Home Search Collections Journals About Contact us My IOPscience

Basic Sturm–Liouville problems

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2005 J. Phys. A: Math. Gen. 38 3775

(http://iopscience.iop.org/0305-4470/38/17/005)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 65.39.15.37
This content was downloaded on 18/03/2015 at 18:37

Please note that terms and conditions apply.

INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS A: MATHEMATICAL AND GENERAL

J. Phys. A: Math. Gen. 38 (2005) 3775–3797 doi:10.1088/0305-4470/38/17/005

Basic Sturm–Liouville problems

M H Annaby and Z S Mansour
Department of Mathematics, Faculty of Science, Cairo University, Giza, Egypt

E-mail: mhannaby@yahoo.com and zeinabs98@hotmail.com

Received 24 December 2004, in final form 3 March 2005

Published 13 April 2005
Online at stacks.iop.org/JPhysA/38/3775

Abstract
This paper is devoted to studying a q-analogue of Sturm–Liouville eigenvalue
problems. We formulate a self-adjoint q-difference operator in a Hilbert space.
Some of the properties of the eigenvalues and the eigenfunctions are discussed.
Green’s function is constructed and the problem in question is inverted into
a q-type Fredholm integral operator with a symmetric kernel. The set of
eigenfunctions is shown to be a complete orthogonal set in the Hilbert space.
Examples involving basic trigonometric functions are involved.

PACS numbers: 02.30.Ks, 02.30.Sa, 02.30.Tb

Mathematics Subject Classification: 33D15, 34L10, 39A13

1. Introduction

Let [a, b] ⊆ R be a finite closed interval and ν(·) be a continuous real-valued function defined
on [a, b]. By a Sturm–Liouville problem we mean the problem of finding a function y(·) and
a number λ ∈ C satisfying the differential equation
Ly := −y  + ν(x)y(x) = λy(x), a
x
b, (1.1)
together with the boundary conditions

U1 (y) := a1 y(a) + a2 y  (a) = 0, (1.2)


U2 (y) := b1 y(b) + b2 y (b) = 0, (1.3)

where ai and bi , i = 1, 2, are real numbers for which

|a1 | + |a2 | = 0 = |b1 | + |b2 |. (1.4)
This problem has been extensively studied. It is known that the differential equation (1.1) and
the boundary conditions (1.2), (1.3) determine a self-adjoint operator in L2 (a, b). There is a
sequence of real numbers {λn }∞n=0 with ∞ as the unique limit point such that corresponding
to each λn there is one and only one linearly independent solution of the problem (1.1)–(1.3).

0305-4470/05/173775+23$30.00 © 2005 IOP Publishing Ltd Printed in the UK 3775

3776 M H Annaby and Z S Mansour

The sequence {λn }∞n=0 is called the sequence of eigenvalues and the sequence of corresponding
solutions {φn (·)}∞
n=0 is said to be a sequence of eigenfunctions. One of the most important
properties of these eigenfunctions is that {φn (·)}∞ 2
n=0 is an orthogonal basis of L (a, b). For
example, let ν(x) ≡ 0 on [a, b]. If we take a = 0, b = π , a1 = 1, a2 = 0, b1 = 1 and b2 = 0,
we get
λn = n2 , φn (x) = sin nx, n = 1, 2, . . . , (1.5)
leading to the well-known fact that {sin nx}∞ 2
is a complete orthogonal set of L (0, π ), while
n=1
taking a = 0, b = π, a1 = 0, a2 = 1, b1 = 0 and b2 = 1, we get
λn = n2 , φn (x) = cos nx, n = 0, 1, 2, . . . , (1.6)
which leads to the completeness of {cos nx}∞
n=0
2
in L (0, π ). We mention here that the Fourier
orthogonal basis {einx }∞ 2
n=0 of L (−π, π ) will not be extracted from this setting but from a
simpler situation, namely the first-order problem
−iy  = λy, y(−π ) = y(π ). (1.7)
Among several references for the above-mentioned facts we mention the monographs of
Coddington and Levinson [12], Estham [13], Levitan and Saragsjan [30, 31], Marchenko [33]
and Titchmarsh [39].
The discrete analogue of the theory outlined above, i.e. when the differential operator d/dx
is replaced by the forward difference operator y(n) = y(n + 1) − y(n) and the backward
operator ∇y(n) = y(n) − y(n − 1) where n is a positive integer belonging to a finite set of
integers of the form {m, m + 1, m + 2, . . . , m + N, m  1}, is treated in Atkinson’s [5] (see
also [27]).
The aim of this paper is to study a basic analogue of Sturm–Liouville systems when
the differential operator is replaced by the q-difference operator Dq (see (2.12), (2.13)).
In [14, chapter 5] and [15], a basic Sturm–Liouville system is defined. It is the system

Dq (r(x)Dq y) + (l(x) + λw(x))y(qx) = 0, a
x
b, (1.8)

h1 y(a) + h2 Dq y(a) = 0, (1.9)
k1 y(b) + k2 Dq y(b) = 0, (1.10)

where r(·), l(·) and w(·) are the real-valued functions which possess appropriate q-derivatives,
h1 , h2 , k1 , k2 are constants. It is proved [14, pp 164–70] that all eigenvalues of this system
are real and the eigenfunctions satisfy an orthogonality relation [14, equation (5.1.5)]. Exton
[14] considered only formal computational aspects of this problem in order to prove certain
orthogonality relations of some q-special functions. There is no attention paid to several
points, which may lead to several mistakes. First the existence of eigenvalues is not proved
and it is not indicated how to determine the eigenvalues and the eigenfunctions. A basic
point here is that if a = 0 = b, then it is not guaranteed that initial conditions at either a
or b determine a unique solution of (1.8) (see [32]). The geometric and algebraic simplicity
of the eigenvalues, which play a major role in proving the reality of the eigenvalues and the
orthogonality of the eigenfunctions, are not proved or even assumed. Moreover the space
where the problem is defined is not specified. If an inner product is defined in the view of
[14, equation (5.1.5)], there will be no orthogonality if h1 , h2 , k1 and k2 are not real. For more
information concerning the monograph [14], see the review by Ismail in [21]. See also the
review of [15] by Hahn in [20]. In the present paper we formulate a self-adjoint basic Sturm–
Liouville eigenvalue problem in a Hilbert space. We prove the existence of a sequence of
real eigenvalues with no finite limit points. The associated Green function is constructed and
Basic Sturm–Liouville problems 3777

the equivalence between the basic Sturm–Liouville problem and a q-type Fredholm integral
operator is proved. Illustrative examples are given at the end of this paper.
There are several physical models involving q-(basic) derivatives, q-integrals, q-functions
and their related problems (see, e.g., [11, 16, 17, 19, 34]). Also the problem of expendability of
functions in terms of q-orthogonal functions, which seems to be first discussed by Carmichael
in [9, 10], has attracted the work of several authors (see, e.g., [7, 8, 23, 35, 36]). However, as
far as we know, there is no study of the general problem as we do in the present setting. At
this point, it is worth mentioning that our work based on the q-difference operator which is
attributed to Jackson, see [25], and a similar study of the Stum–Liouville systems generated
by the Askey–Wilson derivative, cf [4] is very much needed.

2. q-Notation and results

In this section we introduce some of the required q-notation and results. Throughout this
paper q is a positive number with 0 < q < 1. We start with the q-shifted factorial, see [18],
for a ∈ C,


1, n = 0,
(a; q)n := n−1 (2.1)
i=0 (1 − aq ), n = 1, 2, . . . .
i

The multiple q-shifted factorial for complex numbers a1 , . . . , ak is defined by


k
(a1 , a2 , . . . , ak ; q)n := (aj ; q)n . (2.2)
j =1

The limit of (a; q)n as n tends to infinity exists and will be denoted by (a; q)∞ . Let r φs denote
the q-hypergeometric series
    ∞
a1 , . . . , ar  (a1 , . . . , ar ; q)n n
r φs q, x = x (−q (n−1)/2 )n(s+1−r) . (2.3)
b1 , . . . , bs  (q, b1 , . . . , bs ; q) n
n=0

The third type of q-Bessel function is defined by

  
ν (q
ν+1
; q)∞ 0 
(3)
Jν (x; q) := x 1 φ1
2
q; qx , ν > −1, (2.4)
(q; q)∞ q ν+1 
see [22, 24]. This function is called in some literature the Hahn–Exton q-Bessel function (see
[37]). Since the other types of the q-Bessel functions, i.e. Jν(1) (·; q), Jν(2) (·; q), see [22], will
not be used here we use the notation Jν (·; q) for Jν(3) (·; q). The basic trigonometric functions
cos(x; q) and sin(x; q) are defined on C by
∞ 2
q n (x(1 − q))2n
cos(x; q) := (−1)n (2.5)
n=0
(q; q)2n

(q 2 ; q 2 )∞ √
= 2
(x(1 − q))1/2 J−1/2 (x(1 − q)/ q; q 2 ), (2.6)
(q; q )∞

∞
q n(n+1) (x(1 − q))2n+1
sin(x; q) := (−1)n (2.7)
n=0
(q; q)2n+1

(q 2 ; q 2 )∞
= (x(1 − q))1/2 J1/2 (x(1 − q); q 2 ), (2.8)
(q 3 ; q 2 )∞
3778 M H Annaby and Z S Mansour

and they are q-analogues of the cosine and sine functions [2, 18]. It is proved in [29] that
Jν (·, q 2 ) has an infinite number of real and simple zeros only. Moreover, cf [1], if wm
(ν)
, m  1,
2
denote the positive zeros of Jν (·, q ) and q 2ν+2
< (1 − q ) , then
2 2

∞

(ν)
(ν)
wm = q −m+m , m(ν) < ∞, 0
m(ν) < 1. (2.9)
m=1
(ν) ∞
It is also proved in [1] that for any q ∈ (0, 1), the zeros {wm }m=0 of Jν (z; q 2 ) have form (2.9)
for sufficiently large m. From (2.6), (2.8) and (2.9) sin(·; q) and cos(·; q) have only real and
simple zeros {0, ±xm }∞ ∞
m=1 and {±ym }m=1 respectively, where xm , ym > 0, m  1 and

(−1/2)
xm = (1 − q)−1 q −m+m , if q < (1 − q 2 )2 , (2.10)
(1/2)
ym = (1 − q)−1 q −m+1/2+m
, if q 3 < (1 − q 2 )2 . (2.11)

Moreover, for any q ∈ (0, 1), (2.10) and (2.11) hold for sufficiently large m.
For µ ∈ R, a set A ⊆ R is called a µ-geometric set if µx ∈ A for all x ∈ A. If A ⊆ R
is a µ-geometric set, then it contains all geometric sequences {xµn }∞ n=0 , x ∈ A. Let f be a
function, real or complex-valued, defined on a q-geometric set A. The q-difference operator
is defined by
f (x) − f (qx)
Dq f (x) := , x ∈ A/{0}. (2.12)
x − qx
If 0 ∈ A, the q-derivative at zero is defined by
f (xq n ) − f (0)
Dq f (0) := lim , x ∈ A, (2.13)
n→∞ xq n
if the limit exists and does not depend on x. A right inverse to Dq , the Jackson q-integration,
cf [26], is given by
 x ∞

f (t) dq t := x(1 − q) q n f (xq n ), x ∈ A, (2.14)
0 n=0

provided that the series converges, and

 b  b  a
f (t) dq t := f (t) dq t − f (t) dq t, a, b ∈ A. (2.15)
a 0 0

Lemma 2.1. The necessary and sufficient condition for the existence of the q-integral (2.14)
is that limk→∞ xq k f (xq k ) = 0.

Proof. Based on the fact that, for x ∈ C

lim xq k f (xq k ) = 0 ⇐⇒ ∃α ∈ [0, 1[∃C > 0, |f (xq k )|
C|xq k |−α , k ∈ N. 
k→∞

Kac and Cheung [28, p 68] have proved that if x α f (x) is bounded on [0, a] for some
x
0
α < 1, then 0 f (t) dq t exists for all x ∈ [0, a]. It is not hard to see that
 x  a
Dq f (t) dq t = f (x), Dq f (t) dq t = f (a) − f (0). (2.16)
0 0
The second identity in (2.16) occurs when f (·) is q-regular at zero, i.e.
lim f (xq n ) = f (0), for all x ∈ A. (2.17)
n→∞
Basic Sturm–Liouville problems 3779

The non-symmetric Leibniz’ rule

Dq (f g)(x) = g(x)Dq f (x) + f (qx)Dq g(x) (2.18)
holds. Relation (2.18) can be symmetrized using the relation f (qx) = f (x) −
x(1 − q)Dq f (x), giving the additional term x(1 − q)Dq f (x)Dq g(x). Also the rule of
q-integration by parts is nothing but
 a  a
g(x)Dq f (x) dq x = (f g)(a) − lim (f g)(aq n ) − Dq g(x)f (qx) dq x. (2.19)
0 n→∞ 0
If f, g are q-regular at zero, then limn→∞ (f g)(aq n ) on the right-hand side of (2.19) will be
replaced by (f g)(0). A self-contained q-calculus may be found in [32]. In the following we
define Hilbert spaces where our q-Sturm–Liouville problem will be considered. Let L2q (0, a)
be the space of all complex-valued functions defined on [0, a] such that
 a 1/2
f := |f (x)|2 dq x < ∞. (2.20)
0

The space L2q (0, a)is a separable Hilbert space with the inner product
 a
f, g := f (x)g(x) dq x, f, g ∈ L2q (0, a), (2.21)
0
and the orthonormal basis

√ 1 , x = aq n ,
ϕn (x) = x(1 − q) (2.22)

0, otherwise,
n = 0, 1, 2, . . . , cf [3]. The space L2q ((0, a) × (0, a)) is the space of all complex-valued
functions f (x, t) defined on [0, a] × [0, a] such that
 a  a 1/2
f (·, ·) 2 := |f (x, t)|2 dq x dq t < ∞. (2.23)
0 0

Lemma 2.2. L2q ((0, a) × (0, a)) is a separable Hilbert space with the inner product
 a a
f, g2 := f (x, t)g(x, t) dq x dq t. (2.24)
0 0

Proof. Similar to [3, pp 217–8], L2q ((0, a) × (0, a)) is a Banach space. The elements of
L2q ((0, a) × (0, a)) are equivalence classes where f, g are in the same equivalence class if
f (aq m , aq n ) = g(aq m , aq n ), m, n ∈ N. The zero element is the equivalence class of all
functions f (x, t) which satisfy f (aq m , aq n ) = 0, for all m, n ∈ N. To prove separability, it
suffices to prove that
φij (x, t) := φi (x)φj (t), i, j = 1, 2, . . . , (2.25)
is an orthonormal basis of L2q ((0, a) × (0, a)) whenever {φi (·)}∞ i=1 is an orthonormal basis of
L2q (0, a). Indeed,
 a a
φj k , φmn 2 = φj (x)φk (t)φm (x)φn (t) dq x dq t
 a
0 0
 a
= φj (x)φm (x) dq x φk (t)φn (t) dq t = δj m δkn ,
0 0
3780 M H Annaby and Z S Mansour

proving orthogonality. To prove that {φij } is a basis, we prove that if there exists
f ∈ L2q ((0, a) × (0, a)) such that f, φij 2 = 0, then f is the zero element. Indeed,
 a a
0 = f, φij  = f (x, t)φi (x)φj (t) dq x dq t
0
 a 0
 a   a
= φj (t) f (x, t)φi (x) dq x dq t = h(t)φj (t) dq t.
0 0 0

Thus
 a
h(t) := f (x, t)φi (x) dq x (2.26)
0
is orthogonal to the φj which implies that h(aq n ) = 0 for all n ∈ N. So, f (x, aq n ) is
orthogonal to each φi . Consequently, f (aq m , aq n ) = 0, for all m, n ∈ N. 

3. Fundamental solutions

In this section we investigate the fundamental solutions of the basic Sturm–Liouville equation
1
− Dq −1 Dq y(x) + ν(x)y(x) = λy(x), 0
x
a < ∞, λ ∈ C, (3.1)
q
where ν(·) is defined on [0, a] and continuous at zero. Let Cq2 (0, a) be the space of all functions
y(·) defined on [0, a] such that y(·), Dq y(·) are continuous at zero. Clearly, Cq2 (0, a) is a
subspace of the Hilbert space L2q (0, a). By a solution of equation (3.1), we mean a continuous
at zero function that satisfies (3.1) such that the function and its q-derivative have prescribed
values at x = 0. It is proved in [32] that (3.1) has a fundamental set of solutions which consists
of two linearly independent solutions {y1 (·), y2 (·)}. The q-Wronskian of y1 (·), y2 (·) is defined
to be
Wq (y1 , y2 )(x) := y1 (x)Dq y2 (x) − y2 (x)Dq y1 (x), x ∈ [0, a]. (3.2)
{y1 , y2 } forms a fundamental set of solutions if and only if their q-Wronskian does not vanish
at any point of [0, a]. See [32, 38].

Theorem 3.1. For c1 , c2 ∈ C, equation (3.1) has a unique solution in Cq2 (0, a) which satisfies
φ(0, λ) = c1 , Dq −1 φ(0, λ) = c2 , λ ∈ C. (3.3)
−1
Moreover φ(x, λ) is entire in λ for all x ∈ [0, a], where the q -derivative of a function f (x)
at zero is given by
f (xq −n ) − f (0)
Dq −1 f (x) = lim = Dq f (0). (3.4)
n→−∞ xq −n

Proof. The functions


 sin(sx; q) , λ = 0
ϕ1 (x, λ) = cos(sx; q), and ϕ2 (x, λ) = s (3.5)

x, λ = 0,
√
where s := λ is defined with respect to the principal branch, are a fundamental set of
1
Dq −1 Dq y(x) + λy(x) = 0, (3.6)
q
Basic Sturm–Liouville problems 3781

with the q-Wronskian Wq (ϕ1 (·, λ), ϕ2 (·, λ)) ≡ 1 (cf [32]). For all x ∈ [0, a], λ ∈ C, we
define a sequence {ym (·, λ)}∞
m=1 of successive approximations by

y1 (x, λ) = c1 ϕ1 (x, λ) + c2 ϕ2 (x, λ), (3.7)

ym+1 (x, λ) = c1 ϕ1 (x, λ) + c2 ϕ2 (x, λ)

 x
+q {ϕ2 (x, λ)ϕ1 (qt, λ) − ϕ1 (x, λ)ϕ2 (qt, λ)}ν(qt)ym (qt, λ) dq t. (3.8)
0

We prove that for each fixed λ ∈ C the uniform limit of {ym (·, λ)}∞
m=1 as m → ∞ exists and
defines a solution of (3.1) and (3.3). Let λ ∈ C be fixed. There exist positive numbers K(λ)
and A such that
K(λ)
|ν(x)|
A, |ϕi (x, λ)|
; i = 1, 2; x ∈ [0, a]. (3.9)
2


Let K(λ) := (|c1 | + |c2 |) K(λ) 
. Then, from (3.9), |y1 (x, λ)|
K(λ), for all x ∈ [0, a]. Using
2
mathematical induction, we have
m(m+1) (AK(λ)x(1 − q))
m

|ym+1 (x, λ) − ym (x, λ)|
K(λ)q 2 , m = 1, 2, . . . . (3.10)
(q; q)m
Consequently by Weierstrass’ test the series
∞

y1 (x, λ) + ym+1 (x, λ) − ym (x, λ) (3.11)
m=1
converges uniformly in [0, a]. Since the mth partial sum of the series is nothing but ym+1 (·, λ),
then ym+1 (·, λ) approaches a function φ(·, λ) uniformly in [0, a] as m → ∞, where φ(x, λ)
is the sum of the series. We can also prove by induction on m that ym (x, λ) and Dq ym (x, λ)
are continuous at zero, where
Dq ym+1 (x, λ) = c1 Dq ϕ1 (x, λ) + c2 Dq ϕ2 (x, λ)
 x
+q {Dq φ2 (x, λ)ϕ1 (qt, λ) − Dq ϕ1 (x, λ)ϕ2 (qt, λ)}ν(qt)ym (qt, λ) dq t,
0
(3.12)
m = 1, 2, . . . . Hence, both φ(·, λ) and Dq φ(·, λ) are continuous at zero, i.e. φ(·, λ) ∈
Cq2 (0, a). Because of the uniform convergence, letting m → ∞ in (3.8) we obtain
φ(x, λ) = c1 ϕ1 (x, λ) + c2 ϕ2 (x, λ)
 x
+q {ϕ2 (x, λ)ϕ1 (qt, λ) − ϕ1 (x, λ)ϕ2 (qt, λ)}ν(qt)φ(qt, λ) dq t. (3.13)
0
We prove that φ(·, λ) satisfies (3.1) and (3.3). Clearly φ(0, λ) = c1 and
φ(xq n , λ) − φ(0, λ)
Dq −1 φ(0, λ) = lim = c1 Dq −1 φ1 (0, λ) + c2 Dq −1 φ2 (0, λ) = c2 , (3.14)
n→∞ xq n
i.e. φ(x, λ) satisfies (3.3). To prove that φ(·, λ) satisfies (3.1), we distinguish between two
cases, x = 0 and x = 0. If x = 0, then from (2.16),
Dq φ(x, λ) = c1 Dq ϕ1 (x, λ) + c2 Dq ϕ2 (x, λ)
 x
+q {Dq φ2 (x, λ)ϕ1 (qt, λ) − Dq ϕ1 (x, λ)ϕ2 (qt, λ)}ν(qt)φ(qt, λ) dq t
0
(3.15)
3782 M H Annaby and Z S Mansour

and hence
  x 
−1 −1
Dq −1 Dq φ(x, λ) = Dq −1 Dq ϕ1 (x, λ) c1 − ϕ2 (qt, λ)ν(qt)φ(qt, λ) dq t
q q 0
  x 
1
− Dq −1 Dq ϕ2 (x, λ) c2 + ϕ1 (qt, λ)ν(qt)φ(qt, λ) dq t − ν(x)φ(x, λ).
q 0
(3.16)
Substituting from (3.5), (3.6) and (3.13) into (3.16), we conclude that φ(x, λ) satisfies (3.1)
for x = 0. If x = 0, then (3.1) is nothing but
Dq2 y(0) − qν(0)y(0) = −qλy(0). (3.17)
Obviously
Dq2 φ(0, λ) = c1 Dq2 ϕ1 (0, λ) + c2 Dq2 ϕ2 (0, λ) + qν(0)φ(0, λ)
= −c1 qλϕ1 (0, λ) − c2 qλϕ2 (0, λ) + qν(0)φ(0, λ)
= −qλφ(0, λ) + qν(0)φ(0, λ) (3.18)
Hence φ(x, λ) satisfies (3.1). To prove that problem (3.1), (3.3) has a unique solution,
suppose on the contrary that ψi (·, λ), i = 1, 2, are two solutions of (3.1), (3.3). Let
χ (x, λ) = ψ1 (x, λ) − ψ2 (x, λ), x ∈ [0, a]. Then χ (·, λ) is a solution of (3.1) subject to
the initial conditions χ (0, λ) = Dq −1 χ (0, λ) = 0. Applying the q-integration process twice
to (3.1) yields
 x
χ (x, λ) = (x − t)(λ − ν(t))χ (t, λ) dq t. (3.19)
0
Since χ (x, λ), ν(x) are continuous at zero, then there exist positive numbers Nx,λ , Mx,λ
such that
Nx,λ = sup |χ (xq n , λ)|, Mx,λ = sup |λ − ν(xq n )|. (3.20)
n∈N n∈N
Again we can prove by mathematical induction on k that
2 x 2k
|χ (x, λ)|
Nx,λ Mx,λ
k
q k (1 − q)2k , k ∈ N, x ∈ [0, a]. (3.21)
(q; q)2k
Indeed, if (3.21) holds at k ∈ N, then from (3.19)
2k  x
k+1 k 2 (1 − q)
|χ (x, λ)|
Nx,λ Mx,λ q (x − t)t 2k dq t
(q; q)2k 0
k+1 k 2 2k+1 (1 − q)
2k+2
= Nx,λ Mx,λ q q x 2k+2
(q; q)2k+2
k+1 (k+1)2 (1 − q)
2k+2
= Nx,λ Mx,λ q x 2k+2 . (3.22)
(q; q)2k+2
Hence (3.21) holds at k + 1. Consequently (3.21) holds for all k ∈ N because from (3.20)
2 x 2k
it holds at k = 0. Since limk→∞ Mx,λ k
q k (1 − q)2k (q;q) 2k
= 0, then χ (x, λ) = 0, for all
x ∈ [0, a]. This proves the uniqueness. Now, Let M > 0 be arbitrary but fixed. To prove that
φ(x, λ), x ∈ [0, a], is entire in λ, it is sufficient to prove that φ(x, λ) is analytic in each disc
M ; M := {λ ∈ C : |λ|
M}. We prove by induction on m that

for all x ∈ [0, a] ym (x, λ) is analytic on M, (3.23)

∂
for all λ∈ M ym (x, λ) is continuous at (0, λ). (3.24)
∂λ
Basic Sturm–Liouville problems 3783

Clearly, ϕ1 (x, λ), ϕ2 (x, λ) are entire functions of λ for x ∈ [0, a]. Moreover ∂λ ∂
φi (x, λ) is
continuous at (0, λ) for each λ ∈ C. Then (3.23) and (3.24) hold at m = 1. Assume that
(3.23) and (3.24) hold at m  1. Then for x0 ∈ [0, a], λ0 ∈ M , we obtain
   x0
∂ym+1 (x0 , λ)  ∂ϕ2 (x0 , λ) 
 =q  ϕ1 (qt, λ)ym (qt, λ) dq t
∂λ λ=λ0 ∂λ λ=λ0 0
   x0
∂y1 (x0 , λ)  ∂ϕ1 (x0 , λ) 
+  − q  ϕ2 (qt, λ)ym (qt, λ) dq t
∂λ λ=λ0 ∂λ λ=λ0 0
 x0 
∂ 
+ qϕ2 (x0 , λ) ϕ1 (qt, λ)ym (qt, λ) dq t 
∂λ 0 λ=λ
 x0  0
∂ 
− qϕ1 (x0 , λ) ϕ2 (qt, λ)ym (qt, λ) dq t  . (3.25)
∂λ 0 λ=λ0

From (3.24) we conclude that ∂/∂λ(ϕi (qt, λ)ym (qt, λ)), i = 1, 2, are continuous at (0, λ0 ).
Therefore there exist constants C, δ > 0 such that
 
∂ 
 (ϕi (x0 q n , λ)ym (x0 q n , λ))
C, n ∈ N, |λ − λ0 |
δ. (3.26)
 ∂λ 
Hence
 
∂ 

x0 (1 − q)q  (ϕi (x0 q , λ)ym (x0 q λ))
x0 A(1 − q)q n ,
n n+1 n+1
n ∈ N,
∂λ
for all λ in the disc |λ − λ0 |
δ. i.e. the series corresponding to the q-integrals
 x0
∂
(ϕi (qt, λ)ym (qt, λ)) dq t, i = 1, 2 (3.27)
0 ∂λ
are uniformly convergent in a neighbourhood of λ = λ0 . Thus, we can interchange the
differentiation and integration processes in (3.25). Since x0 , λ0 are arbitrary, then
 x
∂ ∂ ∂
ym+1 (x, λ) = y1 (x, λ) − q (ϕ2 (x, λ)ϕ1 (qt, λ)ym (qt, λ))ν(qt) dq t
∂λ ∂λ ∂λ
 x 0
∂
+q (ϕ1 (x, λ)ϕ2 (qt, λ)ym (qt, λ))ν(qt) dq t, (3.28)
0 ∂λ
for all x ∈ [0, a], λ ∈ M . From (3.24) the integrals in (3.28) are continuous at (0, λ).
Consequently ∂λ ∂
ym+1 (x, λ) is continuous at (0, λ). Let x0 ∈ [0, a] be arbitrary. Then there

exists B(x0 ), B(x0 ) > 0 such that
B(x0 )  0 ),
|ϕi (x0 , λ)|
, i = 1, 2, |y1 (x, λ)|
B(x λ∈ M. (3.29)
2
Finally the use of the mathematical induction yields

 0 )q
m(m+1) (AB(x0 )λ(1 − q))m
|ym+1 (x0 , λ) − ym (x0 , λ)|
B(x 2 . (3.30)
(q; q)m
Consequently the series (3.11), with x = x0 , converges uniformly in M to φ(x0 , λ). Hence
φ(x0 , λ) is analytic in M , i.e. it is entire. 

4. The self-adjoint problem

In this section we define a basic Sturm–Liouville problem and prove that it is self-adjoint
in L2q (0, a). The following lemma which is needed in the following indicates that unlike
3784 M H Annaby and Z S Mansour

the classical differential operator d/dx, Dq is neither self-adjoint nor skew self-adjoint.
Equation (4.2) indicates that the adjoint of Dq is − q1 Dq −1 .

Lemma 4.1. Let f (·), g(·) in L2q (0, a) be defined on [0, q −1 a]. Then, for x ∈ (0, a], we have

(Dq g)(xq −1 ) = Dq,xq −1 g(xq −1 ) = Dq −1 g(x), (4.1)

 
−1
Dq f, g = f (a)g(aq −1 ) − lim f (aq n )g(aq n−1 ) + f, Dq −1 g , (4.2)
n→∞ q
 
1
− Dq −1 f, g = lim f (aq n−1 )g(aq n ) − f (aq −1 )g(a) + f, Dq g. (4.3)
q n→∞

Proof. Relation (4.1) follows from

g(x) − g(q −1 x) g(xq −1 ) − g(x)
Dq −1 g(x) = = = (Dq g)(xq −1 ) = Dq,xq −1 g(xq −1 ).
x(1 − q −1 ) xq −1 (1 − q)
(4.4)
Using formula (2.19) of q-integration by parts we obtain
 a
Dq f, g = Dq f (x)g(x) dq x
0
 a
= f (a)g(a) − lim f (aq n )g(aq n ) − f (qt)Dq g(t) dq t
n→∞
 qa
0
1
= f (a)g(a) − lim f (aq n )g(aq n ) − f (t) Dq −1 g(t) dq t
n→∞ 0 q
= f (a)g(a) − lim f (aq n )g(aq n ) + aq −1 (1 − q)f (a)Dq −1 g(a)
n→∞
 a
−1
+ f (t) Dq −1 g(t) dq t
0 q
 
−1
= f (a)g(aq −1 ) − lim f (aq n )g(aq n−1 ) + f, Dq g , (4.5)
n→∞ q
proving (4.2). Equation (4.3) can be proved by the use of (4.2). 

Now consider the basic Sturm–Liouville problem

1
(y) := − Dq −1 Dq y(x) + ν(x)y(x) = λy(x), 0
x
a < ∞, λ ∈ C, (4.6a)
q
U1 (y) := a11 y(0) + a12 Dq −1 y(0) = 0, (4.6b)
U2 (y) := a21 y(a) + a22 Dq −1 y(a) = 0, (4.6c)

where ν(·) is a continuous at zero real-valued function and {aij }, i, j ∈ {1, 2} are arbitrary real
numbers such that the rank of the matrix (aij )1
i,j
2 is 2. Problem (4.6a)–(4.6c) is said to be
formally self-adjoint if for any functions y(·) and z(·) of Cq2 (0, a) which satisfy (4.6b), (4.6c),
y, z = y, z. (4.7)

Theorem 4.2. The basic Sturm–Liouville eigenvalue problem (4.6a)–(4.6c) is formally self-
adjoint.
Basic Sturm–Liouville problems 3785

Proof. We first prove that for y(·), z(·) in L2q (0, a), we have the following q-Lagrange’s
identity
 a
(y(x)z(x) − y(x)z(x)) dq x = [y, z](a) − lim [y, z](aq n ), (4.8)
0 n→∞

where
[y, z](x) := y(x)Dq −1 z(x) − Dq −1 y(x)z(x). (4.9)
Applying (4.3) with f (x) = Dq y(x) and g(x) = z(x), we obtain
 
1
− Dq Dq y(x), z(x) = −(Dq y)(aq −1 )z(a) + lim (Dq y)(aq n−1 )z(aq n ) + Dq y, Dq z
−1
q n→∞

= −Dq −1 y(a)z(a) + lim Dq −1 y(aq n )z(aq n ) + Dq y, Dq z.

n→∞
(4.10)
Applying (4.2) with f (x) = y(x), g(x) = Dq z(x),
 
1
Dq y, Dq z = y(a)Dq z(aq −1 ) − lim y(aq n )Dq z(aq n−1 ) + y, − Dq −1 Dq z
n→∞ q
 
1
= y(a)Dq −1 z(a) − lim y(aq n )Dq −1 z(aq n ) + y, − Dq −1 Dq z . (4.11)
n→∞ q
Therefore,
   
1 1
− Dq −1 Dq y(x), z(x) = [y, z](a) − lim [y, z](aq n ) + y, − Dq −1 Dq z . (4.12)
q n→∞ q
Lagrange’s identity (4.8) results from (4.12) and the reality of ν(x). Letting y(·), z(·) be in
Cq2 (0, a) and assuming that they satisfy (4.6b), (4.6c), we obtain
a11 y(0) + a12 Dq −1 y(0) = 0, a11 z(0) + a12 Dq −1 z(0) = 0. (4.13)
The continuity of y(·), z(·) at zero implies that limn→∞ [y, z](aq ) = [y, z](0). Then (4.12)
n

will be
   
1 1
− Dq −1 Dq y(x), z(x) = [y, z](a) − [y, z](0) + y, − Dq −1 Dq z . (4.14)
q q
Since a11 and a12 are not both zero, it follows from (4.13) that
[y, z](0) = y(0)Dq −1 z(0) − Dq −1 y(0)z(0) = 0.
Similarly,
[y, z](a) = y(a)Dq −1 z(a) − Dq −1 y(a)z(a) = 0.
Since ν(x) is real valued, then
 
1
(y), z = − Dq −1 Dq y(x) + ν(x)y(x), z(x)
q
 
1
= − Dq −1 Dq y(x), z(x) + ν(x)y, z(x)
q
 
1
= y, − Dq −1 Dq z(x) + y, ν(x)z(x) = y, (z),
q
i.e.  is a formally self-adjoint operator. 
A complex number λ∗ is said to be an eigenvalue of the problem (4.6a)–(4.6c) if there is a
non-trivial solution φ ∗ (·) which satisfies the problem at this λ∗ . In this case we say that φ ∗ (·)
3786 M H Annaby and Z S Mansour

is an eigenfunction of the basic Sturm–Liouville problem corresponding to the eigenvalue λ∗ .

The multiplicity of an eigenvalue is defined to be the number of linearly independent solutions
corresponding to it. In particular, an eigenvalue is simple if and only if it has only one linearly
independent solution.

Lemma 4.3. The eigenvalues and the eigenfunctions of the boundary value problem (4.6a)–
(4.6c) have the following properties:
(i) The eigenvalues are real.
(ii) Eigenfunctions that belong to different eigenvalues are orthogonal.
(iii) All eigenvalues are simple from the geometric point of view.

Proof. The proof is similar to that of differential equations, cf [13], and hence it is omitted.


In the following we indicate how to obtain the eigenvalues and the corresponding
eigenfunctions. Let φ1 (·, λ), φ2 (·, λ) be the linearly independent solutions of (4.6a)
determined by the initial conditions
Dqj −1 φi (0, λ) = δij , i, j = 1, 2, λ ∈ C. (4.15)
Thus φ1 (·, λ) is determined by (3.13) by taking c1 = 1, c2 = 0 and φ2 (·, λ) is determined by
taking c1 = 0, c2 = 1. Then, every solution of (4.6a) is of the form
y(x, λ) = A1 φ1 (x, λ) + A2 φ2 (x, λ), (4.16)
where A1 and A2 do not depend on x. A solution y(·, λ) of (4.6a) will be an eigenfunction if
it satisfies the boundary conditions (4.6b)–(4.6c), i.e. if we can find a non-trivial solution of
the linear system
A1 U1 (φ1 ) + A2 U1 (φ2 ) = 0, A1 U2 (φ1 ) + A2 U2 (φ2 ) = 0, (4.17)
Hence, λ ∈ R is an eigenvalue if and only if
 
U1 (φ1 ) U1 (φ2 )
(λ) =   = 0. (4.18)
U2 (φ1 ) U2 (φ2 )
The function (λ) defined in (4.18) is called the characteristic determinant associated with the
basic Sturm–Liouville problem (4.6a)–(4.6c). The zeros of (λ) are exactly the eigenvalues
of the problem. Since φ1 (x, λ) and φ2 (x, λ) are entire in λ for each fixed x ∈ [0, a], then (λ)
is also entire. Thus the eigenvalues of the basic Sturm–Liouville system (4.6a)–(4.6c) are at
most countable with no finite limit points. From lemma 4.3 we know that all eigenvalues are
simple from the geometric point of view. We can prove that the eigenvalues are also simple
algebraically, i.e. they are simple zeros of (λ). Indeed, let θ1 (·, λ) and θ2 (·, λ) be defined by
the relations
θ1 (x, λ) := U1 (φ2 )φ1 (x, λ) − U1 (φ1 )φ2 (x, λ),
(4.19)
θ2 (x, λ) := U2 (φ2 )φ1 (x, λ) − U2 (φ1 )φ2 (x, λ),
Hence, θ1 (·, λ), θ2 (·, λ) are solutions of (4.6a) such that
θ1 (0, λ) = a12 , Dq −1 θ1 (0, λ) = −a11 ; θ2 (a, λ) = a22 , Dq −1 θ2 (a, λ) = −a21 .
(4.20)
One can verify that
Wq (θ1 (·, λ), θ2 (·, λ))(x, λ) = (λ)Wq (φ1 (·, λ), φ2 (·, λ)) = (λ). (4.21)
Basic Sturm–Liouville problems 3787

Let λ0 be an eigenvalue of (4.6a)–(4.6c). Then λ0 is a real number and therefore θi (x, λ0 ) can
be taken to be real valued, i = 1, 2. From (4.21), we conclude that θ1 (x, λ0 ), θ2 (x, λ0 ) are
linearly dependent eigenfunctions. So, there exists a non-zero constant k0 such that
θ1 (x, λ0 ) = k0 θ2 (x, λ0 ). (4.22)
From (4.19) and (4.20)
θ1 (a, λ0 ) = k0 a22 = k0 θ1 (a, λ), Dq −1 θ1 (a, λ0 ) = −k0 a21 = k0 Dq −1 θ1 (a, λ). (4.23)
In the q-Lagrange identity (4.8), taking y(x) = θ1 (x, λ), and z(x) = θ1 (x, λ0 ) implies
 a
(λ − λ0 ) θ1 (x, λ)θ1 (x, λ0 ) dq x = θ1 (a, λ)Dq −1 θ1 (a, λ0 ) − Dq −1 θ1 (a, λ)θ1 (a, λ0 )
0
= k0 (θ1 (a, λ)Dq −1 θ2 (a, λ) − θ2 (a, λ)Dq −1 θ1 (a, λ))
= k0 Wq (θ1 (·, λ), θ2 (·, λ))(q −1 a) = k0 (λ).
Since (λ) is entire in λ,
 a
(λ) 1
 (λ0 ) := lim = θ12 (x, λ0 ) dq x = 0. (4.24)
λ→λ0 λ − λ0 k0 0
Therefore λ0 is a simple zero of (λ).

5. Basic Green’s function

The q-type Green’s function arises when we seek a solution of the nonhomogeneous equation
1
− Dq −1 Dq y(x) + {−λ + ν(x)}y(x) = f (x), x ∈ [0, a], λ ∈ C, (5.1)
q
which satisfies the boundary conditions (4.6b), (4.6c), where f (·) ∈ L2q (0, a) is given. First,
we note that if λ is not an eigenvalue of the Sturm–Liouville problem (4.6a)–(4.6c), then the
solution of (5.1), if it exists, would be unique. To see this, assume that χ1 (x, λ), χ2 (x, λ) are
two solutions of (5.1). Then χ1 (x, λ) − χ2 (x, λ) is a solution of the problem (4.6a)–(4.6c).
So, it is identically zero if λ is not an eigenvalue. Another proof of this assertion is included
in the proof of the next theorem.

Theorem 5.1. Suppose that λ is not an eigenvalue of (4.6a)–(4.6c). Let φ(·, λ) satisfy the
q-difference equation (5.1) and the boundary conditions (4.6b)–(4.6c), where f (·) ∈ L2q (0, a).
Then
 a
φ(x, λ) = G(x, t, λ)f (t) dq t, x ∈ {0, aq m , m ∈ N}, (5.2)
0
where G(x, t, λ) is Green’s function of problem (4.6a)–(4.6c) and it is given by


−1 θ2 (x, λ)θ1 (t, λ), 0
t
x,
G(x, t, λ) = (5.3)
(λ) θ1 (x, λ)θ2 (t, λ), x < t
a.
Conversely the function φ(·, λ) defined by (5.2) satisfies (5.1) and (4.6b)–(4.6c). Green’s
 t, λ) such
function G(x, t, λ) is unique in the sense that if there exists another function G(x,
that (5.2) is satisfied, then
 t, λ),
G(x, t, λ) = G(x, in L2q ((0, a) × (0, a)). (5.4)
If f (·) is q-regular at zero, then (5.2) holds for all x ∈ [0, a].
3788 M H Annaby and Z S Mansour

Proof. Using a q-analogue of the methods of variation of constants, a particular solution of

the non-homogenous equation (5.1) may be given by
φ(x, λ) = c1 (x)θ1 (x, λ) + c2 (x)θ2 (x, λ), (5.5)
where c1 (x), c2 (x) are solutions of the first-order q-difference equations
q q
Dq,x c1 (x) = θ2 (qx, λ)f (qx), Dq,x c2 (x) = − θ1 (qx, λ)f (qx). (5.6)
(λ) (λ)
From lemma 2.1, the functions Dq,x ci (x), i = 1, 2, are q-integrable on [0, t] if and only if
lim tq n θi (tq n+1 , λ)f (tq n+1 ) = 0, i = 1, 2. (5.7)
n→∞

Define the q-geometric set Af by

Af := {x ∈ [0, a]: lim xq n |f (xq n )|2 = 0}. (5.8)
n→∞

Af is a q-geometric set containing {0, aq m , m ∈ N} since f ∈ L2q (0, a). Hence, Dq ci (·),
i = 1, 2, are q-integrable on [0, x] for all x ∈ Af and appropriate solutions of (5.6) are
given by
 x
q
c1 (x) = c1 (0) + θ2 (qt, λ)f (qt) dq t, x ∈ Af (5.9)
(λ) 0
 a
q
c2 (x) = c2 (a) + θ1 (qt, λ)f (qt) dq t, x ∈ Af . (5.10)
(λ) x
That is the general solution of (5.1) is given by
 x
q
φ(x, λ) = c1 θ1 (x, λ) + c2 θ2 (x, λ) + θ1 (x, λ) θ2 (qt, λ)f (qt) dq t
(λ)
 a 0
q
+ θ2 (x, λ) θ1 (qt, λ)f (qt) dq t, (5.11)
(λ) x
where x ∈ Af , and c1 , c2 are arbitrary constants. Now, we determine c1 , c2 for which φ(x, λ)
satisfies (4.6b), (4.6c). It easy to see that
  a 
q
φ(0, λ) = c1 θ1 (0, λ) + c2 + θ1 (qt, λ)f (qt) dq t θ2 (0, λ),
(λ) 0
φ(xq n , λ) − φ(0, λ)
Dq −1 φ(0, λ) = lim
n→∞
x∈Af
xq n
  a 
q
= c1 Dq −1 θ1 (0, λ) + c2 + θ1 (qt, λ)f (qt) dq t Dq −1 θ2 (0, λ).
(λ) 0
The boundary condition a11 φ(0, λ) + a12 Dq −1 φ(0, λ) = 0 implies that
  a 
q
c2 + θ1 (qt, λ)f (qt) dq t Wq (θ1 , θ2 )(0) = 0. (5.12)
(λ) 0
Therefore,

−q a
c2 = θ1 (qt, λ)f (qt) dq t. (5.13)
(λ) 0
Hence,
 x
q
φ(x, λ) = c1 θ1 (x, λ) + (θ1 (x, λ)θ2 (qt, λ) − θ2 (x, λ)θ1 (qt, λ))f (qt) dq t. (5.14)
(λ) 0
Basic Sturm–Liouville problems 3789

Now we compute φ(a, λ) and Dq −1 φ(a, λ). Indeed, from the definition of the q-integration
(2.14) and relation (5.11)
 a
q
φ(a, λ) = c1 θ1 (a, λ) + (θ1 (a, λ)θ2 (qt, λ) − θ2 (a, λ)θ1 (qt, λ))f (qt) dq t
(λ) 0
 q −1 a
q
= c1 θ1 (a, λ) + (θ1 (a, λ)θ2 (qt, λ) − θ2 (a, λ)θ1 (qt, λ))f (qt) dq t
(λ) 0
and  
 q −1 a
q
Dq −1 φ(a, λ) = Dq −1 θ1 (a, λ) c1 + θ2 (qt, λ)f (qt) dq t
(λ) 0
 q −1 a
q
− Dq −1 θ2 (a, λ) θ1 (qt, λ)f (qt) dq t.
(λ) 0

The boundary condition a21 φ2 (a, λ) + a22 Dq −1 φ2 (a, λ) = 0 implies

  q −1 a 
q
c1 + θ2 (qt, λ)f (qt) dq t Wq (θ1 , θ2 )(a) = 0. (5.15)
(λ) 0

Hence
 q −1 a
−q
c1 = θ2 (qt, λ)f (qt) dq . (5.16)
(λ) 0

So for x ∈ Af
 x  q −1 a
−q q
φ(x, λ) = θ2 (x, λ) θ1 (qt, λ)f (qt) dq t − θ1 (x, λ) θ2 (qt, λ)f (qt) dq t
(λ) (λ)
 qx
0
 a x
−1 1
= θ2 (x, λ) θ1 (t, λ)f (t) dq t − θ1 (x, λ) θ2 (t, λ)f (t) dq t
(λ) 0 (λ) qx
 x  a
−1 1
= θ2 (x, λ) θ1 (t, λ)f (t) dq t − θ1 (x, λ) θ2 (t, λ)f (t) dq t.
(λ) 0 (λ) x

proving (5.2), (5.3). Conversely, by direct computations, if φ(x, λ) is given by (5.2), then
it is a solution of (5.1) and satisfies the boundary conditions (4.6b), (4.6c). To prove the
 t, λ), such that
uniqueness, suppose that there exists another function, G(x,
 a
ψ(x, λ) =  t, λ)f (t) dq t,
G(x, (5.17)
0

is a solution of (5.1) which satisfies (4.6b), (4.6c). For convenience, let

G1 (x, t, λ), 0
t
x,
G(x, t, λ) =
G2 (x, t, λ), x
t
a,


1 (x, t, λ),
G 0
t
x,
 t, λ) =
G(x,

G2 (x, t, λ), x
t
a.
By subtraction,
 a
 t, λ)}f (t) dq t = 0,
{G(x, t, λ) − G(x, x ∈ {0, aq m , m ∈ N}, (5.18)
0
3790 M H Annaby and Z S Mansour

 t, λ) and x = aq m ,
for all functions f (t) ∈ L2q (0, a). Let us take f (t) := G(x, t, λ) − G(x,
m ∈ N. Then
 a  aq m
m 
|G(aq , t, λ) − G(aq , t, λ)| dq t =
m 2 1 (aq m , t, λ)|2 dq t
|G1 (aq m , t, λ) − G
0
 a 0

+ 2 (aq m , t, λ)|2 dq t
|G2 (aq m , t, λ) − G
aq m
∞

= a(1 − q)  m , aq n , λ)|2 = 0
q n |G(aq m , aq n , λ) − G(aq (5.19)
n=0
Therefore, from (5.19) we conclude that
 m , aq n , λ),
G(aq m , aq n , λ) = G(aq m, n ∈ N.
If f (·) is q-regular at zero, then Af ≡ [0, a] and (5.2) will be defined for all x ∈ [0, a]. 
Theorem 5.2. Green’s function has the following properties:
(i) G(x, t, λ) is continuous at the point (0, 0).
(ii) G(x, t, λ) = G(t, x, λ).
(iii) For each fixed t ∈ (0, qa], as a function of x, G(x, t, λ) satisfies the q-difference equation
(4.6a) in the intervals [0, t), (t, a] and it also satisfies the boundary conditions (4.6b),
(4.6c).
(iv) Let λ0 be a zero of (λ). Then λ0 can be a simple pole of the function G(x, t, λ), and in
this case
−ψ0 (x)ψ0 (t) 
G(x, t, λ) = + G(x, t, λ), (5.20)
λ − λ0
 t, λ) is an analytic function of λ in a neighbourhood of λ0 and ψ0 (·) is a
where G(x,
normalized eigenfunction corresponding to λ0 .
Proof. (i) Follows from the continuity of θ1 (·, λ), θ2 (·, λ) at zero for each fixed λ ∈ C and (ii)
is easy to be checked. Now, we prove (iii). Let t ∈ (0, qa] be fixed. If x ∈ [0, t], then
1
G(x, t, λ) = θ1 (x, λ)θ2 (t, λ).
(λ)
So,
1 λ
G(x, t, λ) = θ2 (t, λ)θ1 (x, λ) = θ2 (t, λ)θ1 (x, λ) = λG(x, t, λ).
(λ) (λ)
Similarly if x ∈ [t, a]. From (4.20) and (5.3), we have
1
a11 G(0, t, λ) + a12 Dq −1 G(0, t, λ) = θ2 (t, λ){a11 θ1 (0, λ) + a12 Dq −1 θ1 (0, λ)} = 0,
(λ)
1
a21 G(a, t, λ) + a22 Dq −1 G(a, t, λ) = θ1 (t, λ){a21 θ1 (a, λ) + a22 Dq −1 θ1 (a, λ)} = 0.
(λ)
(iv) Let λ0 be a pole of G(x, t, λ), and R(x, t) be the residue of G(x, t, λ) at λ = λ0 . From
(4.22) and (4.24), we obtain
λ − λ0
R(x, t) = lim (λ − λ0 )G(x, t, λ) = k0−1 θ1 (x, λ0 )θ1 (t, λ0 ) lim
λ→λ0 λ→λ0 (λ)
θ1 (x, λ0 )θ1 (t, λ0 )
=− a = −ψ0 (x, λ0 )ψ1 (t, λ0 ).
0 |θ1 (u, λ)| dq u 
2
Basic Sturm–Liouville problems 3791

6. Eigenfunctions expansion formula

In this section, the existence of a countable sequence of eigenvalues of  with no finite limit
points will be proved by using the spectral theorem of compact self-adjoint operators in Hilbert
spaces (see, e.g., [6]). Moreover it will be proved that the corresponding eigenfunctions form
an orthonormal basis of L2q (0, a). We define the operator L : DL → L2q (0, a) to be Ly = y
for all y ∈ DL , where DL is the subspace of L2q (0, a) consisting of those complex-valued
functions y that satisfies (4.6b), (4.6c) such that Dq y(·) is q-regular at zero and Dq2 y(·) lies in
L2q (0, a). Thus L is the difference operator generated by the difference expression  and the
boundary conditions (4.6b), (4.6c). By L(y) = λy, we mean that (y) = λy and y satisfies
(4.6b), (4.6c). The operator L has the same eigenvalues of the basic Sturm–Liouville problem
(4.6a)–(4.6c). We assume without any loss of generality that λ = 0 is not an eigenvalue. Thus
ker L = {0}. From the previous section the solution of the problem
(Ly)(x) = f (x), f ∈ L2q (0, a), (6.1)
is given uniquely in L2q (0, a) by
 a
y(x) = G(x, t)f (t) dq t, (6.2)
0
where


cθ1 (t)θ2 (x), 0
t
x 1
G(x, t) = G(x, t, 0) = c := − . (6.3)
cθ1 (x)θ2 (t), x
t
a, Wq (θ1 , θ2 )
Replacing f (·) by λy(·) in (6.1), then the eigenvalue problem
(Ly)(x) = λy(x), (6.4)
is equivalent to the following basic Fredholm integral equation of the second kind:
 a
y(x) = λ G(x, t)y(t) dq t, in L2q (0, a). (6.5)
0
Let G be the integral operator
 a
G: L2q (0, a) → L2q (0, a), (G f )(x) = G(x, t)f (t) dq t. (6.6)
0
We prove that
(LG )f = f, f ∈ L2q (0, a). (6.7)
We show first that y = G f ∈ DL . From (6.5) and (6.6)
y(x) = (G f )(x) = θ2 (x)y1 (x) + θ1 (x)y2 (x), (6.8)
where
 x  a
y1 (x) = c θ1 (t)f (t) dq t, y2 (x) = c θ2 (t)f (t) dq t.
0 x
Thus, for all x ∈ Af , cf. (5.8),
Dq y(x) = Dq θ2 (x)y1 (qx) + Dq θ1 (x)y2 (qx), (6.9)

Dq2 y(x) = −qν(qx)(y)(qx) − qf (qx) ∈ L2q (0, a). (6.10)

Since Dq θi (x, λ), yi (x), i = 1, 2, are q-regular at zero, then so is Dq y(·) and
y(xq n ) − y(0)
Dq −1 y(0) = Dq y(0) = lim = Dq −1 θ2 (0)y2 (0),
n→∞
x∈Af
xq n
3792 M H Annaby and Z S Mansour

y1 (0) = 0, and y2 (a) = 0, then

a11 y(0) + a12 Dq −1 y(0) = (a11 θ1 (0) + a12 Dq −1 θ1 (0))y2 (0) = 0,
and
a21 y(a) + a22 Dq −1 y(a) = (a21 θ2 (a) + a22 Dq −1 θ2 (a))y1 (a) = 0.
Thus y ∈ DL . It follows from (6.10) that Ly = (LG )(f ) = f . Hence we have established
(6.7). Also, we can see that
(GL)(y) = y, y ∈ DL . (6.11)
Indeed, replacing f in (6.7) by Ly, we get Ly = LGLy. Thus y = GLy since L is assumed
to be injective. It follows from (6.7) and (6.11) that ker G = {0} and φ is an eigenfunction of
G with an eigenvalue µ if and only if φ is an eigenfunction of L with an eigenvalue 1/µ.

Theorem 6.1. The operator G is compact and self-adjoint.

Proof. Let f, h ∈ L2q (0, a). Since G(x, t) is a real-valued function defined on [0, a] × [0, a]
and G(x, t) = G(t, x), then for f, h ∈ L2q (0, a),
 a  a a
G (f ), h = (G f )(x)h(x) dq x = G(x, t)f (t)h(x) dq t dq x
0 0 0
 a  a 
= f (t) G(t, x)h(x) dq x dq t = f, G (h),
0 0

i.e. G is self-adjoint. Let {φij (x, t) = φi (x)φj (t)} be an orthonormal basis of L2q ((0, a) ×
 n
(0, a)). Consequently G = ∞ i,j =1 G, φij φij . For n ∈ Z set Gn =
+
i,j =1 G, φij φij , and
let Gn be the finite rank integral operator defined on Lq (0, a) by
2

 a
Gn (f )(x) := Gn (x, t)f (t) dq t, in L2q (0, a). (6.12)
0
Obviously Gn is compact for all n ∈ N. From Cauchy–Schwarz’ inequality
 a 1/2
(G − Gn )(f ) = |(G − Gn )(f )(x)| dq t
2
0
  a 2 1/2
a  
=  (G − Gn )(x, t)(f )(t) dq t  dq x
 
0 0
 a  a 1/2  a 1/2

|(G − Gn )(x, t)|2 dq t dq x |f (x)|2 dq x
0 0 0
= G − Gn 2 f ,
then
G − Gn
G − Gn 2 →0 as n → ∞.
This completes the proof. 

Corollary 6.2. The eigenvalues of the operator L form an infinite sequence {λk }∞
k=1 of real
numbers which can be ordered so that
|λ1 | < |λ2 | < · · · < |λn | < · · · → ∞ as n → ∞.
The set of all normalized eigenfunctions of L forms an orthonormal basis for L2q (0, a).
Basic Sturm–Liouville problems 3793

Proof. Since G is a compact self-adjoint operator on L2q (0, a), then G has an infinite sequence
of non-zero real eigenvalues {µn }∞ ∞
n=1, ⊆ R, µn → 0 as n → ∞. Let {φn }n=1 denote an
∞
orthonormal set of eigenfunctions corresponding to {µn }n=1 . From the spectral theorem of
compact self-adjoint operators, we have,
∞

G (f ) = λn f, φn φn . (6.13)
n=0

Since the eigenvalues {λn }∞

n=1 of the operator L are the reciprocal of those of G , then
1
|λn | = →∞ as n → ∞. (6.14)
|µn |
Let y ∈ DL . Then, y = G(f ), for some f ∈ L2q (0, a). Consequently,
∞
 ∞

y= λn f, φn φn = µn ly, φn φn
n=0 n=0
∞ ∞
= µn y, φn φn = y, φn φn .
n=0 n=0

If zero is an eigenvalue of L. Then, we can choose r ∈ R such that r is not an eigenvalue of

L. Now, applying the above result on L − rI in place of L yields the corollary. 

Example 1. Consider the q-Sturm–Liouville boundary value problem

1
− Dq −1 Dq y(x) = λy(x), (6.15)
q
with the q-Dirichlet conditions
U1 (y) = y(0) = 0, U2 (y) = y(1) = 0. (6.16)
A fundamental set of solutions of (6.15) is
√
√ sin( λx; q)
φ1 (x, λ) = cos( λx; q), φ2 (x, λ) = √ . (6.17)
λ
Now, the eigenvalues of problem (6.15) are the zeros of the determinant
  √
U1 (φ1 ) U2 (φ1 ) sin( λ; q)
(λ) =   = φ2 (1, λ) = √ . (6.18)
U1 (φ2 ) U2 (φ2 ) λ
√
Hence, the eigenvalues {λn }∞n=1 are the zeros of sin( λ; q). From (2.11),
(−1/2)
λn = (1 − q)−2 q −2n+2n , n  1, (6.19)
 sin(√λn ;q) ∞
for sufficiently large n and the corresponding set of eigenfunctions √ is an
λn n=1
2
orthogonal basis of Lq (0, 1). In the previous notation
√
sin( λx; q)
θ1 (x, λ) = √ , (6.20)
λ
and
√ √
sin( λ; q) √ √ sin( λx; q)
θ2 (x, λ) = √ cos( λx; q) + cos( λ; q) √ . (6.21)
λ λ
3794 M H Annaby and Z S Mansour

So, if λ is not an eigenvalue, Green’s function is given by

√  √ √ 
sin( λt; q) √ sin( λ; q) √ sin( λx; q)
G(x, t, λ) = √ cos( λx; q) √ − cos( λ; q) √ , (6.22)
sin( λ; q) λ λ
for 0
t
x, and
√  √ √ 
sin( λx; q) √ sin( λ; q) √ sin( λt; q)
G(x, t, λ) = √ cos( λt; q) √ − cos( λ; q) √ , (6.23)
sin( λ; q) λ λ
for x
t
1. Since λ = 0 is not an eigenvalue, then Green function G(x, t) is nothing but

t (1 − x), 0
t
x,
G(x, t) = G(x, t, 0) =
x(1 − t), x
t
1.
Hence the boundary value problem (6.15), (6.16) is equivalent to the basic Fredholm integral
equation
 1
y(x) = λ G(x, t)y(t) dq t. (6.24)
0

Example 2. Consider equation (6.15) with the q-Neumann boundary conditions

U1 (y) = Dq −1 y(0) = 0, U2 (y) = Dq −1 y(1) = 0. (6.25)
√
In this case θ1 (x, λ) = cos( λx; q), and
√ √ √ √ √
θ2 (x, λ) = cos( λq −1/2 ; q) cos( λx; q) + q sin( λq −1/2 ; q) sin( λx; q).
√ √
Since (λ) = qλ sin( λq −1/2 ; q). Then λ0 = 0 and for sufficiently large n, the eigenvalue
are given by
(−1/2)
λn = q −2n+1+2n , n1 (6.26)
√
Therefore, {1, cos( λn x; q)}∞ 2
n=1 is an orthogonal basis of Lq (0, 1). If λ is not an eigenvalue,
then Green’s function G(x, t, λ) is defined for x, t ∈ [0, a] × [0, a] by
√
cos( λt; q) √ √
G(x, t, λ) = − √ √ (cos( λq −1/2 ; q) cos( λx; q)
qλ sin( λq −1/2 ; q)
√ √ √
+ q sin( λq −1/2 ; q) sin( λx; q)),
and
√
cos( λx; q) √ √
G(x, t, λ) = − √ √ (cos( λq −1/2 ; q) cos( λt; q)
qλ sin( λq −1/2 ; q)
√ √ √
+ q sin( λq −1/2 ; q) sin( λt; q)), x
t
1.
The operator L associated with problem (6.15), (6.25) is not invertible since zero is an
eigenvalue.
Example 3. Consider (6.15) with the following boundary conditions:
U1 (y) = y(0) = 0, U2 (y) = y(1) + Dq −1 y(1) = 0. (6.27)
√ √ −1/2
Then (λ) = φ2 (1, λ)+Dq −1 φ2 (1, λ) = sin(√ λ;q)
λ
+cos( λq ; q). The eigenvalues {λn }∞
n=1
of this boundary value problem are the solutions of the equation
√
sin( λ; q) √
√ = − cos( λq −1/2 ; q) (6.28)
λ
Basic Sturm–Liouville problems 3795
 sin(√λn ;q) ∞
and the corresponding eigenfunctions are √ . The functions θ1 (x, λ) and θ2 (x, λ)
λn n=1
are
√
sin( λx; q)
θ1 (x, λ) = √ ,
λ
 √ 
√ −1/2 sin( λ; q) 
θ2 (x, λ) = cos( λq ; q) + √ cos( λx; q)
λ
√
 √ −1/2 √ sin( λx; q)
− (− λq sin( λq ; q) + cos( λ; q)) √ .
λ
If λ is not an eigenvalue, then Green’s function G(x, t, λ) is defined to be

√
−1 sin( λt; q)θ2 (x, λ), 0
t
x,
G(x, t, λ) = √ √ √ √
sin( λ; q) + λ cos( λq −1/2 ; q) sin( λx; q)θ2 (t, λ), x
t
1.
and

−1 t (2 − x), 0
t
x,
G(x, t) =
2 x(2 − t), x
t
1.
Therefore the boundary value problem (6.15), (6.27) is equivalent to the basic Fredholm
integral equation
 1
y(x) = λ G(x, t)y(t) dq t.
0

Remark 1. Let r(·) be a real-valued function defined on [0, q −1 a] such that r(x) = 0 for all
x ∈ [0, q −1 a] and Dq −1 r(0) exists. Let w(·) be a real-valued function defined on [0, a] and
positive on {0, aq n , n ∈ N}. The Sturm–Liouville problem (4.6a)–(4.6c) may be defined for
0
x
a < ∞, λ ∈ C to be
−1
M(y) := Dq −1 (r(x)Dq y(x)) + ν(x)y(x) = λw(x)y(x), (6.29a)
q
U1 (y) := a11 y(0) + a12 (rDq y)(0) = 0, (6.29b)
U2 (y) := a21 y(a) + a22 (rDq y)(aq −1 ) = 0, (6.29c)

where the functions ν(·) and the constants {aij }, i, j ∈ {1, 2}, are as in section 4. In this case
we will have the Lagrange identity
 a
((My)(x)z(x) − y(x)(Mz)(x)) dq x = [y, z̄](a) − lim [y, z̄](aq n ), (6.30)
0 n→∞

where
[y, z](x) = y(x)(rDq z)(xq −1 ) − (rDq y)(xq −1 )z(x). (6.31)
If y(·), z(·) and r(·) are q-regular at zero, then Lagrange’s identity (6.30) will be nothing but
 a
((My)(x)z(x) − y(x)(Mz)(x)) dq x = [y, z̄](a) − [y, z̄](0). (6.32)
0

Problem (6.29a)–(6.29c) is formally self-adjoint in L2q ([0, a]; w(·)), where L2q ([0, a]; w(·))
is the Hilbert space
 a
L2q ([0, a]; w(·)) := {f : [0, a] → C : |f (x)|2 w(x) dq x < ∞} (6.33)
0
3796 M H Annaby and Z S Mansour

with the inner product

 a
f, gw := f (x)g(x)w(x) dq x, f, g ∈ L2q ((0, a); w(·)). (6.34)
0
In a way similar to the theory developed in section 5, the eigenvalue problem (6.29a)–(6.29c)
is equivalent to the q-type Fredholm integral equation
 a
y(x) = λ G(x, ξ )y(ξ )w(ξ ) dq ξ, (6.35)
0
where G(x, ξ ) is Green’s function associated with the problem. Consequently all results
established above hold in this setting. In particular, problem (6.29a)–(6.29c) has a countable
set of real simple eigenvalues {λk }∞k=0 with no finite limit points. A corresponding set of
eigenfunctions {φk (·)}∞
k=0 is an orthonormal basis of L2q ((0, a); w(·)).

References

[1] Aberu L D, Bustoz J and Caradoso J L 2003 The roots of the third Jackson q-Bessel functions Int. J. Math.
Math. Sci. 67 4241–8
[2] Andrews G E, Askey R and Roy R 1999 Special Functions (Cambridge: Cambridge University Press)
[3] Annaby M H 2003 q-Type sampling theorems Result. Math. 44 214–25
[4] Askey R A and Wilson J A 1985 Some basic hypergeometric orthogonal polynomials that generalize Jacobi
polynomials Mem. Am. Math. Soc. 319
[5] Atkinson F V 1964 Discrete and Continuous Boundary Problems (New York: Academic)
[6] Berezansky Y M, Sheftl Z G and Us G F 1996 Functional Analysis vol I (Basel: Birkhäuser)
[7] Bustoz J and Cardoso J L 2001 Basic analog of Fourier series on a q-linear grid J. Approx. Theory 112 154–7
[8] Bustoz J and Suslov S K 1998 Basic analog of Fourier series on a q-quadratic grid Methods Appl. Anal. 5 1–38
[9] Carmichael R D 1924 The present state of the difference calculus and the prospect for the future Am. Math.
Month. 31 169–83
[10] Carmichael R D 1933 Systems of linear difference and expansions in series of exponential functions Trans. Am.
Math. Soc. 35 1–28
[11] Chung K, Chung W, Nam S and Kang H 1994 New q-derivative and q-logarithm Int. J. Theor. Phys. 33 2019–29
[12] Coddington E A and Levinson N 1955 Theory of Ordinary Differential Equations (New York: McGraw-Hill)
[13] Eastham M S P 1970 Theory of Ordinary Differential Equations (Princeton, NJ: Van Nostrand-Reinhold)
[14] Exton H 1983 q-Hypergeometric Functions and Applications (Chichester: Ellis-Horwood)
[15] Exton H 1992 Basic sturm-liouville theory Rev. Tecn. Fac. Ingr. Univ. Zulia 1 85–100
[16] Floreanini R and Vinet L 1995 A model for the continuous q-ultraspherical polynomials J. Math. Phys. 36
3800–13
[17] Floreanini R and Vinet L 1995 More on the q-oscillator algebra and q-orthogonal polynomials J. Phys. A: Math.
Gen. 28 L287–293
[18] Gasper G and Rahman M 1990 Basic Hypergeometric Series (New York: Cambridge University Press)
[19] Gray R W and Nelson C A 1990 A completeness relation for the q-analogue coherent states by q-integration
J. Phys. A: Math. Gen. 23 L945–50
[20] Hahn W 1982 Math. Rev. 84i:39002
[21] Ismail M E H 1985 SIAM Rev. 16 279–81
[22] Ismail M E H 1996 On Jackson’s third q-Bessel function Unpublished report
[23] Ismail M E H 2001 Orthognality and completeness of q-Fourier type systems Z. Anal. Anwend. 20 761–75
[24] Jackson F H 1905 The applications of basic numbers to bessel’s and legendre’s equations Proc. Lond. Math.
Soc. 2 192–220
[25] Jackson F H 1910 q-Difference equations Am. J. Math. 32 305–14
[26] Jackson F H 1910 On q-definite integrals Q. J. Pure Appl. Math. 41 193–203
[27] Jirari A Second-order Sturm Liouville difference equations and orthogonal polynomials Mem. Am. Math. Soc.
542 1995
[28] Kac V and Cheung P 2002 Quantum Calculus (New York: Springer)
[29] Koelink H T and Swarttouw R F 1994 On the zeros of the Hahn-Exton q-Bessel function and associated
q-Lommel polynomials J. Math. Anal. Appl. 186 690–710
[30] Levitan B M and Sargsjan I S 1975 Introduction to Spectral Theory: Selfadjoint Ordinary Differential Operators
(Providence, RI: American Mathematical Society)
Basic Sturm–Liouville problems 3797

[31] Levitan B M and Sargsjan I S 1991 Sturm Liouville and Dirac Operators (Dordrecht: Kluwer)
[32] Mansour Z S 2001 q-Difference equations MSc Thesis Faculty of Science, Cairo University
[33] Marchenko V A 1986 Sturm–Liouville Operators and Applications (Basel: Birkhäuser)
[34] Suslov S K 2000 Another addition theorem for the q-exponential function J. Phys. A: Math. Gen. 33 L375–80
[35] Suslov S K 2002 Some expansions in basic Fourier series and related topics J. Approx. Theory 115 289–353
[36] Suslov S K 2003 An Introduction to Basic Fourier Series, vol 9 of Developments in Mathematics (Boston, MA:
Kluwer Academic)
[37] Swarttouw R F 1992 The Hahn-Exton q-Bessel functions PhD Thesis The Technical University of Delft
[38] Swarttouw R F and Meijer H G 1994 A q-analogue of the Wronskian and a second solution of the Hahn–Exton
q-Bessel difference equation Proc. Am. Math. Soc. 120 855–64
[39] Titchmarsh E 1962 Eigenfunction Expansions Asociated with Second Order Differential Equations (Oxford:
Clarendon)