n
=
_
R
t
n
d is the nth moment of d. The standard GramSchmidt process
applied to the sequence
n
with respect to the inner product
f, g) =
_
R
f(t)g(t)d(t)
of L
2
(d) gives a sequence of orthogonal polynomials p
n
n=0
, which satises
p
n
, p
m
) = 0, if n ,= m, and p
n
is a polynomial of degree exactly n. The orthogonal
polynomials are unique up to a constant multiple. They are called orthonormal
if, in addition, p
n
, p
n
) = 1, and we assume that the measure is normalized by
_
R
d = 1 when dealing with orthonormality. If d = w(t)dt, we say that p
n
are associated with the weight function w. The orthogonal polynomials enjoy
many properties, which make them a useful tool in various applications and a
rich source of research problems. A starting point of orthogonal polynomials of
several variables is to extend those properties from one to several variables.
Orthogonal polynomials of several variables 137
To deal with polynomials in several variables we use the standard multiindex
notation. A multiindex is denoted by = (
1
, . . . ,
d
) N
d
0
. For N
d
0
and
x R
d
a monomial in variables x
1
, . . . , x
d
of index is dened by
x
= x
1
1
. . . x
d
d
.
The number [[ =
1
+ +
d
is called the total degree of x
. We denote
by T
d
n
:= spanx
: [[ = n, N
d
0
the space of homogeneous polynomials of
degree n, by
d
n
:= spanx
: [[ n, N
d
0
the space of polynomials of (total)
degree at most n, and we write
d
= R[x
1
, . . . , x
d
] for the space of all polynomials
of d variables. It is well known that
r
d
n
:= dimP
d
n
=
_
n +d 1
n
_
and dim
d
n
=
_
n +d
n
_
.
Let be a positive Borel measure on R
d
with nite moments. For N
d
0
, denote
by
=
_
R
d
x
=
_
R
d
f(x)g(x)d(x)
of L
2
(d) to produce a sequence of orthogonal polynomials of several variables.
One problem, however, appears immediately: orthogonal polynomials of several
variables are not unique. In order to apply the GramSchmidt process, we need
to give a linear order to the moments
1
d
n
are called orthonormal, if P
, P
) =
,
. The space 1
d
n
can have many
dierent bases and the bases do not have to be orthonormal. This nonuniqueness
is at the root of the diculties that we encounter in several variables.
Since the orthogonality is dened with respect to polynomials of dierent degrees,
certain results can be stated in terms of 1
d
0
, 1
d
1
, . . . , 1
d
n
, . . . rather than in terms of
a particular basis in each 1
d
n
. For such results, a degree of uniqueness is restored.
138 Yuan Xu
For example, this allows us to derive a proper analogy of the threeterm relation
for orthogonal polynomials in several variables and proves a Favards theorem.
We adopt this point of view and discuss results of this nature in Section 2.
1.2. Example: orthogonal polynomials on the unit disc. Before we go on
with the general theory, let us consider an example of orthogonal polynomials
with respect to the weight function
W
(x, y) =
2 + 1
2
(1 x
2
y
2
)
1/2
, > 1/2, (x, y) B
2
,
on the unit disc B
2
= (x, y) : x
2
+ y
2
1. The weight function is normalized
so that its integral over B
2
is 1. Among all possible choices of orthogonal bases
for 1
d
n
, we are interested in those for which fairly explicit formulae exist. Several
families of such bases are given below.
For polynomials of two variables, the monomials of degree n can be ordered by
x
n
, x
n1
y, . . . , xy
n1
, y
n
. Instead of using the notation P
, [[ = [
1
+
2
[ = n,
to denote a basis for 1
2
n
, we sometimes use the notation P
n
k
with k = 0, 1, . . . , n.
The orthonormal bases given below are in terms of the classical Jacobi and Gegen
bauer polynomials. The Jacobi polynomials are denoted by P
(a,b)
n
, which are or
thogonal polynomials with respect to (1 x)
a
(1 +x)
b
on [1, 1] and normalized
by P
(a,b)
n
(1) =
_
n+a
n
_
, and the Gegenbauer polynomials are denoted by C
n
, which
are orthogonal with respect to (1 x
2
)
1/2
on [1, 1], and
C
n
(x) = ((2)
n
/( + 1/2)
n
)P
(1/2,1/2)
n
(x),
where (c)
n
= c(c + 1) . . . (c +n 1) is the Pochhammer symbol.
1.2.1. First orthonormal basis. Consider the family
P
n
k
(x, y) = h
k,n
C
k++
1
2
nk
(x)(1 x
2
)
k
2
C
k
_
y
1 x
2
_
, 0 k n,
where h
k,n
are the normalization constants.
Since C
k
(x) is an even function if k is even and is an odd function if k is odd,
P
n
k
are indeed polynomials in
2
n
. The orthogonality of these polynomials can be
veried using the formula
_
B
2
f(x, y)dxdy =
_
1
1
_
1x
2
1x
2
f(x, y)dxdy
=
_
1
1
_
1
1
f(x,
1 x
2
t)
1 x
2
dxdt.
Orthogonal polynomials of several variables 139
1.2.2. Second orthonormal basis. Using polar coordinates x = r cos , y = r sin ,
we dene
h
n
j,1
P
(
1
2
,n2j+
d2
2
)
j
(2r
2
1)r
n2j
cos(n 2j), 0 2j n,
h
n
j,2
P
(
1
2
,n2j+
d2
2
)
j
(2r
2
1)r
n2j
sin(n 2j), 0 2j n 1,
where h
n
j,i
are the normalization constants.
For each n these give exactly n+1 polynomials. That they are indeed polynomials
in (x, y) of degree n can be veried using the relations r = x,
cos m = T
m
(x/x), and sin m/ sin = U
m1
(x/x),
where T
m
and U
m
are the Chebyshev polynomials of the rst and the second
kind. The orthogonality of these polynomials can be veried using the formula
_
B
2
f(x, y)dxdy =
_
1
0
_
2
0
f(r cos , r sin )d r dr.
1.2.3. An orthogonal basis. A third set is given by
P
n
k
(x, y) = C
+1/2
n
_
x cos
k
n + 1
+y sin
k
n + 1
_
, 0 k n.
In particular, if = 1/2, then the polynomials
P
n
k
(x, y) =
1
U
n
_
x cos
k
n + 1
+y sin
k
n + 1
_
, 0 k n,
form an orthonormal basis with respect to the Lebesgue measure on B
2
. The
case = 1/2 rst appeared in [22] in connection with a problem in computer
tomography.
1.2.4. Appells monomial and biorthogonal bases. The polynomials in these bases
are denoted by V
n
k
and U
n
k
for 0 k n (cf. [2]). The polynomials V
n
k
are dened
by the generating function
(1 2(b
1
x +b
2
y) +b
2
)
1/2
=
n=0
n
k=0
b
k
1
b
nk
2
V
n
k
(x, y), b = (b
1
, b
2
),
and they are called the monomial orthogonal polynomials since
V
n
k
(x, y) = x
k
y
nk
+q(x, y),
where q
2
n1
. The polynomials U
n
k
are dened by
U
n
k
(x, y) = (1 x
2
y
2
)
+
1
2
k
x
k
nk
y
nk
(1 x
2
y
2
)
n+1/2
.
140 Yuan Xu
Both V
n
k
and U
n
k
belong to 1
2
n
, and they are biorthogonal in the sense that
_
B
2
V
n
k
(x, y)U
n
j
(x, y)W
(x, y) = 0, k ,= j.
The orthogonality follows from a straightforward computation of integration by
parts.
1.3. Orthogonal polynomials for classical type weight functions. In the
ideal situation, one would like to have fairly explicit formulae for orthogonal poly
nomials and their various structural constants (such as L
2
norm). The classical
orthogonal polynomials of one variable are good examples. These polynomials
include the Hermite polynomials H
n
(t) associated with the weight function e
t
2
on R, the Laguerre polynomials L
a
n
(t) associated with t
a
e
t
on R
+
= [0, ), and
the Jacobi polynomials P
(a,b)
n
(t) associated with (1 t)
a
(1 + t)
b
on [1, 1]. Up
to an ane linear transformation, they are the only families of orthogonal poly
nomials (with respect to a positive measure) that are eigenfunctions of a second
order dierential operator.
One obvious extension to several variables is using tensor product. For 1 j d
let w
j
be the weight function on the interval I
j
R and denote by p
n,j
orthogonal
polynomials of degree n with respect to w
j
. Then for the product weight function
W(x) = w
1
(x
1
) . . . w
d
(x
d
), x I
1
. . . I
d
,
the product polynomials P
(x) =
d
j=1
p
j
,j
(x
j
) are orthogonal with respect to
W. Hence, as extensions of the classical orthogonal polynomials, we can have
product Hermite polynomials associated with
W
H
(x) = e
x
2
, x R
d
,
product Laguerre polynomials associated with
W
L
(x) = x
e
x
, x R
d
+
,
i
> 1,
the product Jacobi polynomials associated with
W
a,b
(x) =
d
i=1
(1 x
i
)
a
i
(1 +x
i
)
b
i
, x [1, 1]
d
, a
i
, b
i
> 1,
as well as the mixed product of these polynomials. Throughout this lecture, the
notation x stands for the Euclidean norm and [x[ stands for the
1
norm of
x R
d
. The product basis in this case is also the monomial basis. There are
other interesting orthogonal bases; for example, for the product Hermite weight
function, another orthonormal basis can be given in polar coordinates.
More interesting, however, are extensions that are not of product type. The
geometry of R
d
is rich, i.e., there are other attractive regular domains; the unit
ball B
d
= x : x 1 and the simplex T
d
= x R
d
: x
1
0, . . . , x
d
Orthogonal polynomials of several variables 141
0, 1 [x[ 0 are two examples. There are orthogonal polynomials on B
d
and
on T
d
for which explicit formulae exist, and their study goes back at least as far
as Hermite (cf. [2] and [11, Vol. 2, Chapt. 12]). The weight functions are
W
B
(x) = (1 x
2
)
1/2
, x B
d
, > 1/2,
and
W
T
(x) =
d
i=1
[x
i
[
i
1/2
(1 [x[)
d+1
1/2
, x T
d
,
i
> 1/2.
In both cases, there are explicit orthonormal bases that can be given in terms
of Jacobi polynomials. In Section 4 and 5 we discuss these two cases and their
extensions.
There is no general agreement on what should be called classical orthogonal poly
nomials of several variables. For d = 2 Krall and Sheer [18] gave a classica
tion of orthogonal polynomials that are eigenfunctions of a second order partial
dierential operator, which shows that only ve such families are orthogonal
with respect to a positive measure: product Hermite, product Laguerre, product
HermiteLaguerre, orthogonal polynomials with respect to W
B
on the triangle T
2
. Clearly these families should be called
classical, but perhaps equally entitled are product Jacobi polynomials and a score
of others.
1.4. Harmonic and hharmonic polynomials. Another classical example of
orthogonal polynomials of several variables is the spherical harmonics. The
Laplace operator on R
d
is dened by
=
2
1
+ +
2
d
,
where
i
= /x
i
. Harmonic polynomials are polynomials that satisfy P = 0,
and spherical harmonics are the restriction of homogeneous harmonic polynomials
on the sphere S
d1
. Let H
d
n
be the set of homogeneous harmonic polynomials of
degree n; H
d
n
= T
d
n
ker . It is known that
P H
d
n
if and only if
_
S
d1
PQd = 0, Q
d
, deg Q < n,
where d is the surface measure of S
d1
. An orthonormal basis for spherical
harmonics can also be given in terms of Jacobi polynomials. The fact that the
Lebesgue measure d is invariant under the orthogonal group O(d) plays an
important role in the theory of spherical harmonics.
An important extension of harmonic polynomials are the hharmonics introduced
by Dunkl [7], in which the role of the rotation group is replaced by a reection
group. The hharmonics are homogeneous polynomials that satisfy
h
P = 0,
142 Yuan Xu
where
h
is a second order dierentialdierence operator. This hLaplacian can
be decomposed as
h
= T
2
1
+. . . +T
2
d
,
where T
i
are the rst order dierentialdierence operators (Dunkls operators)
which commute, that is, T
i
T
j
= T
j
T
i
. The hharmonics are orthogonal with
respect to h
2
d where h is a weight function invariant under the underlying
reection group. Examples include h(x) =
d
i=1
[x
i
[
i
invariant under Z
d
2
and
h(x) =
i<j
[x
i
x
j
[
d
i=1
[x
i
[
i
. We
discuss hharmonics in Section 3, giving special emphasis to the case of prod
uct weight function since it can be developed without prerequisites of reection
groups.
1.5. Fourier orthogonal expansion. Let be a positive measure with nite
moments such that the space of polynomials is dense in L
2
(d). Let P
be a
sequence of orthonormal polynomials with respect to d. Then the standard
Hilbert space theory shows that every f L
2
(d) can be expanded in terms of
P
as
f(x) =
N
d
0
a
(f)P
(x) with a
(f) =
_
R
d
f(x)P
(x)d(x). (1.2)
This is the Fourier orthogonal expansion. Just as the case of the classical Fourier
series, the expansion does not hold pointwisely in general if f is merely a contin
uous function. We dene the nth partial sum of the expansion by
S
n
(f; x) :=
n
k=0
=k
a
(f)P
(x) =
_
R
d
f(y)K
n
(x, y)d(y),
where the rst equation is the denition and the second equation follows from
the formula for a
(f), where
K
n
(x, y) =
n
k=0
P
k
(x, y) with P
k
(x, y) =
=k
P
(x)P
(y);
the function K
n
(x, y) is the reproducing kernel of the space
d
n
in the sense that
_
K
n
(x, y)P(y)d(y) = P(x) for all P
d
n
, and the function P
n
(x, y) is the
reproducing kernel of the space 1
d
n
. In particular, the denition of the kernels
and thus S
n
(f) are independent of the choices of particular orthonormal bases.
As an application, we discuss the convergence of Fourier orthogonal expansions
for the classical type weight functions in Section 7.
Orthogonal polynomials of several variables 143
1.6. Literature. The main early references of orthogonal polynomials of several
variables are Appell and de Feriet [2], and Chapter 12 of Erdelyi et. al. [11],
as well as the inuential survey of Koornwinder on orthogonal polynomials of
two variables [15]. There is also a more recent book of Suetin [30] on orthogonal
polynomials of two variables, which is in the spirit of the above references. We
follow the presentation in the recent book of Dunkl and Xu [10]. However, our
main development for orthogonal polynomials with respect to the classical type
weight functions follows a line that does not require background in reection
groups, and we also include some more recent results. We will not give references
to every single result in the main body of the text; the main references and the
historical notes will appear at the end of the lecture notes.
2. General properties
By general properties we mean those properties that hold for orthogonal polyno
mials associated with weight functions that satisfy some mild conditions but are
not any more specic.
2.1. Threeterm relations. For orthogonal polynomials of one variable, one
important property is the threeterm relation, which states that every system
of orthogonal polynomials p
n
n=0
with respect to a positive measure satises a
threeterm relation
xp
n
(x) = a
n
p
n+1
(x) +b
n
p
n
(x) +c
n
p
n1
(x), n 0, (2.1)
where p
1
(x) = 0 by denition, a
n
, b
n
, c
n
R and a
n
c
n+1
> 0; if p
n
are orthonor
mal polynomials, then c
n
= a
n1
. Furthermore, Favards theorem states that
every sequence of polynomials that satises such a relation must be orthogonal.
Let P
: N
d
0
be a sequence of orthogonal polynomials in d variables and
assume that P
: [[ = n is a basis of 1
d
n
. The orthogonality clearly implies
that x
i
P
(x))
=n
= (P
(1) (x), . . . , P
(r
d
n
)
(x))
T
= G
n
x
n
+. . . ,
where
(1)
, . . . ,
(rn)
is the arrangement of elements in N
d
0
: [[ = n
according to the lexicographical order (or any other xed order), and x
n
=
(x
(1)
, . . . , x
(r
d
n
)
)
T
is the vector of the monomials of degree n; the matrix G
n
of size r
d
n
r
d
n
is called the leading coecient of P
n
, and it is invertible. We note
that if S is a nonsingular matrix of size [[, then the components of SP
n
are also
a basis for 1
d
n
. In terms of P
n
, we have the threeterm relation:
144 Yuan Xu
Theorem 2.1. Let P
_
a
0,n
_ 0
a
1,n
0
.
.
.
.
.
.
_ a
n,n
0
_
_
,
and
A
n,2
=
_
_
e
0,n
d
0,n
_ 0
c
1,n
e
1,n
d
1,n
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
c
n1,n
d
n1,n
0
_ c
n,n
e
n,n
d
n,n
_
_
,
where the coecients can all be computed explicitly.
For d = 1 the relation reduces to the classical threeterm relation. Moreover, let
A
n
= (A
T
n,1
[ . . . [A
T
n,d
)
T
denote the joint matrix of A
n,1
, . . . , A
n,d
, then the following
is an analog of the condition a
n
c
n+1
> 0:
Theorem 2.2. For n 0 and 1 i d, rank A
n,i
= rank C
n+1,i
= r
d
n
. Moreover,
for the joint matrix A
n
of A
n,i
and the joint matrix C
T
n
of C
T
n,i
,
rank A
n
= r
d
n+1
and rank C
T
n+1
= r
d
n+1
.
Proof. Comparing the leading coecients of the both sides of (2.2) shows that
A
n,i
G
n+1
= G
n
L
n,i
, where L
n,i
is the transformation matrix dened by L
n,i
x
n+1
=
x
i
x
n
, which implies that rank L
n,i
= r
d
n
. Hence, rank A
n,i
= r
d
n
as G
n
is in
vertible. Furthermore, let L
n
be the joint matrix of L
n,1
, . . . , L
n,d
. Then the
Orthogonal polynomials of several variables 145
components of L
n
x
n+1
contain every x
, [[ = n + 1. Hence L
n
has full rank,
rank L
n
= r
d
n+1
. Furthermore, A
n
G
n+1
= diagG
n
, . . . , G
n
L
n
from which follows
that rank A
n
= r
d
n+1
. The statement on C
T
n,i
and C
T
n
follows from the relation
A
n,i
H
n+1
= H
n
C
T
n+1,i
.
Just as in the one variable case, the threeterm relation and the rank conditions of
the coecients characterize the orthogonality. A linear functional L is said to be
positive denite if L(p
2
) > 0 for all nonzero polynomials p
d
n
. The following is
the analog of Favards theorem, which we only state for the case of C
n,i
= A
T
n1,i
and P
orthonormal.
Theorem 2.3. Let P
n
n=0
= P
n
: [[ = n, n N
0
, P
0
= 1, be an arbitrary
sequence in
d
. Then the following statements are equivalent.
(1). There exists a linear function L which denes a positive denite linear
functional on
d
and which makes P
n
n=0
an orthogonal basis in
d
.
(2). For n 0, 1 i d, there exist matrices A
n,i
and B
n,i
such that
(a) the polynomials P
n
satisfy the threeterm relation (2.2) with C
n,i
= A
T
n1,i
,
(b) the matrices in the relation satisfy the rank condition in Theorem 2.2.
The proof follows roughly the line that one uses to prove Favards theorem of
one variable. The orthogonality in the theorem is given with respect to a positive
denite linear functional. Further conditions are needed in order to show that the
linear functional is given by a nonnegative Borel measure with nite moments.
For example, if Lf =
_
fd for a measure with compact support in (1) of
Theorem 2.3, then the theorem holds with one more condition
sup
k0
A
k,i

2
< and sup
k0
B
k,i

2
< , 1 i d
in (2). The known proof of such rened results uses the spectral theory of self
adjoint operators.
Although Favards theorem shows that the threeterm relation characterizes or
thogonality, it should be pointed out that the relation is not as strong as in the
case of one variable. In one variable, the coecients of the threeterm relation
(2.1) can be any real numbers satisfying a
n
> 0 (in the case of orthonormal
polynomials c
n
= a
n1
). In several variables, however, the coecients of the
threeterm relations have to satisfy additional conditions.
146 Yuan Xu
Theorem 2.4. The coecients of the threeterm relation of a sequence of or
thonormal polynomials satisfy
A
k,i
A
k+1,j
= A
k,j
A
k+1,i
,
A
k,i
B
k+1,j
+B
k,i
A
k,j
= B
k,j
A
k,i
+A
k,j
B
k+1,i
,
A
T
k1,i
A
k1,j
+B
k,i
B
k,j
+A
k,i
A
T
k,j
= A
T
k1,j
A
k1,i
+B
k,j
B
k,i
+A
k,j
A
T
k,i
,
for i ,= j, 1 i, j d, and k 0, where A
1,i
= 0.
Proof. The relations are obtained from computing the matrices L(x
i
x
j
P
k
P
T
k+2
),
L(x
i
x
j
P
k
P
T
k
), and L(x
i
x
j
P
k
P
T
k+1
) in two dierent ways, using the threeterm re
lation (2.2) to replace x
i
P
n
and x
j
P
n
, respectively.
These equations are called the commuting conditions. Since they are necessary
for polynomials to be orthogonal, we cannot choose arbitrary matrices to gen
erate a family of polynomials satisfying the threeterm relation and hope to get
orthogonal polynomials.
As an application, let us mention that the threeterm relation implies a Christoel
Darboux formula. Recall that reproducing kernel K
n
(x, y) is dened in Section
1.5.
Theorem 2.5. (The ChristoelDarboux formula) For n 0,
K
n
(x, y) =
_
A
n,i
P
n+1
(x)
T
P
n
(y) P
T
n
(x)A
n,i
P
n+1
(y)
x
i
y
i
, 1 i d,
for x ,= y and
K
n
(x, x) = P
T
n
(x)A
n,i
i
P
n+1
(x) [A
n,i
P
n+1
(x)]
T
P
n
(x).
The proof follows just as in the case of one variable. Note, that the right hand
side depends on i, but the left hand side is independent of i.
2.2. Common zeros of orthogonal polynomials. If p
n
is a sequence of
orthogonal polynomials of one variable, then all zeros of p
n
are real and distinct,
and these zeros are the eigenvalues of the truncated Jacobi matrix J
n
.
J
n
=
_
_
b
0
a
0
_
a
0
b
1
a
1
.
.
.
.
.
.
.
.
.
a
n2
b
n1
a
n1
_ a
n1
b
n
_
_
,
where a
n
and b
n
are coecients of the threeterm relation satised by the or
thonormal polynomials. This fact has important applications in a number of
problems.
Orthogonal polynomials of several variables 147
The zero set for a polynomial in several variables can be a point, a curve, and
an algebraic variety in general a dicult object to study. However, using the
threeterm relation, it is possible to study the common zeros of P
n
(x); that is, the
common zeros of all P
_
B
0,i
A
0,i
_
A
T
0,i
B
1,i
A
1,i
.
.
.
.
.
.
.
.
.
A
T
n3,i
B
n2,i
A
n2,i
_ A
T
n2,i
B
n1,i
_
_
, 1 i d.
Then J
n,i
is a square matrix of size N N with N = dim
d
n1
. We say that
= (
1
, . . . ,
d
)
T
R
d
is a joint eigenvalue of J
n,1
, . . . , J
n,d
, if there is a ,= 0,
R
N
, such that J
n,i
=
i
for i = 1, . . . , d; the vector is called a joint
eigenvector associated to .
Theorem 2.6. A point = (
1
, . . . ,
d
)
T
R
d
is a common zero of P
n
if and
only if it is a joint eigenvalue of J
n,1
, . . . , J
n,d
; moreover, a joint eigenvector of
is (P
T
0
(), . . . , P
T
n1
())
T
.
Proof. If P
n
() = 0, then the threeterm relation for P
k
, 0 k n 1, is
the same as J
n,i
=
i
with = (P
T
0
(), . . . , P
T
n1
())
T
. On the other hand,
suppose = (
1
, . . . ,
d
) is an eigenvalue of J
n,1
, . . . , J
n,d
with a joint eigenvector
= (x
T
0
, . . . , x
T
n1
)
T
, x
j
R
r
d
j
. Let us dene x
n
= 0. Then J
n,i
=
i
implies
that x
k
n
k=0
satises the same (rst n1 equations of the) threeterm relation as
P
k
()
n
k=0
does. The rank condition on A
n,i
shows inductively that x
0
,= 0 unless
is zero. But ,= 0 as an eigenvector and we can assume that x
0
= 1 = P
0
. Then
y
k
n
k=0
with y
k
= x
k
P
k
satises the same threeterm relation. But y
0
= 0,
it follows from the rank condition that y
k
= 0 for all 1 k n. In particular,
y
n
= P
n
() = 0.
The main properties of the common zeros of P
n
are as follows:
Corollary 2.1. All common zeros of P
n
are real distinct points and they are
simple. The polynomials in P
n
have at most N = dim
d
n1
common zeros and
P
n
has N zeros if and only if
A
n1,i
A
T
n1,j
= A
n1,j
A
T
n1,i
, 1 i, j d. (2.3)
148 Yuan Xu
Proof. Since the matrices J
n,i
are symmetric, the joint eigenvalues are real. If x
is a common zero, then the ChristoelDarboux formula shows that
P
T
n
(x)A
n,i
i
P
n+1
(x) = K
n
(x, x) > 0,
so that at least one of the partial derivatives of P
n
is not zero at x; that is, the
common zero is simple. Since J
n,i
is an NN square matrix, there are at most N
eigenvalues, and P
n
has at most N common zeros. Moreover, P
n
has N distinct
zeros if and only if J
n,1
, . . . , J
n,d
can be simultaneously diagonalized, which holds
if and only if J
n,1
, . . . , J
n,d
commute,
J
n,i
J
n,j
= J
n,j
J
n,i
, 1 i, j d.
From the denition of J
n,i
and the commuting conditions in Theorem 2.4, the
above equation is equivalent to the condition
A
T
n2,i
A
n2,j
+B
n1,i
B
n1,j
= A
T
n2,j
A
n2,i
+B
n1,j
B
n1,i
.
The third equation of the commuting condition leads to the desired result.
The zeros of orthogonal polynomials of one variable are nodes of the Gaussian
quadrature formula. A similar result can be stated in several variables for the
common zeros of P
n
. However, it turns out that the condition (2.3) holds rarely;
for example, it does not hold for those weight functions that are centrally sym
metric (the support set of W is symmetric with respect to the origin and
W(x) = W(x) for all x in ). Consequently, P
n
does not have N common
zeros in general and the straightforward generalization of Gaussian quadrature
usually does not exist. The relation between common zeros of orthogonal poly
nomials and quadrature formulae in several variables is quite complicated. One
needs to study common zeros of subsets of (quasi)orthogonal polynomials, and
the main problem is to characterize or identify subsets that have a large number
of common zeros. In the language of polynomial ideals and varieties, the prob
lem essentially comes down to characterize or identify those polynomial ideals
generated by (quasi)orthogonal polynomials whose varieties are large subsets of
points, and the size of the variety should equal to the codimension of the ideal.
Although some progress has been made in this area, the problem remains open
for the most part.
3. hharmonics and orthogonal polynomials on the sphere
After the rst subsection on the relation between orthogonal polynomials on the
sphere and those on the ball, we discuss hharmonics in two steps. The main
eort is in a long subsection devoted to the case of the product weight function,
which can be developed without any background in reection groups, and it is
this case that oers most explicit formulae. The theory of hharmonics associated
to general reection groups is summarized in the third subsection.
Orthogonal polynomials of several variables 149
3.1. Orthogonal polynomials on the unit ball and on the unit sphere.
If a measure is supported on the unit sphere S
d1
= x R
d
: x = 1 of R
d
,
then the integrals of the positive polynomials (1 x
2
)
2n
over S
d1
are zero, so
that the general properties of orthogonal polynomials in the previous section no
longer hold. There is, however, a close relation between orthogonal structure on
the sphere and that on the ball, which can be used to study the general properties
for orthogonal polynomials on the sphere.
As a motivating example, recall that in polar coordinates y
1
= r cos and y
2
=
r sin , the spherical harmonics of degree n on S
1
are given by
Y
(1)
n
(y) = r
n
cos n = r
n
T
n
(y
1
/r), Y
(2)
n
(y) = r
n
sin n = r
n
y
2
U
n1
(y
1
/r),
where T
n
and U
n
are the Chebyshev polynomials, which are orthogonal with
respect to 1/
1 x
2
and
1 x
2
on B
1
= [1, 1], respectively. This relation
can be extended to higher dimension. In the following we work with S
d
instead
of S
d1
.
Let H be a weight function dened on R
d+1
and assume that H is nonzero
almost everywhere when restricted to S
d
, even with respect to y
d+1
, and centrally
symmetric with respect to the variables y
= (y
1
, . . . , y
d
); for example, H is even
in each of its variables, H(y) = W(y
2
1
, . . . , y
2
d+1
). Associated with H dene a
weight function W
B
H
on B
d
by
W
B
H
(x) = H(x,
_
1 x
2
), x B
d
. (3.1)
We use the notation 1
d
n
(W) to denote the space of orthogonal polynomials of
degree n with respect to W. Let P
and Q
, y
d+1
) and
y = r(x, x
d+1
), where r = y, (x, x
d+1
) S
d
.
For [[ = n and [[ = n 1 we dene the following polynomials
Y
(1)
(y) = r
n
P
(x) and Y
(2)
(y) = r
n
x
d+1
Q
(x), (3.2)
and dene Y
(2)
,0
(y) = 0. These are in fact homogeneous orthogonal polynomials
with respect to Hd on S
d
:
Theorem 3.1. Let H(x) = W(x
2
1
, . . . , x
2
d+1
) be dened as above. Then Y
(1)
and
Y
(2)
(y)Y
(j)
(y)H(y)d
d
(y) =
,
i,j
, i, j = 1, 2.
150 Yuan Xu
Proof. Since both weight functions W
B
1
and W
B
2
are even in each of its variables,
it follows that P
and Q
(y)
are homogeneous polynomials of degree n in y. Since Y
(1)
, when restricted to S
d
,
is independent of x
d+1
and Y
(2)
and Y
(2)
(for i = 1) or Q
_
n +d 2
d
_
= dimT
d+1
n
dimT
d+1
n2
.
Furthermore, we also have the following orthogonal decomposition:
Theorem 3.2. For each n N
0
and P P
d+1
n
, there is a unique decomposition
P(y) =
[n/2]
k=0
y
2k
P
n2k
(y), P
n2k
H
d+1
n2k
(H).
The classical example is the Lebesgue measure H(x) = 1 on the sphere, which
gives the ordinary spherical harmonics. We discuss a family of more general
weight functions in detail in the following section.
3.2. Orthogonal polynomials for the product weight functions. We con
sider orthogonal polynomials in H
d+1
n
(h
2
) with h
(x) =
d+1
i=1
[x
i
[
i
,
i
> 1, x R
d+1
.
Because of the previous subsection, we consider S
d
instead of S
d1
. The elements
of H
d+1
n
(h
2
is of product type, an
orthonormal basis with respect to h
2
is
d
_
S
d
h
2
(x)d(x) =
(
d+1
2
)
d
2
(
1
+
1
2
) . . . (
d+1
+
1
2
)
([[ +
d+1
2
)
,
where
1
d
=
_
S
d
d = 2
(d+1)/2
/((d + 1)/2). The orthonormal basis is given in
terms of the generalized Gegenbauer polynomials C
(,)
n
dened by
C
(,)
2n
(x) =
( +)
n
_
+
1
2
_
n
P
(1/2,1/2)
n
(2x
2
1),
C
(,)
2n+1
(x) =
( +)
n+1
_
+
1
2
_
n+1
xP
(1/2,+1/2)
n
(2x
2
1),
which are orthogonal with respect to the weight function [x[
2
(1 x
2
)
1/2
on
[1, 1]. It follows that C
(,0)
n
= C
n
, the usual Gegenbauer polynomial. Let
C
(,)
n
denote the corresponding orthonormal polynomial. For d = 1 and h
(x) =
[x
1
[
1
[x
2
[
2
an orthonormal basis for H
2
n
(h
2
) is given by
Y
1
n
(x) = r
n
C
(
2
,
1
)
n
(cos ), Y
2
n
(x) = r
n
1
+
2
+ 1
2
+
1
2
sin
C
(
2
+1,
1
)
n1
(cos ).
For d > 1, we use the following notation: associated to = (
1
, . . . ,
d+1
), dene
j
= (
j
, . . . ,
d+1
), 1 j d + 1.
Since
d+1
consists of only the last element of , write
d+1
=
d+1
. Similarly
dene
j
for N
d
0
.
152 Yuan Xu
Theorem 3.3. Let d 1. In spherical coordinates an orthonormal basis of
H
d+1
n
(h
2
) is given by
Y
n,i
(x) = [A
n
]
1
r
n
d1
j=1
_
C
(a
j
,
j
)
j
(cos
dj+1
)(sin
dj+1
)

j+1

_
Y
i
d
(cos
1
, sin
1
),
where N
d+1
0
, [[ = n, a
j
= [
j+1
[ + [
j+1
[ +
dj
2
, Y
i
d
with i = 1, 2 are
twodimensional hharmonics with parameters (
d1
,
d
) and
[A
n
]
2
=
1
_
[[ +
d+1
2
_
n
d
j=1
_
[
j+1
[ +[
j
[ +
d j + 2
2
_
j
.
This can be veried by straightforward computation, using the integral
_
S
d
f(x)d
d
(x) =
_
0
_
S
d1
f(cos , sin x
)d
d1
(x
) sin
d1
d
inductively and the orthogonality of C
(,)
n
.
3.2.2. hharmonics. There is another way of describing the space H
d+1
n
(h
2
) using
a second order dierentialdierence operator that plays the role of the Laplace
operator for the ordinary harmonics. For
i
0, dene Dunkls operators T
j
by
T
j
f(x) =
j
f(x) +
j
f(x) f(x
1
, . . . , x
j
, . . . , x
d+1
)
x
j
, 1 j d + 1.
It is easy to see that these rst order dierentialdierence operators map T
d
n
into
T
d
n1
. A remarkable fact is that these operators commute, which can be veried
by an easy computation.
Theorem 3.4. The operators T
i
commute: T
i
T
j
= T
j
T
i
, 1 i, j d + 1.
Proof. Let x
j
= (x
1
, . . . , x
j
, . . . , x
d+1
). A simple computation shows that
T
i
T
j
f(x) =
i
j
f(x) +
i
x
i
(
j
f(x)
j
f(x
j
)) +
j
x
j
(
i
f(x)
i
f(x
i
))
+
i
j
x
i
x
j
(f(x) f(x
j
) f(x
i
) f(x
j
i
)),
from which T
i
T
j
= T
j
T
i
is evident.
The operator T
i
plays the role of
i
. The hLaplacian is dened by
h
= T
2
1
+ +T
2
d+1
.
Orthogonal polynomials of several variables 153
Its a second order dierentialdierence operator. If all
i
= 0 then
h
becomes
the classical Laplacian . A quick computation shows that
h
f(x) = f(x) + 2
d+1
j=1
j
x
j
j
f(x)
d+1
j=1
j
f(x) f(x
1
, . . . , x
j
, . . . , x
d
)
x
2
j
.
Let us write
h
= L
h
+ D
h
, where L
h
is the dierential part and D
h
is the
dierence part of the above equation. The following theorem shows that h
harmonics are homogeneous polynomials P satisfying
h
P = 0.
Theorem 3.5. Suppose f and g are homogeneous polynomials of dierent degrees
satisfying
h
f = 0 and
h
g = 0, then
_
S
d
f(x)g(x)h
2
(x)d = 0.
Proof. Assume
i
1 and use analytic continuation to extend the range of
validity to 0. The following formula can be proved using Greens identity:
_
S
d
f
n
gh
2
d =
_
B
d+1
(gL
h
f +f, g))h
2
dx,
where f/n denotes the normal derivative of f. If f is homogeneous, then
Eulers equation shows that f/n = (deg f)f. Hence,
(deg f deg g)
_
S
d
fgh
2
d =
_
B
d+1
(gL
h
f fL
h
g)h
2
dx
=
_
B
d+1
(gD
h
f fD
h
g)h
2
dx = 0,
since the explicit formula of D
h
shows that it is a symmetric operator.
There is a linear operator V
= V
i
, V 1 = 1, V T
n
T
n
.
It follows that
h
V
= V
f(x) =
_
[1,1]
d+1
f(x
1
t
1
, . . . , x
d+1
t
d+1
)
d+1
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt,
where c
= ( + 1/2)/(
_
1
1
f(t)(1 t
2
)
1
d(t) = [f(1) +f(1)]/2.
154 Yuan Xu
Proof. Denote the dierence part of T
i
as
T
i
so that T
i
=
i
+
T
i
. Clearly,
j
V
f(x) =
_
[1,1]
d+1
j
f(x
1
t
1
, . . . , x
d+1
t
d+1
)t
j
d+1
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt.
Taking into account the parity of the integrand, an integration by parts shows
T
j
V
f(x) =
_
[1,1]
d+1
j
f(x
1
t
1
, . . . , x
d+1
t
d+1
)(1 t
j
)
d+1
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt.
Adding the last two equations gives T
i
V
= V
i
.
As one important application, a compact formula of the reproducing kernel can be
given in terms of V
; x, y) of H
d+1
n
(h
2
) is dened
uniquely by the property
_
S
d
P
n
(h
2
; x, y)Q(y)h
2
(y)d(y) = Q(x), Q H
d+1
n
(h
2
).
If Y
is an orthonormal basis of H
d+1
n
(h
2
), then P
n
(h
2
; x, y) =
(x)Y
(y).
Theorem 3.7. For
i
0, and y x = 1,
P
n
(h
2
; x, y) =
n +[[ +
d1
2
[[ +
d1
2
V
_
C
+
d1
2
n
__
,
y
y
___
(x)y
n
.
Proof. Let K
n
(x, y) = V
(x)
(x, y)
n
)/n!. Using the dening property of V
,
it is easy to see that K
n
(x, T
(y)
)f(y) = f(x) for f T
d+1
n
. Fixing y let
p(x) = K
n
(x, y); then P
n
(h
2
; x, y) = 2
n
([[ + d/2)
n
proj
n
p(x), where proj
n
is
the projection operator from T
d+1
n
onto H
d+1
n
(h
2
; x, y) =
0jn/2
_
+
d
2
_
n
2
n2j
_
2 n
d/2)
j
j!
x
2j
y
2j
K
n2j
(x, y).
When x = 1, we can write the right hand side as V
(L
n
(, y/y)))(x), where
the polynomial L
n
is a constant multiple of the Gegenbauer polynomial.
For the classical harmonics ( = 0), these are the socalled zonal harmonics,
P
n
(x, y) =
n +
d1
2
d1
2
C
d1
2
n
(x, y)), x, y S
d
.
As one more application of the intertwining operator, we mention an analogue of
the FunkHecke formula for ordinary harmonics. Denote by w
the normalized
weight function
w
(t) =
( + 1)
( + 1/2)
(1 t
2
)
1/2
, t [1, 1],
Orthogonal polynomials of several variables 155
whose corresponding orthogonal polynomials are the Gegenbauer polynomials.
Theorem 3.8. Let f be a continuous function on [1, 1]. Let Y
h
n
H
d
n
(h
2
).
Then
_
S
d
V
f(x, ))(y)Y
h
n
(y)h
2
(y)d(y) =
n
(f)Y
h
n
(x), x S
d
,
where
n
(f) is a constant dened by
n
(f) =
1
C
+(d1)/2
n
(1)
_
1
1
f(t)C
+
d1
2
n
(t)w
+(d1)/2
(t)dt.
The case = 0 is the classical FunkHecke formula.
3.2.3. Monomial basis. Another interesting orthogonal basis of H
d+1
n
(h
2
) can be
given explicitly. Let V
be polynomials dened by
V
_
(1 2b, ) +b
2
)

d1
2
_
(x) =
N
d+1
0
b
(x), b R
d+1
.
The polynomials P
()
()
(c)

!
x
, , N
d+1
0
, c R,
where the summation is taken over N
d+1
0
. For N
d+1
0
, let [/2] denote
the multiindex whose elements are [
i
/2] where [a] denotes the integer part of
a. The notation ()
0
(
d+1
)
d+1
.
Theorem 3.9. For N
d+1
0
,
P
(x) =
2

([[ +
d1
2
)
!
(
1
2
)
[
+1
2
]
( +
1
2
)
[
+1
2
]
Y
(x),
where Y
are given by
Y
(x) = x
F
B
_
+
_
+1
2
_
,
_
+1
2
_
+
1
2
; [[ [[
d3
2
;
1
x
2
1
, . . . ,
1
x
2
d+1
_
.
Proof. Let = [[ +
d1
2
. The multinomial and binomial formulae show that
(1 2b, x) +b
2
)
2

()

( +)
( )!!
2
2
x
2
.
156 Yuan Xu
Applying the intertwining operator and using the formula
V
x
2
=
(
1
2
)
[
+1
2
]
( +
1
2
)
[
+1
2
]
_
_
+1
2
_
+
1
2
_
_
+1
2
_
+
1
2
_
x
2
completes the proof.
Since (a)
k
= 0 if a < k, Y
(x) = x
are elements of H
d+1

(h
2
).
Proof. We show that Y
are orthogonal to x
for N
d+1
0
and [[ n 1.
If one of the components of is odd, then the orthogonality follows from
changing sign of that component in the integral. Hence, we can assume that
all components of are even and we only have to work with x
for [[ =
[[ 2, since every polynomial of degree n on the sphere is the restriction of a
homogeneous polynomial of degree n. Using the beta integral on the sphere, a
tedious computation shows that
_
S
d
P
(x)x
h
2
(x)d(x) =
( +
1
2
)+
2
([[ +
d+1
2
)

( + [
+1
2
])
([
+1
2
] +
1
2
)
(
+
2
+
1
2
)
!
=
( +
1
2
)+
2
([[ +
d+1
2
)

d+1
i=1
2
F
1
_
i
+ [
i
+1
2
], [
i
+1
2
]
i
+
1
2
i
+
i
2
i
+
1
2
; 1
_
.
Since at least one
i
<
i
, this last term is zero using the ChuVandermonde
identity for
2
F
1
.
A standard Hilbert space argument shows that among all polynomials of the
form x
d)norm and
Y
onto H
d+1
n
(h
2
. For any N
d
, dene homogeneous polynomials H
(cf. [44])
H
(x) = x
2+d2+2
T
_
x
2d+2
_
,
where T
= T
1
1
. . . T
d
d
.
Theorem 3.11. For N
d+1
0
and [[ = n,
H
(x) = (1)
n
2
n
_
[[ +
d 1
2
_
n
Y
(x).
For = 0, the polynomials H
(x) +x
2
T
i
H
,
Orthogonal polynomials of several variables 157
where
i
= (0, . . . , 1, . . . , 0) is the ith standard unit vector in R
d+1
.
3.3. hharmonics for a general reection group. The theory of hharmonics
is established for a general reection group ([6, 7, 8]). For a nonzero vector
v R
d+1
dene the reection
v
by x
v
:= x 2x, v)v/v
2
, x R
d+1
, where
x, y) denotes the usual Euclidean inner product. A nite reection group G is
described by its root system R, which is a nite set of nonzero vectors in R
d+1
such
that u, v R implies u
v
R, and G is the subgroup of the orthogonal group
generated by the reections
u
: u R. If R is not the union of two nonempty
orthogonal subsets, the corresponding reection group is called irreducible. Note,
that Z
d+1
2
is reducible, a product of d + 1 copies of the irreducible group Z
2
.
There is a complete classication of irreducible nite reection groups. The
list consists of root systems of innite families A
d1
with G being the symmetric
group of d objects, B
d
with G being the symmetry group of the hyperoctahedron
1
, . . . ,
d+1
of R
d+1
, D
d
with G being a subgroup of the hyperoctahedral
group for d 4, the dihedral systems I
2
(m) with G being the symmetric group of
regular mgons in R
2
for m 3, and several other individual systems H
3
, H
4
, F
4
and E
6
, E
7
, E
8
.
Fix u
0
R
d+1
such that u, u
0
) ,= 0. The set of positive roots R
+
with respect
to u
0
is dened by R
+
= u R : u, u
0
) > 0 so that R = R
+
(R
+
). A
multiplicity function v
v
of R
+
R is a function dened on R
+
with the
property that
u
=
v
if
u
is conjugate to
v
; in other words, the function is
Ginvariant. Fix a positive root system R
+
. Then the function h
dened by
h
(x) =
vR
+
[x, v)[
v
, x R
d+1
,
is a positive homogeneous function of degree
:=
vR
+
v
and h
(x) is invari
ant under G. The hharmonics are homogeneous orthogonal polynomials on S
d
with respect to h
2
d+1
i=1
[x
i
[
i
, the
most interesting to us are the case A
d
for which R
+
=
i
j
: i > j and
h
(x) =
1i,jd+1
[x
i
x
j
[
, 0,
which is invariant under the symmetric group S
d
, and the case B
d+1
for which
R
+
=
i
j
,
i
+
j
: i < j : 1 i d + 1 and
h
(x) =
d+1
i=1
[x
i
[
1i,jd+1
[x
2
i
x
2
j
[
1
,
0
,
1
0,
which is invariant under the hyperoctahedral group.
158 Yuan Xu
For a nite reection group G with positive roots R
+
and a multiplicity function,
Dunkls operators are dened by
T
i
f(x) :=
i
f(x) +
vR
+
(v)
f(x) f(x
v
)
x, v)
v,
i
), 1 i d + 1,
where
1
, . . . ,
d+1
are the standard unit vectors of R
d+1
. The remarkable fact
that these are commuting operators, T
i
T
j
= T
j
T
i
, holds for every reection
group. The hLaplacian is dened again by
h
= T
2
1
+ + T
2
d+1
, which plays
the role of the Laplacian in the theory of ordinary harmonics. The hharmonics
are the homogeneous polynomials satisfying the equation
h
P = 0 and Theorem
3.5 holds also for a general reection group.
There again exists an intertwining operator V
T
n
T
n
, V 1 = 1, T
i
V
= V
i
, 1 i d.
The representation of the reproducing kernel in Theorem 3.7 in terms of the
intertwining operator holds for a general reection group. It was proved by
Rosler [27] that V
p 0 if p 0.
However, unlike the case of G = Z
d+1
2
, there are few explicit formulae known
for hharmonics with respect to a general reection group. In fact, there is no
orthonormal basis of H
d+1
n
(h
2
: [[ = n, N
d+1
0
,
d+1
= 0 or 1 is an orthogonal basis,
but explicit formulae for H
and its L
2
(h
2
) consists of V
, where Y
is known only
in the case of the symmetric group S
3
on R
3
, and the formula is complicated
and likely not in its nal form. In fact, even in the case of dihedral groups on
R
2
the formula of V
(x) = [x
1
x
2
[
1
[x
2
1
x
2
2
[
2
.
4. Orthogonal polynomials on the unit ball
As we have seen in the Section 3.1, the orthogonal polynomials on the unit ball are
closely related to orthogonal polynomials on the sphere. This allows us to derive,
working with G Z
2
on R
d+1
, various properties for orthogonal polynomials
with respect to the weight function h
(x)(1 x)
1/2
, where h
is a reection
invariant function dened on R
d
. Again we will work with the case of h
being a
Orthogonal polynomials of several variables 159
product weight function,
W
B
,
(x) =
d+1
i=1
[x
i
[
2
i
(1 x)
1/2
,
i
0, > 1/2
for which various explicit formulae can be derived from the results in Section 3.2.
In the case = 0, W
B
= W
B
0,
is the classical weight function on B
d
, which is
invariant under rotations.
4.1. Dierentialdierence equation. For y R
d+1
we use the polar coordi
nates y = r(x, x
d+1
), where r = y and (x, x
d+1
) S
d
. The relation in Sec
tion 3.1 states that if P
(y) = r
n
P
d
i=1
[y
i
[
i
[y
d+1
[
.
Since the polynomials Y
in terms of x B
d
, we choose the following mapping:
y (r, x) : y
1
= rx
1
, . . . , y
d
= rx
d
, y
d+1
= r
_
1 x
2
1
x
2
d
,
which is onetoone from y R
d+1
: y
d+1
0 to itself. We rewrite the h
Laplacian in terms of the new coordinates (r, x). Let
,
h
denote the hLaplacian
associated with the weight function h
,
, and preserve the notation h
for the h
Laplacian associated with the weight function h
(x) =
d
i=1
[x
i
[
i
for x R
d
.
Proposition 4.1. Acting on functions on R
d+1
that are even in y
d+1
, the operator
,
h
takes the form
,
h
=
2
r
2
+
d + 2[[ + 2
r
r
+
1
r
2
,
h,0
in terms of the coordinates (r, x) in y R
d+1
: y
d+1
0, where the spherical
part
,
h,0
, acting on functions in the variables x, is given by
,
h,0
=
h
x, )
2
(2[[ + 2 +d 1)x, ),
in which the operators
h
and = (
1
, . . . ,
d
) are all acting on x variables.
Proof. Since
,
h
is just
(y)
h
acting on functions dened on R
d+1
for h
(y) =
d+1
i=1
[y
i
[
i
(with
d+1
= ), its formula is given in Section 3.2. Writing r and x
i
in terms of y under the change of variables y (r, x) and computing the partial
derivatives x
i
/y
i
, the chain rule implies that
y
i
= x
i
r
+
1
r
_
x
i
x
i
x,
(x)
)
_
, 1 i d + 1,
160 Yuan Xu
where for i = d + 1 we use the convention that x
d+1
=
_
1 x
2
2
and dene
/x
d+1
= 0. A tedious computation gives the second order derivatives
2
y
2
i
=x
2
i
2
r
2
+
1 x
2
i
r
r
+
1
r
2
_
2
x
2
i
(1 x
2
i
)x,
(x)
) x
i
x,
(x)
)
x
i
_
+
_
x
i
x
i
x
2
i
x,
(x)
)
__
1
r
r
1
r
2
_
_
x
i
x
i
x
2
i
x,
(x)
)
_
x,
(x)
).
Using these formulae and the fact that f(y
1
, . . . , y
d+1
) f(y
1
, . . . , y
d+1
) = 0 for
f even in y
d+1
, the stated equation follows from a straightforward computation.
h
x, )
2
(2[[ + 2 +d 1)x, )
_
P = n(n +d + 2[[ + 2 1)P.
Proof. Let P 1
d
n
(W
B
,
). The formula in the Proposition 4.1 applied to the
homogeneous polynomial Y
(y) = r
n
P
(x) gives
0 =
,
h
Y
(y) = r
n2
[n(n +d + 2[[ + 2 1)P
(x) +
,
h,0
P
(x)].
The stated result follows from the formula for
,
h,0
.
For = 0,
h
becomes the ordinary Laplacian and the equation becomes a
dierential equation, which is the classical dierential equation in [2]; note, that
in this case the weight function W
B
(x)(1x
2
)
1/2
for h
(y) = r
n
P
(x), y = r(x, x
d+1
) R
d
and x B
d
, several orthogonal bases
for 1
d
n
(W
B
,
) follow from the bases for the hharmonics in Section 3.2. All or
thonormal bases are with respect to the weight function normalized to have unit
integral. Associated with x = (x
1
, . . . , x
d
) R
d
, dene x
j
= (x
1
, . . . , x
j
) for
1 j d and x
0
= 0.
Theorem 4.2. An orthonormal basis of 1
d
n
(W
B
,
) is given by P
: N
d
0
, [[ =
n dened by
P
(x) = [h
B
]
1
d
j=1
(1 x
j1

2
)
j
2
C
(a
j
,
j
)
j
_
x
j
_
1 x
j1

2
_
,
Orthogonal polynomials of several variables 161
where a
j
= +[
j+1
[ +[
j
[ +
dj
2
and h
B
are given by
[h
B
]
2
=
1
([[ + +
d+1
2
)
n
d
j=1
_
+[
j+1
[ +[
j
[ +
d j + 2
2
_
j
.
Proof. In the spherical coordinates of (x, x
d+1
) S
d
, cos
dj
= x
j+1
/
_
1 x
j

2
and sin
dj
=
_
1 x
j+1

2
/
_
1 x
j

2
. Hence, this basis is obtained from the
hharmonic basis in Theorem 3.3 using the correspondence.
Another orthonormal basis can be given in the polar coordinates. Using
_
B
d
f(x)dx =
_
1
0
r
d1
_
S
d1
f(rx
)d(x
)dr,
the verication is a straightforward computation.
Theorem 4.3. For 0 j n/2 let S
h
n2j,
denote an orthonormal basis of
H
d
n2j
(h
2
C
(,n2j++
d1
2
)
2j
(x)S
h
,n2j
(x)
form an orthonormal basis of 1
d
n
(W
B
,
), in which the constants are given by
[c
B
j,n
]
2
=
([[ + +
d+1
2
)(n 2j +[[ +
d
2
)
([[ +
d
2
)(n 2j +[[ + +
d+1
2
)
.
Another interesting basis, the monomial basis, can be derived from the monomial
hharmonics in Section 3.2.3. This is a basis for which P
(x) = c
+ . . .,
corresponding to
d+1
= 0 of the basis in Denition 3.1. We denote this basis by
P
B
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt
=
N
d
0
b
P
B
(x),
where = [[ + +
d1
2
. This corresponds to the case of b
d+1
= 0 in Denition
3.1. The explicit formulae of these polynomials follow from Theorem 3.9,
P
B
(x) =
2

([[ + +
d1
2
)
!
(
1
2
)
[
+1
2
]
( +
1
2
)
[
+1
2
]
x
F
B
_
+
_
+ 1
2
_
,
_
+ 1
2
_
+
1
2
; [[ [[
d 3
2
;
1
x
2
1
, . . . ,
1
x
2
d+1
_
.
162 Yuan Xu
Clearly the highest degree of P
B
(x) is a multiple of x
, = 0, and using
the fact that
_
+
_
+ 1
2
__
_
+ 1
2
_
+
1
2
_
=
_
2
_
2
+
1
2
_
,
the formula of P
B
can be rewritten as
V
(x) = P
B
(x) =
2

( +
d1
2
)

!
x
F
B
_
2
,
+ 1
2
; [[
d 3
2
;
1
x
2
1
, . . . ,
1
x
2
d+1
_
,
which are Appells monomial orthogonal polynomials. In this case, there is an
other basis dened by
U
(x) =
(1)

(2)

2

( +
1
2
)

!
(1 x
2
)
+
1
2

x
1
1
. . . x
d
d
(1 x
2
)
+
1
2
,
which is biorthogonal to the monomial basis in the following sense:
Theorem 4.4. The polynomials U
and V
are biorthogonal,
w
B
_
B
d
V
(x)U
(x)W
B
(x)dx =
+
d1
2
[[ + +
d1
2
(2)

!
,
.
Proof. Since V
_
B
d
V
(x)U
(x)W
B
(x)dx
= w
B
(2)

2

( +
1
2
)

!
_
B
d
_

x
1
1
. . . x
d
d
V
(x)
_
(1 x
2
)
+
1
2
dx.
However, since V

x
(x) = 0 for [[ > [[, which proves the orthogonality. A simple computation
gives the constant for the case [[ = [[.
Among other explicit formulae that we get, the compact formula for the repro
ducing kernel is of particular interest. Let us denote the reproducing kernel of
1
d
n
(W) by P
n
(W; x, y) as dened in Section 1.5.
Orthogonal polynomials of several variables 163
Theorem 4.5. For x, y B
d
, the reproducing kernel can be written as an integral
P
n
(W
B
,
; x, y) =
n +[[ + +
d1
2
[[ + +
d1
2
_
1
1
_
[1,1]
d
C
++
d1
2
n
_
t
1
x
1
y
1
+ +t
d
x
d
y
d
+s
_
1 x
2
_
1 y
2
_
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt c
(1 s
2
)
1
ds.
Proof. Let h
(y) =
d+1
i=1
[y
i
[
i
with
d+1
= . Then the correspondence in
Section 3.1 can be used to show that
P
n
(W
B
,
; x, y) =
1
2
_
P
n
(h
2
; x, (y,
_
1 [y[
2
)) +P
n
(h
2
; x, (y,
_
1 [y[
2
))
_
,
from which the stated formula follows from Theorem 3.6.
In particular, taking the limit
i
0 for i = 1, . . . , d, we conclude that for the
classical weight function W
B
,
P
n
(W
B
; x, y) = c
n + +
d1
2
+
d1
2
(4.1)
_
1
1
C
+
d1
2
n
_
x, y) +t
_
1 x
2
_
1 y
2
_
(1 t
2
)
1
dt.
Even in this case the formula has been discovered only recently. For d = 1, this
reduces to the classical product formula of the Gegenbauer polynomials:
C
n
(x)C
n
(y)
C
n
(1)
= c
_
1
1
C
n
(xy +t
1 x
2
_
1 y
2
)(1 t
2
)
1
dt.
There is also an analogue of the FunkHecke formula for orthogonal polynomials
on B
d
. The most interesting case is the formula for the classical weight function:
Theorem 4.6. Let f be a continuous function on [1, 1]. Let P 1
d
n
(W
B
).
Then
_
B
d
f(x, y))P(y)W
B
(y)dy =
n
(f)P(x), x = 1,
where
n
(f) is the same as in Theorem 3.8 with [[ replaced by .
As a consequence, it follows that the polynomial C
(+(d1)/2)
n
(x, )) with satis
fying  = 1 is an element of 1
d
n
(W
B
(x)dx =
n
C
+
d1
2
n
(, )),
164 Yuan Xu
where
n
= ( +(d 1)/2)/(n + +(d 1)/2). The basis in Subsection 1.2.3 is
derived from this integral.
Several results given above hold for the weight functions h
2
(x)(1 x
2
)
1/2
,
where h
is one of the weight functions in Section 3.3 that are invariant under
reection groups. Most notably is Theorem 4.2, which holds with [[ replaced by
vR
+
v
. Analogous of Theorems 4.4 and 4.5, in which the intertwining
operator is used, also hold, but the formulae are not really explicit since neither
an explicit basis for H
d
n
(h
2
associated
with the general reection groups.
4.3. Rotation invariant weight function. If (t) is a nonnegative even func
tion on R with nite moments, then the weight function W(x) = (x) is a
rotation invariant weight function on R
d
. Such a function is call a radial func
tion. The classical weight function W
B
02jn
p
(2n4j+d1)
2j
(x)p
(2n4j+d1)
2j
(y)
n +
d1
2
d1
2
x
n2j
y
n2j
C
d1
2
n2j
(x, y)),
where we have used the fact that S
n2j,
are homogeneous and
S
m,
(x)S
m,
(y)
for x, y S
d1
is a zonal polynomial. Using polar coordinates,
S
n
(W; f, x) =
_
0
g(r)
_
S
d1
P
n
(x, ry
)d(y
)r
d1
(r)dr.
Since C
d1
2
n
(x, y)) is a zonal harmonic, its integral over S
d1
is zero for n > 0
and is equal to the surface measure
d1
of S
d1
if n = 0. Hence, we get
S
n
(W; f, x) =
d1
_
0
g(r)p
(d1)
n
(r)w(r)dr p
(d1)
n
(x).
Since orthonormal bases are assumed to be with respect to the normalized weight
function, setting g = 1 shows that there is no constant in front of s
n
(w; g).
5. Orthogonal polynomials on the simplex
Orthogonal polynomials on the simplex T
d
= x R
d
: x
1
0, . . . , x
d
0, 1
[x[ > 0 are closely related to those on the unit ball. The relation depends on
the basic formula
_
B
d
f(y
2
1
, . . . , y
2
d
)dy =
_
T
d
f(x
1
, . . . , x
d
)
dx
x
1
. . . x
d
.
Let W
B
(x) = W(x
2
1
, . . . , x
2
d
) be a weight function dened on B
d
. Associated with
W
B
dene a weight function W
T
on T
d
by
W
T
(y) = W(y
1
, . . . , y
d
)
_
y
1
. . . y
d
, y = (y
1
, . . . , y
d
) T
d
.
Let 1
d
2n
(W
B
; Z
d
2
) denote the subspace of orthogonal polynomials in 1
d
2n
(W
B
) that
are even in each of its variables (that is, invariant under Z
d
2
). If P
1
d
2n
(W
B
; Z
d
2
),
then there is a polynomial R
d
n
such that P
(x) = R
(x
2
1
, . . . , x
2
d
). The
polynomial R
is in fact an element of 1
d
n
(W
T
).
Theorem 5.1. The relation P
(x) = R
(x
2
1
, . . . , x
2
d
) denes a onetoone cor
respondence between an orthonormal basis of 1
2n
(W
B
, Z
d
2
) and an orthonormal
basis of 1
d
n
(W
T
).
166 Yuan Xu
Proof. Assume that R
=n
is an orthonormal polynomial in 1
d
n
(W
T
). If
N
d
0
has one odd component, then the integral of P
(x)x
with respect to W
B
over B
d
is zero. If all components of are even and [[ < 2n, then it can be
written as = 2 with N
d
0
and [[ n 1. The basic integral can be used
to convert the integral of P
(x)x
2
over B
d
to the integral over T
d
, so that the
orthogonality of R
implies that P
is orthogonal to x
.
The classical weight function W
T
on the simplex T
d
is dened by
W
T
(x) =
d
i=1
[x
i
[
i
1/2
(1 [x[)
d+1
1/2
, x T
d
,
i
> 1.
Sometimes we write W
T
as W
T
,
with =
d+1
, as the orthogonal polynomi
als with respect to W
T
,
on T
d
are related to the orthogonal polynomials with
with respect to W
B
,
on B
d
. Below we give explicit formulas for these classical
orthogonal polynomials on the simplex.
Theorem 5.2. The orthogonal polynomials in 1
d
n
(W
T
i=1
x
i
(1 x
i
)
2
P
x
2
i
2
1i<jd
x
i
x
j
2
P
x
i
x
j
+
d
i=1
_
_
i
+
1
2
_
_
[[ +
d 1
2
_
x
i
_
P
x
i
= n
_
n +[[ +
d 1
2
_
P.
Proof. For functions that are even in each of its variables, the hLaplacian
h
for
the product weight function becomes a dierential operator (see the formula in
Section 3.2.2). Consequently, for the orthogonal polynomials P
1
d
2n
(W
B
,
; Z
d
2
),
the dierentialdierence equation in Theorem 4.1 become a dierential equation.
Changing variables x
i
z
i
gives
x
i
= 2
z
i
z
i
and
2
x
2
i
= 2
_
z
i
+ 2z
i
2
z
2
i
_
,
from which the equation for P
(x) = R
(x
2
1
, . . . , x
2
d
) translates into an equation
satised by R
1
d
n
(W
T
,
).
The theorem and its proof can be extended to the case of the weight function
d
i=1
x
0
1/2
i
1i<jd
[x
i
x
j
[
1
(1 [x[)
1/2
,
and the dierential equation becomes a dierentialdierence equation.
Next we give explicit formulae for orthogonal bases. Let P
be the orthonormal
basis with respect to W
B
,
given in Theorem 4.2; then it is easy to check that
Orthogonal polynomials of several variables 167
P
2
: N
d
0
, [[ = n forms an orthonormal basis for 1
d
2n
(W
B
,
; Z
d
2
). Hence,
using the fact that C
(,)
2n
is given in terms of Jacobi polynomial, we get
Theorem 5.3. With respect to W
T
, the polynomials
P
(x) = [h
T
]
1
d
j=1
_
1 [x
j
[
1 [x
j1
[
_

j+1

p
(a
j
,b
j
)
j
_
2x
j
1 [x
j1
[
1
_
,
where a
j
= 2[
j+1
[ + [
j+1
[ +
dj1
2
and b
j
=
j
1
2
, are orthonormal and the
normalization constants h
T
are given by
[h
T
]
2
=
([[ +
d+1
2
)
2
d
j=1
(2[
j+1
[ +[
j
[ +
dj+2
2
)
2
j
.
On the simplex T
d
, it is often convenient to dene x
d+1
= 1[x[ and work with the
homogeneous coordinates (x
1
, . . . , x
d+1
). One can also derive a basis of orthogonal
polynomials that are homogeneous in the homogeneous coordinates. In fact, the
relation between orthogonal polynomials on B
d
and S
d
allows us to work with
hharmonics. Let us denote by H
d+1
2n
(h
2
, Z
d+1
2
) the subspace of hharmonics in
H
d+1
2n
(h
2
(x
2
1
, . . . , x
2
d+1
) : [[ =
n, N
d
0
be an orthonormal basis of H
d
2n
(h
2
, Z
d+1
2
). Then S
n
(x
1
, . . . , x
d+1
) :
[[ = n, N
d
0
forms an orthonormal homogeneous basis of 1
d
n
(W
T
).
For x T
d
let X = (x
1
, . . . , x
d
, x
d+1
) denote the homogeneous coordinates. Let
Y
)
in the homogeneous coordinates X := (x
1
, . . . , x
d+1
) with x
d+1
= 1 [x[,
P
(x) = X
F
B
_
, +
1
2
; 2[[ [[
d 3
2
;
1
x
1
, . . . ,
1
x
d+1
_
.
Furthermore, changing summation index shows that the above Lauricella function
of type B can be written as a constant multiple of the Lauricella function of type
A dened by
F
A
(c, ; ; x) =
(c)

()
()
!
x
, , N
d+1
0
, c R,
where the summation is taken over N
d+1
0
. This gives the following:
Theorem 5.4. For each N
d+1
0
with [[ = n, the polynomials
R
(x) = F
A
([[ +[[ +d, ; +1; X), x T
d
are orthogonal polynomials in 1
d
n
(W
T
) and
R
= (1)
n
(n +[[ +d)
n
( +1)
+p
, p
d
n1
,
168 Yuan Xu
where 1 = (1, . . . , 1) R
d+1
.
The polynomial R
for N
d+1
0
. The set R
: N
d+1
0
, [[ = n clearly has more elements
than it is necessary for a basis of 1
d
n
(W
T
). Let us write V
T
(x) = R
(,0)
(x). Then V
T
: N
d
0
, [[ = n is
the monomial basis of 1
d
n
(W
T
). The notation V
.
Theorem 5.5. For N
d
, [[ = n, the polynomials U
dened by
U
(x) = x
1
+1/2
1
. . . x
d
+1/2
d
(1 [x[)
d+1
+1/2

x
1
1
. . . x
d
d
x
1
+
1
1/2
1
. . . x
d
+
d
1/2
d
(1 [x[)
+
d+1
1/2
.
are polynomials in 1
d
n
(W
T
,
_
T
d
V
(x)U
(x)W
T
(x)dx =
( + 1/2)
(
d+1
+ 1/2)

([[ + (d + 1)/2)
2
!
,
.
Proof. It follows from the denition that U
_
T
d
V
(x)U
(x)W
T
(x)dx
= w
T
_
T
d
_

x
1
1
. . . x
d
d
V
(x)
_
d
i=1
x
i
+
i
1
2
i
(1 [x[)
+
d+1
1
2
dx.
Since V

x
1
1
. . . x
d
d
V
(x) = !
,
,
from which the stated formula follows. The fact that U
are biorthogonal to V
.
The correspondence between orthogonal polynomials in Theorem 5.1 also gives
an explicit formula for the reproducing kernel associated with W
T
.
Orthogonal polynomials of several variables 169
Theorem 5.6. Let = [[ + (d 1)/2. Then
P
n
(W
T
,
, x, y) =
2n +
_
[1,1]
d+1
C
2n
_
x
1
y
1
t
1
+ +
x
d+1
y
d+1
t
d+1
_
d+1
i=1
c
i
(1 +t
i
)(1 t
2
i
)
i
1
dt,
where x
d+1
= 1 [x[ and y
d+1
= 1 [y[.
Proof. Recall that we write W
T
,
for W
T
with
d+1
= . Using the correspondence
in Theorem 5.1, it can be veried that
P
n
(W
T
,
; x, y) =
1
2
d
Z
d
2
P
2n
_
W
B
,
; (
1
x
1
, . . . ,
d
x
d
), (
y
1
, . . . ,
y
d
)
_
.
Hence the stated formula follows from Theorem 4.5.
6. Classical type product orthogonal polynomials
As mentioned in the introduction, if W(x) is a product weight function
W(x) = w
1
(x
1
) . . . w
d
(x
d
), x R
d
,
then an orthonormal basis with respect to W is given by the product orthogonal
polynomials P
(x) = p
1
,1
(x
1
) . . . p
d
,d
(x
d
), where p
m,i
is the orthogonal polyno
mial of degree m with respect to the weight function w
i
. In this section we discuss
the product classical polynomials and some of their extensions.
6.1. Multiple Jacobi polynomials. Recall that the product Jacobi weight
functions is denoted by W
a,b
in Section 1.3. One orthonormal basis is given
by
P
(x) = p
(a
1
,b
1
)
1
(x
1
) . . . p
(a
d
,b
d
)
d
(x
d
),
where p
(a,b)
m
is the mth orthonormal Jacobi polynomial. Although this basis of
multiple Jacobi polynomials are simple, there is no close formula for the repro
ducing kernel P
n
(W
a,b
; x, y) in general. There is, however, a generating function
for P
n
(W
a,b
; x, 1), where 1 = (1, 1, . . . , 1) [1, 1]
d
. It is based on the generating
function (Poisson formula) of the Jacobi polynomials,
G
(a,b)
(r; x) :=
k=0
p
(a,b)
k
(1)p
(a,b)
k
(x)r
n
=
1 r
(1 +r)
a+b+2
2
F
1
_a+b+2
2
,
a+b+3
2
b + 1
;
2r(1 +x)
(1 +r)
2
_
, 0 r < 1,
170 Yuan Xu
which gives a generating function for the multiple Jacobi polynomials that can
be written in term of the reproducing kernel P
n
(W
a,b
) as
n=0
P
(W
a,b
; x, y)r
n
=
d
i=1
G
(a
i
,b
i
)
(r; x
i
) := G
(a,b)
d
(r; x).
Moreover, in some special cases, we can derive an explicit formula for P
n
(W; x, 1).
Let us consider the reproducing kernel for the case that a = b and a
i
are half
integers. This corresponds to the multiple Gegenbauer polynomials with respect
to the weight function
W
(x) =
d
i=1
(1 x
i
)
i
1/2
,
i
> 1/2, x [1, 1]
d
,
Recall that the generating function of the Gegenbauer polynomials is given by
1 r
2
(1 2rx +r
2
)
+1
=
n=0
+n
n
(x)r
n
.
Since
+n
n
(x) =
C
n
(1)
n
(x), it follows that P
n
(W; x, 1) satises a generating
function relation
(1 r
2
)
d
d
i=1
(1 2rx
i
+r
2
)
i
+1
=
n=0
P
n
(W
; x, 1)r
n
.
If N
0
, then the reproducing kernel can be given in terms of the divided
dierence, dened inductively by
[x
0
]f = f(x
0
), [x
0
, . . . x
m
]f =
[x
1
, . . . x
m
]f [x
0
, . . . x
m1
]f
x
m
x
0
.
The divided dierence [x
1
, . . . , x
d
]f is a symmetric function of x
1
, . . . , x
d
.
Theorem 6.1. Let
i
N
d
0
. Then
P
n
(W
; x, 1) = [
1
+1
..
x
1
, . . . , x
1
, . . . ,
d
+1
..
x
d
, . . . , x
d
]G
n
,
with
G
n
(t) = (1)
[
d+1
2
]
2(1 t
2
)
d1
2
_
T
n
(t) for d even,
U
n1
(t) for d odd.
Proof. In the case of
i
= 0, the left hand side of the generating function for
P
n
(W
0
; x, 1) can be expanded as a power series using the formula
[x
1
, . . . , x
d
]
1
a b()
=
b
d1
d
i=1
(a bx
i
)
,
Orthogonal polynomials of several variables 171
which can be proved by induction on the number of variables; the result is
(1 r
2
)
d
d
i=1
(1 2rx
i
+r
2
)
=
(1 r
2
)
d
(2r)
d1
[x
1
, . . . , x
d
]
1
1 2r() +r
2
=
(1 r
2
)
d
(2r)
d1
[x
1
, . . . , x
d
]
n=d1
U
n
()r
n
using the generating function of the Chebyshev polynomials of the second kind.
Using the binomial theorem to expand (1 r
2
)
d
and the fact that U
m+1
(t) =
sin m/ sin and sin m = (e
im
e
im
)/(2i) with t = cos , the last term can
be shown to be equal to
r=0
[x
1
, . . . , x
d
]G
n
r
n
.
For
i
> 1, we use the fact that (d
k
/dx
k
)C
n
(x) = 2
k
()
k
C
+k
nk
(x) and + n =
( +k) + (n k), which implies
d
k
dx
k
P
n
(W
; x, 1) = 2
k
( + 1)
k
P
nk
(W
+k
; x, 1),
so that the formula
d
dx
1
[x
1
, x
2
, . . . , x
d
]g = [x
1
, x
1
, x
2
, x
3
, . . . , x
d
]g
and the fact that the divided dierence is a symmetric function of its knots can
be used to nish the proof.
For multiple Jacobi polynomials, there is a relation between P
n
(W
a,b
; x, y) and
P
n
(W
a,b
; x, 1). This follows from the product formula of the Jacobi polynomials,
P
(,)
n
(x
1
)P
(,)
n
(x
2
)
P
(,)
n
(1)
=
_
0
_
1
0
P
(,)
n
(2A
2
(x
1
, x
2
, r, ) 1)dm
,
(r, ),
where > > 1/2,
A(x
1
, x
2
, r, )
=
1
2
_
(1 +x
1
)(1 +x
2
) + (1 x
1
)(1 x
2
)r
2
+ 2
_
1 x
2
1
_
1 x
2
2
r cos
_
1/2
and
dm
,
(r, ) = c
,
(1 r
2
)
1
r
2+1
(sin )
2
drd,
in which c
,
is a constant so that the integral of dm
,
over [0, 1] [0, ] is 1.
Sometimes the precise meaning of the formula is not essential, and the following
theorem of [12] is useful.
Theorem 6.2. Let a, b > 1. There is an integral representation of the form
p
(a,b)
n
(x)p
(a,b)
n
(y) = p
(a,b)
n
(1)
_
1
1
p
(a,b)
n
(t)d
(a,b)
x,y
(t), n 0,
172 Yuan Xu
with the real Borel measures d
(a,b)
x,y
on [1, 1] satisfying
_
1
1
[d
(a,b)
x,y
(t)[dt M, 1 < x, y < 1,
for some constant M independent of x, y, if and only if a b and a + b 1.
Moreover, the measures are nonnegative, i.e., d
(a,b)
x,y
(t) 0, if and only if b
1/2 or a +b 0.
The formula can be extended to the multiple Jacobi polynomials in an obvious
way, which gives a relation between P
n
(W
a,b
; x, y) and P
n
(W
a,b
; x, 1). Hence the
previous theorem can be used to give a formula of P
n
(W
Z
d
2
P
n
(W
0
; cos( +), 1),
where x = cos = (cos
1
, . . . , cos
d
) and y = cos = (cos
1
, . . . , cos
d
). The
vector + has components
i
+
i
i
.
6.2. Multiple Laguerre polynomials. The multiple Laguerre polynomials are
orthogonal with respect to W
L
(x) = x
e
x
with x R
d
+
. One orthonormal basis
is given by the multiple Laguerre polynomials P
(x) = L
1
. . . L
d
. Let us denote
this basis by P
(W
L
(W
T
,
; x) (set
d+1
= ).
Theorem 6.3. The multiple Laguerre polynomials associated to W
L
P
n
(W
T
+1/2,
; x/) =
L
1
(x
1
) . . .
L
d
(x
d
),
where
L
satisfy the
partial dierential equation
d
i=1
x
i
2
P
x
2
i
+
d
i=1
_
(
i
+ 1) x
i
_
P
x
i
_
= nP.
Proof. We make a change of variables x x/ in the equation satised by the
orthogonal polynomials with respect to W
T
,
and then divide by and take the
limit .
The limit relation, however, does not give an explicit formula for the reproducing
kernel. Just as in the case of multiple Jacobi polynomials, there is no explicit for
mula for the kernel P
n
(W
L
) satises
P
n
(W
L
; x, 0) = L
+d1
n
([x[), x R
d
+
.
Proof. The generating function of the Laguerre polynomials is
(1 r)
a1
exp
_
xr
1 r
_
=
n=0
L
a
n
(x)r
n
, [r[ < 1.
Since L
a
n
(0) = (a + 1)
n
/n! = L
a
n

2
2
, where the norm is taken with respect to
the normalized weight function x
a
e
x
, multiplying the formula gives a generating
function for P
n
(W
L
; x, 0),
n=0
P
n
(W
L
; x, 0)r
n
=
n=0
=n
(0)
(x)r
n
= (1 r)
d
e
xr/(1r)
,
which gives the stated formula.
There is also a product formula for the multiple Laguerre polynomials which gives
a relation between P
n
(W
L
; x, y) and P
n
(W
L
n
(x)L
n
(y) =
(n + + 1)2
(n + 1)
2
_
0
L
n
(x +y + 2
xy cos )e
xy cos
j
1
2
(
xy sin ) sin
2
d,
where j
i=1
x
0
i
1i<jd
[x
i
x
j
[
1
e
x
, x R
d
+
.
6.3. Multiple generalized Hermite polynomials. By generalized Hermite
polynomials we mean orthogonal polynomials with respect to the weight function
[x[
e
x
2
on R. For 0 the generalized Hermite polynomial H
n
(x) is dened
by
H
2n
(x) = (1)
n
2
2n
n!L
1/2
n
(x
2
),
H
2n+1
(x) = (1)
n
2
2n+1
n!xL
+1/2
n
(x
2
).
The normalization is chosen such that the leading coecient of H
n
is 2
n
. For
several variables, we consider the multiple generalized Hermite polynomials with
respect to W
H
(x) =
d
i=1
[x
i
[
i
e
x
2
,
i
0; evidently an orthogonal basis is
given by H
1
(x
1
) . . . H
d
(x
d
), and another basis can be given in polar coordinates
in terms of hspherical harmonics associated with h
(x) =
d
i=1
[x
i
[
i
. One can
also dene analogous of Appells biorthogonal bases.
Much of the information about these polynomials can be derived from the or
thogonal polynomials on the unit ball, since
lim
n/2
C
n
_
x
_
=
1
n!
H
n
(x).
Indeed, denote the orthonormal polynomials with respect to W
B
,
on B
d
in The
orem 4.2 by P
n
(W
B
,
; x), then it is easy to see the following:
Theorem 6.6. Let
H
n
denote the orthonormal generalized Hermite polynomials.
Then
lim
P
n
(W
B
,
; x/
) =
H
1
(x
1
) . . .
H
d
(x
d
).
Using this limit relation, it follows from the equation in Theorem 4.1 that the
polynomials in 1
d
n
(W
H
; x, y),
not even when y takes a special value. In fact, there is no special point to be
taken for R
d
and no convolution structure. What can be proved is a generating
function for the reproducing kernel (Mehler type formula):
Orthogonal polynomials of several variables 175
Theorem 6.7. For 0 < z < 1 and x, y R
d
n=0
P
n
(W
H
; x, y)z
n
=
1
(1 z
2
)
+d/2
exp
_
z
2
(x
2
+y
2
)
1 z
2
_
V
_
exp
_
2zx, )
1 z
2
__
(y),
where V
(x) = R
(x
2
1
, . . . , x
2
d
), just
as the relation between the orthogonal polynomials on B
d
and those on T
d
. Much
of the properties for the multiple Laguerre polynomials can be derived from this
relation and properties of the multiple Hermite polynomials. For example, there
is a counterpart of Mehlers formula for the multiple Laguerre polynomials.
The limiting relation similar to that in Theorem 6.6 holds for the orthogonal
basis in polar coordinates, which implies that the dierential equations also hold
for orthogonal polynomials with respect to the weight functions h
2
(x)e
x
2
. For
example, it holds for the type A weight functions
1i<jd
[x
i
x
j
[
2
e
x
2
, x R
d
and the type B weight functions
d
i=1
[x
i
[
2
0
1i<jd
[x
2
i
x
2
j
[
2
e
x
2
, x R
d
.
These two cases are related to the Schrodinger equations of the CalogeroSuther
land systems; these are exactly solvable models of quantum mechanics involving
identical particles in a one dimensional space. The eigenfunctions can be ex
pressed in terms of a family of homogeneous polynomials, called the nonsymmet
ric Jack polynomials, which are simultaneous eigenfunctions of a commuting set
of selfadjoint operators. Although there is no explicit orthogonal basis known for
these weight functions, there is a uniquely dened basis of orthogonal polynomials
for which the L
2
norm of the polynomials can be computed explicitly. The ele
ments of this remarkable family are labeled by partitions, and their normalizing
constants are proved using the recurrence relations and algebraic techniques.
176 Yuan Xu
7. Fourier orthogonal expansion
The nth partial sums of the Fourier orthogonal expansion do not converge for
continuous functions pointwisely or uniformly. It is necessary to consider summa
bility methods of the orthogonal expansions, such as certain means of the partial
sums. One important method are the Ces`aro (C, ) means.
Denition 7.1. Let c
n
n=0
be a given sequence. For > 0, the Ces`aro (C, )
means are dened by
s
n
=
n
k=0
(n)
k
(n )
k
c
k
.
The sequence c
n
is (C, ) summable by Ces`aros method of order to s if s
n
converges to s as n .
If = 0, then s
n
is the nth partial sum of c
n
and we write s
0
n
as s
n
.
7.1. hharmonic expansions. We start with the hharmonics. Using the re
producing kernel P
n
(h
2
n
(h
2
n
(h
2
; f, x) = c
h
_
S
d
f(y)K
n
(h
2
; x, y)h
2
(y)d(y),
where c
h
is the normalization constant of h
2
and K
n
(h
2
; x, y). If = 0, then K
n
(h
2
) is the nth
partial sum of P
k
(h
2
).
Proposition 7.1. Let f C(S
d
). Then the (C, ) means S
n
(h
2
; f, x) converge
to f(x) if
I
n
(x) := c
h
_
S
d
[K
n
(h
2
; x, y)[h
2
(y)d < ;
the convergence is uniform if I
n
(x) is uniformly bounded.
Proof. First we show that if p is a polynomial then S
n
(h
2
; p) converge uniformly to
p. Indeed, let S
n
(h
2
n
). It follows that
S
n
(h
2
; f) =
+n
n
k=0
(n)
k
(1 n)
k
S
k
(h
2
; f).
Assume p
d
m
. By denition, S
n
(h
2
; p) = p if n m. Hence,
S
n
(h
2
; p, x) p(x) =
+n
m1
k=0
(n)
k
(1 n)
k
_
S
k
(h
2
; p, x) p(x)
_
,
Orthogonal polynomials of several variables 177
which is of size O(n
1
) and converges to zero uniformly as n . Now the
denition of I
n
(x) shows that [S
n
(h
2
; f, x)[ I
n
(x)f
, where f
is the
uniform norm of f taken over S
d
. The triangular inequality implies
[S
n
(h
; f, x) f(x)[ (1 +I(x))f p
+[S
n
(h
; p, x) p(x)[.
Since f C(S
d
), we can choose p such that f p
< .
Recall that the explicit formula of the reproducing kernel is given in terms of the
intertwining operator. Let p
n
(w
(x) = (1x
2
)
1/2
. Then
p
n
(w
n
(x). Let = [[ +(d 1)/2, it follows
from Theorem 3.7 that
K
n
(h
2
; x, y) = V
_
p
n
(w
; x, ), 1)
(y).
Theorem 7.1. If 2[[+d, then the (C, ) means of the hharmonic expansion
with respect to h
2
is
positive if 2 + 1 (cf [3, p. 71]), and V
is a positive operator.
The positivity shows that I
n
(x) = 1 for all x, hence it implies that S
n
(h
2
; f)
converges uniformly to the continuous function f. For convergence, however,
positivity is not necessary. First we state an integration formula for the inter
twining operator.
Theorem 7.2. Let V
f(x)h
2
(x)d(x) = A
_
B
d+1
f(x)(1 x
2
)
1
dx,
for f L
2
(h
2
; S
d
) such that both integrals are nite. In particular, if g : R R
is a function such that all integrals below are dened, then
_
S
d
V
g(x, ))(y)h
2
(y)d(y) = B
_
1
1
g(tx)(1 t
2
)
+
d2
2
dt,
where A
and B
P
n
with respect to h
2
over S
d
is zero for n > 0,
and 1 for n = 0, so that the integral of V
f over S
d
is equal to the constant term
in the orthogonal expansion on B
d+1
. Of particular interest to us is the second
formula, which can also be derived from the FunkHecke formula in Theorem
3.8. It should be mentioned that this theorem holds for the intertwining operator
178 Yuan Xu
with respect to every reection group (with [[ replaced by
), even though an
explicit formula for the intertwining operator is unknown in general. The formula
plays an essential role in the proof of the following theorem.
Theorem 7.3. Let f C(S
d
). Then the Ces`aro (C, ) means of the hharmonic
expansion of f converge uniformly on S
d
provided > [[ + (d 1)/2.
Proof. Using the fact that V
_
[p
n
(w
; x, ), 1)[
(y)h
2
(y)d(y) = b
_
1
1
[p
n
(w
; 1, t)[w
(t)dt,
where b
converge if and
only if > nishes the proof.
The above theorem and its proof in fact hold for hharmonics with respect to
any reection group. It reduces the convergence of the hharmonics to that of
the Gegenbauer expansion. Furthermore, since sup
x
[I
n
(x)[ is also the L
1
norm
of S
n
(h
2
n
(h
2
; f) converges in
L
p
(h
2
; S
d
), 1 p < , in norm if > [[ + (d 1)/2.
It is natural to ask if the condition on is sharp. With the help of Theorem 7.2,
the above proof is similar to the usual one for the ordinary harmonics in the sense
that the convergence is reduced to the convergence at just one point. For the
ordinary harmonics, the underlying group is the orthogonal group and S
d
is its
homogeneous space, so reduction to one point is to be expected. For the weight
function h
(x) =
d+1
i=1
[x
i
[
i
, however, the underlying group Z
d+1
2
is a subgroup
of the the orthogonal group, which no longer acts transitively on S
d
. In fact, in
this case, the condition on is not sharp. The explicit formula of the reproducing
kernel for the product weight function allows us to derive a precise estimate for
the kernel K
n
(h
2
; x, y): for x, y S
d
and > (d 2)/2,
[K
n
(h
2
; x, y)[ c
_
d+1
j=1
([x
j
y
j
[ +n
1
[ x y[ +n
2
)
j
n
(d2)/2
([ x y[ +n
1
)
+
d+1
2
+
d+1
j=1
([x
j
y
j
[ +[ x y[
2
+n
2
)
j
n([ x y[ +n
1
)
d+1
_
,
where x = ([x
1
[, . . . , [x
d+1
[) and y = ([y
1
[, . . . , [y
d+1
[). This estimate allows us to
prove the sucient part of the following theorem:
Theorem 7.4. The (C, ) means of the hharmonic expansion of every continu
ous function for h
(x) =
d+1
i=1
[x
i
[
i
converge uniformly to f if and only if
> (d 1)/2 +[[ min
1id+1
i
.
Orthogonal polynomials of several variables 179
The sucient part of the proof follows from the estimate, the necessary part
follows from that of the orthogonal expansion with respect to W
B
,
in Theorem
7.7, see below. For = 0, the order (d 1)/2 is the socalled classical index
for the ordinary harmonics. The proof of the theorem shows that the maximum
of I
n
(x) appears on the great circles dened by the intersection of S
d
and the
coordinate planes. An estimate that takes the relative position of x S
d
into
consideration proves the following result:
Theorem 7.5. Let f C(S
d
). If > (d 1)/2, then S
n
(h
2
, f; x) converges to
f(x) for every x S
d
int
dened by
S
d
int
= x S
d
: x
i
,= 0, 1 i d + 1.
This shows that the great circles x S
d
: x
i
= 0 are like a boundary on the
sphere for summability with respect to h
2
; f) as n .
Theorem 7.6. Let = [[ + (d 1)/2. Then as n ,
S
n
(h
2
; ) = sup
f1
S
n
(h
2
; f)
= O
_
n
_
.
In particular, if f C
[]+1
(S
d
), then S
n
(h
2
; f) converge to f uniformly.
Proof. By the denition of S
n
(h
2
; f, x)[ b
_
1
1
k=0
k +
k
(t)
(t)dt f
.
The integral of the partial sum of the Gegenbauer polynomials is known to be
bounded by O(n
cn
[]1
, from which the convergence follows
from the fact that S
n
(h
; P) = P.
Again, this theorem holds for h
(x) =
d+1
i=1
[x
i
[
i
, the statement in Theorem 7.6
suggests the following conjecture:
S
n
(h
2
; ) = O
_
n
_
with =
d 1
2
+[[ min
1id+1
i
;
furthermore, for x S
d
int
, [S
n
(h
2
n
(h
2
n
(W; f). It follows that
S
n
(W; f, x) = c
_
n
(W; x, y)f(y)W(y)dy,
where = B
d
or T
d
, c is the constant dened by c
1
=
_
W(y)dy, and K
n
(W)
is the (C, ) means of the sequence P
n
(W). Using the correspondence in Theo
rem 3.1, most of the results for hharmonics can be extended to the orthogonal
expansions with respect to W
B
,
on the ball B
d
.
Theorem 7.7. The (C, ) means of the orthogonal expansion of every continuous
function with respect to W
B
,
converge uniformly to f if and only if
> (d 1)/2 +[[ + min
1
, . . . ,
d
, . (7.1)
Furthermore, for f continuous on B
d
, S
n
(W
B
,
, f; x) converges to f(x) for every
x B
d
int
, where
B
d
int
= x B
d
: x 1 and x
i
,= 0, 1 i d
provided > (d 1)/2.
Proof. The sucient part of the rst and the second statement follow from
Theorems 7.4 and 7.5, upon using the fact that
K
n
(W
B
,
; x, y) =
1
2
_
K
n
(h
2
; x, (y,
_
1 [y[
2
)) +K
n
(h
2
; x, (y,
_
1 [y[
2
))
_
and the elementary integration formula in the proof of Theorem 3.1. The neces
sary part of the rst statement uses the fact that the expansion reduces to the gen
eralized Gegenbauer expansion at certain points. Let w
,
(t) = [t[
2
(1 t
2
)
1/2
.
Denote by K
n
(w
,
; t, s) the (C, ) means of the kernel for the generalized Gegen
bauer expansion. Let = [[ + + (d 1)/2. From the explicit formula of the
reproducing kernel, it follows that
K
n
(W
B
,
; x, 0) = K
n
(w
,
; x, 0) and K
n
(W
B
,
; x,
i
) = K
n
(w
i
,
i
; x
i
, 1)
for i = 1, 2, . . . , d, so that the necessary condition can be derived from the con
vergence of the (C, ) means of the generalized Gegenbauer expansion s
n
(w
,
; f)
at 0 and 1. In fact, for continuous functions g on [1, 1], s
n
(w
,
; g) converge
uniformly to g if and only if > max, .
We can also state that the order of growth of S
n
(W
B
,
; )
is bounded by n
with = [[ ++(d 1)/2, just as in the Theorem 7.6, and conjecture that the
Orthogonal polynomials of several variables 181
sharp order is =
d1
2
+[[ + min
1id+1
i
with
d+1
= . For the classical
orthogonal polynomials with respect to W
B
= (1 x
2
)
1/2
, we have
S
n
(W
B
; ) = sup
f1
S
n
(W
B
; f)
n
+
d1
2
, 0.
The operator f S
n
(W
B
c n
d1
2
, d 2,
where c is a constant depending only on d. It turns out that the minimal norm
is obtained for the weight function W
B
; ) n
d1
2
.
That is, S
n
(W
B
,
i
< 0. It is
easy to conjecture, but likely hard to prove, that S
n
(W
B
,
; ) n
d1
2
if 0
and
i
0, 1 i d.
For orthogonal expansions with respect to W
T
Z
d
2
P
2n
_
W
B
,
; (
1
x
1
, . . . ,
d
x
d
), (
y
1
, . . . ,
y
d
)
_
,
the (C, ) means of the left hand side does not relate directly to that of the
(C, ) means of the right hand side. Much of the dierence can be seen already
in the case of d = 1, for which the weight function W
B
,
(t) = [t[
2
(1 t)
1/2
for t [1, 1] and W
T
,
(t) = t
1/2
(1 t)
1/2
for t [0, 1], and the latter one
is the classical Jacobi weight function (1 + t)
1/2
(1 t)
1/2
when converting
to t [1, 1]; thus, it is the dierence between the generalized Gegenbauer
expansion and the Jacobi expansion on [1, 1].
182 Yuan Xu
Theorem 7.9. Suppose the parameters of W
T
,
satisfy the conditions
d+1
i=1
_
2
i
[
i
]
_
1 + min
1id+1
i
with =
d+1
, (7.2)
where [x] stands for the largest integer part of x. Then the (C, ) means of the
orthogonal expansion of every continuous function with respect to W
T
,
converge
uniformly to f on T
d
if and only if (7.1) holds.
The necessary part of the theorem follows from the (C, ) means of the Jacobi
expansion without the additional condition (7.2). The proof of the sucient
part uses an explicit estimate of the kernel K
n
(W
T
,
; x, y) just as in the proof
of Theorem 7.7. However, there is an additional diculty for the estimate of
K
n
(W
T
,
; x, y), and condition (7.2) is used to simplify the matter. We note, that
the condition excludes only a small range of parameters. Indeed, if one of the
parameters, say
1
or , is 1/2, or if one of the parameters is 1, then the
condition holds. In particular, it holds for the unit weight function (
1
= . . . =
d+1
= 1/2). Naturally, we expect that the theorem holds for all
i
0 without
the condition.
For pointwise convergence, a theorem similar to that of Theorem 7.7 can be stated
for the interior of the simplex, but the proof in [21] puts a stronger restriction on
the parameters. We only state the case for the unit weight function.
Theorem 7.10. If f C(T
d
), then the (C, ) means of the orthogonal expansion
of f converge to f uniformly on each compact set contained in the interior of T
d
if > (d 1)/2.
For the unit weight function, the uniform convergence of the (C, ) means on T
d
holds if and only if > d 1/2.
7.3. Product type weight functions. The Fourier expansions for the product
type weight functions are quite dierent from those on the ball and on the simplex.
As we pointed out in Section 6, there is no explicit formula for the kernel function.
In the case of multiple Jacobi polynomials and multiple Laguerre polynomials, the
product formulae of the orthogonal polynomials lead to a convolution structure
that can be used to study the Fourier expansions.
Let us consider the multiple Jacobi polynomials. Denote the kernel function of
the Ces`aro means by K
n
(W
a,b
; x, y).
Theorem 7.11. In order to prove the uniform convergence of the (C, ) means of
the multiple Jacobi expansions for a continuous function, it is sucient to prove
Orthogonal polynomials of several variables 183
that, for a
j
b
j
> 1, a
j
+b
j
> 1,
_
[1,1]
d
[K
n
(W
a,b
; 1, y)[W
a,b
(y)dy c,
where c is a constant independent of n.
Proof. We know that the convergence of the (C, ) means follows from
_
[1,1]
d
[K
n
(W
a,b
; x, y)[W
a,b
(y)dy c, x [1, 1]
d
, n 0.
The product formula in Theorem 6.2 shows that that
K
n
(W
a,b
; x, y) =
_
[1,1]
d
K
n
(W
a,b
; t, 1)d
(a,b)
x,y
(t),
where the measure
(a,b)
x,y
is the product measure given in Theorem 6.2. This leads
to a convolution structure which gives the stated result using the corresponding
result for one variable.
This shows that uniform convergence is reduced to convergence at one point.
Multiplying the generating function of the multiple Jacobi polynomials by (1
r)
1
=
n=0
_
n+
n
_
r
n
gives
n=0
_
n +
n
_
K
n
(W
a,b
; x, 1)r
n
= (1 r)
1
G
(a,b)
d
(r; x).
This is the generating function of K
n
(W
a,b
; x, 1), which does not give the explicit
formula. What can be used to study the orthogonal expansion is the following:
Theorem 7.12. For d 1 and 0 r < 1,
K
n
(W
a,b
; x, 1) =
_
n +
n
_
1
1
r
n
_
(1 re
i
)
1
G
(a,b)
d
(re
i
; x)e
in
d.
Proof. Since both sides are analytic functions of r for [r[ < 1, the generating
function for K
n
(W
a,b
; x, y) holds for r being complex numbers. Replacing r by
re
i
, we get
n=0
_
n +
n
_
K
n
(W
a,b
; x, 1)r
n
e
in
= (1 re
i
)
1
G
(a,b)
d
(re
i
; x).
Hence, we see that
_
n+
n
_
K
n
(W
a,b
; x, 1)r
n
is the nth Fourier coecient of the
function (of ) in the right hand side.
With r = 1 n
1
, this expression allows us to derive a sharp estimate for the
kernel K
n
(W
a,b
; x, 1), which can be used to show that the integral in Theorem
7.13 is nite for larger than the critical index. One result is as follows:
184 Yuan Xu
Theorem 7.13. Let a
j
, b
j
1/2. The Ces`aro (C, ) means of the multiple
Jacobi expansion with respect to W
a,b
are uniformly convergent in the norm of
C([1, 1]
d
) provided >
d
j=1
maxa
j
, b
j
+
d
2
.
Similar results also hold for the case a
j
> 1, b
j
> 1 and a
j
+ b
j
1, 1
j d, with a properly modied condition on . In particular, if a
j
= b
j
= 1/2,
then the convergence holds for > 0.
The dierence between the Fourier expansion on [1, 1]
d
and the expansion on
B
d
or T
d
is best explained from the behavior of the multiple Fourier series on T
d
,
f
(f)e
ix
, where a
(f) =
_
T
d
f(x)e
ix
dx,
On the one hand, we have seen that the orthogonal expansion with respect to
the weight function (1 x
2
)
1/2
on B
d
is closely related to the spherical har
monic expansion, which is known to behave like summability of spherical multiple
Fourier series; that is, sums are taken over the 2 ball,
S
(2)
n
(f; x) =
n
a
(f)e
ix
= (D
(2)
n
f)(x),
where f g means the convolution of f and g and the Dirichlet kernel D
(2)
n
(x) =
g
n
(x) is a radial function (g
n
is given in terms of the Bessel function). In this
case, it is known that the (C, ) means converge if > (d 1)/2, the socalled
critical index.
On the other hand, the usual change of variables x
i
= cos
i
shows that the
summability in the case of
d
i=1
(1x
2
i
)
1/2
on [1, 1]
d
corresponds to summability
of multiple Fourier series in the 1 sense; that is,
S
(1)
n
(f; x) =

1
n
a
(f)e
ix
= (D
(1)
n
f)(x),
the Dirichlet kernel D
(1)
n
is given by (recall Theorem 6.1)
D
(1)
n
(x) = [cos x
1
, . . . , cos x
d
]G
n
.
In this case, the (C, ) means converge if > 0, independent of the dimension.
For the multiple Laguerre polynomials, there is also a convolution structure which
allows us to reduce the convergence of the (C, ) means to just one point, x = 0;
the proof is more involved since the measure is not positive. The result is as
follows:
Theorem 7.14. Let
i
0, 1 i d, and 1 p . The Ces`aro (C, )
means of the multiple Laguerre expansion are uniformly convergent in the norm
of C(R
d
+
) if and only if > [[ +d 1/2.
Orthogonal polynomials of several variables 185
For both multiple Jacobi expansions and multiple Laguerre expansions, the uni
form convergence is reduced to a single point, the corner point of the support
set of the weight function. In the case of the multiple Hermite expansions, the
support set is R
d
and there is no nite corner point. In fact, the convergence
in this case cannot be reduced to just a single point. Only the situation of the
classical Hermite expansions, that is, the case
i
= 0, is studied, see [32].
8. Notes and Literature
Earlier books on the subject are mentioned at the end of the Section 1. Many
historical notes on orthogonal polynomials of two variables can be found in Koorn
winder [15] and in Suetin [30]. The references given below are for the results in the
text. We apologize for any possible omission and refer to [10] for more detailed
references.
Section 2: The study of the general properties of orthogonal polynomials in
several variables appeared in Jackson [14] of 1936. In the paper [18] of 1967, Krall
and Sheer suggested that some of the properties can be restored if orthogonality
is taken in terms of orthogonal subspaces instead of a particular basis. The rst
vectormatrix form of the threeterm relation and Favards theorem appeared
in Kowalski [16, 17]; the present form and the theorem appeared in Xu [34,
35]. This form adopted the point of view of Krall and Sheer. Further studies
have been conducted in a series of papers; see the survey in [37] and the book
[10]. The study of Gaussian cubature formulae started with the classical paper
of Radon [26]. Signicant results on cubature formulae and common zeros of
orthogonal polynomials were obtained by Mysovskikh and his school [24] and
Moller [23]. Further study appeared in [36, 43]. The problem can be studied
using the language of polynomial ideals and varieties.
Section 3: Section 3.1 is based on [40]. Ordinary spherical harmonics appeared
in many books, for example, [1, 28, 33]. The hharmonics are introduced and
studied by Dunkl in a number of papers; see [6, 7, 8] and the references in [10].
A good reference for reection groups is [13]. The account of the theory of h
harmonics given in [10] is selfcontained. The case of the product weight function
in Section 3.2 is studied in [38], while the monomial basis contained in Subsection
3.2.3 is new [49].
Section 4, 5 and 6: The relation between orthogonal polynomials with respect
to (1x
2
)
(m1)/2
on B
d
and spherical harmonics on S
d+m
can be traced back to
the work of Hermite, Didon, Appell and Kampe de Feriet; see Chapt. XII, Vol. II,
of [11]. In the general setting, the relation is studied in [40] and further properties
are given in [45, 46]. In various special cases the explicit formulae for the classical
orthogonal polynomials on B
d
and on T
d
have appeared in the literature. The
relation between orthogonal polynomials on the simplex and those on the ball
186 Yuan Xu
or on the sphere has also appeared in special cases. It is studied in the general
setting in [41]. Except for the multiple Jacobi polynomials, all other classical type
orthogonal polynomials can be studied using hharmonics; see [47]. Apart from
some two dimensional examples (cf. [15]), classical and product type orthogonal
polynomials are the only cases for which explicit formulae are available.
The Hermite type polynomials of type A and type B are studied by Baker and
Forrester [4], Lassalle [19], Dunkl [9], and several other people. The commuting
selfadjoint operators that are used to dene the nonsymmetric Jack polynomials
are due to Cherednik. They are related to Dunkl operators. The nonsymmetric
Jack polynomials are dened by Opdam [25]. There are many other papers
studying these polynomials and CalogeroSutherland models.
Section 7: Summability of orthogonal expansion is an old topic, but most of
the results in this section are obtained only recently. See [42] for the expansion
of classical orthogonal polynomials on the unit ball, [20] for the product Jacobi
polynomials, [46] and [21] for hharmonics expansions and expansions on the unit
ball and on the simplex. The integration formula of the intertwining operator and
its application to summability appeared in [39]. The topic is still in its initial
stage, apart from the problems on the growth rate of the partial sums, many
questions such as those on L
p
and almost everywhere convergence have not been
studied.
Acknowledgement: I would like to thank the organizers of the workshop for inviting me
to give these lectures, and especially Wolfgang zu Castell for carefully reading through these
notes and making numerous corrections.
References
[1] G.E. Andrews, R. Askey, and R. Roy, Special functions, Encyclopedia of Mathematics and
its Applications 71, Cambridge University Press, Cambridge, 1999.
[2] P. Appell and J.K. de Feriet, Fonctions hypergeometriques et hyperspheriques, Polynomes
dHermite, GauthierVillars, Paris, 1926.
[3] R. Askey, Orthogonal polynomials and special functions, Regional Conference Series in
Applied Mathemathics 21, SIAM, Philadelphia, 1975.
[4] T.H. Baker and P.I. Forrester, Nonsymmetric Jack polynomials and integral kernels, Duke
Math. J. 95 (1998), 150.
[5] H. Berens and Yuan Xu, Fejer means for multivariate Fourier series, Math. Z. 221 (1996),
449465.
[6] C.F. Dunkl, Reection groups and orthogonal polynomials on the sphere, Math. Z. 197
(1988), 3360.
[7] , Dierentialdierence operators associated to reection groups, Trans. Amer.
Math. Soc. 311 (1989), 167183.
[8] , Integral kernels with reection group invariance, Canad. J. Math. 43 (1991),
12131227.
Orthogonal polynomials of several variables 187
[9] , Orthogonal polynomials of types A and B and related Calogero models, Comm.
Math. Phys. 197 (1998), 451487.
[10] C.F. Dunkl and Yuan Xu, Orthogonal Polynomials of Several Variables, Encyclopedia of
Mathematics and its Applications 81, Cambridge University Press, Cambridge, 2001.
[11] A. Erdelyi, W. Magnus, F. Oberhettinger and F. G. Tricomi, Higher transcendental func
tions, Vol. 2, McGrawHill, New York, 1953.
[12] G. Gasper, Banach algebras for Jacobi series and positivity of a kernel, Ann. Math. 95
(1972), 261280.
[13] J.E. Humphreys, Reection groups and Coxeter groups, Cambridge Studies in Advanced
Mathematics 29, Cambridge University Press, Cambridge, 1990.
[14] D. Jackson, Formal properties of orthogonal polynomials in two variables, Duke Math. J.
2 (1936), 423434.
[15] T. Koornwinder, Twovariable analogues of the classical orthogonal polynomials, In: R.
Askey (ed.), Theory and applications of special functions, 435495, Academic Press, New
York, 1975.
[16] M.A. Kowalski, The recursion formulas for orthogonal polynomials in n variables, SIAM
J. Math. Anal. 13 (1982), 309315.
[17] , Orthogonality and recursion formulas for polynomials in n variables, SIAM J.
Math. Anal. 13 (1982), 316323.
[18] H.L. Krall and I.M. Sheer, Orthogonal polynomials in two variables, Ann. Mat. Pura
Appl. 76(4) (1967), 325376.
[19] M. Lassalle, Polynomes de Hermite generalises. (French) C. R. Acad. Sci. Paris Ser. I
Math. 313 (1991), 579582.
[20] Zh.K. Li and Yuan Xu, Summability of product Jacobi expansions, J. Approx. Theory
104 (2000), 287301.
[21] , Summability of orthogonal expansions on spheres, balls and simplices, J. Approx.
Theory 122 (2003), 267333.
[22] B. Logan and I. Shepp, Optimal reconstruction of a function from its projections, Duke
Math. J. 42 (1975), 649659.
[23] H.M. Moller, Kubaturformeln mit minimaler Knotenzahl Numer. Math. 25 (1976), 185
200.
[24] I.P. Mysovskikh, Interpolatory cubature formulas (in Russian), Nauka, Moscow, 1981.
[25] E.M. Opdam, Harmonic analysis for certain representations of graded Hecke algebras, Acta
Math. 175 (1995), 75121.
[26] J. Radon, Zur mechanischen Kubatur, Monatsh. Math. 52 (1948), 286300.
[27] M. Rosler, Positivity of Dunkls intertwining operator, Duke Math. J., 98 (1999), 445463.
[28] E.M. Stein and G. Weiss, Introduction to Fourier analysis on Euclidean spaces, Princeton
Univ. Press, Princeton, NJ, 1971.
[29] A. Stroud, Approximate calculation of multiple integrals, PrenticeHall, Englewood Clis,
NJ, 1971.
[30] P.K. Suetin, Orthogonal polynomials in two variables, translated from the 1988 Russian
original by E. V. Pankratiev, Gordon and Breach, Amsterdam, (1999).
[31] B. S undermann, On projection constants of polynomial space on the unit ball in several
variables, Math. Z. 188 (1984), 111117.
[32] S. Thangavelu, Lectures on Hermite and Laguerre expansions, Princeton University Press,
Princeton, NJ, 1993.
[33] N.J. Vilenkin, Special functions and the theory of group representations, American Math
ematical Society Translation of Mathematics Monographs 22, American Mathematical
Society, Providence, RI, 1968.
188 Yuan Xu
[34] Yuan Xu, On multivariate orthogonal polynomials, SIAM J. Math. Anal. 24 (1993), 783
794.
[35] , Multivariate orthogonal polynomials and operator theory, Trans. Amer. Math.
Soc. 343 (1994), 193202.
[36] , Common zeros of polynomials in several variables and higher dimensional quad
rature, Pitman Research Notes in Mathematics Series 312, Longman, Harlow, 1994.
[37] , On orthogonal polynomials in several variables, in Special functions, qseries and
related topics, 247270, The Fields Institute for Research in Mathematical Sciences, Com
munications Series 14, American Mathematical Society, Providence, RI, 1997.
[38] , Orthogonal polynomials for a family of product weight functions on the spheres,
Canad. J. Math. 49 (1997), 175192.
[39] , Integration of the intertwining operator for hharmonic polynomials associated to
reection groups, Proc. Amer. Math. Soc. 125 (1997), 29632973.
[40] , Orthogonal polynomials and cubature formulae on spheres and on balls, SIAM J.
Math. Anal. 29 (1998), 779793.
[41] , Orthogonal polynomials and cubature formulae on spheres and on simplices, Meth
ods Anal. Appl. 5 (1998), 169184.
[42] , Summability of Fourier orthogonal series for Jacobi weight on a ball in R
d
, Trans.
Amer. Math. Soc. 351 (1999), 24392458.
[43] , Cubature formulae and polynomial ideals, Adv. Appl. Math. 23 (1999), 211233.
[44] , Harmonic polynomials associated with reection groups, Canad. Math. Bull. 43
(2000), 496507.
[45] , FunkHecke formula for orthogonal polynomials on spheres and on balls, Bull.
London Math. Soc. 32 (2000), 447457.
[46] , Orthogonal polynomials and summability in Fourier orthogonal series on spheres
and on balls, Math. Proc. Cambridge Phil. Soc. 31 (2001), 139155.
[47] , Generalized classical orthogonal polynomials on the ball and on the simplex,
Constr. Approx., 17 (2001), 383412.
[48] , Representation of reproducing kernels and the Lebesgue constants on the ball, J.
Approx. Theory, 112 (2001), 295310.
[49] , Monomial orthogonal polynomials of several variables, submitted.