Opdam, Harmonic analysis for certain representations of graded Hecke algebras,
Acta Math. 175 (1995), 75-121.
[38] S. N. M. Ruijsenaars, Complete integrability of relativistic Calogero-Moser systems
and elliptic function identities, Comm. Math. Phys. 110 (1987), 191–213.
[39] J. Sekiguchi, Zonal spherical functions on some symmetric spaces, Publ. Res. Inst.
Math. Sci. 12 Suppl. (1977), 455–459.
[40] I. G. Sprinkhuizen, Orthogonal polynomials in two variables. A further analysis of the
polynomials orthogonal on a region bounded by two lines and a parabola, SIAM J.
Math. Anal. 7 (1976), 501–518.
[41] J. V. Stokman & T. H. Koornwinder, Limit transitions for BC type multivariable
orthogonal polynomials, Canad. J. Math. 49 (1997), 373–404.
[42] L. Vretare Elementary spherical functions on symmetric spaces, Math. Scand. 39
(1976), 343–358.