J. Math. Anal. Appl.

284 (2003) 104–117

www.elsevier.com/locate/jmaa

Sampling of vector-valued transforms associated

with Green’s matrix of Dirac systems
M.Z. Abd-Alla and M.H. Annaby ∗
Department of Mathematics, Faculty of Science, Cairo University, Giza, Egypt
Received 12 March 2001
Submitted by H.M. Srivastava

Abstract
Recently a sampling theorem associated with regular one-dimensional Dirac systems has been
developed. Integral transforms whose kernels are solutions of the systems have been reconstructed
from their values at the eigenvalues. In the present article the situation is considered where the kernels
are replaced by Green’s matrix of the problem. Thus a vector-valued sampling theorem is established
for vector-valued transforms. A family of examples where the systems have variable coefficients and
a vector-valued Kramer-type sampling lemma are given at the end of the paper.
 2003 Published by Elsevier Inc.

Keywords: Regular Dirac’s system; Kramer’s sampling theorem; Lagrange interpolations

1. Introduction

The classical sampling theorem [7,11,12,22] states:

If g(x) ∈ L2 (−π, π) and

π
1
f (λ) √ g(x)eiλx dx, λ ∈ C, (1.1)
2π
−π

then f (λ) can be recovered from its values at the integers via the sampling representation

* Corresponding author.
E-mail addresses: [email protected] (M.Z. Abd-Alla), [email protected]
(M.H. Annaby).

0022-247X/$ – see front matter  2003 Published by Elsevier Inc.

doi:10.1016/S0022-247X(03)00260-9
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 105

∞
 sin π(λ − n)
f (λ) = f (n) . (1.2)
n=−∞
π(λ − n)
The sampling series (1.2) converges uniformly and absolutely on R and on compact sets
of C.

The above theorem is related to the names of Whittaker [21], Kotel’nikov [13], and
Shannon [19] and is known in the literature as the Whittaker–Kotel’nikov–Shannon (WKS)
sampling theorem, cf. [7,15]. This theorem has many applications in signal processing,
where analog signals of the form (1.1), i.e., band limited signals, can be recovered from
(1.2), allowing us to be in the digital setting.
There are many extensions of the WKS sampling theorem. One of those generaliza-
tions based on the observation that the kernel eiλx of (1.1) is a solution of the first-order
differential equation of the eigenvalue problem −iy  = λy, y(−π) = y(π), while the sam-
pling points, the integers, are the eigenvalues. Thus one might think of a sampling theorem
for more general transforms arising from higher-order eigenvalue problems. This was first
considered by Weiss [20] and followed by Kramer [14], who first introduced the following
lemma, see [5,11,12].

Lemma 1.1. Let {λn }n∈Z be a sequence of real numbers and K(·, λ) be a function defined
on I × C, for some closed interval I . Let K(·, λ) ∈ L2 (I ) for all λ ∈ C and the sequence
{K(·, λn )}n∈Z be an orthogonal basis of L2 (I ). Then, the transform


f (λ) = g(x)K(x, λ) dx, g(x) ∈ L2 (I ), (1.3)
I
can be reconstructed from its values at {λn }n∈Z through
∞ 
 K(x, λ)K(x, λn ) dx
f (λ) = f (λn ) I  . (1.4)
I |K(x, λn )| dx
2
n=−∞
The convergence of the sampling series is absolute on C and uniform wherever
K(·, λ) L2 (I ) , as a function of λ, is bounded.

Obviously the WKS sampling theorem is a special case of Lemma 1.1. In Weiss’ note
[20] and Kramer’s paper [14] the kernel K(·, λ) is taken to be a solution of a second-order
self-adjoint regular Sturm–Liouville eigenvalue problem with separate type boundary con-
ditions. The sampling points in this case are the eigenvalues of the problem. Kramer has
introduced an extension of this idea to nth-order regular self-adjoint eigenvalue problems,
which has been proved in [1,4] to be incorrect via a counterexample. This example in-
dicates that Kramer’s extension is not correct when the eigenvalue problem has multiple
eigenvalues. In [23] the sampling expansion of Sturm–Liouville transforms is shown to
be nothing but a Lagrange-type interpolation one and in [24] the singular case is stud-
ied. Also [8] considered some extensions of the situation using more general boundary
conditions. The mixed-type boundary conditions case is treated in [1]. The second-order
multipoint case is considered in [28]. The nth-order self-adjoint and nonself-adjoint prob-
lems are studied extensively, see, e.g., [3,4,6,25].
106 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

In [29], a sampling theorem is introduced where sampling representations are derived

for integral transforms whose kernels are solutions of one-dimensional regular Dirac sys-
tems. This result is mentioned in Section 2. In the present paper we study the possibility
of sampling transforms that have kernels expressed in terms of the Green’s matrix of one-
dimensional regular Dirac systems. Since the Green’s matrix is a 2 × 2 matrix, we derive a
two-dimensional sampling theorem for vector-valued transforms. The proof is given in two
different ways. The first is using the eigenfunctions expansion, while the second uses the
theory of residues, putting the results in the frame of Haddad–Yao–Thomas’ theorem [9],
the first that connected sampling theory and the Green’s function; see also [22]. Other pa-
pers of sampling theory and Green’s function and integral equations are [2,26]. In Section 4
we give an example illustrating the main result of the paper. The Dirac system associated
with this example has variable coefficients that include a wide family of problems. We
mention that, except for first-order problems, examples of sampling series associated with
regular Sturm–Liouville and other high-order problems were derived only when the coef-
ficients are constants or step functions. To the best of our knowledge, Example 2 of [8,
pp. 117–119] is the first example of a sampling series associated with a Sturm–Liouville
problem whose coefficients are step functions. However, [28] is the first general treatment.
Finally, we give a vector-valued Kramer’s lemma, where the kernel of the sampled vector-
valued transform is a matrix of functions.

2. Preliminaries

Consider the first-order system

BY + P(x)Y = λY, −∞ < a
x
b < ∞, λ ∈ C, (2.1)

where
     
0 1 p11 (x) p12 (x) y1 (x)
B= , P(x) = , Y(x) = . (2.2)
−1 0 p21 (x) p22 (x) y2 (x)
Here pij (·) are real-valued continuous functions on [a, b]. Using an orthogonal transfor-
mation [17, pp. 48–50], [18, pp. 185–187], Eq. (2.1) can be transformed into
   
0 1  p(x) 0
Z + Z = λZ, (2.3)
−1 0 0 r(x)
p(·) and r(·) are continuous functions and Z is the result of acting the transformation on Y.
Hence [17, p. 50], we consider the eigenvalue problem (one-dimensional Dirac system),

y2 − p(x)y1 = λy1 , y1 + r(x)y2 = −λy2 , (2.4)

y1 (a) sin α + y2 (a) cos α = 0, (2.5)
y1 (b) sin β + y2 (b) cos β = 0. (2.6)
Here α, β ∈ [0, π[. A complex number λ∗
is said to be an eigenvalue of the one-
dimensional Dirac system (2.4)–(2.6) if there is a nontrivial vector-valued function Y∗ (·) =
(y1 (·), y2 (·))T which satisfies the differential system (2.4) when λ = λ∗ and the boundary
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 107

conditions (2.5)–(2.6). In this case we say that Y∗ (x) is an eigenfunction (or a vector-
valued eigenfunction) of problem (2.4)–(2.6) corresponding (belonging) to λ∗ . Here YT
denotes the matrix transpose of Y. Let H be the Hilbert space
   
y1 (x)
H := Y(x) = : y1 (x), y2 (x) ∈ L (a, b) .
2
(2.7)
y2 (x)
The inner product of H is defined by

b
Y, ZH = YT (x)Z̄(x) dx. (2.8)
a
Let
   
y1 (·, λ) z1 (·, λ)
Y(·, λ) = and Z(·, λ) =
y2 (·, λ) z2 (·, λ)
be the solutions of (2.4) determined by the initial conditions
   
y1 (a, λ) y2 (a, λ) cos α − sin α
= . (2.9)
z1 (b, λ) z1 (b, λ) cos β − sin β
We notice that Y(·, λ) satisfies (2.4) and (2.5) while Z(·, λ) satisfies (2.4) and (2.6).
Thus the eigenvalues of problem (2.4)–(2.6) are either the zeros of ω1 (λ) or those of ω2 (λ),
where
   
ω1 (λ) y1 (b, λ) sin β + y2 (b, λ) cos β
= . (2.10)
ω2 (λ) z1 (a, λ) sin α + z2 (a, λ) cos α
Let W (Y, Z)(·, λ) denote the Wronskian of Y(·, λ) and Z(·, λ), i.e. [18, p. 194],
 
 y (x, λ) y2 (x, λ) 
W (Y, Z)(x, λ) :=  1 .
z1 (x, λ) z2 (x, λ) 
The Wronskian W (Y, Z)(x, λ) is independent of x [18, p. 194]. If w(λ) := W (Y, Z), then
a simple calculation leads to

ω1 (λ) = −ω(λ), ω2 (λ) = ω(λ). (2.11)

The eigenvalues of problem (2.4)–(2.6) are the zeros of ω(λ). All zeros (eigenvalues)
are simple from algebraic and geometric points of view [17, pp. 51–52, 55–56]. Let {λk }∞
−∞
denote the sequence of eigenvalues; then [17, p. 57]
 
c 1
λn = + n + O as |n| → ∞, (2.12)
π n
where c is a constant. The corresponding set of eigenfunctions is {Y(·, λn )}−∞ n=−∞ (or
equivalently {Z(·, λn )}−∞
n=−∞ ). A set of eigenfunctions is an orthogonal basis of H [17,
p. 58, ff.]. Since the eigenvalues are all real we can take the eigenfunctions to be real
valued.
Before defining the Green’s matrix of (2.4)–(2.6), we mention Zayed–García’s result
[29, p. 592].
108 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

f1 (x)

Theorem 2.1. Let a = 0, b = π . Let f(x) = f2 (x) ∈ H. The integral transform

π
F (λ) = fT (x)Y(x, λ) dx (2.13)
0

admits the sampling representation

∞
 ω(λ)
F (λ) = F (λn ) . (2.14)
n=−∞
(λ − λn )ω (λn )

Series (2.14) converges uniformly on compact subsets of C. The function F (λ) is an entire
function of exponential type at most π .

Remark 2.2. Similar to the results of Zayed [24], a sampling theorem may be derived for
the transform

π
∗
F (λ) = fT (x)Z(x, λ) dx. (2.15)
0

Since for λ ∈ C, Y(x, λ) is a continuous vector-valued function of x, then

π
π
       
F (λ)
f1 (x)y1 (x, λ) dx + f2 (x)y2 (x, λ) dx
0 0
     

2 max f1 L1 , f2 L1 max y1 (x, λ), y2 (x, λ) , λ ∈ C. (2.16)
a
x
b

The above inequality and [18, Lemma 7.2.1, p. 191] prove that F (λ) has order one.

Let f(·) = (f1 (·), f2 (·))T be a continuous vector-valued function. Then the inhomoge-
neous eigenvalue problem consisting of the differential system
   
y2 − λ + p(x) y = f1 (x), y1 + λ + r(x) y2 = −f2 (x), (2.17)
and the boundary conditions (2.5)–(2.6) has the unique solution

b
Yf (x, λ) = G(x, ξ, λ)f(ξ ) dξ, λ = λn , n ∈ Z, (2.18)
a

where

1 Z(x, λ)YT (ξ, λ) if a
ξ
x
b,
G(x, ξ, λ) = (2.19)
ω(λ) Y(x, λ)ZT (ξ, λ) if a
x
ξ
b.
Thus, concretely [17, p. 69],
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 109

 

 1
y (x, λ)z1 (ξ, λ) y1 (x, λ)z2 (ξ, λ)
if a
x
ξ
b,
1  y2 (x, λ)z1 (ξ, λ) y2 (x, λ)z2 (ξ, λ)
G(x, ξ, λ) =  
ω(λ) 

 y1 (ξ, λ)z1 (x, λ) y2 (ξ, λ)z1 (x, λ)
if a
ξ
x
b;
y1 (ξ, λ)z2 (x, λ) y2 (ξ, λ)z2 (x, λ)
(2.20)
see [17, p. 67, ff.]. The matrix G(x, ξ, λ) is called the Green’s matrix of problem (2.4)–
(2.6). Obviously G(x, ξ, λ) is a meromorphic function of λ, for every (x, ξ ) ∈ [a, b]2,
which has simple poles only at the eigenvalues. Notice that both Y(x, λ) and Z(x, λ) are
entire functions of λ for x ∈ [a, b].
When λ is not an eigenvalue, the solution of the inhomogeneous system (2.17) and the
boundary conditions (2.5)–(2.6) given by (2.18) can be written in the form [17, pp. 75–77]
∞
an Yn (x)
Yf (x, λ) = , (2.21)
n=−∞
λn − λ
 b T
where f(x) = ∞ n=−∞ an Yn (x), an = a Yn (ξ )f(ξ ) dξ is the Fourier expansion of f in
terms of an orthonormal basis of eigenfunctions, i.e.,
Y(x, λn ) Z(x, λn )
Yn (x) = or Yn (x) = , n ∈ Z.
Y(·, λn ) 2H Z(·, λn ) 2H
Also the vector-valued function f(·) can be represented in the following form:

1
f(x) = − lim Yf (x, λ) dλ, (2.22)
2πi m→∞
γm

where {γm } is a sequence of closed contours such that the eigenvalues (poles) λ−m , λ−m+1 ,
. . . , λ0 , λ1 , . . . , λm lie in the interior of the contour γm .

3. The main result

The main result of this paper, Theorem 3.1 below, is a vector-valued sampling theo-
rem of vector-valued transforms whose kernels are the Green’s matrix introduced above,
multiplied by the Wronskian ω(λ). The use of Green’s function in this setting leads to a
vector-valued sampling theorem for vector-valued transforms, which as far as we know, has
not been considered before. Moreover, Green’s matrix is a function of three variables and
this gives us the opportunity to sample a family of vector-valued integral transforms. As is
mentioned before, we introduce two ways of proof, one using the eigenfunction expansion
and the other using the theory of residues. Let {Y(·)}∞ −∞ denote a complete orthonormal
set of eigenfunctions. Let x0 ∈ [a, b] be fixed. Define the matrix

T(ξ, λ) = ω(λ)G(x0 , ξ, λ). (3.1)

By the properties of the Green’s matrix, T(ξ, λ) is an entire-matrix function of λ for every
ξ ∈ [a, b] of order one and type not exceeding b − a, cf. [18, p. 191].
110 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

f1 (ξ )
F1 (λ)

Theorem 3.1. Let f(ξ ) = f2 (ξ ) ∈ H. Let F(λ) = F2 (λ) be the vector-valued transform

b
F(λ) = T(ξ, λ)f(ξ ) dξ. (3.2)
a

Then F(λ) is a vector-valued function entire function of order one and type at most b − a
that admits the vector-valued sampling expansion
∞
 ω(λ)
F(λ) = F(λn ) . (3.3)
n=−∞
(λ − λn )ω (λn )

The vector-valued series (3.3) converges uniformly on compact subsets of C and absolutely
on C. Here (3.3) means
 ∞ ∞
T
 ω(λ)  ω(λ)
F(λ) = F1 (λn ) , F2 (λn ) .
n=−∞
(λ − λn )ω (λn ) n=−∞ (λ − λn )ω (λn )

A proof using the eigenfunction expansion. Let λ = λn , n ∈ Z. Since F(λ) is nothing

but the unique solution (multiplied by ω(λ)) of the inhomogeneous Dirac system (2.17),
(2.5)–(2.6) when x = x0 , then [17, p. 77] F(λ) has the expansion

∞
b
an ω(λ)
F(λ) = Yn (x0 ), an = fT (ξ )Yn (ξ ) dξ, (3.4)
λ −λ
n=−∞ n a

and {Yn (·)}∞

−∞ is a complete orthonormal set of eigenfunctions. Thus
∞
an ω(λ) Y(x0 , λn )
F(λ) = . (3.5)
λ − λ Y(·, λn ) 2H
n=−∞ n

Now we calculate Y(·, λn ) 2H using the Green’s identity [17, p. 51]. Let λ, µ ∈ C be
different and µ = λn = λ; then

b
 b
(λ − µ) YT (x, λ)Z̄(x, µ) dx = y1 (x, λ)z̄2 (x, µ) − y2 (x, λ)z̄1 (x, µ) a , (3.6)
a

where Y(·, λ), Z(·, µ) are the solutions defined above. By the initial conditions (2.9), we
have

b
 
(λ − µ) YT (x, λ)Z̄(x, µ) dx = − sin βy1 (b, λ) + cos βy2 (b, λ)
a
 
− sin αz̄1 (a, µ) + cos αz̄2 (a, µ) . (3.7)
Letting µ = λn , for some n, and noting that Z(x, λn ) is a real-valued eigenfunction, we
have
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 111

b
 
(λ − λn ) YT (x, λ)Z(x, λn ) dx = − sin βy1 (b, λ) + cos βy2 (b, λ) . (3.8)
a

From (2.10) and (2.11), one obtains

b
ω(λ)
YT (x, λ)Z(x, λn ) dx = . (3.9)
λ − λn
a

Taking the limit in (3.9) when λ → λn ,

b
YT (x, λn )Z(x, λn ) dx = ω (λn ). (3.10)
a

Since ω(λn ) = 0, there is a nonzero constant Cn , such that Z(x, λn ) = Cn Y(x, λn ). Hence

b
 2
Cn Y(·, λn )H = Cn YT (x, λn )Y(x, λn ) dx = ω (λn ). (3.11)
a

Substituting in (3.5), we get the following series, which holds for λ is not an eigenvalue:
∞
 ω(λ)Cn
F(λ) = an Y(x0 , λn ). (3.12)
n=−∞
(λ − λn )ω (λn )

To prove (3.3), it suffices to prove that F(λn ) = an Cn Y(x0 , λn ). Indeed, (3.2) is nothing,
but
 
 y1 (x0 , λn )z1 (ξ, λn )f1 (ξ ) + y1 (x0 , λn )z2 (ξ, λn )f2 (ξ )


b  if x0
ξ,
y2 (x0 , λn )z1 (ξ, λn )f1 (ξ ) + y2 (x0 , λn )z2 (ξ, λn )f2 (ξ )
F(λn ) =  

 y1 (ξ, λn )z1 (x0 , λn )f1 (ξ ) + y2 (ξ, λn )z1 (x0 , λn )f2 (ξ )
a  if x0  ξ.
y1 (ξ, λn )z2 (x0 , λn )f1 (ξ ) + y2 (ξ, λn )z2 (x0 , λn )f2 (ξ )
(3.13)
Thus
 
F1 (λn )
F(λn ) =
F2 (λn )
 x b 
z1 (x0 , λn ) a 0 YT (ξ, λn )f(ξ ) dξ + y1 (x0 , λn ) x0 ZT (ξ, λn )f(ξ ) dξ
= x b .
z2 (x0 , λn ) a 0 YT (ξ, λn )f(ξ ) dξ + y2 (x0 , λn ) x0 ZT (ξ, λn )f(ξ ) dξ
(3.14)
Now we prove that

F1 (λn ) = an Cn y1 (x0 , λn ), F2 (λn ) = an Cn y2 (x0 , λn ). (3.15)

Since there is a nonzero constant Cn such that Z(·, λn ) = Cn Y(·, λn ), then
112 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

b
F1 (λn ) = Cn y1 (x0 , λn ) YT (ξ, λn )f(ξ ) dξ
a

b
= Cn y1 (x0 , λn ) fT (ξ )Y(ξ, λn ) dξ = Cn an y1 (x0 , λn ). (3.16)
a

The same for F2 (λn ) and the proof of (3.3) is complete with a pointwise convergence
on C. For the proof of the uniform convergence, we use the same technique established in
[29], but we have to mention that the definition of the inner product of H has to include the
complex conjugate, cf. (2.8) above. Otherwise the constant C(K) of [29, p. 595] may be
negative or even complex. As for the absolute convergence on C, let λ ∈ C. From (3.5), we
have
∞  
∞  
 ω(λ)   an ω(λ) Y(x0 , λn ) 
F(λn ) =  
 (λ − λ )ω  (λ )   (λ − λ) Y(·, λ ) 2 
n=−∞ n n n=−∞ n n H
 ∞   1/2  ∞  1/2
  a 2   ω(λ)Y(x0 , λn ) 2

 n    < ∞,
 Y(·, λ )   (λ − λ) Y(·, λ ) 
n=−∞ n H n=−∞ n n H

(3.17)
since the last two series are nothing but the norms f(·) H , Y(·, λ) H , proving the ab-
solute convergence. The uniform convergence on compact sets of C implies the analyticity
of F(λ) on C and the growth properties of F(λ) can be established as we explained in a
previous remark. ✷

A proof using residues. First we denote by Yn (·) and Z(·), respectively, to Y(·, λn ) and
Z(·, λn ) for any integer n. By (3.1), F(λ) has the form

b
F(λ) = ω(λ)G(x0 , ξ, λ)f(ξ ) dξ = ω(λ)Yf (x0 , λ), (3.18)
a

where Yf (x0 , λ) is the solution defined by (2.18). From Eq. (2.22) and since non of the
λn ’s lie on the circles γm , we have

1
f(x0 ) = − lim Yf (x0 , λ) dλ
2πi m→∞
γm


b
1
=− lim G(x0 , ξ, λ)f(ξ ) dξ dλ
2πi m→∞
γm a

b   
1
=− lim G(x0, ξ, λ) dλ f(ξ ) dξ. (3.19)
2πi m→∞
a γm
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 113

But from the Theory of residues [10, p. 241] and noting that the singularities of G(x0 , ξ, λ)
inside γm are the simple poles {λn }m
n=−m ,
 ∞


lim G(x0 , ξ, λ) dλ = 2πi Res G(x0 , ξ, λ), λn

m→∞
γm n=−∞
 
 y (x )z (ξ )
 n1 0 n1
yn1 (x0 )zn2 (ξ )
if x0
ξ,
1  yn2 (x0 )zn1 (ξ )
∞
yn2 (x0 )zn2 (ξ )
= 2πi   (3.20)
ω (λn ) 

n=∞  yn1 (ξ )zn1 (x0 ) yn2 (ξ )zn1 (x0 )
if ξ
x0 .
yn1 (ξ )zn2 (x0 ) yn2 (ξ )zn2 (x0 )
The convergence in (3.19) and (3.20) and the interchange of the limit and the integration are
guaranteed because of the asymptotics derived in [17, p. 52, ff.], while the interchange of
the integrals in (3.19) is a simple application of Fubini theorem since the Green’s function
satisfies its conditions. Using the relations Z(x, λn ) = Cn Y(x, λn ), n ∈ Z, and (3.18)–
(3.20), we obtain
∞

b  
Cn yn1 (x0 )yn1 (ξ ) yn1 (x0 )yn2 (ξ )
f(x0 ) = −  (λ )
F(ξ ) dξ
ω n yn2 (x0 )zn1 (ξ ) yn2 (x0 )yn2 (ξ )
n=−∞ a
∞

b
Cn
=− 
Yn (x0 ) YTn (ξ )f(ξ ) dξ
n=−∞
ω (λn )
a
∞
 Cn an
=− Yn (x0 ). (3.21)
n=−∞
ω (λn )

Hence the solution Yf (x0 , λ) has the expansion

∞
 Cn an
Yf (x0 , λ) = Yn (x0 ).
n=−∞
(λ − λn )ω (λn )

Therefore
∞
 ω(λ)
F(λ) = Cn an Yn (x0 ). (3.22)
n=−∞
(λ − λn )ω (λn )

The last equation and relations (3.15) complete the proof. ✷

Remark 3.2. (i) The matrix T(ξ, λ) may be defined in another way replacing ω(λ) by the
canonical product
∞  
λ λ/λk
G(λ) = (λ − λ0 ) 1− e . (3.23)
λn
n=−∞, n=0

This product converges defining an entire function of order one, cf. [16]. In this case in the
sampling theorem, G(λ) appears instead of ω(λ).
114 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

(ii) The above sampling theorem is of vector-valued type. But for special choices of x0
it becomes of a scalar type as follows.
(iii) From the definition of G(x, ξ, λ), (2.19), we notice that when x0 = a,

T(ξ, λ) = Y(a, λ)ZT (ξ, λ) (3.24)

and expansion (3.3) becomes in a scalar form. It is the expansion of the integral transform
(2.15) when it is multiplied by the constant Y(a, λ). Similarly if we take x0 = b,

T(ξ, λ) = Z(b, λ)YT (ξ λ), (3.25)

leading to the results of [29] since Z(b, λ) is a constant matrix.

4. An example and a vector-valued Kramer-type lemma

This section is devoted to introduce an example illustrating Theorem 3.1 above and a
derivation of a vector-valued Kramer-type sampling theorem, where Theorems 2.1 and 3.1
may be considered as special cases.
In the next example we give a sampling expansion associated with a Dirac system with
variable coefficients and the most general separate type boundary conditions. In this case,
the eigenvalues, sampling points, are equidistant. We would like to mention here that the
unique example of the one-dimensional theorem of [29, Section 4] contains a Dirac system
with trivial coefficients, i.e., when r(x) ≡ 0 ≡ p(x).

Example 4.1. Consider the system

y2 − p(x)y1 = λy1 , y1 − p(x)y2 = −λy2 , 0
x
π, (4.1)

y1 (0) sin α + y2 (0) cos α = 0, (4.2)
y1 (π) sin β + y2 (π) cos β = 0, (4.3)
where p(x) is a continuous function on [0, π]. For simplicity we define

x
x
P (x) := p(t) dt, H (x) := p(t) dt, 0
x
π.
0 π

In the notations of Section 2, the solutions Y(·, λ) and Z(·, λ) are

 
cos{λx + P (x) − α}
Y(x, λ) = ,
sin{λx + P (x) − α}
 
cos{λ(x − π) + H (x) − β}
Z(x, λ) = . (4.4)
sin{λ(x − π) + H (x) − β}
The eigenvalues are the solutions of the equation
   
cos λπ + P (π) − α sin β + sin λπ + P (π) − α cos β = 0. (4.5)
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 115

Taking γ = P (π) − α + β, the eigenvalues of the problem are λn = n − γ /π , n ∈ Z. We

assume, without any loss of generality, that zero is not an eigenvalue. The Green’s matrix
of the problem is
1
G(x, ξ, λ) = −
sin{λπ + γ }
 

 cos{λx + P (x) − α} cos{λ(ξ − π ) + H (ξ ) − β} cos{λx + P (x) − α} sin{λ(ξ − π ) + H (ξ ) − β}

 ,

 sin{λx + P (x) − α} cos{λ(ξ − π ) + H (ξ ) − β} sin{λx + P (x) − α} sin{λ(ξ − π ) + H (ξ ) − β}
 x
ξ,
×  
 cos{λξ + P (ξ ) − α} cos{λ(x − π ) + H (x) − β}
 cos{λξ + P (ξ ) − α} sin{λ(x − π ) + H (x) − β}

 ,

 sin{λξ + P (ξ ) − α} cos{λ(x − π ) + H (x) − β}
 sin{λξ + P (ξ ) − α} sin{λ(x − π ) + H (x) − β}
x  ξ.
(4.6)

The sampling result in this case can be formulated in the following lemma.

F1 (λ)

Lemma 4.2. Let f(x) = ff12 (x)(x) ∈ H. Let F(λ) = F2 (λ) be the vector-valued transform
 
F1 (λ)
F(λ) =
F2 (λ)
 x  
cos{λx0 + H (x0 ) − β} 0 0 YT (ξ, λ)f(ξ ) dξ + cos{λx0 + P (x0 ) − α} xπ ZT (ξ, λ)f(ξ ) dξ
=  x0 T  π0 T ,
sin{λx0 + H (x0 ) − β} 0 Y (ξ, λ)f(ξ ) dξ + sin{λx0 + P (x0 ) − α} x Z (ξ, λ)f(ξ ) dξ
0
(4.7)
where x0 ∈ [0, π]. Then
∞
  
γ sin π(λ − n + γ /π)
F(λ) = F n− . (4.8)
n=−∞
π π(λ − n + γ /π)

Proof. Since ω(λ) = − sin{λπ + γ }, then ωλ (n − γ /π) = π(−1)n+1 . Applying Theo-
rem 3.1, Eq. (4.8) is proved. ✷

Remark 4.3. If we take P (x) ≡ 0, the corresponding sampling representation will have
the following vector-valued form of the WKS sampling series:
∞
 sin π(λ − n)
F(λ) = F(n) . (4.9)
n=−∞
π(λ − n)

Kramer’s lemma, Lemma 1.1 above, gave an extension of sampling theorems associ-
ated with Sturm–Liouville problems, see, e.g., [8,23,24], or those associated with general
self-adjoint problems, see, e.g., [4,6,25]. A biorthogonal version of Kramer’s lemma has
been developed [11], giving the extension of sampling and nonself-adjoint problems [4,
11]. A multivariate analog of the same lemma is derived [27] extending sampling theorems
associated with Dirichlet problems. It is our aim now to give a vector-valued Kramer-type
lemma. Thus we have Lemma 4.4 below. Since the proof is similar to the classical one
116 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117

which may be found in [11,14,22], it is not given here. An interested reader can deduce the
proof using Parseval’s equality and uniform convergence techniques.

In the following Hν denote the Hilbert space

 
Hν := Y(x) = (y1 (x), y2 (x), . . . , yν (x))T : y1 , . . . , yν ∈ L2 (a, b) .
The inner product is defined by

b
Y, ZHν = YT (x)Z̄(x) dx.
a

Lemma 4.4. Let {λn }∞

n=−∞ be a sequence of numbers and K(·, λ) be the matrix
 
k11 (x, λ) k12 (x, λ) . . . k1ν (x, λ)

T  k21 (x, λ) k22 (x, λ) . . . k2ν (x, λ) 
K(·, λ) := k1 (x, λ), . . . , km (x, λ) =  ,
... ... ... ...
km1 (x, λ) km2 (x, λ) . . . kmν (x, λ)
(4.10)
where K(·, λ) is defined on [a, b] × C. Suppose that ki (·, λ) ∈ Hν for all λ and all i, 1

i
ν, and for all i, {ki (·, λn )}∞
n=−∞ is an orthogonal basis of Hν . Hence the vector-valued
transform

b
F(λ) = K(x, λ)f(x) dx, f ∈ Hν , (4.11)
a
can be recovered via the pointwise convergent vector-valued sampling representation
∞ b T
 K (x, λ)K̄(x, λn ) dx
F(λ) = F(λn )  ba . (4.12)
T
n=−∞ a K (x, λn )K̄(x, λn ) dx
b
The sampling series converges uniformly on the subsets of the λ-domain where a KT (x, λ)
× K̄(x, λ) dx is bounded.

It is not hard to see that Theorem 3.1 above as well as the results of [29] are special
cases of Lemma 4.4 above.

References

[1] M.H. Annaby, H.A. Hassan, A sampling theorem associated with boundary-value problems with not neces-
sarily simple eigenvalues, Internat. J. Math. Math. Sci. 21 (1998) 571–580.
[2] M.H. Annaby, A.I. Zayed, On the use of Green’s function in sampling theory, J. Integral Equations Appl. 10
(1998) 117–139.
[3] M.H. Annaby, G. Freiling, On sampling associated with Kamke problems, Math. Z. 234 (2000) 163–189.
[4] M.H. Annaby, Interpolation expansions associated with nonselfadjoint boundary value problems, Comm.
Appl. Anal. 3 (1999) 545–561.
 M.Z. Abd-Alla, M.H. Annaby / J. Math. Anal. Appl. 284 (2003) 104–117 117

[5] P.L. Butzer, A survey of the Whittaker–Shannon sampling theorem and some of its extensions, J. Math. Res.
Exposition 3 (1983) 185–212.
[6] P.L. Butzer, G. Schöttler, Sampling theorems associated with fourth and higher order self-adjoint eigenvalue
problems, J. Comput. Appl. Math. 51 (1994) 159–177.
[7] P.L. Butzer, J.R. Higgins, R. Stens, Sampling theory of signal analysis, to appear.
[8] W.N. Everitt, G. Schöttler, P.L. Butzer, Sturm–Liouville boundary value problems and Lagrange interpola-
tion series, Rend. Mat. Appl. 14 (1994) 87–126.
[9] A. Haddad, K. Yao, J. Thomas, General methods for the derivation of the sampling theorem, IEEE Trans.
Inform. Theory 13 (1967) 227–230.
[10] E. Hille, Analytic Function Theory, Vol. 1, Ginn, Boston, 1962.
[11] J.R. Higgins, Sampling Theory in Fourier and Signal Analysis: Foundations, Oxford Univ. Press, Oxford,
1996.
[12] A.J. Jerri, The Shannon sampling theorem—its various extensions and applications: A tutorial review, Proc.
IEEE 65 (1977) 1565–1598.
[13] V. Kotel’nikov, On the carrying capacity of the “ether” and wire in telecommunications, in: Material for the
All-Union Conference on Questions, Izd. Red. Upr. Svyazi RKKA, Moscow, 1933.
[14] H.P. Kramer, A generalized sampling theorem, J. Math. Phys. 38 (1959) 68–72.
[15] R. Lasser, Introduction to Fourier Series, Dekker, New York, 1996.
[16] B. Levin, Distribution of Zeros of Entire Functions, American Mathematical Society, Providence, RI, 1964.
[17] B.M. Levitan, I.S. Sergsjan, Introduction to Spectral Theory: Selfadjoint Ordinary Differential Operators, in:
Translations of Mathematical Monographs, Vol. 39, American Mathematical Society, Providence, RI, 1975.
[18] B.M. Levitan, I.S. Sergsjan, Sturm–Liouville and Dirac Operators, Kluwer Academic, Dordrecht, 1991.
[19] C.E. Shannon, Communications in the presence of noise, Proc. IRE 37 (1949) 10–21.
[20] P. Weiss, Sampling theorems associated with Sturm–Liouville systems, Bull. Amer. Math. Soc. 163 (1957)
242.
[21] E. Whittaker, On the functions which are represented by the expansion of the interpolation theory, Proc.
Roy. Soc. Edinburgh Sect. A 35 (1915) 181–194.
[22] A.I. Zayed, Advances in Shannon’s Sampling Theory, CRC Press, Boca Raton, 1993.
[23] A.I. Zayed, G. Hinsen, P.L. Butzer, On Lagrange interpolations and Kramer-type sampling theorems asso-
ciated with Sturm–Liouville problems, SIAM J. Appl. Math. 50 (1990) 893–909.
[24] A.I. Zayed, On Kramer’s sampling theorem associated with general Sturm–Liouville problems, SIAM J.
Appl. Math. 52 (1991) 575–604.
[25] A.I. Zayed, M.A. EL-Sayed, M.H. Annaby, On Lagrange interpolations and Kramer’s sampling theorem
associated with self-adjoint boundary value problems, J. Math. Anal. Appl. 158 (1991) 269–284.
[26] A.I. Zayed, A new role of Green’s functions in interpolation and sampling theory, J. Math. Anal. Appl. 175
(1993) 222–238.
[27] A.I. Zayed, Kramer’s sampling theorem for multidimensional signals and its relationship with Lagrange-
type interpolation, J. Multidimensional Syst. Signal Process. 3 (1992) 323–340.
[28] A.I. Zayed, A.G. García, Kramer’s sampling theorem with discontinuous kernels, Results Math. 34 (1998)
197–206.
[29] A.I. Zayed, A.G. García, Sampling theorem associated with a Dirac operator and Hartley transform, J. Math.
Anal. Appl. 214 (1997) 587–598.