See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/243066508

A generalization of the prolate spheroidal wave functions

Article in Proceedings of the American Mathematical Society · July 2007

DOI: 10.1090/S0002-9939-07-08739-4

CITATIONS READS
39 153

1 author:

Ahmed Zayed
DePaul University
157 PUBLICATIONS 2,942 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Linear integral transformations View project

Special issue on Geometrical Wavelet transforms: Theory and Application View project

All content following this page was uploaded by Ahmed Zayed on 27 November 2014.

The user has requested enhancement of the downloaded file.

 PROCEEDINGS OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 135, Number 7, July 2007, Pages 2193–2203
S 0002-9939(07)08739-4
Article electronically published on March 2, 2007

A GENERALIZATION
OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS

AHMED I. ZAYED

(Communicated by Carmen C. Chicone)

Abstract. Many systems of orthogonal polynomials and functions are bases

of a variety of function spaces, such as the Hermite and Laguerre functions
which are orthogonal bases of L2 (−∞, ∞) and L2 (0, ∞), and the Jacobi poly-
nomials which are an orthogonal basis of a weighted L2 (−1, 1). The associated
Legendre functions, and more generally, the spheroidal wave functions are also
an orthogonal basis of L2 (−1, 1).
The prolate spheroidal wave functions, which are a special case of the spher-
oidal wave functions, possess a very surprising and unique property. They are
an orthogonal basis of both L2 (−1, 1) and a subspace of L2 (−∞, ∞), known
as the Paley-Wiener space of bandlimited functions. They also satisfy a dis-
crete orthogonality relation. No other system of classical orthogonal functions
is known to possess this strange property. This raises the question of whether
there are other systems possessing this property.
The aim of the article is to answer this question in the affirmative by
providing an algorithm to generate such systems and then demonstrating the
algorithm by a new example.

1. Introduction
Orthogonal polynomials and functions are an important tool in studying func-
tion spaces because they provide orthogonal bases for many of these spaces. For
example, the Hermite and Laguerre functions, hn (x) = Hn (x) exp(−x2 /2) and
Lαn (x) = x
α/2
exp(−x/2)Ln (x), n = 0, 1, · · · ; α > −1, are orthogonal bases of
L (−∞, ∞) and L2 (0, ∞), respectively, where Hn (x) and Ln (x) are the Hermite
2
(α,β)
and Laguerre polynomials, while the Jacobi polynomials Pn (x), α, β > −1, are
an orthogonal basis of L2 (−1, 1) with respect to the weight function (1−x)α (1+x)β .
The associated Legendre functions, Pnm (x), and more generally, the spheroidal wave
functions, P sm 2 2
n (x, γ ), are also an orthogonal basis of L (−1, 1).
The prolate spheroidal wave functions, P s0n (x, γ 2 ), which are a special case of
the spheroidal wave functions, possess a very surprising and unique property. They
are an orthogonal basis of both L2 (−1, 1) and a subspace of L2 (−∞, ∞), known
as the Paley-Wiener space of bandlimited functions. They also satisfy a discrete
orthogonality relation.

Received by the editors October 20, 2005 and, in revised form, March 27, 2006.
2000 Mathematics Subject Classification. Primary 33C47, 44A05; Secondary 42C05, 33C45.
Key words and phrases. Prolate and oblate spheroidal wave functions, orthogonal polynomials
and functions, reproducing-kernel Hilbert spaces, bandlimited functions.

2007
c American Mathematical Society
Reverts to public domain 28 years from publication

2193

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 2194 AHMED I. ZAYED

No other system of orthogonal functions is known to possess this strange prop-

erty. This raises the question of whether other systems of functions possessing this
property do exist. The purpose of this article is to answer this question. We shall
show that the answer is in the affirmative by providing an algorithm to generate
such systems, and then we shall demonstrate the idea by giving a new example.
To explain the mystery surrounding the prolate spheroidal wave functions, it
is worth noting that the orthogonality of all the aforementioned systems can be
proved using the fact that they are eigenfunctions of Sturm-Liouville boundary-
value problems on their respective intervals. The same, of course, holds for the
orthogonality of the prolate spheroidal wave functions on the interval (−1, 1); how-
ever, this is not enough to establish their orthogonality on (−∞, ∞). The latter
was proved using the fact that the corresponding Sturm-Liouville differential oper-
ator commutes with an integral operator whose eigenfunctions are also the prolate
spheroidal wave functions. This commutativity property was described as a lucky
accident by Slepian [10] and more recently by Walter [14].
Our approach differs from the classical approach in avoiding the theory of Sturm-
Liouville boundary-value problems all together. Instead, we use the theory of
reproducing-kernel Hilbert spaces and compact operators on Hilbert spaces. In
fact, we shall show at the end of the article that if one uses the differential operator
approach, then the double orthogonality relation of the prolate spheroidal functions
is indeed a lucky accident.
The article is organized as follows. In Section 2 we give a brief introduction
to the Paley-Wiener space of bandlimited functions and to the prolate spheroidal
wave functions. In Section 3 we discuss the theory of reproducing-kernel Hilbert
spaces and summarize the main results that will be needed in the sequel. The main
result is introduced in Section 4, and in Section 5 we conclude the article with a
new example.

2. Prolate spheroidal wave functions

In this section we give a brief introduction to the prolate spheroidal wave func-
tions and some of their important properties. Most of the material here is collected
from different sources [2, 3, 5, 6, 11, 12, 16]; see also [1, 4, 15, 17].
Let Bσ2 be the Paley-Wiener space of functions bandlimited to [−σ, σ], which
consists of entire functions of exponential type, with type σ, that belong to L2 (R)
when restricted to the real axis. It follows from a theorem, due to Paley and Wiener,
that f ∈ Bσ2 if and only if

σ
1
f (z) = √ F (t)eizt dt, for some F ∈ L2 (−σ, σ).
2π −σ
The space Bσ2 is a reproducing-kernel Hilbert space so that for any f ∈ Bσ2 [18],

∞
sin σ(t − x)
(2.1) f (x) dx = f (t).
−∞ π(t − x)
Now consider the spheroidal differential equation (see [2, 3, 13])
   
d   dw µ2
(2.2) 1 − x2 + λ + γ 2 (1 − x2 ) − w = 0,
dx dx 1 − x2
with real parameters λ, γ 2 , and µ. For µ = m = 0, 1, 2, ..., the spheroidal wave
functions, P sm 2
n (x, γ ), are solutions of Eq. (2.2) that are bounded on (−1, 1). They

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 GENERALIZATION OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS 2195

exist only for special values of λ, namely, the eigenvalues λm 2

n (γ ). When γ = 0,
m
they reduce to the associated Legendre functions with λn (0) = n(n + 1).
The spheroidal wave functions are orthogonal on (−1, 1) and satisfy the relation

1
2 (n + m)!
P sm 2 m 2
n (x, γ )P sk (x, γ )dx = δk,n .
−1 2n + 1 (n − m)!
Throughout this article, we shall confine ourselves to the case where µ = 0, i.e.,
we will focus on P s0n (x, γ 2 ) with γ 2 = 0. If γ 2 > 0, P s0n (x, γ 2 ) are called prolate
spheroidal wave functions and when γ 2 < 0, they are called oblate spheroidal   wave
functions. Let γ = τ σ, with τ, σ > 0 and define ϕn,σ,τ (t) = αn P s0n t/τ, γ 2 , n =
0, 1, 2, . . ., for some constants αn > 0. It is known that ϕn,σ,τ satisfy a number of
interesting relations, chief among them being the double orthogonality relations

τ
(2.3) ϕn,σ,τ (t)ϕm,σ,τ (t)dt = an δm,n
−τ

and

∞
(2.4) ϕn,σ,τ (t)ϕm,σ,τ (t)dt = δm,n .
−∞

They also form an orthogonal basis for L2 (−τ, τ ) and an orthonormal basis for
a subspace of L2 (R), namely, the space of functions Bσ2 that are bandlimited to
(−σ, σ). It should be noted that one may renormalize {ϕn,σ,τ } so that they are
orthonormal on L2 (−τ, τ ) instead of L2 (R), i.e., for some positive constants µn ,

τ
∞
(2.5) ϕn,σ,τ (t)ϕm,σ,τ (t)dt = δm,n and ϕn,σ,τ (t)ϕm,σ,τ (t)dt = µn δm,n .
−τ −∞

It is also known that the prolate spheroidal wave functions, ϕn,σ,τ , are eigen-
functions of the differential operator
d2 ϕn,σ,τ dϕn,σ,τ
(2.6) (τ 2 − t2 ) − 2t − σ 2 t2 ϕn,σ,τ = νn,σ,τ ϕn,σ,τ ,
dt2 dt
where νn,σ,τ are the eigenvalues, while the oblate spheroidal wave functions, ψn,σ,τ ,
are eigenfunctions of the differential operator
d2 ψn,σ,τ dψn,σ,τ
(2.7) (τ 2 − t2 ) − 2t + σ 2 t2 ψn,σ,τ = ν̃n,σ,τ ψn,σ,τ .
dt2 dt
The prolate spheroidal wave functions are also eigenfunctions of two integral equa-
tions, the first of which is

τ
sin σ(t − x)
(2.8) ϕn,σ,τ (x) dx = λn ϕn,σ,τ (t),
−τ π(t − x)
where λn = λn,τ,σ are the eigenvalues, and the second is

τ
(2.9) ϕn,σ,τ (t)e−iσwt/τ dt = γn,σ,τ ϕn,σ,τ (w).
−τ

Because the prolate spheroidal wave functions are bandlimited to (−σ, σ), we have
from (2.1),

∞
sin σ(t − x)
(2.10) ϕn,σ,τ (x) dx = ϕn,σ,τ (t),
−∞ π(t − x)

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 2196 AHMED I. ZAYED

from which we obtain

∞
sin σ(t − x) 
(2.11) = ϕn,σ,τ (t)ϕn,σ,τ (x),
π(t − x) n=0

and also the discrete orthogonality relation

∞

(2.12) ϕn,σ,τ (kπ/σ)ϕn,σ,τ (mπ/σ) = δk,m .
n=0

If the prolate spheroidal wave functions are normalized according to Eq. (2.5), then
Eq. (2.11) becomes
∞
sin σ(t − x)  ϕn,σ,τ (t)ϕn,σ,τ (x)
(2.13) = .
π(t − x) n=0
µn
The Fourier transform of the prolate spheroidal wave functions satisfies the re-
lations

∞ 
(2.14) e−itw ϕn,σ,τ (t)dt = (−i)n 2πτ /(σλn ) ϕn,σ,τ (τ w/σ)χσ (w)
−∞

and

τ 
(2.15) e−itw ϕn,σ,τ (t)dt = (i)n 2πτ λn /σ ϕn,σ,τ (τ w/σ),
−τ

where χσ (w) is the characteristic function of (−σ, σ).

To simplify the notation, we normalize the prolate spheroidal wave functions so
that they are orthonormal on (−1, 1), i.e., we set τ = 1 and hence γ = σ and denote
1
ϕn,σ,1 by ϕn . Hence, from (2.15) we obtain −1 e−itσw ϕn (t)dt = γn ϕn (w), where

γn = (i)n 2πτ λn /σ, which gives the less familiar formula

(2.16) e−itσw = γn ϕn (w)ϕn (t), t, w ∈ [−1, 1].
n

3. Reproducing-kernel Hilbert spaces

Let H denote a Hilbert space of functions defined on a set E. We denote its inner
product and norm by  , H and −H , respectively. A function K(x, y), x, y ∈ E
is called a reproducing kernel of H if: i) for every y ∈ E, K(x, y) ∈ H as a function
of x; ii) for every y ∈ E and every f ∈ H, f (y) = f (x), K(x, y)H .
Here it is understood that the inner product applies to functions in x. If a
reproducing kernel exists, it must be unique. Moreover, the reproducing kernel
exists if and only if for every y ∈ E, the point evaluation f (y) is a continuous linear
functional on H.
A function h(x, y) is said to be Hermitian if h(x, y) = h(y, x) and positive definite
n n
if i,j=1 h(yi , yj )ci c̄j ≥ 0 for any set of finite points {yi }i=1 and complex numbers
n
{ci }i=1 . It is easy to see that if K(x, y) is a reproducing kernel of H, then K(x, y)
is Hermitian and positive definite. Conversely, if K(x, y) is a positive definite
Hermitian function defined on E, then there exists a uniquely determined Hilbert
space HK of functions on E admitting the reproducing kernel K(x, y). If K(x, y)
is a reproducing kernel, then the following is true:
K(x, x) ≥ 0, K(x, y) = K(y, x) and |K(x, y)|2 ≤ K(x, x)K(y, y).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 GENERALIZATION OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS 2197

If K(x, y) is a reproducing kernel of H and {gn }∞ n=0 is an orthonormal basis of H,

then
∞
(3.1) K(x, y) = gn (x)gn (y) , x, y ∈ E.
n=0

Let E be an arbitrary set and F(E) be the linear space of all complex-valued
functions defined on E. Let H be a Hilbert space with inner product  , H , and
h : E → H be a vector-valued function from E into H. Consider the linear mapping
L from H into F(E) defined by
(3.2) f (p) = (LF )(p) = F, h(p)H ,
where LF = f, F ∈ H, f ∈ F(E).
Let H̃ and N (L) denote the range and the null space of L. Let M = H
N (L),
and denote by PM the orthogonal projection from H into M. It has been shown
that [8, 9] H̃,  , H̃ is a Hilbert space that is isometric to (M,  , H ) , where

(3.3) f, gH̃ = LF , LGH̃ = PM F , PM GH .

Moreover, H̃,  , H̃ is a reproducing-kernel Hilbert space. Now let

(3.4) K(p, q) = h(p), h(q)H , p, q ∈ E.

Because for any complex numbers {αi }ni=1 ,

n 
n 
n
0≤ αi h(pi ), αj h(pj ) = αi αj h(pi ), h(pj )H
i=1 j=1 H i,j=1

n
= αi , αj K(pi , pj ),
i,j=1

it follows that K(p, q) is positive definite and Hermitian, and hence it is the repro-
ducing kernel of some reproducing-kernel Hilbert space (possibly finite-dimensional
space). In fact, that space is exactly H̃.
We have
(3.5) f H̃ = inf F H ≤ F H ,
F

where the infimum is taken over all F such that LF = f. Furthermore, L is an

isometry between H and H̃ if and only if {h(p) : p ∈ E} is complete in H and in
this case N (L) = {0} and hence
(3.6) f H̃ = PM F H = F H .
Now we apply these general results to a specific case. Let dµ be a σ-finite positive
measure and T be a dµ -measurable set. Consider the Hilbert Space H = L2 (T, dµ)
consisting of all complex-valued functions F such that


2
F L2 (T,dµ) = |F (t)|2 dµ(t) < ∞.
T

Let E be an arbitrary set and h(t, p) be a fixed complex-valued function on T × E,

such that
h(t, p) ∈ L2 (T, dµ) for any p ∈ E.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 2198 AHMED I. ZAYED

Let L be the linear mapping L : L2 (T, dµ) → F(E) defined by

(3.7) f (p) = (LF )(p) = F (t)h̄(t, p)dµ(t), F ∈ L2 (T, dµ).
T

Then, the set of all such f ’s is a reproducing-kernel Hilbert space H̃ = HK with

reproducing kernel


(3.8) K(p, q) = h(t, q)h(t, p)dµ(t),
T

with f (q) = f, K(. , q)H̃ and

2
(3.9) f HK = inf |F (t)|2 dµ(t) ≤ |F (t)|2 dµ(t),
T T

where the infimum is taken over all F such that LF = f. Moreover, the integral
transform (3.7) is an isometry between L2 (T, dµ) and HK if and only if {h(t, p)}p∈E
is complete in L2 (T, dµ).

4. The main result

Now we state and prove our main result.
Theorem 4.1. Let T be an interval of the form [a, b], where −∞ < a < b < ∞,
and E be a measurable subset of R containing T, i.e., T ⊂ E. Let h(t, p) be real and
continuous on T × E, and assume that {h(t, p)}p∈E is complete in L2 (T, dµ). Then
there exist a reproducing-kernel Hilbert space, H̃, comprising functions defined on
∞
E and an orthonormal basis {ϕn }n=0 of L2 (T, dµ) with the property that ϕn can be
naturally extended to functions Φn defined on E such that Φn ∈ H̃ and
i)



Φm (t)Φn (t)dµ(t) = ϕm (t)ϕn (t)dµ(t) = δm,n ;
T T
ii)
Φm , Φn H̃ = µn δm,n , with µn = 1/|λ|2n ,
where {λn } are the eigenvalues of a compact operator on L2 (T, dµ);
iii) {ϕn }∞
n=0 are solutions of the Fredholm integral equation of the second kind

b
(4.1) µn ϕn (x)K(x, p)dµ(x) = ϕn (p), p ∈ T,
a

where K(x, p) is the reproducing kernel of H̃;

iv)
∞

h(t, p) = λn ϕn (t)ϕn (p), t, p ∈ T ;
n=0

v)
∞
Φn (p)Φn (q)
K(p, q) = ;
n=0
µn

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 GENERALIZATION OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS 2199

vi) in addition, if there exists a sequence of points {pn }∞ n=0 ⊂ E such that
{h(t, pn )} is an orthonormal basis of L2 (T, dµ), then for any f ∈ H̃,
∞

(4.2) f (p) = f (pn )K(p, pn )
n=0

and
∞
Φn (pk )Φn (pm )
(4.3) = δk,m .
n=0
µn

Proof. i) Define

b
(4.4) f (p) = (LF )(p) = F (t)h(t, p)dµ(t), p∈E
a

for all F ∈ L2 (T, dµ) = H. Since h(t, p) is continuous on T × E, it is in L2 (T, dµ)

for any fixed p ∈ E; hence, (LF )(p) is well defined for any p ∈ E. Moreover, we
have 
 

b b
2
|(LF )(p)| ≤ |F (t)|2 dµ(t) |h(t, p)|2 dµ(t) .
a a

Since h(t, p) is continuous on the closed square Q = T × T, we have

b


2
|(LF )(p)|2 dµ(p) ≤ F H |h(t, p)|2 dµ(t)dµ(p) < ∞ ,
a
Q

which implies that when LF (p) is restricted to T, it is in L2 (T, dµ), i.e., L is a

continuous linear transformation on L2 (T, dµ). Because h(t, p) is real, it is easy to
see that L is self-adjoint. Moreover, by standard arguments one can show that L
is compact.
Since L is a self-adjoint, compact transformation on L2 (T, dµ) = H, it has a
sequence of eigenvalues {λn }∞ ∞
n=1 and eigenfunctions {ϕn }n=1 such that Lϕn =
λn ϕn for all n, i.e.,

b
(4.5) ϕn (t)h(t, p)dµ(t) = ϕn (·), h(. , p)H = λn ϕn (p).
a

If {ϕ̃k } is an orthogonal basis of the null space of L, then the set {ϕ̃k , ϕk } is
an orthogonal basis of L2 (T, dµ). But from the assumption that {h(t, p)}p∈E is
complete in L2 (T, dµ), it follows that the null space of L is {0} , and hence the
eigenfunctions {ϕn } are an orthogonal basis of L2 (T, dµ). From now on we shall
normalize them so that ϕm , ϕn H = δm,n , and hence, they form an orthonormal
basis of L2 (T, dµ).
But from (3.7) and the discussion following it, it follows that L also maps H =
L2 (T, dµ) into a vector subspace, H̃, of the space, F(E), of all functions defined on
E. The space H̃, which consists of all functions of the form (3.7), is a reproducing-
kernel Hilbert space with inner product given by (3.3) and a reproducing kernel
given by (3.8).
Let us denote the image of ϕn (t) under L by Φn (p), p ∈ E. Evidently, Φn (p) =
λn ϕn (t) for p ∈ T , and as a result, we may conclude that Φn (p) is an extension of
ϕn (t) from T into E.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 2200 AHMED I. ZAYED

ii) From (3.3) we have

(4.6) Φm , Φn H̃ = λm ϕm , λn ϕn H̃ = PM ϕm , PM ϕn H .
But since {h(t, p)}p∈E is complete, we have
(4.7) λm λn ϕm , ϕn H̃ = ϕm , ϕn H = δm,n ,
which implies
1
(4.8) ϕn 2H̃ = µn , where µn = .
|λn |2
iii) We also have from (4.5),

b


b b
λn ϕn (t)h(t, p)dµ(t) = h(t, p)dµ(t) ϕn (x)h(x, t)dµ(x)
a a a



b b
(4.9) = ϕn (x) h(t, p)h(x, t)dµ(t) dµ(x)
a a

b
= ϕn (x)K(x, p)dµ(x).
a
But once more from (4.5), we have

b
λ2n ϕn (p) = λn ϕn (t)h(t, p)dµ(t),
a
which, when combined with (4.9), leads to the integral equation

b
1
ϕn (x)K(x, p)dµ(x) = λ2n ϕn (p) = ϕn (p),
a µn
which is equivalent to (4.1).
iv) This follows from (4.5) and the fact that {ϕn } is an orthonormal basis for
L2 (T, dµ).  √ 
v) This follows from (3.1) and the fact that ϕn (x)/ µn is an orthonormal
basis of H̃.
v) Let us denote h(t, pn ) by hn (t). Since {hn (t)} is an orthonormal basis of
∞
L2 (T, dµ), we have F (x) = n=0 F, hn hn (x). By applying the transformation L
to this relation and noting that f (pn ) = F, hn , we obtain (4.2).
Finally, because {hn (t)} is orthonormal, it follows that K(pm , pn ) = δm,n , and
hence (4.3) follows from (iv). 
Remarks. 1) Clearly, (i), (ii), and (iii) generalize (2.5) and (2.8), while (iv), (v) and
(4.3) generalize (2.16), (2.13) and (2.12).
2) Theorem 4.1 is essentially an existence theorem. Finding explicit examples is
not easy because it depends on finding explicit solutions of integral equations.
3) Since finding the eigenvalues and eigenfunctions of boundary-value problems
involving differential operators is easier than finding those of integral equations, one
may be tempted to derive the double orthogonality property by first starting with
a boundary-value problem and then converting it into an integral equation. This
approach is bound to fail because in this case the kernel of the integral equation,
which is the Green’s function of the problem G(x, t), does not satisfy the hypotheses
of Theorem 4.1, the reason being that the Green’s function is generally given by
two expressions, one for a ≤ x ≤ t and one for t ≤ x ≤ b and therefore, it does not

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 GENERALIZATION OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS 2201

have a natural extension for either x or t beyond the original interval of definition
[a, b].

5. Examples
First, we show that the prolate spheroidal wave functions are a special case of
Theorem 4.1.
Example 5.1. First, let us observe that if f ∈ Bσ2 , then f can be written in the
1
form f (t) = −1 eiσxt F (x) dx, for some F ∈ L2 (−1, 1). This may suggest that we
take the function h(x, t) in the theorem to be eiσxt ; however this function does not
satisfy the hypotheses of the theorem. Therefore, we will first take h(x, t) = cos σxt,
and consider the integral operator

1
(K1 F ) (t) = F (x) cos(σxt) dx = f (t),
−1

acting on the space L2even (T ),

of all even functions in L2 (T ), where T = [−1, 1].
That h(x, t) is complete follows from the Riemann-Lebesgue lemma.
Because ϕn is real and even when n is even, it follows from (2.9) that the eigen-
functions of the integral equation

1
φ(x) cos(σxt) dx = λφ(t)
−1
are {ϕ2n (x)} with eigenvalues γ2n . Hence, the extension of {ϕ2n (x)} to the whole
real line is an orthogonal basis for the space of all even bandlimited functions.
Secondly, we take h(x, t) = sin σxt, and consider the integral operator

1
(K2 F ) (t) = F (x) sin(σxt) dx = f (t),
−1

acting on the space L2odd (T ).

By the same reasoning, we prove that the extension
of {ϕ2n+1 (x)} to the whole real line is an orthogonal basis for the space of all odd
bandlimited functions. The result now follows since L2 (T ) = L2even (T ) ⊕ L2odd (T ).
Example 5.2. Let h(x, t) = eγxt, , γ = 0 and consider the operators K defined by

1
(5.1) (KF )(t) = eγtx F (x)dx = f (t).
−1
Although one can use standard methods for solving integral equations to find the
eigenfunctions of this operator, we will use an easier, but indirect method, to do
so. Now consider the differential operator Lx = dx d
(1 − x2 ) dx
d
+ γ 2 x2 . It is easy to
2  
see that Lt (KF )(t) = (1 − t )f − 2tf + γ t f , or
2 2

1
 
(5.2) Lt (KF )(t) = eγtx F (x) γ 2 x2 (1 − t2 ) − 2γtx + γ 2 t2 dx.
−1
On the other hand, we also have

1  
d  

K(Lx F )(t) = eγtx
(1 − x )F (x) + γ x F (x) dx = I1 + I2 ,
2 2 2
−1 dx
where

1  

d
I1 = eγtx (1 − x2 )F  (x) dx and I2 = γ 2 eγtx x2 F (x)dx.
−1 dx

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 2202 AHMED I. ZAYED

By integrating I1 by parts twice, we obtain

1
1
d    
I1 = γt F (x) (1 − x2 )eγtx dx = γt eγtx F (x) γt(1 − x2 ) − 2x dx.
−1 dx −1

Thus, by adding I1 and I2 we have

1
 
(5.3) K(Lx F )(t) = eγtx F (x) γ 2 t2 (1 − x2 ) − 2γtx + γ 2 x2 dx.
−1

By comparing (5.2) and (5.3), we conclude that K and L commute; hence, they have
the same eigenfunctions. But from (2.7), we conclude that the oblate spheroidal
wave functions are the eigenfunctions of K.
It is easy to see that the space H̃,  , H̃ consists of all functions f (t) of the
form (5.1) and the inner product is given by [7, p. 62]

∞
1
f , gH̃ = f (it)g(it) dt.
2π −∞
The reproducing kernel is given by Kγ (t, η) = sinh γ(t − η)/π(t − η). It follows
that the oblate spheroidal wave functions are orthogonal on (−1, 1) and also on
the imaginary axis. Moreover, they are an orthogonal basis of L2 (−1, 1) and of the
Hilbert space H̃ because the exponential functions are complete.

References
[1] J. Boyd, Approximation of an analytic function on a finite interval by a bandlimited function
and conjectures on properties of prolate spheroidal functions, Appl. Comput. Harmon. Anal.,
Vol. (2003), 168-176. MR2007058 (2004g:41027)
[2] A. Erdelyi, W. Magnus, F. Oberhettinger, F. Tricomi, Higher Transcendental Functions,Vol.
III, Bateman Manuscript, McGraw-Hill, New York (1955). MR0066496 (16:586c)
[3] C. Flammer, Spheroidal Wave Functions, Stanford University Press, California (1957).
MR0089520 (19:689a)
[4] K. Khare and N. George, Sampling theory approach to prolate spheroidal wavefunctions, J.
Phys. Ser. A, Vol. 36(2003), 10011-10021. MR2024509 (2004m:33029)
[5] H. Landau and H. O. Pollak, Prolate spheroidal wave functions, Fourier analysis and
uncertainty-III, Bell Syst. Tech. J. Vol. 41 (1962), pp. 1295 - 1336. MR0147686 (26:5200)
[6] H. Landau and H. O. Pollak, Prolate spheroidal wave functions, Fourier analysis and
uncertainty-II, Bell Syst. Tech. J. Vol. 40 (1961), pp. 65 - 84. MR0140733 (25:4147)
[7] S. Saitoh, Integral Transforms, Reproducing Kernels and Their Applications, Pitman Re-
search Notes in Math., Vol. 369 (1997), Addison Wesley Longman, United Kingdom.
MR1478165 (98k:46041)
[8] S. Saitoh, The Theory of Reproducing Kernels and Its Applications, Pitman Research Notes
in Math., Vol. 189 (1988), Longman Scientific & Technical, United Kingdom. MR0983117
(90f:46045)
[9] S. Saitoh, Hilbert spaces induced by Hilbert space valued functions, Proc. Amer. Math.
Soc., Vol. 89 (1983), pp. 74 -78. MR0706514 (84h:44001)
[10] D. Slepian, Some comments on Fourier analysis, uncertainty and modeling, SIAM Rev., Vol.
25, (1983), pp. 379 - 393. MR0710468 (84i:94016)
[11] D. Slepian, Prolate spheroidal wave functions, Fourier analysis and uncertainty-IV: Exten-
sions to many dimensions; Generalized prolate spheroidal functions , Bell Syst. Tech. J.
Vol. 43 (1964), pp. 3009 -3058. MR0181766 (31:5993)
[12] D. Slepian and H. O. Pollak, Prolate spheroidal wave functions, Fourier analysis and
uncertainty-I, Bell Syst. Tech. J. Vol. 40 (1961), pp. 43 -64. MR0140732 (25:4146)
[13] H. Volkmer, Spheroidal Wave Functions, in Handbook of Mathematical Functions, Nat.
Bureau of Stds., Applied Math. Series 2004.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 GENERALIZATION OF THE PROLATE SPHEROIDAL WAVE FUNCTIONS 2203

[14] G. Walter, Differential operators which commute with characteristic functions with ap-
plications to a lucky accident, J. Complex Variables, Vol. 18 (1992), 7-12. MR1157019
(93a:34101)
[15] G. Walter and X. Shen, Wavelets based on prolate spheroidal wave functions, J. Fourier
Anal. Appls., Vol. 10 (2004), 1-26. MR2045522 (2005c:42041)
[16] G. Walter and X. Shen, Sampling with prolate spheroidal wave functions, J. Sampl. Theory:
Signal & Image Process., Vol. 2 (2003), 25-52. MR2002855 (2004k:42064)
[17] H. Xiao, V. Rokhlin, and N. Yarvin, Prolate spheroidal wavefunctions, quadrature and
interpolation, J. Inverse Problems, Vol. 17 (2001), 805-838. MR1861483 (2002h:41049)
[18] A. I. Zayed, Advances in Shannon’s Sampling Theory, CRC Press, Boca Raton, Fl (1993).
MR1270907 (95f:94008)

Department of Mathematical Sciences, DePaul University, Chicago, Illiniois 60614

E-mail address: [email protected]

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

View publication stats