SQUARE SUMMABILITY WITH GEOMETRIC WEIGHT FOR

CLASSICAL ORTHOGONAL EXPANSIONS

Dmitrii Karp
Institute of Applied Mathematics
7 Radio Street, Vladivostok, 690041, Russia
dmkrp@yandex.ru

Abstract Let fk be the k-th Fourier coefficient of a function f in terms of the orthonormal Hermite, Laguerre or
P
Jacobi polynomials. We give necessary and sufficient conditions on f for the inequality k |fk |2 θk < ∞ to
hold with θ > 1. As a by-product new orthogonality relations for the Hermite and Laguerre polynomials are
found. The basic machinery for the proofs is provided by the theory of reproducing kernel Hilbert spaces.

Keywords: Hermite polynomials, Laguerre polynomials, Jacobi polynomials, square summability, orthogonality, reproducing
kernel, Szegö space, spaces of entire functions.

1. Introduction. The goal of this paper is to find necessary and sufficient conditions to be imposed
on a function f for its Fourier coefficients in terms of classical orthogonal polynomials to satisfy the
inequality
X∞
|fk |2 θk < ∞ (1)
k=0
with θ > 1. So, we have three problems corresponding to the following three definitions of fk :
Z∞
2
fk = f (x)Hk (x)e−x dx, (2)
−∞

Z∞
fk = f (x)Lνk (x)xν e−x dx, (3)
0
and
Z1
fk = f (x)Pα,β α β
k (x)(1 − x) (1 + x) dx. (4)
−1

Here Hk , Lνk and Pα,β

k is the k-th orthonormal polynomial of Hermite, Laguerre and Jacobi, respectively
[Szego91]. For the sake of convenience we use orthonormal instead of standardly normalized versions of
the classical polynomials. In each case f is defined on the interval of orthogonality of the corresponding
system of polynomials. We will use ϕk as a generic notation for either of the three types of polynomials.
Classes of functions with rapidly decreasing Fourier coefficients in classical orthogonal polynomials
have been extensively studied. We only mention a few contributions without any attempt to make a
survey. The series of papers [Hille39]-[Hille80] by E. Hille is devoted to the Hermite expansions. Among
other things Hille studied expansions with fk vanishing as quick as exp(−τ (2k + 1)1/2 ). The functions
possessing such coefficients are holomorphic in the strip |=z| < τ . Hille provided an exact description
of the linear vector space with compact convergence topology formed by these functions where the set
{Hk : k ∈ N ∪ 0} is a basis. Its members are characterized by a suitable growth condition. In addition,
he studied the convergence on and analytic continuation through the boundary of the strip.

1
2

The convergence domain of the Laguerre series with fk ∼ exp(−τ k 1/2 ) is the interior of the parabola
<(−z)1/2 = τ /2. Rusev in [Rusev84] described the linear vector space with compact convergence
topology formed by functions holomorphic in {z : <(−z)1/2 < τ /2} where the set {Lνk : k ∈ N ∪ 0} is a
basis, and Boyadjiev in [Boyad92] studied the behavior of the Laguerre series on the boundary of the
convergence domain.
If coefficients (2) or (3) decrease as fast as exp(−τ k η ) with η > 1/2 the function f is entire. The
spaces comprising such functions for Hermite expansions have been characterized by Janssen and van
Eijndhoven in [JanEijnd90]. If the Fourier-Hermite coefficients fk decline quicker than any geometric
progression, the suitable characterization is provided by Berezanskij and Kondratiev in [BerKondr95]
(see Corollary 1.1 below). Byun was the first to study the Hermite expansions with condition (1) -
see remark after Theorem 1. For the Laguerre expansions with lim supk |fk |1/k < 1 Zayed related the
P
singularities of the Borel transform of f with those of F (z) = k fk z k in [Zayed81].
For the Legendre expansions with lim supk |fk |1/k = ζ < 1 Nehari relates the singularities of f on
P
the boundary of convergence domain {z : |z + 1| + |z − 1| < ζ + ζ −1 } to those of F (z) = k fk z k
in [Nehari56]. Gilbert in [Gilbert64] and Gilbert and Howard in [GilbHoward66] generalized the results
of Nehari to the Gegenbauer expansions and to expansions in eigenfunctions of a Sturm-Liouville
operator.
1 1
Functions satisfying (1) apparently form a proper subclass of functions with lim supk |fk | k ≤ θ− 2 . On
the other hand the condition (1) itself cannot be expressed in terms of asymptotics of fk . Consequently,
our criteria for the validity of (1) are of different character from those contained in the above references.
Although they also describe the growth of f for the Hermite and Laguerre expansions and its boundary
behaviour for the Jacobi expansions, our growth conditions are given in terms of existence of certain
weighted area integrals of f and cannot be expressed by an estimate of the modulus, while the restriction
on the boundary behaviour is given on the whole boundary and not in terms of analysis of individual
singularities.

2. Preliminaries. Throughout the paper the following standard notation will be used: N, R, R+
and C will denote positive integers, real numbers, positive real numbers and the finite complex plane,
respectively. Since coefficients (2)-(4) do not change if we modify f on a set of Lebesgue measure zero,
all our statements about the properties of f should be understood to hold almost everywhere. If, for
instance, we say that f is the restriction of a holomorphic function to (some part of) the real axis, it
means that f is allowed to differ from such restriction on a set of zero measure.
All proofs in the paper hinge on the theory of reproducing kernel Hilbert spaces (RKHS), so for
convenience we briefly outline the basic facts of the theory we will make use of.
For a Hilbert space H comprising complex-valued functions on a set E, the reproducing kernel
K(p, q) : E × E → C is a function that belongs to H as a function of p for every fixed q ∈ E and
possesses the reproducing property
f (q) = (f, K(·, q))H
for every f ∈ H and for any q ∈ E. If a Hilbert space admits the reproducing kernel then this kernel
is unique and positive definite on E × E:
n
X
K(pi , pj )ci cj ≥ 0 (5)
i,j=1

for an arbitrary finite complex sequence {ci } and any points pi ∈ E. The theorem of Moore and
Aronsjain [Aron50] states that the converse is also true: every positive definite kernel K on E × E
uniquely determines a Hilbert space H admitting K as its reproducing kernel. This fact justifies the
notation HK for the Hilbert space H induced by the kernel K. The following propositions can be found
in [Aron50, Saitoh97].
Square summability with geometric weight for classical orthogonal expansions 3

Proposition 1 If HK is a Hilbert space of functions E → C and s is an arbitrary non-vanishing

function on E, then
Ks (p, q) = s(p)s(q)K(p, q) (6)
is the reproducing kernel of the Hilbert space HKs comprising all functions on E expressible in the form
fs (p) = s(p)f (p) with f ∈HK and equipped with inner product
µ ¶
fs gs
(fs , gs )HKs = , . (7)
s s HK

Proposition 2 Let E1 ⊂ E and K1 be the restriction of a positive definite kernel K to E1 × E1 . Then

the RKHS HK1 comprises all restrictions to E1 of functions from HK and has the norm given by

kf1 kHK1 = min {kf kHK ; f |E1 = f1 , f ∈ HK } . (8)

Proposition 3 If RKHS HK is separable and {ψk : k ∈ N} is a complete orthonormal system in HK ,

then its reproducing kernel is expressed by
X
K(p, q) = ψk (p)ψk (q), (9)
k

where the series (9) converges absolutely for all p, q ∈ E and uniformly on every subset of E, where
K(q, q) is bounded.

The relation K1 ¿ K2 will mean that K2 − K1 is positive definite. This relation introduces partial
ordering into the set of positive definite kernels on E × E. Inclusion HK1 ⊂ HK2 and equality HK1 =
HK2 will be understood in the set-theoretic sense, which implies, however, that the same relations hold
in the topological sense as stated in the following two propositions.

Proposition 4 Inclusion HK1 ⊂ HK2 takes place iff K1 ¿ M K2 for a constant M > 0. In this case
M 1/2 kf k1 ≥ kf k2 for all f ∈ HK1 . (Here kf k1 and kf k2 are the norms in HK1 and HK2 , respectively).

Proposition 5 Equality HK1 = HK2 takes place iff mK2 ¿ K1 ¿ M K2 for some positive constants
m, M . In this case m1/2 kf k1 ≤ kf k2 ≤ M 1/2 kf k1 .

The kernels satisfying Proposition 5 are said to be equivalent which is denoted by K1 ≈ K2 .

It is shown in [Aron50] that the RKHS HK induced by the kernel K(p, q) = K1 (p, q)K2 (p, q) consists
of all restrictions to the diagonal of E 0 = E × E (i.e. the set of points of the form {p, p}) of the elements
of the tensor product H 0 = HK1 ⊗ HK2 . The space HK is characterized by

Proposition 6 Let K(p, q) = K1 (p, q)K2 (p, q) and let {ψk : k ∈ N} be a complete orthonormal set in
HK2 . Then the RKHS HK comprises the functions of the form
∞
X ∞
X
f (p) = fk1 (p)ψk (p), fk1 ∈ HK1 , kfk1 k21 < ∞. (10)
k=1 k=1

The norm in HK is given by ( )

∞
X
kf k2HK = min kfk1 k21 ,
k=1

where the minimum is taken over all representations of f in the form (10) and is attained on one such
representation.
4

3. Results for the Hermite and Laguerre expansions. Functions satisfying (1) form a Hilbert
space with inner product
∞
X
(f, g) = fk gk θk . (11)
k=0
This space will be denoted by Hθ for the Hermite expansions and by Lνθ for the Laguerre expansions.
The sets {θ−k/2 Hk }k∈N∪0 and {θ−k/2 Lνk }k∈N∪0 constitute orthonormal bases of the spaces Hθ and Lνθ ,
respectively. For each space we can form the reproducing kernel according to (9):
X
K(z, u) = ϕk (z)ϕk (u)θ−k . (12)
k

The explicit formulae for these kernels are known to be [Bateman53]:

X∞ µ ¶
def θ 2zuθ − z 2 − u2
HKθ (z, u) = Hk (z)Hk (u)θ−k = p exp (13)
π(θ2 − 1) θ2 − 1
k=0

(Mehler’s formula) and

∞ µ ¶ Ã √ !
def
X θν/2+1 z+u θzu
− ν2
LKθν (z, u) = Lνk (z)Lνk (u)θ−k = exp − (zu) Iν 2 (14)
θ−1 θ−1 θ−1
k=0

(Hardy-Hille’s formula). Here Iν is the modified Bessel function. The kernels (13) and (14) are entire
functions of both z and u. The space Hθ comprises functions on R, while the space Lνθ comprises
functions on R+ . Hence, we consider the restrictions of the kernels (13) and (14) to R and R+ ,
respectively. Applying Proposition 2 with E = C and E 0 = R or E 0 = R+ we conclude that the spaces
Hθ and Lνθ are formed by all restrictions to R and R+ , respectively, of entire functions from the spaces
generated by the kernels (13) and (14). We can drop minimum in (8) due to uniqueness of analytic
extension and, consequently, the norm induced by inner product (11) equals the norm in HHKθ or in
HLKθν .
Next, we observe that both kernels (13) and (14) are of the form s(z)s(u)K(zu) with non-vanishing
2 2
functions sH (z) = e−z /(θ −1) and sL (z) = e−z/(θ−1) . Hence we are in the position to apply Proposition 1.
In compliance with (7) the norms in HHKθ and HLKθν are known once we have found the norms in the
spaces induced by the kernels
µ ¶
g θ 2zuθ
HK θ (zu) = p exp (15)
π(θ2 − 1) θ2 − 1
and à √ !
f νθ (zu) θν/2+1 ν θzu
LK = (zu)− 2 Iν 2 . (16)
θ−1 θ−1
Both of them depend on the product zu and are rotation invariant thus. Rotation invariance of the
kernel implies radial symmetry of the measure with respect to which the integral representing the norm
is taken. For the kernel (15) the measure and the space are well-known. It is the Fischer-Fock (or the
Bargmann-Fock) space Fθ of entire functions with finite norms
Z µ ¶
2 2 2 2θ|z|2
kf kFθ = p |f (z)| exp − 2 dσ, (17)
π(θ2 − 1) θ −1
C

where the integration is with respect to Lebesgue’s area measure. By Proposition 1 the final result for
the Hermite expansions now becomes straightforward.
Theorem 1 Inequality (1) with fk defined by (2) holds true for all restrictions to R of the entire
functions with Z · µ ¶¸
2 (<z)2 (=z)2
|f (z)| exp −2 + dσ < ∞ (18)
θ+1 θ−1
C
Square summability with geometric weight for classical orthogonal expansions 5

and only for them.

For inner product (11) this leads to the expression

Z · µ ¶¸
2 (<z)2 (=z)2
(f, g)Hθ = (f, g)HHKθ = p f (z)g(z) exp −2 + dσ. (19)
π(θ2 − 1) θ+1 θ−1
C

Since polynomials Hk are orthogonal with respect to this inner product, we obtain the following
orthogonality relation for the standardly normalized Hermite polynomials Hk :
Z · ¸
2 2(<z)2 2(=z)2
p Hk (z)Hm (z) exp − − dσ = δk,m π(2θ)k k!. (20)
(θ2 − 1) θ+1 θ−1
C

These results for the Hermite expansions have been essentially proved by Du-Wong Byun in [Byun93],
although the emphasis in his work is different and it seems that the orthogonality relation (20) has not
been noticed.
The following corollary is an immediate consequence of (18).

Corollary 1.1 Inequality (1) with fk defined by (2) holds true for all θ > 1 iff f is the restriction to R
2
of an entire function F satisfying |F (z)| ≤ Ceε|z| for all ε > 0 and a constant C = C(ε) independent
of z.

A direct proof is given in [BerKondr95].

For the Laguerre expansions the situation is a bit more complicated. The kernel (16) is a particular
case of a much more general hypergeometric kernel. The spaces generated by hypergeometric kernels
are studied in depth in [Karp03]. For the kernel (16) we get the space of entire functions with finite
norms à √ !
2θ −ν/2 Z 2 θ|z|
kf k2LK
fν = |f (z)|2 |z|ν Kν dσ, (21)
θ π(θ − 1) θ−1
C

where Kν is the modified Bessel function of the second kind (or the McDonald function). Application
of Proposition 1 brings us to our final result for the Laguerre expansions.

Theorem 2 Inequality (1) with fk defined by (3) holds true for all restrictions to R+ of the entire
functions with à √ !
Z µ ¶
2 2<z ν 2 θ|z|
|f (z)| exp |z| Kν dσ < ∞ (22)
θ−1 θ−1
C
and only for them.

The orthogonality relation for the standardly normalized Laguerre polynomials Lνk that follows from
this result is given by
Z µ ¶ Ã √ !
2θ−ν/2 ν 2<z ν 2 θ|z| Γ(k + ν + 1)θk
Lk (z)Lνm (z) exp |z| Kν dσ = δk,m . (23)
π(θ − 1) θ−1 θ−1 k!
C

Corollary 2.1 Inequality (1) with fk defined by (3) holds true for all θ > 1 iff f is the restriction
to R+ of an entire function F satisfying |F (z)| ≤ Ceε|z| for all ε > 0 and a constant C = C(ε)
independent of z.

This corollary can be easily derived from (22) with the help of asymptotic relation [Bateman53]
r
π −x
Kν (x) ' e , x → ∞.
2x
6

4. Results for the Jacobi expansions. The space of complex-valued functions on (−1, 1) whose
Fourier-Jacobi coefficients (4) satisfy (1) will be denoted by Jθα,β . Pursuing the same line of argument
as in the previous section, we form the reproducing kernel of this space found by Baily’s formula
[Bateman53]
X∞
def 1 θα+β+1 (θ − 1)
θ
JKα,β (z, u) = Pα,β α,β
k (z)Pk (u) θ k = τ (α, β)(θ + 1)α+β+2 ×
k=0
µ ¶
α β α β 3 θ(1 − z)(1 − u) θ(1 + z)(1 + u)
×F4 + + 1, + + ; α + 1, β + 1; , , (24)
2 2 2 2 2 (θ + 1)2 (θ + 1)2
where
∞
X
0 (a)n+m (b)n+m m n
F4 (a, b; c, c ; t, s) = t s (25)
(c)n (c0 )m m!n!
m,n=0

is Appel’s hypergeometric function and τ (α, β) = 2α+β+1 Γ(α + 1)Γ(β + 1)/Γ(α + β + 2). Define the
ellipse Eθ by
Eθ = {z : |z − 1| + |z + 1| < θ1/2 + θ−1/2 }. (26)
Our first observation here is that both the series on the right hand side and on the left hand side of
(24) converge absolutely and uniformly on compact subsets of Eθ × Eθ (see [Karp00]). The implication
of the uniform convergence is holomorphy of the kernel (24) in Eθ × Eθ with respect to both variables.
Application of Proposition 2 with E = Eθ , E 0 = (−1, 1) leads to the assertion that the space Jθα,β is
formed by restrictions to the interval (−1, 1) of functions holomorphic in Eθ thereby the norm in Jθα,β
equals the norm in HJK θ due to uniqueness of analytic continuation.
α,β
Our main result for the Jacobi expansions will be derived from its particular case α = β = λ − 1/2
thereby the orthonormal Jacobi polynomials reduce to the orthonormal Gegenbauer polynomials Cλk .
The reproducing kernel (24) reduces in this case to
∞
X
def 1
GKλθ (z, u) = Cλk (z)Cλk (u) =
θk
k=0
µ ¶
θ2λ (θ2 − 1) λ+1 λ+2 1 4θ2 (1 − z 2 )(1 − u2 )
= 2 F1 , ;λ + ; , (27)
τ (λ)(θ2 − 2θzu + 1)λ+1 2 2 2 (θ2 − 2θzu + 1)2
√
where 2 F1 is the Gauss hypergeometric function and τ (λ) = πΓ(λ+1/2)/Γ(λ+1). The series on both
sides of (27) again converge absolutely and uniformly on compact subsets of Eθ × Eθ . Formula (27) can
be obtained from (24) by using a reduction formula for F4 and applying a quadratic transformation to
the resulting hypergeometric function. Details are in [Karp00].
Let ∂Eθ denote the boundary of the ellipse Eθ . We introduce the weighted Szegö space AL2 (∂Eθ ; ρ)
with continuous positive weight ρ(z) defined on ∂Eθ as the set of functions holomorphic in Eθ ,
possessing non-tangential boundary values almost everywhere on ∂Eθ and having finite norms
Z
kf k2AL2 (∂Eθ ;ρ) = |f (z)|2 ρ(z)|dz| < ∞. (28)
∂Eθ

We will write AL2 (∂Eθ ) for AL2 (∂Eθ ; 1).

For λ = 0, the orthonormal Gegenbauer polynomials reduce to the orthonormal Chebyshev polynomials
of the first kind Tk :
1 1
C0k (z) = Tk (z) = (2/π) 2 Tk (z) = (2/π) 2 cos(k arccos z), k ∈ N,
1 1
(29)
C00 (z) = T0 (z) = (1/π) 2 T0 (z) = (1/π) 2 ,

where Tk is the k-th Chebyshev polynomial of the first kind in standard normalization.
Square summability with geometric weight for classical orthogonal expansions 7
1
Lemma 1 Polynomials (θk + θ−k )−1/2 Tk /2 form orthonormal basis of the space AL2 (∂Eθ ; |z 2 − 1|− 2 ).

Proof. The boundary ∂Eθ is an analytic arc which implies the completeness of the set of all polynomials
1
in AL2 (∂Eθ ; |z 2 −1|− 2 ) (see for instance [Gaier80]). To prove orthogonality we will need some properties
√
of the Zhukowskii function z = (w + w−1 )/2. This function maps the annulus 1 < |w| < θ one-to-
√
one and conformally onto Eθ cut along the interval (−1, 1) thereby the circle |w| = θ corresponds
√
to ∂Eθ . The inverse function is given by w = z + z 2 − 1, where the principal value of the square
root is to be chosen. We see by differentiation that the infinitesimal arc lengths are connected by the
relation |dz| = |w2 − 1||dw|/(2|w|2 ). Note also that z 2 − 1 = (w2 − 1)2 /(4w2 ). Applying the identity
Tk (z(w)) = wk + w−k we get by the substitution z = (w + w−1 )/2:
Z Z µ ¶µ ¶
1 2 − 12 1 k 1 m 1 |dw|
Tk (z)Tm (z)|z − 1| |dz| = w + k w + m =
2π 2π √
w w |w|
∂Eθ |w|= θ

Z2π −(k+m)/2 Z
2π
(k−m)/2 Z
2π
θ(k+m)/2 θ θ
= eiϕ(k−m) dϕ + eiϕ(m−k) dϕ + eiϕ(k+m) dϕ+
2π 2π 2π
0 0 0

Z2π 
 0, k 6= m,
θ(m−k)/2 −iϕ(k+m) k −k
+ e dϕ = θ + θ , k = m 6= 0,
2π 

0 4, k = m = 0.
Combined with (29) this proves the lemma.¤
λ−1/2,λ−1/2
Denote Gθλ = Jθ . We are ready to formulate our main result for the Gegenbauer expansions.

Theorem 3 Let λ ≥ 0. The space Gθλ is formed by all restrictions of the elements of AL2 (∂Eθ ) to the
interval (−1, 1). The norms in Gθλ and AL2 (∂Eθ ) are equivalent.

Proof. The proof will be divided in three steps.

Step 1. For the space HGK θ induced by the kernel (27) with λ = 0 we want to prove that
0

HGK θ = AL2 (∂Eθ ). (30)

0

1
The weight |z 2 − 1|−1/2 is positive and continuous on ∂Eθ so the norms in AL2 (∂Eθ ; |z 2 − 1|− 2 ) and
AL2 (∂Eθ ) are equivalent and these spaces coincide elementwise. According to Lemma 1 and formula
1
(9) the space AL2 (∂Eθ ; |z 2 − 1|− 2 ) admits the reproducing kernel given by
∞
1 X Tk (z)Tk (u)
Rθ (z, u) = .
4 θk + θ−k
k=0

This kernel is equivalent to the kernel GK0θ due to (29) and inequalities
n n X
∞ n
1 X X Tk (zi )Tk (zj ) X
GK0θ (zi , zj )ci cj ≤ c c
i j ≤ GK0θ (zi , zj )ci cj
2 θk + θ−k
i,j=0 i,j=0 k=0 i,j=0

satisfied for any choice of n ∈ N, ci ∈ C and zi ∈ Eθ . Hence by Proposition 5 our claim is proved.
Step 2. Consider the following auxiliary kernel:
µ ¶
θ π −1 (θ2 − 1) λ+1 λ+2 1 4θ2 (1 − z 2 )(1 − u2 )
K̂λ (z, u) = 2 2 F1 , ;λ + ; . (31)
θ − 2θzu + 1 2 2 2 (θ2 − 2θzu + 1)2
It is positive definite as will be shown below. Substitution λ = 0 yields the identity

K̂0θ (z, u) = GK0θ (z, u). (32)

8

We want to prove that for all λ, µ > − 12

HK̂ θ = HK̂ θ . (33)

λ µ

According to Proposition 5 we need to show that K̂λθ ≈ K̂µθ . Following the definition of the positive
definite kernel (5), choose n ∈ N, a finite complex sequence ci and points zi ∈ Eθ , i = 1, n. Positive
definiteness of the kernel [4θ2 (1 − z 2 )(1 − u2 )]k /(θ2 − 2θzu + 1)2k+1 due to its reproducing property in
the Hilbert space of functions representable in the form f (z) = (1 − z 2 )k g(z), where g belongs to the
Bergman-Selberg space generated by the kernel (36) with λ = 2k + 1, and interchange of the order of
summations justified by absolute convergence, lead to the estimates
n
X ∞ n
θ2 − 1 X µ X [4θ2 (1 − zi2 )(1 − zj 2 )]k
0≤ K̂µθ (zi , zj )ci cj = ak ci cj
π (θ2 − 2θzi zj + 1)2k+1
i,j=1 k=0 i,j=1
½ µ¾X∞ X n ½ µ¾ X n
2
θ −1 ak λ [4θ (1 − zi2 )(1 − zj 2 )]k
2 ak
≤ sup ak ci cj = sup K̂λθ (zi , zj )ci cj ,
π k∈N0 aλk (θ2 − 2θzi zj + 1)2k+1 k∈N0 a λ
k
k=0 i,j=1 i,j=1

where
([λ + 1]/2)k ([λ + 2]/2)k Γ(λ + 1/2)Γ((λ + 1)/2 + k)Γ((λ + 2)/2 + k)
aλk = = > 0.
(λ + 1/2)k k! Γ((λ + 1)/2)Γ((λ + 2)/2)Γ(λ + 1/2 + k)k!

This shows the positive definiteness of the kernel K̂µθ . Using the asymptotic relation [Bateman53]

Γ(a + z)
= z a−b (1 + O(z −1 )), |z| → ∞, | arg z| < π, (34)
Γ(b + z)
we obtain ½ ¾
aµ aµk
lim kλ = 1 ⇒ 0 < sup < ∞.
k→∞ a
k k∈N0 aλk
The estimate from below is obtained in the same fashion with sup {aµk /aλk } substituted by inf {aµk /aλk }.
k∈N0 k∈N0
This proves equality (33). Combined with (32) and (30) this gives:

HK̂ θ = AL2 (∂Eθ ) (35)

λ

for all λ > −1/2.

Step 3. According to (27) and (31), the kernel GKλθ is related to the kernel K̂λθ by

GKλθ (z, u) = Bλθ (z, u)K̂λθ (z, u),

where
πθ2λ hθλ πθ2λ
Bλθ (z, u) = = , hθ
λ = . (36)
τ (λ)(θ2 − 2θzu + 1)λ (1 − θ22θ+1 zu)λ τ (λ)(θ2 + 1)λ
For λ > 0 the function Bλθ (z, u) is the reproducing kernel of the Bergman-Selberg space HB θ [Saitoh97].
λ
This space comprises functions holomorphic in the disk |z| < [(θ2 + 1)/2θ]1/2 and having finite norms
" ∞ · ¸k # 12
1 X 2 k! θ2 + 1
kf kHBθ = |fk | ,
λ hθλ k=0 (λ)k 2θ

where fk is the k-th Taylor coefficient of f . The functions

· ¸1 · ¸k
θ 21 (λ)k 2θ
2 2
def
γk (z) = [hλ ] zk
k! θ2 + 1

constitute a complete orthonormal system in HB θ . Note further that the closed ellipse Eθ is contained
λ
p 1 1
2
in the disk |z| < [(θ + 1)/2θ]1/2 due to inequality (θ2 + 1)/(2θ) > (θ 2 + θ− 2 )/2, the right hand
Square summability with geometric weight for classical orthogonal expansions 9

side of which equals the big semi-axis of the ellipse Eθ . As stated in Proposition 6 the space HGK θ is
λ
obtained by restricting the elements of the tensor product HB θ ⊗ HK̂ θ to the diagonal of Eθ × Eθ and
λ λ
comprises the functions of the form
∞
X ∞
X
f (z) = gk (z)γk (z), gk ∈HK̂ θ , kgk k2H < ∞.
λ K̂ θ
λ
k=1 k=1

By (35) we can put AL2 (∂Eθ ) instead of HK̂ θ here. For any g ∈ AL2 (∂Eθ ) consider the estimate
λ
Z
kgγk k2AL2 (∂Eθ ) = |g(z)γk (z)|2 |dz| ≤ max |γk (z)|2 kgk2AL2 (∂Eθ ) , (37)
z∈∂Eθ
∂Eθ

which shows that every product gγk belongs to AL2 (∂Eθ ) and hence so does a finite sum of such
products. Denote
· ¸1 · ¸k · ¸k
def 1 (λ)k 2 2θ 2 θ+1
αλθ (k) = max |γk (z)| = [hθλ ] 2 2
√ =
z∈∂Eθ k! θ +1 2 θ
· ¸1 " #k
1 (λ)k 2 θ+1 θ+1
= [hθλ ] 2 p , p < 1.
k! 2(θ2 + 1) 2(θ2 + 1)
The sequence
n
X ∞
X
Sn (z) = gk (z)γk (z), gk ∈AL2 (∂Eθ ), kgk k2AL2 (∂Eθ ) < ∞,
k=1 k=1

is a Cauchy sequence in AL2 (∂Eθ ). Indeed, using (37) and the Cauchy-Schwarz inequality we get
° M °2
°X ° M X
X M
° °
kSM − SN k2AL2 (∂Eθ ) =° gk γk ° = (gk γk , gl γl )AL2 (∂Eθ ) ≤
° °
k=N AL2 (∂Eθ ) k=N l=N

X M
M X M X
X M
≤ kgk γk kAL2 (∂Eθ ) kgl γl kAL2 (∂Eθ ) ≤ kgk kAL2 (∂Eθ ) αλθ (k)kgl kAL2 (∂Eθ ) αλθ (l) =
k=N l=N k=N l=N
" M
#2 M M
X X X
= kgk kAL2 (∂Eθ ) αλθ (k) ≤ kgk k2AL2 (∂Eθ ) [αλθ (k)]2 .
k=N k=N k=N
P 2
P θ 2
Since both k kgk kAL2 (∂Eθ )
and k [αλ (k)]
converge, the above estimates prove that the sequence
Sn is Cauchy. It follows that HGK θ ⊂ AL2 (∂Eθ ). Inverse inclusion AL2 (∂Eθ )⊂HGK θ is obvious, since
λ λ
I(z) ≡ 1 belongs to HB θ and so for any g∈AL2 (∂Eθ ), the product Ig = g∈HGK θ . ¤
λ λ
Now it is not difficult to establish our main result for Jacobi expansions.

Theorem 4 Let α, β ≥ − 12 . Inequality (1) with fk defined by (4) holds true for all restrictions to the
interval (−1, 1) of the elements of AL2 (∂Eθ ) and only for them.

Proof. Choose γ > max{α, β}, then

θ θ
JKα,β (z, u) ¿ M JKγ,γ (z, u)

for some constant M > 0. Indeed, for an arbitrary n ∈ N, complex numbers ci and points zi ∈ Eθ ,
i = 1, n, estimate using (24):
n
X ∞
X n
X θk+l (1 − zi )k (1 − zj )k (1 + zi )l (1 + zj )l
θ
JKα,β (zi , zj )ci cj = aα,β
k,l ci cj ≤
(θ + 1)2k+2l
i,j=1 k,l=0 i,j=1
10
( )
aα,β
k,l
∞
X n
X θk+l (1 − zi )k (1 − zj )k (1 + zi )l (1 + zj )l
≤ sup aγ,γ ci cj =
k,l∈N0 aγ,γ
k,l
k,l (θ + 1)2k+2l
k,l=0 i,j=1
( )
aα,β
k,l
n
X
θ
= sup JKγ,γ (zi , zj )ci cj ,
k,l∈N0 aγ,γ
k,l i,j=1

where
θα+β+1 (θ − 1) ([α + β]/2 + 1)k+l ([α + β + 3]/2)k+l
aα,β
k,l = > 0.
τ (α, β)(θ + 1)α+β+2 (α + 1)k (β + 1)l k!l!
Interchange of the order of summations is justified by absolute convergence of the series (24).
Application of formula (34) yields as k, l → ∞

aα,β
k,l
³ ´ ³ ´
= O (k + l)α+β−2γ k γ−α lγ−β = O (1 + l/k)α−γ (1 + k/l)β−γ = O(1).
aγ,γ
k,l

Therefore sup {aα,β γ,γ

k,l /ak,l } is positive and finite. Similarly by choosing −1 < η < min{α, β} we can
k,l∈N0
prove that
θ θ
mJKη,η ¿ JKα,β .
θ = GK θ , where λ = β + 1/2, and GK θ is defined by (27). Now Theorem 3
It is left to note that JKβ,β λ λ
gives the desired result. ¤
When α and/or β belongs to (−1, −1/2) step 3 of the proof of the Theorem 3 breaks and the problem
remains open.

Corollary 4.1 Condition (1) for the Fourier-Jacobi coefficients (4) of a function f is satisfied for all
θ > 1 iff f is the restriction of an entire function to the interval (−1, 1).

The last theorem and Szegö’s theory [Szego91] suggest that the following much more general
conjecture might be true.
Conjecture. Inequality (1) holds true for the Fourier coefficients in polynomials orthonormal on
R1 √
(−1, 1) with respect to a weight w that satisfies Szegö’s condition −1 ln w(x)dx/ 1 − x2 > −∞ if and
only if f belongs to AL2 (∂Eθ ).

Acknowledgments
The author thanks Professor Martin Muldoon and the York University in Toronto for hospitality and
support during the Fourth ISAAC Congress in August 2003 and Professor Saburou Saitoh of Gunma
University in Kiryu, Japan, whose book [Saitoh97] on reproducing kernels was the main inspiration for
this research.
References

[Aron50] Aronszajn N. Theory of reproducing kernels, Trans. Amer. Math. Soc. 68, 1950, pp.337-404.

[BerKondr95] Berezanskij Yu.M., Kondratiev Yu.G. Spectral methods in infinite-dimensional analysis. Vol. 1, 2.
Mathematical Physics and Applied Mathematics. 12. Dordrecht: Kluwer Academic Publishers. xvii, 1995.

[Bateman53] Bateman Manuscript Project (edited by A. Erdelyi), Higher transcendental functions, Vol. 1-3, McGraw-
Hill Book Company, Inc., New York, 1953.

[Boyad92] Boyadjiev L. I. On series representation in Laguerre polynomials. C. R. Acad. Bulg. Sci., 45, no.4, 1992,
pp.13-15.

[Byun93] Byun D-W. Inversions of Hermite Semigroup. Proc. Amer. Math Soc. 118, no. 2, 1993, pp.437-445.

[Gaier80] Gaier, D. Vorlesungen über Approximation im Komplexen, Basel, Boston, Stuttgart: Birkhäuser Verlag., 1980.

[Gilbert64] Gilbert R. P. Bergman’s integral operator method in generalized axially symmetric potential theory, J. Math.
Phys. 5, 1964, pp.983-997.

[GilbHoward66] Gilbert R. P., Howard H. C. A generalization of a theorem of Nehari, Bull. Am. Math. Soc. 72, 1966,
pp.37-39.

[Hille39] Hille E. Contributions to the theory of Hermitian series, Duke Math. J. 5, 1939, pp.875-936.

[Hille40] Hille E. Contributions to the theory of Hermitian series. II. The representation problem., Trans. Am. Math.
Soc. 47, 1940, pp.80-94.

[Hille80] Hille E. Contributions to the theory of Hermitians series III: Mean values., Inter. Jour. of Math. and Math.
Sci. 3, no. 3, 1980, pp.407-421.

[JanEijnd90] Janssen A. J. E. M., van Eijndhoven S. J. L. Spaces of type W, growth of Hermite coefficients, Wigner
distribution and Bargmann transform, J. Math. Anal. Appl., 152, 1990, pp.368-390.

[Karp00] Karp D.B. Square summability with geometric weight for Jacobi expansions, Far Eastern Journal of
Mathematics, no.1, 2000, pp.16-27.

[Karp01] Karp D.B. Holomorphic spaces related to orthogonal polynomials and analytic continuation of functions, in:
Analytic Extenstion Formulas and their Applications, Kluwer Academic Publishers, 2001, pp.169-188.

[Karp03] Karp D.B. Hypergeometric reproducing kernels and analytic continuation from a half-line., Journal of Integral
Transforms and Special Functions, 2003 (in print).

[Nehari56] Nehari Z. On the singularities of Legendre Expansions, J. Rational Mech. Anal. 5, no. 6, 1956, pp.987-992.

[Rusev84] Rusev P.Analytic functions and classical orthogonal polynomials, Sofia, 1984.

[Saitoh97] S. Saitoh, Integral transforms, reproducing kernels and their applications, Pitman Research Notes in
Mathematics Series 369, Addison Wesley Longman, Harlow, 1997.

[Szego91] G. Szegö, Orthogonal polynomials, AMS Colloquium Publications 23, 8th printing, 1991.

[Zayed81] Zayed A. I. On Laguerre series expansions of entire functions. Indian Jour. of Pure and Appl. Math. 12, no.
11, 1981, pp.1319-1325.

11