PHD Thesis
PHD Thesis
N
Z
:= Z 0
R := (, )
R
+
:= (0, )
C is the set of all complex numbers.
f.s. is abbreviation for fundamental set.
RLF is abbreviation for RiemannLiouville fractional.
v
Preface
This thesis is concerned with the fast growing area of basic special functions and basic
dierence equations. It consists of three parts. The rst one is concerned with some
basic analogues of classical results. This part contains a self contained q-calculus,
which has considered before in [9, 90]. We also derive some new basic type identities
which will be needed throughout the thesis. The rst part contains two new results
concerning zeros of basic functions and q-integral transforms. We study the q-sine
and q-cosine functions, their asymptotics and zeros. Then we derive a q-analogue of
a theorem of George Polya in Chapter 1. In Chapter 2 we study the zeros of a nite
q-Hankel transform. As an application of the results of this chapter we introduce
another q-analogue of the theorem of George Polya of Chapter 1.
Part II is devoted to study q-dierence equations. We start this part with Chapter
3 which contains some known results in this subject, see e.g. [8, 9, 90], in particular
existence and uniqueness theorems as well as, q-linear equations. Our new contri-
bution lie in Chapter 4. We establish a q-SturmLiouville theory and investigate
the asymptotic behavior of eigenvalues and eigenfunctions of basic SturmLiouville
eigenvalue problems.
In the last part we study q-analogues of fractional calculus and fractional dier-
ential equations. In Chapter 5 we give a brief account about the history of fractional
calculi and some results which we are intended in deriving their q-analogues. Basic
RiemannLiouville fractional calculus introduced in [10,11,13] is studied in a rigorous
analytic way. We derive q-analogues of other types of fractional derivatives, namely
vi
vii
Gr unwaldLetinkov derivative and Caputo fractional derivative. q-type fractional dif-
ference equations are introduced in Chapter 7. The solutions of fractional q-dierence
equations are also studied in this part. For this task we dene q-Mittag-Leer func-
tions, and study some of their properties. We also solve q-fractional dierence equa-
tions with constant coecients by using q-type Laplace transform.
Part I
Classical Results
1
Chapter 1
Basic Cosine and Sine Functions
This chapter includes some q-notations and results. It starts with a q-calculus and
including some q-notations and functions. We also consider q-sine and q-cosine functions,
some of their properties and their zeros. Then we end up with a basic analogue of a theorem
of George Polya.
1.1 A q-calculus
Throughout this thesis unless otherwise stated q is a positive number less than 1 and by
the word basic we mean q-analogue. In this section we introduce some of the q-notations
and results. We start with the q-shifted factorial, see [53], for a C,
(a; q)
n
:=
_
_
1, n = 0,
n1
i=0
(1 aq
i
), n = 1, 2, . . . .
(1.1.1)
The limit of (a; q)
n
as n tends to innity exists and will be denoted by (a; q)
. Moreover
(a; q)
n=0
(1)
n
q
n(n1)
2
a
n
(q; q)
n
. (1.1.2)
The multiple q-shifted factorial for complex numbers a
1
, . . . , a
k
is dened by
(a
1
, a
2
, . . . , a
k
; q)
n
:=
k
j=1
(a
j
; q)
n
. (1.1.3)
2
3
We also use the following notations for the q-binomial coecients
_
0
_
q
= 1,
_
n
_
q
=
(1 q
)(1 q
1
) . . . (1 q
n+1
)
(q; q)
n
, n Z
+
, R. (1.1.4)
For C, aq
,= q
n
, n N, we dene (a; q)
to be
(a; q)
:=
(a; q)
(aq
; q)
. (1.1.5)
The -function is dened for z C 0 , 0 < [q[ < 1 to be
(z; q) :=
n=
q
n
2
z
n
. (1.1.6)
The following identity is introduced by C.G.J. Jacobi in 1829, and it is called Jacobi triple
product identity, see [53]
n=
q
n
2
z
n
=
_
q
2
, qz, qz
1
; q
2
_
n=0
(1)
n
_
m
n
_
q
q
n(n1)/2
u
n
= (u; q)
m
. (1.1.8)
Let
r
s
denote the q-Hypergeometric series
r
s
_
_
_
a
1
, . . . , a
r
b
1
, . . . , b
s
q, z
_
_
_
=
n=0
(a
1
, . . . , a
r
; q)
n
(q, b
1
, . . . , b
s
; q)
n
z
n
(q
(n1)/2
)
n(s+1r)
. (1.1.9)
The q-Gamma function, [53, 71], is dened by
q
(z) :=
(q; q)
(q
z
; q)
(1 q)
1z
, z C, 0 < [q[ < 1. (1.1.10)
Here we take the principal values of q
z
and (1q)
1z
. Then
q
(z) is a meromorphic function
with poles at z = n, n N. Because
q
(z) has no zeros,
1
q
(z)
is an entire function with
zeros at z = n, n N.
q
(z) is the unique function that satises the functional equation
q
(z + 1) =
1 q
z
1 q
q
(z),
q
(1) = 1, (1.1.11)
4
and logarithmically convex. See [14, 53]. Hence,
q
(n) = (q; q)
n1
_
(1 q)
n1
. The q-Beta
function is dened by
B
q
(a, b) :=
_
1
0
x
a1
(qx; q)
b1
d
q
x, a, b > 0, (1.1.12)
where the integration is dened in (1.1.31) below. Using the q-binomial theorem, cf. [14,
p. 488], we have
n=0
(a; q)
n
(q; q)
n
z
n
=
(az; q)
(z; q)
q
(a +b)
. (1.1.14)
The third type of q-Bessel functions is dened for z C by, cf. e.g. [63, 67, 68],
J
(3)
(z; q) := z
(q
+1
; q)
(q; q)
1
_
_
_
0
q
+1
q; qz
2
_
_
_
= z
(q
+1
; q)
(q; q)
n=0
(1)
n
q
n(n+1)/2
z
2n
(q; q)
n
(q
+1
; q)
n
(1.1.15)
This function is called in some literature the Hahn-Exton q-Bessel function, see [78, 112].
It is also called the
1
1
q-Bessel function, cf. [77]. Since the other types of the q-Bessel
functions, i.e. J
(1)
(; q), J
(2)
(; q), see e.g. [63, 69, 104], will not be used throughout this
thesis we use the notation J
(; q) for J
(3)
n=0
(1)
n
q
n
2
(z(1 q))
2n
(q; q)
2n
=
(q
2
; q
2
)
(q; q
2
)
(zq
1/2
(1 q))
1/2
J
1/2
_
z(1 q)/
q ; q
2
_
,
(1.1.16)
sin(z; q) :=
n=0
(1)
n
q
n(n+1)
(z(1 q))
2n+1
(q; q)
2n+1
=
(q
2
; q
2
)
(q; q
2
)
(z(1 q))
1/2
J
1/2
_
z(1 q); q
2
_
.
(1.1.17)
5
They are q-analogues of the cosine and sine functions, [14, 53]. The basic hyperbolic
trigonometric functions cosh(z; q) and sinh(z; q) are dened for z C by
cosh(z; q) := cos(iz; q), sinh(z; q) := i sin(iz; q). (1.1.18)
A q-analogue of the exponential function is the function
E(z; q) :=
n=0
q
n
2
/4
z
n
q
(n + 1)
, z C. (1.1.19)
One can easily see that
E(iz; q) = cos(z; q) +iq
1/4
sin(z; q), (1.1.20)
and
E([z[; q) = cosh([z[; q) +q
1/4
sin([z[; q). (1.1.21)
F. H. Jackson [70, 1904] introduced the functions e
q
(z) and E
q
(z) as q-analogues of the
exponential functions, where
E
q
(z) :=
n=0
q
(
n
2
)
z
n
(q; q)
n
= (z; q)
, z C, (1.1.22)
e
q
(z) :=
n=0
z
n
(q; q)
n
, [z[ < 1. (1.1.23)
E
q
(z) is an entire function with simple zeros at the points q
n
, n N. Since
e
q
(z)E
q
(z) 1, [z[ < 1, (1.1.24)
see e.g. [53, 70], then the domain of the function e
q
(z) can be extended to C by dening
e
q
(z), z C, to be
e
q
(z) :=
1
n=0
_
1 +q
n
z
_
. (1.1.25)
Hence, the relation (1.1.24) holds in C, and the function e
q
(z) has simple poles at the points
q
n
, n N. Let the basic trigonometric functions sin
q
z, cos
q
z, Sin
q
z, and Cos
q
z be
6
dened by
sin
q
z :=
e
q
(iz) e
q
(iz)
2i
, cos
q
z :=
e
q
(iz) e
q
(iz)
2
, [z[ < 1, (1.1.26)
Sin
q
z :=
E
q
(iz) E
q
(iz)
2i
, Cos
q
z :=
E
q
(iz) E
q
(iz)
2
, z C. (1.1.27)
In the next section we prove that the functions cos(z; q) and sin(z; q) have only real and
simple zeros, which will be denoted throughout this thesis by x
m
m=1
, and 0, y
m
m=1
respectively, where x
m
and y
m
, m 1, are the positive zeros, cf. [7, 28, 78].
If R, a subset A of R is called a -geometric set if x A for all x A. If a subset
A of R is a -geometric, then it contains all geometric sequences x
n
n=0
, x A. Let
f be a function, real or complex valued, dened on a q-geometric set A. The q-dierence
operator is dened by
D
q
f(x) :=
f(x) f(qx)
x qx
, x A 0 . (1.1.28)
If 0 A, the q-derivative at zero is dened by
D
q
f(0) := lim
n
f(xq
n
) f(0)
xq
n
, x A 0 , (1.1.29)
if the limit exists and does not depend on x. In some literature the q-derivative at zero is
dened to be f
(0) if it exists, cf. [78, 112], but the above denition is more suitable for our
approaches. Also the q
1
-derivative of a function f(x) at zero is dened by
D
q
1f(0) := lim
n
f(xq
n
) f(0)
xq
n
= D
q
f(0), x A 0 , (1.1.30)
provided that the limit exists and does not depend on x. A right inverse to D
q
, the Jackson
q-integration, cf. [72], is
_
x
0
f(t) d
q
t := x(1 q)
n=0
q
n
f(xq
n
), x A, (1.1.31)
provided that the series converges, and
_
b
a
f(t) d
q
t :=
_
b
0
f(t) d
q
t
_
a
0
f(t) d
q
t, a, b A. (1.1.32)
7
If A is q
1
-geometric, then the q-integration over [x, ), x A, is dened by
_
x
f(t) d
q
t :=
n=1
xq
n
(1 q)f(xq
n
). (1.1.33)
Lemma 1.1.1. The q-integrals (1.1.31) and (1.1.33) exist if
lim
n
xq
n
f(xq
n
) = 0 and lim
n
xq
n
f(xq
n
) = 0,
respectively.
Proof. Based on the facts that, for x C
lim
k
xq
k
f(xq
k
) = 0 = [0, 1) C > 0, [f(xq
k
)[ C[xq
k
[
, k N, (1.1.34)
lim
k
xq
k
f(xq
k
) = 0 = (1, ) C > 0, [f(xq
k
)[ C[xq
k
[
, k N. (1.1.35)
Kac and Cheung [75, p. 68] have proved that if x
_
f(a) lim
n
f(aq
n
)
. (1.1.39)
The non-symmetric Leibniz rule is
D
q
(fg)(x) = g(x)D
q
f(x) +f(qx)D
q
g(x). (1.1.40)
Relation (1.1.40) can be symmetrized using the relation f(qx) = f(x) x(1 q)D
q
f(x),
giving the additional term x(1 q)D
q
f(x)D
q
g(x). Also the rule of q-integration by parts is
_
a
0
g(x)D
q
f(x) d
q
x = (fg)(a) lim
n
(fg)(aq
n
)
_
a
0
D
q
g(x)f(qx) d
q
x. (1.1.41)
If f, g are q-regular at zero, then lim
n
(fg)(aq
n
) on the right hand side of (1.1.41) will
be replaced by (fg)(0).
In the following we dene Hilbert spaces where a q-analogue of SturmLiouville problems
will be considered. Let L
2
q
(0, a) be the space of all complex valued functions dened on [0, a]
such that
|f| :=
__
a
0
[f(x)[
2
d
q
x
_
1/2
< . (1.1.42)
9
The space L
2
q
(0, a) is a separable Hilbert space with the inner product
f, g) :=
_
a
0
f(x)g(x) d
q
x, f, g L
2
q
(0, a), (1.1.43)
and the orthonormal basis
n
(x) =
_
_
1
x(1q)
, x = aq
n
,
0, otherwise,
(1.1.44)
n = 0, 1, 2, . . ., cf. [15]. The space L
2
q
_
(0, a) (0, a)
_
is the space of all complex valued
functions f(x, t) dened on [0, a] [0, a] such that
|f(, )|
2
:=
__
a
0
_
a
0
[f(x, t)[
2
d
q
xd
q
t
_
1/2
< . (1.1.45)
The elements of L
2
q
_
(0, a) (0, a)
_
are equivalence classes where f, g are in the same equiv-
alence 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.
Lemma 1.1.3. L
2
q
_
(0, a) (0, a)
_
is a separable Hilbert space with the inner product
f, g)
2
:=
_
a
0
_
a
0
f(x, t)g(x, t) d
q
xd
q
t. (1.1.46)
Proof. Similar to [15, pp. 217-218], L
2
q
_
(0, a) (0, a)
_
is a Banach space. To prove
separability, it suces to prove that
ij
(x, t) :=
i
(x)
j
(t), i, j = 1, 2, . . . , (1.1.47)
is an orthonormal basis of L
2
q
_
(0, a) (0, a)
_
whenever
i
()
i=1
is an orthonormal basis
of L
2
q
(0, a). Indeed,
jk
,
mn
)
2
=
_
a
0
_
a
0
j
(x)
k
(t)
m
(x)
n
(t) d
q
xd
q
t
=
_
a
0
j
(x)
m
(x) d
q
x
_
a
0
k
(t)
n
(t) d
q
t =
jm
kn
,
10
proving orthogonality. To prove that
ij
is a basis, we prove that if there exists
f L
2
q
_
(0, a) (0, a)
_
such that f,
ij
)
2
= 0, then f is the zero element. Indeed,
0 = f,
ij
) =
_
a
0
_
a
0
f(x, t)
i
(x)
j
(t) d
q
xd
q
t
=
_
a
0
j
(t)
_
_
a
0
f(x, t)
i
(x) d
q
x
_
d
q
t =
_
a
0
h(t)
j
(t) d
q
t.
Thus
h(t) :=
_
a
0
f(x, t)
i
(x) d
q
x (1.1.48)
is orthogonal to the
j
s which implies that h(aq
n
) = 0, for all n N. So, f(x, aq
n
) is
orthogonal to each
i
. Consequently, f(aq
m
, aq
n
) = 0, for all m, n N.
By L
1
q
(0, a), a > 0, we mean the Banach space of all complex valued functions dened
on (0, a] such that
|f| :=
_
a
0
[f(t)[ d
q
t < . (1.1.49)
Let L
1
q
(0, a) denote the space of all functions f dened on (0, a] such that f L
1
q
(0, x) for
all x (qa, a].
1.2 Asymptotics of zeros of cos(z; q) and sin(z; q)
There are studies on the zeros of q-trigonometric functions, see references below. In this
section we establish other formulae for the asymptotic behavior of the zeros of cos(z; q) and
sin(z; q), i.e. x
m
and y
m
respectively. Our asymptotic formulae t with the study for the
eigenvalues of q-SturmLiouville problems as well as for deriving a q-analogue of a theorem
of G. Polya. We also state and prove some inequalities for expressions involving x
m
and
y
m
. We rst state the asymptotic formulae in the existing literature, namely [7, 28]. We
start with the following preliminaries concerning entire functions, cf. e.g [27, 61, 84]. Let
f, g be entire functions , we say that
f(z) = O
_
g(z)
_
, as z , (1.2.1)
11
if f(z)
_
g(z) is bounded in a neighborhood of . On the other hand we write
f(z) g(z), as z , if lim
z
[f(z)[
[g(z)[
= 1. (1.2.2)
Denition 1.2.1. Let f(z) :=
n=0
a
n
z
n
be an entire function. The maximum modulus
is dened by
M(r; f) := sup [f(z)[ : [z[ = r , r > 0. (1.2.3)
The order of f, (f), is dened by, cf. [27, 84]
(f) := limsup
r
log log M(r, f)
log r
= limsup
n
nlog n
log a
1
n
. (1.2.4)
Theorem 1.2.1. [27] If (f) is nite and is not equal to a positive integer, then f has
innitely many zeros, or is a polynomial.
One can easily see that J
(; q
2
) is an entire functions of order zero. Therefore, from
Theorem 1.2.1, J
(; q
2
) has innitely many zeros. Koelink and Swarttouw proved that
J
(; q
2
) has only real and simple zeros, cf. [78, 1994]. Let w
()
m
, > 1, denote the positive
zeros of J
(; q
2
) in an increasing order of m Z
+
. In [7] L.D. Abreu et al. proved that if
q
2+2
< (1 q
2
)
2
, then
w
()
m
(q) = q
m+2
m
()
, 0 <
m
() <
log
_
1
q
2m+2
1q
2m
_
2 log q
,
m=0
m
() =
log(q
2+2
; q
2
)
4 log q
.
(1.2.5)
Moreover w
()
m
q
m
when m without the restriction q
2+2
< (1 q
2
)
2
. Applying
these formulae to the cases = 1/2 respectively and using (1.1.16) and (1.1.17), we obtain
the following estimates
if q < (1 q
2
)
2
, then x
m
=
q
m+1/2+2
(1/2)
m
1 q
, m 1, (1.2.6)
if q
3
< (1 q
2
)
2
, then y
m
=
q
m+2
(1/2)
m
1 q
, m 1. (1.2.7)
12
Moreover
x
m
q
m+1/2
(1 q)
, y
m
q
m
(1 q)
as m (1.2.8)
without further restrictions on q. The results for y
m
coincide with those of Bustoz and
Cardoso [28]. A study of asymptotic formulae of zeros of another class of basic trigonometric
functions is given in [110, 111]. This section also includes some results involving x
m
, y
m
,
which will be needed for the derivation of the asymptotic behavior of eigenvalues of the
basic SturmLiouville problems. In the following we introduce some results of Bergweiler
et al. [26], which are essential in our investigations. These results are concerned with the
functional equation
l
j=0
a
j
(z)f(c
j
z) = Q(z), (1.2.9)
where c C, 0 < [c[ < 1, is xed, Q and the a
j
s are polynomials. Set
p
j
:= deg (a
j
), and d
j
:= p
j
p
0
, j = 0, 1, . . . , l. (1.2.10)
The NewtonPuiseux diagram associated with equation (1.2.9), P, is dened to be the
convex hull of
l
_
j=0
_
(x, y) R
2
: x j and y d
j
_
. (1.2.11)
Let K 0 be the number of vertices of P. Thus the vertices of P are (j
k
, d
j
k
), k = 0, . . . , K,
where
0 = j
0
< j
1
< . . . < j
K
l.
It is proved in [26, Theorem 1.1] that if (1.2.9) has a transcendental entire solution, then
there exists j 1, . . . , l such that d
j
> 1. In other words, in this case the NewtonPuiseux
diagram of (1.2.9) has at least one vertex which is not (0, 0), i.e. K 1. For k = 1, . . . , K
set
k
:=
d
j
k
d
j
k1
j
k
j
k1
.
13
Then
1
>
2
> . . . >
K
> 0. The
k
s are the slopes of the segments which together with
the negative part of the y-axis form the boundary of P. It is shown in [26, Theorem 1.2]
and also in [105] that if f is an entire transcendental solution of (1.2.9), then there exists
k
0
1, . . . , K such that
log M(r; f)
k
0
2 log [c[
(log r)
2
as r . (1.2.12)
The following example is a special case of (1.2.9) and is important in our study since the
solutions are nothing but basic trigonometric functions dened in (1.1.16) and (1.1.17).
Example 1.2.1. Consider the functional equation
qy(z) +
_
(1 +q) +q
2
z
2
(1 q)
2
_
y(qz) +y(q
2
z) = 0, z C. (1.2.13)
The functions cos(z; q), sin(z; q) are transcendental entire solutions of (1.2.13). Using the
previous notations we have
l = 2, a
0
(z) = q, a
1
(z) =
_
(1 +q) +q
2
z
2
(1 q)
2
_
, a
2
(z) = 1, Q(z) = 0. (1.2.14)
Therefore
p
0
= p
2
= 0, p
1
= 2, d
0
= d
2
= 0, d
1
= 2. (1.2.15)
Hence P is the minimal convex set containing
_
(x, y) R
2
: x 0, y 0
_
_
(x, y) R
2
: x 1, y 2
_
_
(x, y) R
2
: x 2, y 0
_
. (1.2.16)
Therefore P is the set in R
2
whose boundary consists of the negative part of the y-axis, the
line segment y = 2x, 0 x 1 and the line y = 2, x 1. This convex hull is illustrated in
Figure 1 below. Clearly P has only two vertices (0, 0) and (1, 2). Thus there is only one line
segment contained in the boundary of P with a slope,
1
,
1
= 2. So from (1.2.12) it follows
that the functions cos(z; q), sin(z; q) and E(z; q) have the following asymptotic behavior as
r
log M
_
r; cos(z; q)
_
, log M
_
r; sin(z; q)
_
, log M(r; E(z; q))
1
log q
(log r)
2
. (1.2.17)
14
Figure 1.1: The Newton-Puiseux diagram P of (1.2.13).
Notice that the asymptotic of E(z; q) is obtained via (1.1.20).
The following lemma is taken from [25] and will be needed in the sequel. It is concerned
with the asymptotic behavior of the -function dened in (1.1.6).
Lemma A. Suppose that 0 < [q[ < 1 and that z C, [z[ 1. For m Z
+
let A
m
be the
annulus dened by
A
m
:=
_
z C, q
2m+2
[z[ < [q[
2m
_
. (1.2.18)
Then we have, uniformly as m ,
log [(z; q)[ =
(log [z[)
2
4 log [q[
+ log [1 +q
2m1
z[ +O(1), z A
m
. (1.2.19)
Using the technique of Bergweiler and Hayman [25], we will derive the asymptotic
formulae for the zeros of cos(z; q) and sin(z; q) in the following theorem.
Theorem 1.2.2. If x
m
are the positive zeros of cos(z; q) and y
m
are the positive zeros
15
of sin(z; q), then we have for suciently large m,
x
m
= q
m+1/2
(1 q)
1
_
1 +O(q
m
)
_
, (1.2.20)
y
m
= q
m
(1 q)
1
_
1 +O(q
m
)
_
. (1.2.21)
Proof. We prove the theorem only for the zeros of cos(z; q) since the proof for the zeros of
sin(z; q) is similar. Let R(z) be the dierence
R(z) := cos(z; q)
1
(q; q)
(z
2
(1 q)
2
; q), z C 0 . (1.2.22)
Hence
R(z) =
k=
2k
z
2k
,
2k
:=
_
_
(1)
k
q
k
2
(1 q)
2k
(q; q)
_
(q
2k+1
; q)
1
_
, k 0,
(1)
k+1
q
k
2
(1 q)
2k
(q; q)
, k < 0.
(1.2.23)
We shall prove that
(q
2k+1
; q)
1 = O(q
2k
) as k +. (1.2.24)
Indeed from (1.1.2), we have for k Z
+
(q
2k+1
; q)
j=1
(1)
j
q
j(j1)/2
q
(2k+1)j
(q; q)
j
q
2k+1
j=1
q
j(j1)/2
(q; q)
j
q
2k+1
(q; q)
j=1
q
(j1)/2
=
q
2k+1
(1
q)(q; q)
. (1.2.25)
This proves (1.2.24). We conclude from (1.2.23) and (1.2.25) that
[
2k
[
q
k
2
+2k
(1 q)
2k
(1
q)(q; q)
2
, k Z. (1.2.26)
Thus for z C 0,
[R(z)[
1
(1
q)(q; q)
2
q
k
2
+2k
(1 q)
2k
[z
2k
[
=
1
(1
q)(q; q)
2
(q
2
(1 q)
2
[z[
2
; q).
(1.2.27)
16
Taking the logarithm of both sides of (1.2.27) yields
log [R(z)[ log (q
2
(1 q)
2
[z[
2
; q) + log
1
(1
q)(q; q)
2
, z C 0 . (1.2.28)
We deduce from (1.2.19) that there exists m
0
N and a constant C > 0 such that if
q
2m+2
[z[ < q
2m
, m m
0
,
then
(log [z[)
2
4 log q
+log [1+q
2m1
z[C log [(z; q)[
(log [z[)
2
4 log q
+log [1+q
2m1
z[+C. (1.2.29)
Thus if
q
2m+2
q
2
[z[
2
(1 q)
2
< q
2m
, m m
0
, (1.2.30)
then replacing z in (1.2.29) by q
2
(1 q)
2
[z[
2
, we obtain for m m
0
log [(q
2
(1 q)
2
[z[
2
; q)[
(log [z[
2
(1 q)
2
q
2
)
2
4 log q
+log
1 +q
2m+1
(1 q)
2
[z[
2
+C. (1.2.31)
Since log(1 +x) < x, x > 0, then (1.2.30) implies
log
1 +q
2m+1
(1 q)
2
[z[
2
q
2m+1
(1 q)
2
[z[
2
q
1
. (1.2.32)
Substituting in (1.2.31), we obtain for z C, which satises (1.2.30),
log [(q
2
(1 q)
2
[z[
2
; q)[
(log [z[(1 q)q)
2
log q
+q
1
+C. (1.2.33)
Consequently, by (1.2.28), for z satises (1.2.30)
log [R(z)[
_
log [z[(1 q)q
_
2
log q
+q
1
+C + log
1
(1
q)(q; q)
2
=
_
log [z[(1 q)
_
2
log q
log [z[
2
+C
1
,
(1.2.34)
where
C
1
:= C log(1
q)(q; q)
2
2 log(1 q) log q +q
1
. (1.2.35)
17
Notice that (1.2.34) holds uniformly for z satises (1.2.30), m m
0
and C
1
is positive. Let
A
m
, m m
0
, be the annulus dened by
A
m
:=
_
z C : q
2m
[z[
2
(1 q)
2
< q
2m2
_
. (1.2.36)
The other side of (1.2.29) implies for z
A
m
_
log r(1 q)
_
2
log q
log
_
[((1 q)
2
z
2
; q)[
_
log
1 q
2m+1
z
2
(1 q)
2
+C. (1.2.37)
Therefore by (1.2.34), z
A
m
,
log [R(z)[ log
_
(q; q)
1
[((1q)
2
z
2
; q)[
_
log
1q
2m+1
z
2
(1 q)
2
log r
2
+C
2
, (1.2.38)
where C
2
= C +C
1
+ log(q; q)
. If z
A
m
, m m
0
, then
[1 q
2m+1
z
2
(1 q)
2
[ q
2m+1
[z[
2
(1 q)
2
1 = q
1
(1 q), [z[ =
q
m1
1 q
, (1.2.39)
[1 q
2m+1
z
2
(1 q)
2
[ 1 q
2m+1
[z[
2
(1 q)
2
= (1 q), [z[ =
q
m
1 q
. (1.2.40)
That is
log [1 q
2m+1
z
2
(1 q)
2
[ log(1 q), z
A
m
. (1.2.41)
Substituting in (1.2.38) we obtain
log [R(z)[ log
_
(q; q)
1
((1 q)
2
z
2
; q)
_
log(1 q) log r
2
+C
2
, z
A
m
. (1.2.42)
We can also choose m
0
suciently large such that log(1 q) log r
2
+C
2
< 0. Hence
[R(z)[
_
(q; q)
1
[((1 q)
2
z
2
; q)[
_
, z
A
m
, m m
0
. (1.2.43)
Hence an application of Rouche theorem shows that cos(z; q) and ((1q)
2
z
2
; q) have the
same number of zeros in
A
m
. Since ((1q)
2
z
2
; q) has two simple symmetric zeros in
A
m
,
we deduce that cos(z; q) has exactly two zeros in
A
m
. Since cos(z; q) is an even function,
18
then these two zeros are real, symmetric and simple. Assume now that x
m
, m m
0
, is the
positive zero of cos(z; q) in
A
m
. Then from (1.2.22)
[R(x
m
)[ =
_
(q; q)
1
[((1 q)
2
x
2
m
; q)[
_
1 q
2m+1
x
2
m
(1 q)
2
2 log x
m
+C
2
. (1.2.44)
In other words
log
1 q
m+(1/2)
x
m
(1 q)
+ log
1 +q
m+(1/2)
x
m
(1 q)
2 log x
m
+C
2
. (1.2.45)
Thus, either
log
1 q
m+(1/2)
x
m
(1 q)
log x
m
+
C
2
2
, (1.2.46)
or
log
1 +q
m+(1/2)
x
m
(1 q)
log x
m
+
C
2
2
. (1.2.47)
We can assume without any loss of generality that (1.2.46) holds. Then for all m m
0
1 q
m+(1/2)
x
m
(1 q)
= O(x
1
m
). (1.2.48)
This implies that x
m
q
m1/2
(1 q)
1
for suciently large m. More precisely, we have
x
m
= q
m1/2
(1 q)
1
_
1 +O(q
m
)
_
, as m . (1.2.49)
Lemma 1.2.3. As m , we have
q
2k
x
2
m+k
x
2
m
1
= O(q
m
), k N. (1.2.50)
The O-term is uniform in k N.
19
Proof. From (1.2.20) we see that for m
x
m
=
q
m+1/2
1 q
_
1 +O(q
m
)
_
. (1.2.51)
Hence there exists a constant C > 0 and m
0
N such that
[q
m1/2
(1 q)x
m
1[ Cq
m
, for all m m
0
. (1.2.52)
Consequently for all k N we have
[q
m+k1/2
(1 q)x
m+k
1[ Cq
m+k
, for all m m
0
. (1.2.53)
We can choose m
0
suciently large such that 1Cq
m
> 0 for all m m
0
. So from (1.2.52)
and (1.2.53) we obtain
(1 Cq
m+k
)
2
(1 +Cq
m
)
2
q
2k
x
2
m+k
x
2
m
(1 +Cq
m+k
)
2
(1 Cq
m
)
2
, m m
0
. (1.2.54)
Thus
C(1 +q
k
)(2 +Cq
m
(1 q
k
))
(1 +Cq
m
)
2
_
q
2k
x
2
m+k
x
2
m
1
_
q
m
C(1 +q
k
)(2 Cq
m
(1 q
k
))
(1 Cq
m
)
2
. (1.2.55)
That is
4C(1 +Cq
m
)
(1 +Cq
m
)
2
_
q
2k
x
2
m+k
x
2
m
1
_
q
m
4C(1 +Cq
m
)
(1 Cq
m
)
2
, (1.2.56)
holds for all k N and m m
0
. Hence
_
q
2k
x
2
m+k
x
2
m
1
_
q
m
max
_
4C(1 +Cq
m
)
(1 +Cq
m
)
2
,
4C(1 +Cq
m
)
(1 Cq
m
)
2
_
=
4C(1 +Cq
m
)
(1 Cq
m
)
2
, (1.2.57)
for all m m
0
and k N. Thus the sequence
_
_
q
2k
x
2
m+k
x
2
m
1
_
q
m
_
m=1
is uniformly
bounded for k N since
4C(1 +Cq
m
)
(1 Cq
m
)
2
4C as m . (1.2.58)
20
Lemma 1.2.4. For n N, if z C and x R are such that [zx[
q
n
1 q
, then for r Z
+
,
we have
r
m=1
x
2
m
+q
2n
[z[
2
x
2
x
2
m
+[z[
2
x
2
q
2nr
r
m=1
(1 +x
2
m
). (1.2.59)
Proof. Assume that z C, x R such that [zx[
q
n
1 q
, n N. Then
x
2
m
+q
2n
[z[
2
x
2
x
2
m
+[z[
2
x
2
q
2n
=
(1 q
2n
)x
2
m
x
2
m
+[zx[
2
x
2
m
[zx[
2
q
2n
x
2
m
. (1.2.60)
Consequently
r
m=1
x
2
m
+q
2n
[z[
2
x
2
x
2
m
+[z[
2
x
2
r
m=1
q
2n
(1 +x
2
m
) = q
2nr
r
m=1
(1 +x
2
m
), (1.2.61)
proving the lemma.
Similarly we have
Lemma 1.2.5. For n N, if z C and y R are such that [zy[
q
n
1 q
, then for r Z
+
,
we have
r
m=1
y
2
m
+q
2m
[z[
2
y
2
y
2
m
+[z[
2
y
2
q
2nr
r
m=1
(1 +y
2
m
). (1.2.62)
1.3 Asymptotics of cos(z; q) and sin(z; q)
In this section, we study the asymptotic behavior of the basic trigonometric functions
cos(z; q), and sin(z; q). The results of this section are contained in [19]. We start our investi-
gations with the following lemma that exhibits a major dierence between the trigonometric
functions and their basic analogues.
Lemma 1.3.1. The basic trigonometric functions sin(x; q) and cos(x; q) are not bounded
on R.
21
Proof. Daalhuis in [38] and Chen et al. in [35] proved that J
(x; q
2
) has the asymptotic
behavior
J
(x; q
2
)
[x[
(x
2
q
2
; q
2
)
(q
2
; q
2
)
, as [x[ . (1.3.1)
But (x
2
; q
2
)
k=n+2
(1 r
2
n
q
2k
)
k=n+2
(1 q
2k2n2
) =
j=1
(1 q
2j
) = (q
2
; q
2
)
, (1.3.4)
and
n1
k=0
(1 r
2
n
q
2k
)
n1
k=0
(1 q
2n2k
) = (q
2
; q
2
)
n
> (q
2
; q
2
)
. (1.3.5)
But
(r
2
n
; q
2
)
k=0
[1 r
2
n
q
2k
[
= (r
2
n
q
2n
1)(1 r
2
n
q
2n+2
)
n1
k=0
[1 r
2
n
q
2k
[
k=n+2
(1 r
2
n
q
2k
)
=
q
2
16
(1 + 2q 3q
2
)(3 2q q
2
)r
2n
n
q
n(n1)
n1
k=0
(1 r
2
n
q
2k
)
k=n+2
(1 r
2
n
q
2k
).
(1.3.6)
Combining equations (1.3.3)(1.3.6) yields
(r
2
n
; q
2
)
q
2
16
(1 + 2q 3q
2
)(3 2q q
2
)(q
2
; q
2
)
2
q
n
2
n
r
2n
n
=
q
2
16
(1 + 2q 3q
2
)(3 2q q
2
)(q
2
; q
2
)
2
q
n
2
3n
_
1 +q
2
_
2n
.
(1.3.7)
But
_
1 +q
2
_
2n
q
2n
, for all n N. (1.3.8)
22
Thus
[(r
2
n
; q
2
)
[
q
2
16
(1 + 2q 3q
2
)(3 2q q
2
)(q
2
; q
2
)
2
q
n
2
n
. (1.3.9)
Hence
(r
2
n
; q
2
)
as n .
This proves that (x
2
; q
2
)
(x; q
2
) is also not bounded on
R. Applying this result to the cases = 1/2 respectively and using (1.1.16) and (1.1.17),
we deduce that the functions cos(x; q) and sin(x; q) are not bounded on R.
The following two lemmas will be needed in the sequel.
Lemma 1.3.2. For a positive sequence z
m
m=1
, if z
m
1 as m , then inf
mN
z
m
> 0.
Proof. Since z
m
1 as m , then for 0 < < 1, there exists m
0
N such that
[z
m
1[ < for all m m
0
. Hence z
m
1 for all m m
0
. That is inf
mm
0
z
m
1 > 0.
Therefore
inf
mN
z
m
= min
_
z
1
, z
2
, . . . , z
m
0
, inf
mm
0
z
m
_
> 0.
Lemma 1.3.3. If z C such that 0 [z[ < 1, then
log [1 z[
log
1
1 [z[
. (1.3.10)
Proof. Let z C such that 0 [z[ < 1. Since [1 z[ 1 [z[, then we have
log [1 z[ log(1 [z[) = log
1
1 [z[
. (1.3.11)
Also [1 z[ 1 +[z[, then
log [1 z[(1 [z[) log(1 [z
2
[) 0.
Consequently
log [1 z[ log(1 [z[) = log
1
(1 [z[)
. (1.3.12)
23
Combining (1.3.11) and (1.3.12) we obtain (1.3.10).
In Lemma A the asymptotic behavior of log [(z; q)[ as z is introduced by studying
the asymptotic behavior of log [(z; q)[ in the set of annuli A
m
m=1
dened in (1.2.18) when
m is large enough. Similarly we shall study the asymptotic behavior of log [ cos(z; q)[ and
log [ sin(z; q)[. We begin with the q-cosine function. We divide the complex plane except
for a nite disk D with center at zero into an innite set of annuli A
c
m
m=1
, where the
disk D will be determined later. Then we study the asymptotic behavior of log [ cos(z; q)[
as z via the study of its behavior in each A
c
m
, for suciently large m. Let
m
:= log (x
m
/x
m+1
)/ log q, m Z
+
. (1.3.13)
Then
m
> 0 for all m Z
+
. Moreover, from (1.2.20)
m
= 1 +
_
log
_
1 +O(q
m
)
1 +O(q
m+1
)
___
log q,
for suciently large m. That is lim
m
m
= 1. Then from Lemma 1.3.2
0 < := inf
mZ
+
m
1. (1.3.14)
We dene two sequences, (a
m
)
m=1
, (b
m
)
m=1
, to be
a
m
:=
_
m
+
2
, if
m
,= ,
2
, if
m
= ,
(1.3.15)
and
b
1
:=
2
, b
m+1
:=
_
2
, if
m
,= ,
2
, if
m
= ,
(1.3.16)
where m 1. Then we have
x
m
q
a
m
= x
m+1
q
b
m+1
, m 1. (1.3.17)
Thus the set of annuli A
c
m
m=1
where A
c
m
is dened to be
A
c
m
:=
_
z C : x
m
q
b
m
[z[ < x
m
q
a
m
_
, m 1 (1.3.18)
24
Figure 1.2: The annulus A
c
2
divide the region
_
z C : [z[ q
/2
x
1
_
.
In the following we shall study the asymptotic behavior of log [ cos(z; q)[ in each of the
annuli described above when m is suciently large.
Theorem 1.3.4. Let be the positive number dened in (1.3.14). Assume that [z[ q
/2
x
1
and let A
c
m
, m 1, be the annulus dened by (1.3.18). Then we have, uniformly as m ,
the asymptotic relation
log [ cos(z; q)[ =
(log [z[)
2
log q
+ log [1
z
2
x
2
m
[ +O(1), z A
c
m
. (1.3.19)
Proof. First of all, it should be noted that z A
c
m
, m 1, if and only if
[z[ = x
m
q
g
m
(t)
, g
m
(t) := (a
m
+b
m
)t +b
m
, t [0, 1). (1.3.20)
Let m 1 and z A
c
m
be xed. Then there exists t [0, 1) such that
r := [z[ = x
m
q
g
m
(t)
= x
m
q
(a
m
+b
m
)t+b
m
. (1.3.21)
25
Since cos(z; q) is an entire function of order zero, then, see e.g. [27],
cos(z; q) =
n=1
_
1
z
2
x
2
n
_
. (1.3.22)
Thus
log
cos(z; q)
1 (
z
x
m
)
2
=
m1
n=1
log
1
z
2
x
2
n
n=m+1
log
1
z
2
x
2
n
=
m1
n=1
log [
z
x
n
[
2
+
m1
n=1
log
1
x
2
n
z
2
n=m+1
log
1
z
2
x
2
n
.
(1.3.23)
Set
n
:= q
n
1
2
(1 q)x
n
1, n = 1, 2, . . . . (1.3.24)
From (1.2.20),
n
= O(q
n
) as n . Dene the constants l, L to be
l := inf
nN
[1 +
n
[, L := sup
nN
[1 +
n
[. (1.3.25)
Since 1 +
n
> 0, for all n N and 1 +
n
1 as n , then l > 0. We dene constants
K
1
and K
2
as follows
K
1
:=
L
l
, K
2
:= q
a
; a = sup
nZ
+
a
n
. (1.3.26)
Then q
g
m(t)
q
a
m
K
2
for all t [0, 1). From (1.3.24), we have for n m+ 1
z
x
n
=
x
m
q
g
m
(t)
x
n
=q
nm
1 +
m
1 +
n
q
g
m
(t)
K
1
q
a
m
q
nm
K
1
K
2
q
nm
.
(1.3.27)
Consequently
n=m+1
log
1
z
2
x
2
n
n=m+1
log
_
1 +
z
x
n
2
_
n=m+1
z
x
n
2
K
2
1
K
2
2
n=m+1
q
2(nm)
K
2
1
K
2
2
q
2
1 q
2
. (1.3.28)
As for n = 1, 2, . . . , m1,
x
n
z
=
x
n
q
g
m
(t)
x
m
= q
mn
1 +
n
1 +
m
q
g
m
(t)
K
1
q
b
m
q
mn
. (1.3.29)
26
From (1.3.15) and (1.3.16) we obtain
q
b
m
= q
q
a
m1
q
q
a
= K
2
q
. (1.3.30)
Combining (1.3.29) and (1.3.30) yields
x
n
z
K
1
K
2
q
mn
. (1.3.31)
Moreover
x
n
z
=
x
n
q
g
m
(t)
x
m
x
m1
q
g
m
(t)
x
m
, n = 1, 2, . . . , m1. (1.3.32)
But from (1.3.17)
x
m1
x
m
= q
a
m1
+b
m
, and q
g
m
(t)
= q
a
m
t(1t)b
m
q
b
m
.
Then
x
n
z
q
a
m1
< q
< 1, z A
c
m
. (1.3.33)
So from (1.3.10) we obtain
log
1
x
2
n
z
2
log
1
_
1 [
x
2
n
z
2
[
_ = log
_
_
1 +
j=1
[
x
n
z
[
2j
_
_
= log
_
1 +
[
x
n
z
[
2
1 [
x
n
z
[
2
_
[
x
n
z
[
2
1 [
x
n
z
[
2
[
x
n
z
[
2
1 q
2
K
2
1
K
2
2
1 q
2
q
2(mn)
,
(1.3.34)
Therefore
m1
n=1
log
1
x
2
n
z
2
K
2
1
K
2
2
1 q
2
m1
n=1
q
2(mn)
K
2
1
K
2
2
q
2
(1 q
2
)(1 q
2
)
. (1.3.35)
Hence from (1.3.23), (1.3.28), and (1.3.35) we obtain
log
cos(z; q)
1 (
z
x
m
)
2
=
m1
n=1
log [
z
x
n
[
2
+O(1), as z . (1.3.36)
27
Now
m1
n=1
log
z
x
n
2
=
m1
n=1
log [z[
2
m1
n=1
log [x
n
[
2
= 2(m1) log [z[ 2
m1
n=1
(n + 1/2) log q 2
m1
n=1
log [1 +
n
[
= 2(m1) log [z[ + (m1)
2
log q 2
m1
n=1
log [1 +
n
[. (1.3.37)
Since
n
= O(q
n
) as n , then there exists a constant K
3
> 0 and n
0
N such that
[
n
[ K
3
q
n
, for all n n
0
. Let
K
4
:= max
_
q
1
[
1
[, . . . , q
m
0
+1
[
n
0
1
[, K
3
_
.
Thus [
n
[ K
4
q
n
for all n Z
+
, and
m1
n=1
log [1 +
n
[
m1
n=1
log
_
1 +[
n
[
_
m1
n=1
[
n
[
n=1
[
n
[ K
4
n=1
q
n
K
4
q
1 q
.
(1.3.38)
Since r =
q
m+1/2
(1q)
(1 +
m
)q
g
m
(t)
, then
m1 =
log r
log q
+d
m
, d
m
:= 1/2 +
log(1 +
m
)
log q
+g
m
(t)
log(1 q)
log q
. (1.3.39)
A simple computation yields
2(m1) log r + (m1)
2
log q =
(log r)
2
log q
+d
2
m
log q =
(log r)
2
log q
+O(1), (1.3.40)
as m since d
m
m=1
is a bounded sequence. Hence the theorem follows by combining
(1.3.36), (1.3.37), and (1.3.40).
Corollary 1.3.5.
M(r; cos(z; q)) = O
_
(exp
_
(log [z[)
2
log q
__
, as z . (1.3.41)
28
Proof. From (1.3.19) we conclude that there exists m
0
N such that for all m m
0
we
have
(log r)
2
log q
+ log [1
z
2
x
2
m
[ log [ cos(z; q)[
(log r)
2
log q
+ log [1
z
2
x
2
m
[ +, z A
c
m
.
(1.3.42)
From (1.3.18) we obtain
q
2b
m
[z[
2
x
2
m
q
2a
m
. (1.3.43)
Hence
log
1
z
2
x
2
m
log
1 +
[z[
2
x
2
m
log q
2a
m
, m 1, (1.3.44)
Since the sequence a
m
m=1
is bounded, then there exists a constant R > 0 such that
log [ cos(z; q)[
(log [z[)
2
log q
+R, z A
c
m
, m m
0
. (1.3.45)
Thus
[ cos(z; q)[ e
R
e
(log [z[)
2
log q
, z A
c
m
, m m
0
, (1.3.46)
proving the corollary.
The same study may be carried out for sin(z; q) and for this purpose we dene suitable
annuli. Let
m
:= log (y
m
/y
m+1
)/ log q, m Z
+
. (1.3.47)
Since
m
1 as m , then
:= inf
mZ
+
m
> 0. (1.3.48)
Let (c
m
)
m=1
and (d
m
)
m=1
be the sequence dened by
c
m
:=
_
m
+
2
, if
m
,= ,
2
, if
m
= ,
(1.3.49)
29
and
d
1
:=
2
, d
m+1
:=
_
2
, if
m
,= ,
2
, if
m
= ,
(1.3.50)
where m 1. Then we have
y
m
q
c
m
= y
m+1
q
d
m+1
, m 1. (1.3.51)
Thus the set of annuli A
s
m
m=1
where A
s
m
is dened to be
A
s
m
:=
_
z C : y
m
q
d
m
[z[ < y
m
q
c
m
_
, m 1 (1.3.52)
divide the region
_
z C : [z[ q
/2
y
1
_
.
Theorem 1.3.6. Let be the positive number dened in (1.3.48). Assume that [z[ q
/2
x
1
.
Let A
s
m
, m 1, be the annulus dened in (1.3.52). Then we have, uniformly as m
log [ sin(z; q)[ =
(log [z[)
2
log q
+ log [1
z
2
y
2
m
[ +O(1), z A
s
m
. (1.3.53)
The proof is similar to that of Theorem 1.3.4, and so it will not be given here.
Also we have the following corollary
Corollary 1.3.7.
M(r; sin(z; q)) = O
_
(exp
_
(log [z[)
2
log q
__
, as z . (1.3.54)
Lemma 1.3.8. The function tanh([z[; q) :=
sinh([z[; q)
cosh([z[; q)
is bounded on C.
Proof. From (1.3.19) and (1.3.53) we conclude that there exists m
0
N such that for all
m m
0
we have
(log r)
2
log q
+ log [1 +
[z[
2
x
2
m
[ log [ cosh([z[; q)[
(log r)
2
log q
+ log [1 +
[z[
2
x
2
m
[ +, z A
c
m
(1.3.55)
30
and
(log r)
2
log q
+log
1 +
[z[
2
y
2
m
1 +
[z[
2
y
2
m
+, z A
s
m
. (1.3.56)
But if z A
s
m
, m 1, then [z[
2
/y
2
m
q
2c
m
, where the sequence c
m
m=1
dened in
(1.3.49) is bounded. Hence there exists a constant L > 0 such that
1 +
[z[
2
y
2
m
L, for all m Z
+
.
Let r
0
:= max
_
q
b
m
0
x
m
0
, q
d
m
0
y
m
0
_
. Thus by (1.3.55) and (1.3.56) we obtain for all z C,
[z[ r
0
sinh([z[; q)
cosh([z[; q)
e
2
[1 +
[z[
2
y
2
m
[ Le
2
. (1.3.57)
1.4 Basic analogues of a theorem of Polya
In this section we give a q-analogue of Theorem 1.4.1 below of George Polya, cf. [102, 103].
These q-analogues are the main results of [18].
Theorem 1.4.1. If the function f L
1
(0, 1) is positive and increasing, then
1. the zeros of the entire functions of exponential type
U(z) =
_
1
0
f(t) cos(zt) dt, V (z) =
_
1
0
f(t) sin(zt) dt (1.4.1)
are real, innite and simple.
2. U(z) is an even function having no zeros in [0,
2
), and its positive zeros are situated in
the intervals (k/2, k+/2), 1 k < , one in each. The odd function V (z) has only
one zero z = 0 in [0, ), and its positive zeros are situated in the intervals (k, (k + 1)),
1 k < , one zero in each interval.
31
Theorem 1.4.2. If f L
1
q
(0, a) and U
f
(z) is dened for z C by
U
f
(z) :=
_
1
0
f(t) cos(tz; q) d
q
t, 0 < q < 1. (1.4.2)
Then U
f
(z) has at most a nite number of non real zeros and it has an innite number of
real zeros
m
m=1
,
m
> 0, such that
m
x
m
as m . More precisely
m
= x
m
(1 +O(q
m
)) as m . (1.4.3)
Proof. From the denition of the q-integral (1.1.31) U
f
(z) can be written as
U
f
(z) = H(z) +R(z), (1.4.4)
where
H(z) := (1 q)f(1) cos(z; q), and R(z) :=
k=1
q
k
(1 q)f(q
k
) cos(q
k
z; q). (1.4.5)
From (1.3.41) we deduce that there exists constants r
0
, C > 0 such that
max
|z|=r
[ cos(z; q)[ Ce
(log r)
2
/log q
, for all r > r
0
. (1.4.6)
Let z C such that [z[ > q
1
r
0
, then
[R(z)[ Ce
(log qr)
2
log q
k=1
q
k
(1 q)[f(q
k
)[ = q
1
C
e
(log r)
2
/ log q
r
2
_
q
0
[f(t)[ d
q
t. (1.4.7)
Thus
log [R(z)[ C
1
2 log r
_
log r
_
2
log q
, (1.4.8)
where C
1
:= log
_
q
1
C
_
q
0
[f(t) d
q
t
_
. Let A
c
m
, m 1, be the annuls dened in (1.3.18).
Then
log [ cos(z; q)[ =
_
log r
_
2
log q
+ log [1
z
2
x
2
m
[ +O(1), z A
c
m
(1.4.9)
uniformly as m . Hence there exists a constant C
2
> 0 and m
0
N such that for all
z A
c
m
, m m
0
we have
_
log r
_
2
log q
+log
1
z
2
x
2
m
C
2
log [ cos(z; q)[
_
log r
_
2
log q
+log
1
z
2
x
2
m
+C
2
. (1.4.10)
32
Consequently by (1.4.8)
log [R(z)[ C
1
+C
2
2 log r + log [ cos(z; q)[ log
1
z
2
x
2
m
1
z
2
x
2
m
+C
3
, (1.4.11)
where C
3
:= C
1
+C
2
log(1 q)[f(1)[. Let D
m
0
be the disk dened by
D
m
0
:=
_
z C : [z[ q
b
m
0
x
m
0
_
. (1.4.12)
Clearly
C := D
m
0
mm
0
A
m
.
If z D
m
0
, i.e. [z[ = q
b
m
0
x
m
0
, then by (1.4.11)
log [R(z)[ log [H(z)[ 2 log r log
1 q
2b
m
0
[ +C
3
. (1.4.13)
We can choose the m
0
suciently large such that 2 log r log
1q
2b
m
0
+C
3
< 0. That is
log [R(z)[ log [H(z)[, [z[ = q
b
m
0
x
m
0
. (1.4.14)
So applying Rouche theorem on D
m
0
, we conclude that H(z) and U
f
(z) have the same
number of zeros inside D
m
0
. As cos(z; q) has exactly 2m
0
2 zeros inside D
m
0
, then U
f
(z)
has at most 2m
0
2 zeros inside D
m
0
.
If z A
c
m
, m m
0
, then
1
z
2
x
2
m
1 q
2b
m
; [z[ = q
b
m
x
m
,
1
z
2
x
2
m
q
2a
m
1; [z[ = q
a
m
x
m
. (1.4.15)
Thus
log [1
z
2
x
2
m
[ inf
mm
0
_
log(1 q
2b
m
) log(q
2a
m
1)
_
, z A
c
m
. (1.4.16)
Since the sequences a
m
m=1
and b
m
m=1
are bounded and positive, then there exits a
constant C
4
> 0 such that
log [1
z
2
x
2
m
[ C
4
, z A
c
m
, m m
0
. (1.4.17)
33
That is
log [R(z)[ log [H(z)[ 2 log r +C
3
C
4
. (1.4.18)
Again, we choose the m
0
large enough such that if r = [z[ A
c
m
, m m
0
, then 2 log r +
C
3
C
4
< 0. That is
[R(z)[ [H(z)[, z A
c
m
, m m
0
.
Applying Rouche theorem on A
c
m
, m m
0
, we conclude that H(z) and U
f
(z) have the
same number of zeros inside A
c
m
. As cos(z; q) has two simple symmetric zeros there, we
deduce that the even function U
f
(z) has only two real, symmetric and simple zeros inside
A
c
m
, m m
0
.
Now we give the asymptotic behavior of such zeros. Let
m
, be a positive zero of U
f
(z)
in A
c
m
, then log [H(
m
)[ = log [R(
m
)[. Consequently by (1.4.11)
log
1
2
m
x
2
m
C
3
2 log [
m
[, m m
0
. (1.4.19)
Thus either
log
1
m
x
m
C
3
2
log [
m
[, (1.4.20)
or
log
1 +
m
x
m
C
3
2
log [
m
[. (1.4.21)
We can assume without any loss of generality that
m
satises (1.4.20), then
1
m
x
m
= O([
m
[
1
), as m . (1.4.22)
Thus
m
x
m
for all suciently large m. More precisely
m
= x
m
(1 +O(q
m
)), as m . (1.4.23)
Since U
f
(z) is even function then
m
are both zeros of U
f
(z) in A
c
m
. This completes the
proof.
Similarly we have the following theorem when we replace cos(tz; q) by sin(tz; q) in (1.4.2).
34
Theorem 1.4.3. If f L
1
q
(0, a) and V
f
(z) be dened for z C by
V
f
(z) :=
_
1
0
f(t) sin(tz; q) d
q
t, 0 < q < 1. (1.4.24)
Then V
f
(z) has at most a nite number of non real zeros and it has an innite number
of real zeros
m
m=1
,
m
> 0, for all m Z
+
such that
m
y
m
as m . More
precisely
m
= y
m
(1 +O(q
m
)) as m . (1.4.25)
Chapter 2
Zeros of Finite q-Hankel
Transforms
In this chapter we prove that the basic nite Hankel transform whose kernel is the third-
type Jackson q-Bessel function has innitely many real and simple zeros under certain
conditions on q. We also study the asymptotic behavior of the zeros. The obtained results
are applied to investigate the zeros of q-Bessel functions as well as zeros of q-trigonometric
functions. A basic analogue of a theorem of G. Polya (1918) on the zeros of sine and cosine
transformations is given as a special case.
2.1 Introduction and preliminaries
This chapter is devoted to a study the zeros of the q-type nite Hankel transform
F(z) =
_
1
0
f(t)J
(tz; q
2
) d
q
t, z C, (2.1.1)
where J
(; q
2
) is the third Jackson q-Bessel function dened in (1.1.15), cf. [17]. Throughout
this chapter we assume that > 1. Next we state and prove the main results of this
chapter. These results prove that, under certain conditions on f and q, the nite q-Hankel
transform (2.1.1) has only real and simple zeros and there are innitely many zeros. We
also investigate the distribution of these zeros. Section 2.3 contains some applications of
35
36
the results of Section 2.2. Among these applications we derive another basic analogue of a
theorem of Polya. We also derive in a manner dierent from the analysis of [7, 28, 63, 112]
the distribution of zeros of q-Bessel and q-trigonometric functions. It is worth mentioning
that q-type Hankel transforms are studied in [80] where the Hankel transform is dened on
(0, ) simulating the classical situation. The following lemma is needed for the proof of
the main results of this chapter. It is taken from [103, p. 143].
Lemma A. Let g
1
(z), g
2
(z), . . . , g
n
(z), . . . be entire functions which have real zeros only.
If
lim
n
g
n
(z) = g(z),
uniformly in any compact subset of C, then g(z) is an entire function with only real zeros.
We use the following convenient notations. For f L
1
q
(0, 1), we denote by A
k
(f) the
q-moments of f, i.e.
A
k
(f) :=
_
1
0
t
k
f(t) d
q
t, k N, (2.1.2)
see (1.1.31) for the denition of the q-integration. Let also c
k,
(f), b
k,
(f), > 1, k N,
denote the numbers
c
k,
(f) :=
A
2k+2
(f)
A
2k
(f)(1 q
2k+2+2
)(1 q
2k+2
)
, k N, (2.1.3)
and
b
k,
(f) :=
A
2k+3
(f)
A
2k+1
(f)(1 q
2k+2+2
)(1 q
2k+2
)
, k N, (2.1.4)
We start our investigations with the following lemma.
Lemma 2.1.1. Let f L
1
q
(0, 1) be positive on 0, q
n
, n N. Then the numbers
c
,f
:= inf
kN
c
k,
(f), C
,f
:= sup
kN
c
k,
(f), (2.1.5)
and
b
,f
:= inf
kN
b
k,
(f), B
,f
:= sup
kN
b
k,
(f), (2.1.6)
37
exist and they are nite positive numbers.
Proof. We will prove only (2.1.5). The proof of (2.1.6) is similar. First of all it should
be noted that the sequence A
k
(f)
k=0
is positive and strictly decreasing. Using Cauchy-
Schwarz inequality, for k N, k 2, we obtain
_
A
2k
(f)
_
2
=
__
1
0
t
2k
f(t) d
q
t
_
2
=
_
_
j=0
q
j
(1 q)q
2kj
f(q
j
)
_
_
2
j=0
q
j
(1 q)q
(2k+2)j
f(q
j
)
j=0
q
j
(1 q)q
(2k2)j
f(q
j
)
= A
2k+2
(f)A
2k2
(f).
That is,
_
A
2k+2
(f)
A
2k
(f)
_
k=0
is an increasing sequence of positive numbers. Since A
2k
(f)
k=0
is strictly decreasing, then the above inequality implies
A
2
(f)
A
0
(f)
A
2k+2
(f)
A
2k
(f)
< 1, k = 0, 1, . . . . (2.1.7)
Therefore, for k N we have
A
2
(f)
A
0
(f)
A
2k+2
(f)
A
2k
(f)
A
2k+2
(f)
A
2k
(f)(1 q
2k+2+2
)(1 q
2k+2
)
= c
k,
(f)
<
1
(1 q
2k+2+2
)(1 q
2k+2
)
1
(1 q
2+2
)(1 q
2
)
.
(2.1.8)
Inequalities (2.1.8) and the BolzanoWeierstrass theorem guarantee the existence of the
positive numbers c
,f
, C
,f
.
2.2 Main results
In this section we will investigate the zeros of the nite q-Hankel Transforms.
U
,f
(z) := z
_
1
0
t
f(t)J
(tz; q
2
) d
q
t, z C, (2.2.1)
V
,f
(z) := z
1
_
1
0
t
1
f(t)J
(tz; q
2
) d
q
t, z C. (2.2.2)
38
It is known that J
_
1
0
t
f(t)J
(tz; q
2
) d
q
t
=
(q
2+2
; q
2
)
(q
2
; q
2
)
k=0
(1)
k
q
k
2
+k
z
2k
(q
2+2
; q
2
)
k
(q
2
; q
2
)
k
_
1
0
t
2k
f(t) d
q
t
=
(q
2+2
; q
2
)
(q
2
; q
2
)
k=0
(1)
k
q
k
2
+k
A
2k
(f)
z
2k
(q
2+2
; q
2
)
k
(q
2
; q
2
)
k
=
k=0
(1)
k
a
k
z
2k
, (2.2.3)
where a
k
is the sequence of positive numbers
a
k
:=
(q
2+2
; q
2
)
(q
2
; q
2
)
q
k
2
+k
A
2k
(f)
(q
2+2
; q
2
)
k
(q
2
; q
2
)
k
, k N. (2.2.4)
Since the radius of convergence of (2.2.3) is
lim
k
a
k
a
k+1
= lim
k
A
2k
A
2k+2
q
2k2
(1 q
2k+2
)(1 q
2+2k+2
)
= lim
k
A
2k
A
2k+2
q
2k2
= ,
(2.2.5)
U
,f
(z) is entire.
Now we show that U
,f
(z) has order zero. From (1.2.3)
(U
,f
) = limsup
k
k log k
log a
1
k
. (2.2.6)
From (2.2.4) we obtain
log
1
a
k
= log
_
(q
2
; q
2
)
(q
2
; q
2
)
k
(q
2+2k+2
; q
2
)
A
2k
(f)
_
+ (k
2
+k) log
1
q
, (2.2.7)
39
where we have used
(q
2+2
; q
2
)
k
(q
2+2
; q
2
)
=
_
_
q
2+2k+2
; q
2
_
_
1
.
Since A
k
(f)
k=0
is a bounded decreasing sequence, then from the continuity of the loga-
rithmic function, it follows that
log
_
(q
2
; q
2
)
(q
2
; q
2
)
k
(q
2+2k+2
; q
2
)
A
2k
(f)
_
log
_
(q
2
; q
2
)
2
A(f)
_
, as k ,
where A(f) := lim
k
A
k
(f). Recalling that 0 < q < 1, we deduce from (2.2.7)
lim
k
log
1
a
k
k log k
= lim
k
(k
2
+k) log
1
q
k log k
= . (2.2.8)
Thus (U
,f
) = 0.
The following is the rst main result of this chapter. The second is Theorem 2.2.3 below.
So far we only assumed 0 < q < 1. In our main results as well as those in the existing similar
studies, cf. e.g. [7, 28], more restrictions are imposed on q.
Theorem 2.2.2. Let > 1 and f L
1
q
(0, 1) be positive on 0, q
n
, n N. If
q
1
(1 q)
c
,f
C
,f
> 1, (2.2.9)
then the zeros of U
,f
(z) are real, simple and innite in number. Moreover, U
,f
(z) is an
even function with no zeros in
_
0, q
1
/
_
C
,f
_
, and its positive zeros lie in the intervals
_
q
r+1/2
_
C
,f
,
q
r1/2
_
C
,f
_
, r = 1, 2, 3, . . . , (2.2.10)
one zero in each interval.
Proof. We prove the theorem in three steps.
1. We show that U
,f
(z) has no zeros in B
R
(0), where R := 1/q
_
C
,f
and
B
R
(0) := z C : [z[ R, z ,= R .
40
First
0 <
a
1
a
0
= q
2
c
0,
(f) q
2
C
,f
= R
2
,
0 <
a
k+1
a
k
= q
2k+2
c
k,
(f) < q
2
C
,f
= R
2
, k 1.
(2.2.11)
Therefore, a
0
R
2
a
1
0 and a
k
R
2
a
k+1
> 0 for all k 1. If z B
R
(0), then by (2.2.3)
(R
2
+z
2
)U
,f
(z)
a
0
R
2
k=1
(1)
k
z
2k
(a
k1
R
2
a
k
)
a
0
R
2
k=1
(1)
k
z
2k
(a
k1
R
2
a
k
)
. (2.2.12)
If z C is not real, i.e. z = re
i
, ,= k, k Z, then for c
0
, c
1
> 0,
[c
0
+c
1
z[
2
= c
2
0
+ 2c
0
c
1
r cos +c
2
1
r
2
< c
2
0
+ 2c
0
c
1
r +c
2
1
r
2
= (c
0
+c
1
[z[)
2
. (2.2.13)
Consequently [c
0
+c
1
z[ < c
0
+c
1
[z[. Thus for non real z B
R
(0), we obtain
k=1
(1)
k
z
2k
(a
k1
R
2
a
k
)
z
4
(a
1
a
2
R
2
) +z
6
(a
2
a
3
R
2
)
k=1
k=2,3
[z[
2k
(a
k1
R
2
a
k
)
< [z
4
[
_
(a
1
a
2
R
2
) +[z[
2
(a
2
a
3
R
2
)
_
+
k=1
k=2,3
[z[
2k
(a
k1
R
2
a
k
) (2.2.14)
=
k=1
[z[
2k
(a
k1
R
2
a
k
)
k=1
(R
2k
a
k1
R
2k+2
a
k
)
= a
0
R
2
lim
m
R
2m+2
a
m
.
From (2.2.11)
m1
k=0
a
k+1
a
k
=
m1
k=0
q
2k+2
c
k,
(f)
m1
k=0
q
2k+2
C
,f
= q
m
2
+m
C
m
,f
. (2.2.15)
Thus
a
m
q
m
2
+m
C
m
,f
a
0
, R
2m+2
a
m
q
m
2
m2
a
0
C
,f
. (2.2.16)
41
That is lim
m
R
2m+2
a
m
= 0, and therefore we have for a non real z B
R
(0), z ,= iR,
(R
2
+z
2
)U
,f
(z)
> lim
m
R
2m+2
a
m
= 0. (2.2.17)
If z = iR, then by (2.2.3)
U
,f
(iR) =
k=0
a
k
R
2k
> 0. (2.2.18)
Thus U
,f
(z) > 0 for all non real z B
R
(0). On the other hand if 0 < [z[ < R, z is real,
then
(R
2
+z
2
)U
,f
(z)
a
0
R
2
k=1
(1)
k
z
2k
(a
k1
R
2
a
k
)
a
0
R
2
k=1
[z[
2k
(a
k1
R
2
a
k
)
> a
0
R
2
k=1
R
2k
(a
k1
R
2
a
k
),
(2.2.19)
since if 0 <
n
<
n
, and
n=0
n
< , then
n=0
n
<
n=0
n
. Moreover
a
0
R
2
k=1
R
2k
(a
k1
R
2
a
k
) = a
0
R
2
_
a
0
R
2
lim
m
R
2m+2
a
m
_
= 0.
Since
U
,f
(0) =
(q
2+2
; q
2
)
(q
2
; q
2
)
A
0
(f) ,= 0,
then
(R
2
+ z
2
)U
,f
(z)
(q
2
; q
2
)
2n
k=0
(1)
k
q
k
2
+k
A
2k
(f)
(q
2+2
; q
2
)
k
(q
2
; q
2
)
k
z
2k
, n N, z C. (2.2.20)
Obviously U
2n,,f
(z) approaches U
,f
(z) uniformly as n on any compact subset of
C. Take z
r
:= q
r1/2
/
_
C
,f
, r N. We prove that U
2n,,f
(z) has a zero in the interval
42
(z
r1
, z
r
), r = 1, 2, . . . , 2n. Indeed,
U
2n,,f
(z
r
) =
2n
k=0
(1)
k
k
(r), (2.2.21)
where
k
(r) is the positive sequence
k
(r) :=
(q
2+2
; q
2
)
(q
2
; q
2
)
q
k
2
2kr
C
k
,f
(q
2
; q
2
)
k
(q
2+2
; q
2
)
k
A
2k
(f), k N. (2.2.22)
If 0 k r 1, then from (2.2.9)
k+1
(r)
k
(r)
= q
2k2r+1
c
k,
(f)
C
,f
c
,f
C
,f
q
1
> (1 q)
1
> 1, (2.2.23)
and
k+1
(r)
k
(r)
q < 1, if k r. (2.2.24)
From (2.2.23), and (2.2.24), we obtain for r 1
r+1
(r)
r
(r)
+
r1
(r)
r
(r)
< q + (1 q) = 1. (2.2.25)
Therefore, we have the following inequality
r+1
(r) <
r
(r)
r1
(r), r = 1, 2, 3, . . . . (2.2.26)
Now we prove that
sign U
2n,,f
(z
r
) = (1)
r
, r = 0, 1, . . . , 2n. (2.2.27)
If r = 0, then from (2.2.24)
k+1
(0)
k
(0)
< 1 for all k 0. Consequently
U
2n,,f
(z
0
) =
n1
j=0
_
2j
(0)
2j+1
(0)
_
+
2n
(0) > 0.
Take r = 2m, 1 m n. Then
U
2n,,f
(z
r
) =
m1
k=1
(
2k
(r)
2k1
(r)) +
n1
k=m+1
(
2k
(r)
2k+1
(r))
+
0
(r) +
2n
(r) +
_
r
(r)
r1
(r)
_
r+1
(r).
(2.2.28)
43
Hence from (2.2.23), (2.2.24), and (2.2.26),
U
2n,,f
_
z
r
_
r
(r)
r1
(r)
_
r+1
(r) > 0.
Similarly if r = 2m1, 1 m n, we have
U
2n,,f
_
z
r
_
=
m2
k=0
_
2k+1
(r)
2k
(r)
_
k=m+1
_
2k1
(r)
2k
(r)
_
r
(r)
r1
(r)
_
+
r+1
(r) < 0.
(2.2.29)
Thus, U
2n,,f
(z) has at least one zero in each of the intervals (z
r1
, z
r
), r = 1, 2, . . . , 2n,
i.e. it has at least 2n positive zeros. Since U
2n,,f
(z) is an even polynomial of degree 4n,
then it also has 2n negative zeros. So, the 4n zeros of the polynomial U
2n,,f
(z) are all
real and simple. Consequently, from Lemma A, U
,f
(z) has only real zeros. Since
k
(r) is
independent of n, the same argument holds when n is replaced by , and noting that all
series converge absolutely, shows that
sign U
,f
(z
r
) = (1)
r
, r N. (2.2.30)
proving that U
,f
(z) has an innite number of real zeros w
()
r
, w
()
r
(z
r
, z
r+1
), r N.
3. Finally we prove that the zeros of U
,f
(z) are simple. Consider the annulus T
r
dened
for r Z
+
by
T
r
:= z C : z
r1
[z[ z
r
. (2.2.31)
Since U
,f
(z) has no zeros on the boundary T
r
of T
r
, where [z[ =
q
r(1/2)
C
,f
, then there
exists > 0 such that [U
,f
(z)[ > on T
r
. For this we can nd N
0
N, N
0
> r such
that
[U
2m,,f
(z) U
,f
(z)[ , m N
0
, z T
r
. (2.2.32)
Hence,
[U
2m,,f
(z) U
,f
(z)[ < [U
,f
(z)[, m N
0
, z T
r
. (2.2.33)
44
Applying Rouches Theorem, we conclude that U
,f
(z) and U
2m,,f
(z), m N
0
, have the
same numbers of zeros inside T
r
. Since for all m N
0
, U
2m,,f
(z) has exactly two symmetric
simple real zeros inside T
r
, then so is U
,f
(z). This completes the proof of the theorem.
Remark 2.2.1. Step 2 of the proof of Theorem 2.2.2 includes a proof of the fact that the
function U
,f
has innitely many zeros. However, according to Theorem 1.2.1 we do not
need to give a proof because (U
,f
) = 0.
Theorem 2.2.3. Let > 1 and f L
1
q
(0, 1) be positive on 0, q
n
, n N. If
q
1
(1 q)
b
,f
B
,f
> 1, (2.2.34)
then the zeros of V
,f
(z) are real, simple and innite in number. The odd function V
,f
(z)
has only one zero z = 0 in [0,
q
1
B
,f
), and its positive zeros are situated in the intervals
(
q
r+1/2
_
B
,f
,
q
r1/2
_
B
,f
), r = 1, 2, 3, . . . , (2.2.35)
one zero in each interval.
Proof. The proof is similar to that of the previous theorem.
Corollary 2.2.4. Let > 1 and f L
1
q
(0, 1) be positive and decreasing on 0, q
n
, n N.
If
q
1
(1 q)
2
(1 q
2
)(1 q
2+2
)
f
2
(1)
f
2
(0)
> 1, (2.2.36)
then the zeros of U
,f
(z) are real, simple and innite in number. Moreover, the positive
zeros lie in the intervals
_
q
r+1/2
__
C
,f
, q
r1/2
__
C
,f
_
, r = 1, 2, . . . ,
one zero in each interval. We have the same result if f is increasing and positive on
0, q
n
, n N and q satises
q
1
(1 q)
2
(1 q
2
)(1 q
2+2
)
f
2
(0)
f
2
(1)
> 1, (2.2.37)
instead of (2.2.36).
45
Proof. We notice that if
q
1
(1 q)
l
,f
L
,f
> 1, (2.2.38)
where l
,f
and L
,f
are lower and upper bounds of the sequence c
k,
(f)
k=0
respectively,
then (2.2.9) is satised and the results of Theorem 2.2.2 hold. This observation suces to
prove the corollary because if f is decreasing and positive on 0, q
n
, n N, then by (2.1.2)
f(1)(1 q)
(1 q
2k+1
)
A
2k
(f)
f(0)(1 q)
(1 q
2k+1
)
, k N. (2.2.39)
Hence
f(1)(1 q
2k+1
)
f(0)(1 q
2k+3
)
A
2k+2
(f)
A
2k
(f)
f(0)(1 q
2k+1
)
f(1)(1 q
2k+3
)
, k N. (2.2.40)
Now the sequence
_
(1 q
2k+1
)
((1 q
2k+2
)(1 q
2k+3
)(1 q
2k+2+2
))
_
k=0
is bounded above by 1/
_
(1 q
2
)(1 q
2+2
)
_
and it is bounded below by (1 q). Let l
,f
and L
,f
be the numbers
l
,f
=
f(1)(1 q)
f(0)
, L
,f
=
f(0)
f(1)(1 q
2
)(1 q
2+2
)
. (2.2.41)
Then from (2.1.3), l
,f
and L
,f
are lower and upper bounds of c
k,,f
respectively. A
direct substitution with l
,f
and L
,f
of (2.2.41) in (2.2.38) yields (2.2.36). Similarly if f is
increasing and positive on 0, q
n
, n N, then l
,f
and L
,f
can be taken to be
l
,f
=
f(0)(1 q)
f(1)
, L
,f
=
f(1)
f(0)(1 q
2
)(1 q
2+2
)
.
A similar result holds for V
,f
when f is positive and monotonic on 0, q
n
, n N.
2.3 Applications
In this section we introduce some applications of the results of Section 2.2. First we inves-
tigate the zeros of J
(z; q
2
) and the q-trigonometric functions.
46
Lemma 2.3.1. If q (0, 1) satises the inequality
q
1
(1 q)(1 q
2
)(1 q
2+2
) > 1, > 1, (2.3.1)
then the zeros of the third Jackson q-Bessel function J
(; q
2
) are real, simple, and innite
in number. Moreover J
(; q
2
) has a zero at z = 0 if and only if > 0. The positive zeros
of J
(; q
2
) belong to the intervals
_
q
r+1/2
_
(1 q
2+2
)(1 q
2
), q
r1/2
_
(1 q
2+2
)(1 q
2
)
_
, r = 1, 2, . . . , (2.3.2)
where each interval contains only one zero.
Proof. Applying Theorem 2.2.2 with
f(t) :=
_
_
(1 q)
1
, t = 1,
0, otherwise,
then U
,f
(z) = z
(z; q
2
). For k N, we have
A
k
(f) 1, k N, C
,f
=
1
(1 q
2
)(1 q
2+2
)
, c
,f
= 1.
Therefore condition (2.2.9) is nothing but (2.3.1) and the lemma is proved.
Remark 2.3.1. In [66] asymptotics of the q-Bessel functions J
(1)
(; q) and J
(2)
(; q) is given.
Also, a study for the zeros of the second Jackson q-Bessel function J
(2)
(x; q) is given in
[60].
Corollary 2.3.2. i. If 0 < q <
0
, where
0
0.429052 is the root of
(1 q)(1 q
2
)(1 q
3
) q, q (0, 1),
then the zeros of sin(z; q) are real, innite and simple and the positive zeros belong to the
intervals
_
q
r+1/2
_
(1 q
2
)(1 q
3
), q
r1/2
_
(1 q
2
)(1 q
3
)
_
, r = 1, 2, . . . , (2.3.3)
47
and each interval contains only one zero.
ii. If 0 < q <
0
, where
0
0.38197 is the zero of (1 q)
3
(1 + q) q in 0 < q < 1, then
the zeros of cos(z; q) are real and simple and the positive zeros belong to the intervals
_
q
r
(1 q)
_
1 +q, q
r1
(1 q)
_
1 +q
_
, r = 0, 1, 2, . . . , (2.3.4)
and each interval contains only one zero.
Proof. To prove i, substitute with = 1/2 in (2.3.1). Then (2.3.1) becomes
q
1
(1 q)(1 q
2
)(1 q)
3
> 1
which holds if 0 < q <
0
. Since sin(z; q) has the representation (1.1.17), then applying
Lemma 2.3.1 with = 1/2, we conclude that if 0 < q <
0
, the zeros of sin(z; q) are real,
innite and simple and they lie in the intervals (2.3.3). The proof of ii. follows similarly by
considering = 1/2.
Remark 2.3.2. In [78] a study on the zeros of J
r=0
of the q-function
V
t
(z) :=
_
1
0
t sin(tz; q) d
q
t =
1
z
2
_
sin(z; q) z cos(q
1/2
z; q)
_
, 0 < q <
0
, (2.3.5)
are real, innite and simple where
0
0.441751 is the zero in 0 < q < 1 of
(1 q)(1 q
2
)(1 q
5
) q.
48
Moreover,
v
r
_
q
r+1/2
_
(1 q
2
)(1 q
5
), q
r1/2
_
(1 q
2
)(1 q
5
)
_
, r = 0, 1, . . . . (2.3.6)
As another example, dene a function f on [0, 1] to be
f(t) :=
_
_
t
_
(1 q), t = 1, q,
0, otherwise,
, (2.3.7)
> 1. If q
1
(1 q)c
,f
_
C
,f
> 1, then the zeros u
r
r=0
of the q-function
U
,f
:= z
_
1
0
t
f(t)J
(tz; q
2
) d
q
t = z
_
J
(z; q
2
) +qJ
(qz; q
2
)
_
, (2.3.8)
are real, innite, simple and u
r
_
q
r+1/2
/
_
C
,f
, q
r1/2
/
_
C
,f
_
, r = 0, 1, 2, . . ..
In the following we introduce another q-analogue of Theorem 1.4.1.
Theorem 2.3.3. Let f L
1
q
(0, 1) be positive on 0, q
n
, n N. If
q
1
(1 q)
c
1/2,f
C
1/2,f
> 1, (2.3.9)
then the zeros of the entire function of order zero
U
f
(z) :=
_
1
0
f(t) cos(tz; q) d
q
t, z C, (2.3.10)
are real, simple and innite. Moreover U
f
(z) is an even function with no zeros in the
interval
_
0, q
1/2
/
__
C
1/2,f
(1 q)
__
, and its positive zeros lie in the intervals
_
q
r
/
_
(1 q)
_
C
1/2,f
), q
r1
/
_
(1 q)
_
C
1/2,f
)
_
, r = 0, 1, . . . , (2.3.11)
one zero in each interval.
Proof. From (1.1.16) we conclude that
U
f
(z) =
(q
2
; q
2
)
(q; q
2
)
U
1/2,f
(z(1 q)q
1/2
). (2.3.12)
49
Applying Theorems 2.2.2 to U
1/2,f
(z(1 q)q
1/2
) where q-satises (2.3.9), proves the
theorem.
Similarly, we have the following theorem.
Theorem 2.3.4. Let f L
1
q
(0, 1) be positive on 0, q
n
, n N. If
q
1
(1 q)
b
1/2,f
B
1/2,f
> 1, (2.3.13)
then the zeros of the entire function of order zero
V
f
(z) :=
_
1
0
f(t) sin(tz; q) d
q
t, z C, (2.3.14)
are real, simple and innite. Moreover V
f
(z) is an odd function with exactly one zero, z = 0,
in the interval [0,
q
1
B
1/2,f
(1q)
), and its positive zeros are located in the intervals
_
q
r+1/2
(1 q)
_
B
1/2,f
,
q
r1/2
(1 q)
_
B
1/2,f
_
, r = 0, 1, . . . , (2.3.15)
one in each interval.
A remarkable dierence between Polyas theorem and its basic analogue here is that in
Theorems 2.3.3 and 2.3.4 we do not have the monotonicity condition. However the price
for this are the restrictions (2.3.9) and (2.3.13) on q. This allowed us to prove that U
f
(z)
and V
f
(z) have no non-real zeros.
Part II
Basic Dierence Equations
50
Chapter 3
Introduction
In this introductory chapter we briey introduce the main concepts and theorems of linear
q-dierence equations as stated in [8,9,90]. We start with the existence and uniqueness the-
orem of q-Cauchy problems based on q-analogue of the celebrated successive approximation
method. Then we state some results concerning q-linear dierence equations.
3.1 q-successive approximations
In this section we dene the q-successive approximations associated with q-Cauchy problems.
This technique is the main method we use to prove the theorems of existence as in [9, 90].
Proofs will not be given here. The interested reader could nd them in [9, 90].
Denition 3.1.1. Let r, s, and n
i
, i = 0, 1, . . . , r, be positive integers and let
N = (n
0
+ 1) +. . . + (n
r
+ 1) 1.
Let F
j
(x, y
0
, y
1
, . . . , y
N
) , j = 0, 1, . . . , s, be real or complexvalued functions, where x is a
real variable lying in some interval I and each y
i
is a complex variable lying in some region
D
i
of the complex plane. If there are a subinterval J of I and functions
i
, 0 i r,
dened in J such that
51
52
1.
i
has n
i
q-derivatives in J for 0 i r ;
2. D
m
q
i
exists and lies in the region D
i
for all x in J, 0 m n
i
, and 0 i r, for
which the lefthand side in equation (3.1.1) below is dened;
3. for all x in J and 0 j s, the following equations hold
F
j
_
x,
0
(x), D
q
0
(x), . . . , D
n
0
q
0
(x), . . . ,
r
(x), . . . , D
n
r
q
r
(x)
_
= 0; (3.1.1)
then we say that
i
r
i=0
is a solution of the system of the q-dierence equations
F
j
_
x, y
0
(x), D
q
y
0
(x), . . . , D
n
0
q
y
0
(x), . . . , y
r
(x), . . . , D
n
r
q
y
r
(x)
_
= 0, (3.1.2)
0 j s, valid in J, or that the set
i
r
i=0
satises (3.1.2) in J. If there are no such J
and functions
i
, we say that the system (3.1.2) has no solutions.
System (3.1.2) is said to be of order n, where n := max
0ip
n
i
. If the functions F
j
are
such that the equations (3.1.2) can be solved for the D
n
i
q
y
i
in the form
D
n
i
q
y
i
(x) = f
i
(x, y
0
(x), D
q
y
0
(x), . . . , y
1
(x) . . .) , 0 i p, (3.1.3)
or in the form
D
n
i
q
y
i
(x) = f
i
(qx, y
0
(qx), D
q,qx
y
0
(qx), . . . , y
1
(qx) . . .) , 0 i p, (3.1.4)
then the systems (3.1.3) and (3.1.4) will be called normal systems. The following systems
are examples of normal rst order systems,
D
q
y
i
(x) = f
i
_
qx, y
0
(qx), y
1
(qx), . . . , y
p
(qx)
_
, i = 0, 1, . . . , p, (3.1.5)
and
D
q
y
i
(x) = f
i
_
x, y
0
(x), y
1
(x), . . . , y
p
(x)
_
, i = 0, 1, . . . , p. (3.1.6)
In a more restrictive manner, rst order normal linear systems have been dened in [34].
The existence of a solution of system (3.1.5) or (3.1.6) in a neighborhood of a zero is
53
established in [9, 90] by using a q-analogue of the Picard-Lindelof method of successive
approximations , see e.g. [37, 41]. The following is a description of this q-analogue.
Dene sequences of functions
i,m
m=1
, i = 0, 1, . . . , p, by the equations
i,m
(x) =
_
_
b
i
, m = 1
b
i
+
_
x
0
f
i
(t,
0,m
(t), . . . ,
p,m
(t)) d
q
t, m 2,
(3.1.7)
where b
i
are constants which lie in D
i
. It is worth mentioning here that the constants
b
i
p
i=0
should be replaced by q-periodic functions if we are looking for solutions which are
not necessarily continuous at zero.
As is stated in Theorem 3.2.2 below under suitable conditions on the f
i
s, there are
functions
i
and an interval J contained in I and containing zero such that
i,m
(x)
i
(x)
on J as m , where by we mean the uniform convergence. Then, it will be possible
to apply Lemma 3.2.1 below to obtain
i
(x) = b
i
+
_
x
0
f
i
(t,
0
(t), . . . ,
p
(t)) d
q
t. (3.1.8)
Hence
i
(0) = b
i
, and
i
p
i=0
is a solution of (3.1.6). Similarly for (3.1.5).
3.2 q-Initial value problems
Here we state an existence and uniqueness theorem of q-Cauchy problems as treated in
[9, 90].
Denition 3.2.1. Let b
i
D
i
be arbitrary values. By a q-initial value problem in a
neighborhood of zero we mean the problem of nding functions y
i
p
i=0
, continuous at zero
and satisfying the system (3.1.5) or (3.1.6) and the initial conditions
y
i
(0) = b
i
, b
i
D
i
, 0 i p. (3.2.1)
The proof of the following lemma is straightforward and will be omitted.
54
Lemma 3.2.1. Let I and J be intervals containing zero, such that J I. Let f
n
, f be
functions dened in I, n N, such that
lim
n
f
n
(t) = f(t), when t I, and f
n
f on J. (3.2.2)
Then,
lim
n
_
x
0
f
n
(t) d
q
t =
_
x
0
f(t) d
q
t, for all x I. (3.2.3)
Theorem 3.2.2. Let I be an interval containing zero and let E
r
be disks of the form
E
r
:= y C : [y b
r
[ < , > 0, b
r
C,
and r = 0, 1, . . . , p. Let f
i
(x, y
0
, y
1
, . . . , y
p
), i = 0, 1, . . . , p, be functions dened on I E
0
. . . E
p
for which:
(i) for y
r
E
r
, 0 r p, f
i
(x, y
0
, y
1
, . . . , y
p
) is continuous at x = 0, 0 i p,
(ii) there is a positive constant A such that, for x I and y
r
, y
r
E
r
, 0 r, i p, the
following Lipschiz condition is fullled
f
i
(x, y
0
, . . . , y
p
) f
i
(x, y
0
, . . . , y
p
)
A
_
[ y
0
y
0
[ +. . . +[ y
p
y
p
[
_
. (3.2.4)
Then, if 0 is not an end point of I, there exists a positive h such that the q-Cauchy
problem (3.1.6), (3.2.1) has a unique solution valid for [x[ h. Moreover if 0 is the left or
right end point of I, the result holds, except that the interval [h, h] is replaced by [0, h] or
[h, 0], respectively
Theorem 3.2.3 (Range of validity). Assume that all conditions of Theorem 3.2.2 are
satised with E
r
= C for all r; r = 0, 1, . . . , p. Then problem (3.1.6), (3.2.1) has a unique
solution valid at least in I
_
1
A(p+1)(1q)
,
1
A(p+1)(1q)
_
.
Remark 3.2.1. Theorem 3.2.2 holds for the q-Cauchy problem (3.1.5), (3.2.1), but the so-
lution will be valid throughout the whole interval I if f
i
s satisfy the conditions (i), (ii) of
Theorem 3.2.2 with E
r
= C, 0 r, i p.
55
Example 3.2.1. It is known, [53], that both of the functions e
q
(x) and E
q
(x) are q-analogues
of the exponential function. While E
q
(x) is entire, e
q
(x) is analytic only when [x[ < 1. They
are dened and related by
e
q
(x) =
1
E
q
(x)
=:
1
k=0
_
1 xq
k
_, x C
_
q
n
, n N
_
.
Theorem 3.2.2 and Remark 3.2.1 gave us an explanation since e
q
(x) and E
q
(x) are the
solutions of the q-Cauchy problems
D
q
y(x) =
1
1 q
y(x), y(0) = 1; D
q
y(x) =
1
1 q
y(qx), y(0) = 1, (3.2.5)
respectively. From Theorem 3.2.3 it follows that the solution of the rst problem is valid at
most in [x[ < 1 and from Remark 3.2.1 the solution of the second one is valid in R.
3.3 Linear q-dierence equations
This section contains a brief study of linear q-dierence equations of order n, n Z
+
.
Complete treatments could be found in [8, 9, 90].
For n Z
+
, b
i
C, Theorem 3.2.2 can be applied to discuss the existence and uniqueness
of the n-th order q-initial value problem
_
_
D
n
q
y(x) = f
_
x, y(x), D
q
y(x), . . . , D
n1
q
y(x)
_
D
i1
q
y(0) = b
i
, b
i
C; 1 i n .
(3.3.1)
Corollary 3.3.1. Let p, I, E
r
, b
r
be as in the previous theorem. Let f(x, y
0
, y
1
, . . . , y
p
) be
a function dened on I E
0
. . . E
p
such that the following conditions are satised
(i) For any xed values of the y
r
in E
r
, f(x, y
0
, y
1
, . . . , y
p
) is continuous at zero.
(ii) f satises Lipschitz condition with respect to y
0
, . . . , y
p
.
56
If 0 is an interior point of I, then there exists a positive h such that q-Cauchy problem
(3.3.1) has a unique solution valid for [x[ h. If 0 is a left or right endpoint of I the
result holds, except that the interval [x[ h is replaced by [0, h] or [h, 0] respectively.
Proof. Assume that 0 is an interior point of I. The q-Cauchy problem (3.3.1) is equivalent
to the rst order q-initial value problem
D
q
y
i
(x) = f
i
(x, y
0
, . . . , y
n1
) y
i
(0) = b
i
, ; 1 i n,
(3.3.2)
in the sense that
i
n1
i=0
is a solution of (3.3.2) if and only if
0
is a solution of (3.3.1).
Here f
i
are the functions
f
i
(x, y
0
, . . . , y
n1
) =
_
_
y
i+1
, 0 i n 2
f(x, y
0
, . . . , y
n1
), i = n.
(3.3.3)
By Theorem (3.2.2) there exists h > 0 such that system (3.3.2) has a unique solution valid
in [x[ h.
Corollary 3.3.2. Consider the q-Cauchy problem
_
_
a
0
(x)D
n
q
y(x) +a
1
(x)D
n1
q
y(x) +. . . +a
n
(x)y(x) = b(x)
D
i
q
(0) = b
i
, 0 i n 1 .
(3.3.4)
Let the a
j
(x), 0 j n, and b(x) be dened on an interval I containing zero such that
a
0
(x) ,= 0 for all x in I and the functions a
j
(x), b(x) are continuous at zero and bounded
on I. Then for any complex numbers b
r
, there exists a subinterval J of I containing zero
such that (3.3.4) has a unique solution valid in J.
Proof. Dividing by a
0
(x), we get the equation
D
n
q
y(x) = A
n
(x)D
n1
q
y(x) +. . . +A
1
(x)y(x) +B(x), (3.3.5)
57
where A
j
(x) = a
j
(x)/a
0
(x), 1 j n, and B(x) = b(x)/a
0
(x). Equation (3.3.5) is of the
form (3.3.2) with
f(x, y
0
, . . . , y
n1
) = A
n
(x)y
0
+. . . +A
1
(x)y
n1
+B(x).
Since A
j
(x) and B(x) are continuous at zero and bounded on I, then f(x, y
0
, . . . , y
p
) satises
the conditions of Theorem 3.3.1. Hence, there exists a subinterval J of I containing zero
such that equation (3.3.5) has a unique solution valid in J.
Consider the non-homogeneous linear q-dierence equation of order n
a
0
(x)D
n
q
y(x) +a
1
(x)D
n1
q
y(x) +. . . +a
n
(x)y(x) = b(x), x I, (3.3.6)
for which a
i
, 0 i n, and b are continuous at zero functions dened on I and a
0
(x) ,= 0
for all x I . From Corollary 3.3.2 equation (3.3.6) together with the initial conditions
D
i1
q
y(0) = b
i
, b
i
C, i = 1, . . . , n, (3.3.7)
has a unique solution which is continuous at zero in a subinterval J of I, J = [h, h], h > 0.
The corresponding n-th order homogeneous linear equation of (3.3.6) is
a
0
(x)D
n
q
y(x) +a
1
(x)D
n1
q
y(x) +. . . +a
n
(x)y(x) = 0, x I. (3.3.8)
Let M denote the set of solutions of (3.3.8) valid in a subset J I, 0 J. Then,cf. [8],
M is a linear space over C of dimension n. Also from the existence and uniqueness of the
solutions, if M and D
i
q
(0) = 0, 0 i n 1, then (x) 0 on J. Moreover,
_
D
i
q
_
n1
i=0
are continuous at zero for any M. A set of n solutions of (3.3.8) is said to
be a fundamental set (f.s.) for (3.3.8) valid in J or a f.s. of M if it is linearly independent
in J. Moreover, as in dierential equations, if b
ij
, 1 i, j n, are numbers, and, for each
j,
j
is the unique solution of (3.3.8) which satises the initial conditions
D
i1
q
j
(0) = b
ij
, 1 i n,
then
j
n
j=1
is a f.s. of (3.3.8) if and only if det(b
ij
) ,= 0.
58
3.4 A q-type Wronskian
In the following we give a q-analogue of the Wronskian for n-th order qdierence equations.
A study for a second order Wronskian is established by R.F. Swarttouw and H.G. Meijer
in [113]. The results of the present section are also included in [8, 90].
Denition 3.4.1. Let y
i
, 1 i n, be functions dened on a q-geometric set A. The
q-Wronskian of the functions y
i
which will be denoted by W
q
(y
1
, . . . , y
n
)(x) is dened to be
W
q
(y
1
, . . . , y
n
)(x) :=
y
1
(x) . . . y
n
(x)
D
q
y
1
(x) . . . D
q
y
n
(x)
.
.
.
.
.
.
.
.
.
D
n1
q
y
1
(x) . . . D
n1
q
y
n
(x)
, (3.4.1)
provided that the derivatives exist in I. For convenience we write W
q
(x) instead of
W
q
(y
1
, . . . , y
n
)(x).
Theorem 3.4.1. If y
1
, y
2
, . . . , y
n
are solutions of (3.3.8) in J I, then their q-Wronskian
satises the rst order q-dierence equation
D
q
W
q
(x) = R(x)W
q
(x), x J 0 ; R(x) =
n1
k=0
(x qx)
k
a
k+1
(x)
a
0
(x)
. (3.4.2)
The following theorems gives a q-type Liouvilles formula for the q-Wronskian.
Theorem 3.4.2. Suppose that x(1 q)R(x) ,= 1 when x J. Then the q-Wronskian of
any set of solutions
i
n
i=1
of equation (3.3.8) is given by
W
q
(x) = W
q
(
1
, . . . ,
n
)(x) =
1
k=0
(1 +x(1 q)q
k
R(xq
k
))
W
q
(0), x J. (3.4.3)
Corollary 3.4.3. Let
i
n
i=1
be a set of solutions of (3.3.8) in some subinterval J of I
which contains zero. Then W
q
(x) is either never zero or identically zero in I. The rst
case occurs when
i
n
i=1
is a fundamental set of (3.3.8) and the second when it is not.
59
Example 3.4.1. In this example we calculate the q-Wronskian of
1
q
D
q
1D
q
y(x) +y(x) = 0, x R. (3.4.4)
The solutions of (3.4.4) subject to the initial conditions
y(0) = 0, D
q
y(0) = 1 and y(0) = 1, D
q
y(0) = 0,
are sin(x; q), cos(x; q), x R, respectively. Since (3.4.4) can be written as
D
2
q
y(x) +qx(1 q)D
q
y(x) qy(x) = 0, (3.4.5)
then a
0
(x) 1, a
1
(x) = qx(1 q) and a
2
(x) = q. Thus
R(x) 0 on R and W
q
(x) W
q
(0).
But
W
q
(0) = W
q
_
cos(; q), sin(; q)
_
(0)
=
_
cos(x; q) cos(
qx; q) +
q sin(x; q) sin(
qx; q)
_
x=0
= 1.
Then, W
q
(x) 1 for all x R.
Example 3.4.2. We calculate the q-Wronskian of solutions of the q-dierence equation
D
2
q
y(x) +y(x) = 0, x R. (3.4.6)
The functions sin
q
x(1 q), cos
q
x(1 q), [x[(1 q) < 1, are solutions of (3.4.6) subject
to the initial conditions
y(0) = 0, D
q
y(0) = 1 and y(0) = 1, D
q
y(0) = 0,
respectively. Here R(x) = x(1 q). So, x(1 q)R(x) ,= 1 for all x in R. Hence,
W
q
(x) =
W
q
(0)
n=0
(1 +q
2n
x(1 q)
2
)
, [x[(1 q) < 1.
But
W
q
(0) = W
q
_
cos
q
, sin
q
_
(0) =
_
cos
q
2
x + sin
q
2
x
_
x=0
= 1. (3.4.7)
Therefore, W
q
(x) 1/
n=0
(1 +q
2n
x(1 q)
2
), [x[(1 q) < 1.
Chapter 4
Basic Sturm-Liouville Problems
This chapter is devoted to a study a q-analogue of the Sturm-Liouville eigenvalue prob-
lem. We formulate a formally self adjoint q-dierence operator in a Hilbert space. Some of
the properties of the eigenvalues and the eigenfunctions are discussed. The Greens func-
tion 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 will
be given. A study for the asymptotics of the eigenvalues and eigenfunctions will be given
at the end of this Chapter. The results of this Chapter also exists in [16, 19].
4.1 Introduction
Let [a, b] be a nite closed interval in R and let be a continuous real-valued function
dened on [a, b]. By a Sturm-Liouville problem we mean the problem of nding a function
y and a number in C satisfying the dierential equation
Ly := y
(a) = 0, (4.1.2)
60
61
U
2
(y) := b
1
y(b) +b
2
y
(b) = 0, (4.1.3)
where a
i
and b
i
, i = 1, 2 are real numbers for which
[a
1
[ +[a
2
[ , = 0 ,= [b
1
[ +[b
2
[. (4.1.4)
This problem has been extensively studied. It is known that the dierential equation (4.1.1)
and the boundary conditions (4.1.2)(4.1.3) determine a self adjoint operator in L
2
(a, b).
There is a sequence of real numbers
n
n=0
with lim
n
n
= +, such that corre-
sponding to each
n
there is one and only one linearly independent solution of the problem
(4.1.1)(4.1.3). The sequence
n
n=0
is the sequence of eigenvalues and the sequence of
corresponding solutions
n
n=0
is said to be a sequence of eigenfunctions. One of the
most important properties of these eigenfunctions is that,
n
n=0
is an orthogonal basis
of L
2
(a, b). For example, let (x) 0 on [a, b]. If we take a = 0, b = , a
1
= 1, a
2
= 0,
b
1
= 1, and b
2
= 0, we get
n
= n
2
,
n
(x) = sin nx, n = 1, 2, . . . , (4.1.5)
leading to the well known fact that sin nx
n=1
is a complete orthogonal set of L
2
(0, ),
while taking a = 0, b = , a
1
= 0, a
2
= 1, b
1
= 0, and b
2
= 1, we get
n
= n
2
,
n
(x) = cos nx, n = 0, 1, 2, . . . , (4.1.6)
which leads to the completeness of cos nx
n=0
in L
2
(0, ). We mention here that the
Fourier orthogonal basis
_
e
inx
_
n=0
of L
2
(, ) does not arise in this setting but arises
from a simpler situation, namely the rst order problem
iy
1
q
D
q
1D
q
y(x) +(x)y(x) = y(x), 0 x a < , C, (4.2.1)
where is dened on [0, a] and continuous at zero. Let C
2
q
(0, a) be the space of all functions
y dened on [0, a] such that y, D
q
y are continuous at zero. Clearly, C
2
q
(0, a) is a subspace
of the Hilbert space L
2
q
(0, a).
Theorem 4.2.1. For c
1
, c
2
C, equation (4.2.1) has a unique solution in C
2
q
(0, a) which
64
satises
(0, ) = c
1
, D
q
1(0, ) = c
2
, C. (4.2.2)
Moreover (x, ) is entire in for all x [0, a].
Proof. The functions
1
(x, ) = cos(sx; q), and
2
(x, ) =
_
_
sin(sx; q)
s
, ,= 0
x, = 0
, (4.2.3)
where s :=
1
(, ),
2
(, )
_
1 on [0, a], C. For all x [0, a], C, we dene a
sequence y
m
(, )
m=1
of successive approximations by
y
1
(x, ) = c
1
1
(x, ) +c
2
2
(x, ), (4.2.5)
y
m+1
(x, ) = c
1
1
(x, ) +c
2
2
(x, )
+q
_
x
0
2
(x, )
1
(qt, )
1
(x, )
2
(qt, ) (qt) y
m
(qt, ) d
q
t. (4.2.6)
We shall prove that for each xed C the uniform limit of y
m
(, )
m=1
as m
exists and denes a solution of (4.2.1) and (4.2.2). Let C be xed. There exist positive
numbers K() and A such that
[(x)[ A, [
i
(x, )[
_
K()
2
; i = 1, 2; x [0, a]. (4.2.7)
Let
K() :=
_
[c
1
[ + [c
2
[
_
_
K()
2
. Then, from (4.2.7), [y
1
(x, )[
K(), for all x [0, a].
Using mathematical induction, we have
[y
m+1
(x, ) y
m
(x, )[
K()q
m(m+1)
2
(AK()x(1 q))
m
(q; q)
m
, m = 1, 2, . . . . (4.2.8)
Consequently by Weierstrass test the series
y
1
(x, ) +
m=1
y
m+1
(x, ) y
m
(x, ) (4.2.9)
65
converges uniformly in [0, a]. Since the m-th partial sums of the series is nothing but
y
m+1
(, ), then y
m+1
(, ) approaches a function (, ) uniformly in [0, a] as m ,
where (x, ) is the sum of the series. We can also prove by induction on m that y
m
(x, )
and D
q
y
m
(x, ) are continuous at zero, where
D
q
y
m+1
(x, ) = c
1
D
q
1
(x, ) +c
2
D
q
2
(x, )
+q
_
x
0
D
q
2
(x, )
1
(qt, ) D
q
1
(x, )
2
(qt, ) (qt)y
m
(qt, ) d
q
t,
(4.2.10)
m Z
+
. Hence, both (, ) and D
q
(, ) are continuous at zero. i.e. (, ) C
2
q
(0, a).
Because of the uniform convergence, letting m in (4.2.6) we obtain
(x, ) = c
1
1
(x, ) +c
2
2
(x, )
+q
_
x
0
2
(x, )
1
(qt, )
1
(x, )
2
(qt, ) (qt) (qt, ) d
q
t.(4.2.11)
We prove that (, ) satises (4.2.1) and (4.2.2). Clearly (0, ) = c
1
and
D
q
1(0, ) = lim
n
(xq
n
, ) (0, )
xq
n
= c
1
D
q
1
1
(0, ) +c
2
D
q
1
2
(0, ) = c
2
, (4.2.12)
i.e. (, ) satises (4.2.2). To prove that (, ) satises (4.2.1), We distinguish between
two cases, x ,= 0 and x = 0. If x ,= 0, then from (1.1.28),
D
q
(x, ) = c
1
D
q
1
(x, ) +c
2
D
q
2
(x, )
+q
_
x
0
D
q
2
(x, )
1
(qt, ) D
q
1
(x, )
2
(qt, ) (qt)(qt, ) d
q
t
(4.2.13)
and hence
1
q
D
q
1D
q
(x, ) =
1
q
D
q
1D
q
1
(x, )
_
c
1
_
x
0
2
(qt, )(qt)(qt, ) d
q
t
_
1
q
D
q
1D
q
2
(x, )
_
c
2
+
_
x
0
1
(qt, )(qt)(qt, ) d
q
t
_
(x)(x, ).
(4.2.14)
Substituting from (4.2.3), (4.2.4) and (4.2.11) in (4.2.14), we conclude that (x, ) satises
(4.2.1) for x ,= 0. If x = 0, then (4.2.1) is nothing but
D
2
q
y(0) q(0)y(0) = qy(0). (4.2.15)
66
From (1.1.29), one can see that
D
2
q
(0, ) = c
1
D
2
q
1
(0, ) +c
2
D
2
q
2
(0, ) +q(0)(0, )
= c
1
q
1
(0, ) c
2
q
2
(0, ) +q(0)(0, )
= q(0, ) +q(0)(0, ). (4.2.16)
Consequently (4.2.1) is satised at x = 0.
To prove that problem (4.2.1), (4.2.2) has a unique solution, assume the contrary, that
is
i
(, ), i = 1, 2, are two solutions of (4.2.1), (4.2.2). Let
(x, ) =
1
(x, )
2
(x, ), x [0, a].
Then (, ) is a solution of (4.2.1) subject to the initial conditions
(0, ) = D
q
1(0, ) = 0.
Applying the q-integration process on (4.2.1) twice yields
(x, ) =
_
x
0
(x t)
_
(t)
_
(t, ) d
q
t. (4.2.17)
Since (x, ), (x) are continuous at zero, then there exist positive numbers N
x,
, M
x,
such that
N
x,
= sup
nN
[(xq
n
, )[, M
x,
= sup
nN
(xq
n
)
. (4.2.18)
Again we can prove by mathematical induction on k that
[(x, )[ N
x,
M
k
x,
q
k
2
(1 q)
2k
x
2k
(q; q)
2k
, k N, x [0, a]. (4.2.19)
Indeed, if (4.2.19) holds at k N, then from (4.2.17)
[(x, )[ N
x,
M
k+1
x,
q
k
2 (1 q)
2k
(q; q)
2k
_
x
0
(x t)t
2k
d
q
t
= N
x,
M
k+1
x,
q
k
2
q
2k+1
(1 q)
2k+2
(q; q)
2k+2
x
2k+2
= N
x,
M
k+1
x,
q
(k+1)
2 (1 q)
2k+2
(q; q)
2k+2
x
2k+2
. (4.2.20)
67
Hence (4.2.19) holds at k+1 and therefore (4.2.19) holds for all k N because from (4.2.18)
it holds at k = 0. Since lim
k
M
k
x,
q
k
2
(1q)
2k
x
2k
(q; q)
2k
= 0, then (x, ) = 0, for all x [0, a].
This proves the uniqueness. Now, let M > 0 be arbitrary but xed. To prove that (x, ),
x [0, a], is entire in , it is sucient to prove that (x, ) is analytic in each disk
M
;
M
:= C : [[ M. We prove by induction on m that
for all x [0, a] y
m
(x, ) is analytic on
M
, (4.2.21)
for all
M
y
m
(x, ) is continuous at (0, ). (4.2.22)
Clearly,
1
(x, ),
2
(x, ) are entire functions of for x [0, a]. Moreover
i
(x, ) is
continuous at (0, ) for each C. Then (4.2.21) and (4.2.22 ) holds at m = 1. Assume
that (4.2.21) and (4.2.22) hold at m 1. Then for x
0
[0, a],
0
M
, we obtain
y
m+1
(x
0
, )
=
0
= q
2
(x
0
, )
=
0
_
x
0
0
1
(qt, )y
m
(qt, ) d
q
t
+
y
1
(x
0
, )
=
0
q
1
(x
0
, )
=
0
_
x
0
0
2
(qt, )y
m
(qt, ) d
q
t
+q
2
(x
0
, )
_
_
x
0
0
1
(qt, )y
m
(qt, ) d
q
t
_
=
0
q
1
(x
0
, )
_
_
x
0
0
2
(qt, )y
m
(qt, ) d
q
t
_
=
0
(4.2.23)
From (4.2.22) we conclude that /
_
i
(qt, )y
m
(qt, )
_
, i = 1, 2, are continuous at (0,
0
).
Therefore there exist constants C, > 0 such that
i
(x
0
q
n
, )y
m
(x
0
q
n
, )
_
C, n N, [
0
[ . (4.2.24)
Hence
x
0
(1 q)q
n
i
(x
0
q
n+1
, )y
m
(x
0
q
n+1
)
_
x
0
A(1 q)q
n
, n N,
for all in the disk [
0
[ , i.e. the series corresponding to the q-integrals
_
x
0
0
i
(qt, )y
m
(qt, )
_
d
q
t, i = 1, 2 (4.2.25)
68
are uniformly convergent in a neighborhood of =
0
. Thus, we can interchange the
dierentiation and integration processes in (4.2.23). Since x
0
,
0
are arbitrary, then
y
m+1
(x, ) =
y
1
(x, ) q
_
x
0
2
(x, )
1
(qt, )y
m
(qt, )
_
(qt) d
q
t
+q
_
x
0
1
(x, )
2
(qt, )y
m
(qt, )
_
(qt) d
q
t,
(4.2.26)
for all x [0, a],
M
. From (4.2.22) the integrals in (4.2.26) are continuous at (0, ).
Consequently
y
m+1
(x, ) is continuous at (0, ). Let x
0
[0, a] be arbitrary. Then there
exists B(x
0
),
B(x
0
) > 0 such that
[
i
(x
0
, )[
_
B(x
0
)
2
, i = 1, 2, [y
1
(x, )[
B(x
0
),
M
. (4.2.27)
Finally the use of the mathematical induction yields
[y
m+1
(x
0
, ) y
m
(x
0
, )[
B(x
0
)q
m(m+1)
2
(AB(x
0
)(1 q))
m
(q; q)
m
. (4.2.28)
Consequently the series (4.2.9), with x = x
0
, converges uniformly in
M
to (x
0
, ). Hence
(x
0
, ) is analytic in
M
, i.e. it is entire.
4.3 The self adjoint problem
In this section we dene a basic Sturm-Liouville problem and prove that it is formally self
adjoint in L
2
q
(0, a). The following lemma which is needed in the sequel indicates that unlike
the classical dierential operator d/dx, D
q
is neither self adjoint nor skew self adjoint.
Equation (4.3.2) below indicates that the adjoint of D
q
is
1
q
D
q
1.
Lemma 4.3.1. Let f, g in L
2
q
(0, a) be dened on [0, q
1
a]. Then, for x (0, a], we have
(D
q
g)(xq
1
) = D
q,xq
1g(xq
1
) = D
q
1g(x), (4.3.1)
D
q
f, g) = f(a)g(aq
1
) lim
n
f(aq
n
)g(aq
n1
) +
f,
1
q
D
q
1g
_
, (4.3.2)
1
q
D
q
1f, g
_
= lim
n
f(aq
n1
)g(aq
n
) f(aq
1
)g(a) +f, D
q
g) . (4.3.3)
69
Proof. Relation (4.3.1) follows from
D
q
1g(x) =
g(x) g(q
1
x)
x(1 q
1
)
=
g(xq
1
) g(x)
xq
1
(1 q)
= (D
q
g)(xq
1
) = D
q,xq
1g(xq
1
). (4.3.4)
Using the formula (1.1.41) of q-integration by parts we obtain
D
q
f, g) =
_
a
0
D
q
f(x)g(x) d
q
x
= f(a)g(a) lim
n
f(aq
n
)g(aq
n
)
_
a
0
f(qt)D
q
g(t) d
q
t
= f(a)g(a) lim
n
f(aq
n
)g(aq
n
)
_
qa
0
f(t)
1
q
D
q
1g(t) d
q
t
= f(a)g(a) lim
n
f(aq
n
)g(aq
n
) +aq
1
(1 q)f(a)D
q
1g(a)
+
_
a
0
f(t)
1
q
D
q
1g(t) d
q
t
= f(a)g(aq
1
) lim
n
f(aq
n
)g(aq
n1
) +
f,
1
q
D
q
g
_
, (4.3.5)
proving (4.3.2). Equation (4.3.3) can be proved by a use of (4.3.2).
Now consider the basic Sturm-Liouville problem
(y) :=
1
q
D
q
1D
q
y(x) +(x)y(x) = y(x), 0 x a < , C, (4.3.6)
U
1
(y) := a
11
y(0) +a
12
D
q
1y(0) = 0, (4.3.7)
U
2
(y) := a
21
y(a) +a
22
D
q
1y(a) = 0, (4.3.8)
where is a continuous at zero real valued function and a
ij
, i, j 1, 2 are arbitrary real
numbers such that the rank of the matrix (a
ij
)
1i,j2
is 2. Problem (4.3.6)(4.3.8) is said to
be formally self adjoint if for any functions y and z of C
2
q
(0, a) which satisfy (4.3.7)(4.3.8),
y, z) = y, z) . (4.3.9)
Theorem 4.3.2. The basic Sturm- Liouville eigenvalue problem (4.3.6)(4.3.8) is formally
self adjoint.
70
Proof. We rst prove that for y, z in L
2
q
(0, a), we have the following q-Lagranges identity
_
a
0
_
y(x)z(x) y(x)z(x)
_
d
q
x = [y, z](a) lim
n
[y, z](aq
n
), (4.3.10)
where
[y, z](x) := y(x)D
q
1z(x) D
q
1y(x)z(x). (4.3.11)
Applying (4.3.3) with f(x) = D
q
y(x) and g(x) = z(x), we obtain
1
q
D
q
1D
q
y(x), z(x)
_
= (D
q
y)(aq
1
)z(a) + lim
n
(D
q
y)(aq
n1
)z(aq
n
) +
D
q
y, D
q
z
_
= D
q
1y(a)z(a) + lim
n
D
q
1y(aq
n
)z(aq
n
) +
D
q
y, D
q
z
_
.
(4.3.12)
Applying (4.3.2) with f(x) = y(x), g(x) = D
q
z(x),
D
q
y, D
q
z
_
= y(a)D
q
z(aq
1
) lim
n
y(aq
n
)D
q
z(aq
n1
) +
y,
1
q
D
q
1D
q
z
_
= y(a)D
q
1z(a) lim
n
y(aq
n
)D
q
1z(aq
n
) +
y,
1
q
D
q
1D
q
z
_
. (4.3.13)
Therefore,
1
q
D
q
1D
q
y(x), z(x)
_
= [y, z](a) lim
n
[y, z](aq
n
) +
y,
1
q
D
q
1D
q
z
_
. (4.3.14)
Lagranges identity (4.3.10) results from (4.3.14) and the reality of (x). Letting y, z in
C
2
q
(0, a) and assuming that they satisfy (4.3.7)(4.3.8), we obtain
a
11
y(0) +a
12
D
q
1y(0) = 0, a
11
z(0) +a
12
D
q
1z(0) = 0. (4.3.15)
The continuity of y, z at zero implies that lim
n
[y, z](aq
n
) = [y, z](0). Then (4.3.14) will
be
1
q
D
q
1D
q
y(x), z(x)
_
= [y, z](a) [y, z](0) +
y,
1
q
D
q
1D
q
z
_
. (4.3.16)
Since a
11
and a
12
are not both zero, it follows from (4.3.15) that
[y, z](0) = y(0)D
q
1z(0) D
q
1y(0)z(0) = 0.
71
Similarly,
[y, z](a) = y(a)D
q
1z(a) D
q
1y(a)z(a) = 0.
Since (x) is real valued, then
(y), z) =
1
q
D
q
1D
q
y(x) +(x)y(x), z(x)
_
=
1
q
D
q
1D
q
y(x), z(x)
_
+(x)y, z(x))
=
y,
1
q
D
q
1D
q
z(x)
_
+
y, (x)z(x)
_
= y, (z)) ,
i.e. (4.3.6)(4.3.8) is a formally self adjoint operator.
A complex number
0
_
_
a
0
[y(x)[
2
d
q
x = 0. (4.3.18)
72
Since y
0
() is non-trivial then
0
=
0
, which proves i.
ii. Let , be two (real) distinct eigenvalues with corresponding eigenfunctions y(), z(),
respectively. Again, we make use of (4.3.9) to get
( )
_
a
0
y(x)z(x) d
q
x = 0. (4.3.19)
Since ,= , then y() and z() are orthogonal, proving ii.
iii. Let
0
be an eigenvalue with two eigenfunctions y
1
() and y
2
(). We prove that they
are linearly dependent by proving that their q-Wronskian vanishes at x = 0, see [8] for a
complete treatment of linear q-dierence equations . Indeed,
W
q
(y
1
, y
2
)(0) = y
1
(0)D
q
y
2
(0) y
2
(0)D
q
y
1
(0) = 0, (4.3.20)
since both y
1
and y
2
satisfy (4.3.7).
In the following we indicate how to obtain the eigenvalues and the corresponding eigen-
functions. Let
1
(, ),
2
(, ) be the linearly independent solutions of (4.3.6) determined
by the initial conditions
D
j1
q
i
(0, ) =
ij
, i, j = 1, 2, C. (4.3.21)
Thus
1
(, ) is determined by (4.2.5) by taking c
1
= 1, c
2
= 0 and
2
(, ) is determined
by taking c
1
= 0, c
2
= 1. Then, every solution of (4.3.6) is of the form
y(x, ) = A
1
1
(x, ) +A
2
2
(x, ), (4.3.22)
where A
1
and A
2
does not depend on x. A solution y(, ) of (4.3.6) will be an eigenfunction
if it satises the boundary conditions (4.3.7)(4.3.8), i.e. if we can nd a non trivial solution
of the linear system
A
1
U
1
(
1
) +A
2
U
1
(
2
) = 0,
A
1
U
2
(
1
) +A
2
U
2
(
2
) = 0,
(4.3.23)
73
Hence, in R is an eigenvalue if and only if
() =
U
1
(
1
) U
1
(
2
)
U
2
(
1
) U
2
(
2
)
= 0. (4.3.24)
The function () dened in (4.3.24) is called the characteristic determinant associated
with the basic Sturm- Liouville problem (4.3.6)(4.3.8). The zeros of () are exactly the
eigenvalues of the problem. Since
1
(x, ) and
2
(x, ) are entire in for each xed x in
[0, a], then () is also entire. Thus the eigenvalues of the basic Sturm-Liouville system
(4.3.6)(4.3.8) are at most countable with no nite limit points. From Lemma 4.3.3 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 (). Indeed, let
1
(, ) and
2
(, ) be dened by the relations
1
(x, ) := U
1
(
2
)
1
(x, ) U
1
(
1
)
2
(x, ),
2
(x, ) := U
2
(
2
)
1
(x, ) U
2
(
1
)
2
(x, ).
(4.3.25)
Hence,
1
(, ),
2
(, ) are solutions of (4.3.6) such that
1
(0, ) = a
12
, D
q
1
1
(0, ) = a
11
;
2
(a, ) = a
22
, D
q
1
2
(a, ) = a
21
. (4.3.26)
One can verify that
W
q
_
1
(, ),
2
(, )
_
(x, ) = ()W
q
_
1
(, ),
2
(, )
_
= (). (4.3.27)
Let
0
be an eigenvalue of (4.3.6)(4.3.8). Then
0
is a real number and therefore
i
(x,
0
)
can be taken to be real valued, i = 1, 2. From (4.3.27), we conclude that
1
(x,
0
),
2
(x,
0
)
are linearly dependent eigenfunctions. So, there exists a non zero constant k
0
such that
1
(x,
0
) = k
0
2
(x,
0
). (4.3.28)
From (4.3.25) and (4.3.26)
1
(a,
0
) = k
0
a
22
= k
0
1
(a, ), D
q
1
1
(a,
0
) = k
0
a
21
= k
0
D
q
1
1
(a, ). (4.3.29)
74
In the q-Lagrange identity (4.3.10), taking y(x) =
1
(x, ), and z(x) =
1
(x,
0
) implies
_
0
_
_
a
0
1
(x, )
1
(x,
0
) d
q
x =
1
(a, )D
q
1
1
(a,
0
) D
q
1
1
(a, )
1
(a,
0
)
= k
0
_
1
(a, )D
q
1
2
(a, )
2
(a, )D
q
1
1
(a, )
_
= k
0
W
q
(
1
(, ),
2
(, ))(q
1
a) = k
0
().
Since () is entire in ,
(
0
) := lim
0
()
0
=
1
k
0
_
a
0
2
1
(x,
0
) d
q
x ,= 0. (4.3.30)
Therefore
0
is a simple zero of ().
4.4 Basic Greens function
The q-type Greens function arises when we seek a solution of the nonhomogeneous equation
1
q
D
q
1D
q
y(x) + +(x) y(x) = f(x), x [0, a], C, (4.4.1)
which satises the boundary conditions (4.3.7)(4.3.8), where f L
2
q
(0, a) is given. First,
we note that if is not an eigenvalue of the Sturm-Liouville problem (4.3.6)(4.3.8), then
the solution of (4.4.1), if it exists, would be unique. To see this, assume that
1
(x, ),
2
(x, ) are two solutions of (4.4.1). Then
1
(x, )
2
(x, ) is a solution of the problem
(4.3.6)(4.3.8). So, it is identically zero if is not an eigenvalue. Another proof of this
assertion is included in the proof of the next theorem.
Theorem 4.4.1. Suppose that is not an eigenvalue of (4.3.6)(4.3.8). Let (, ) sat-
isfy the q-dierence equation (4.4.1) and the boundary conditions (4.3.7)(4.3.8), where
f L
2
q
(0, a). Then
(x, ) =
_
a
0
G(x, t, )f(t) d
q
t, x 0, aq
m
, m N , (4.4.2)
75
where G(x, t, ) is the Greens function of problem (4.3.6)(4.3.8) and it given by
G(x, t, ) =
1
()
_
2
(x, )
1
(t, ) , 0 t x,
1
(x, )
2
(t, ) , x < t a.
(4.4.3)
Conversely the function (, ) dened by (4.4.2) satises (4.4.1) and (4.3.7)(4.3.8). Greens
function G(x, t, ) is unique in the sense that if there exists another function
G(x, t, ) such
that (4.4.2) is satised, then
G(x, t, ) =
G(x, t, ), in L
2
q
_
(0, a) (0, a)
_
. (4.4.4)
If f is q-regular at zero, then (4.4.2) holds for all x [0, a].
Proof. Using a q-analogue of the methods of variation of constants, a particular solution
of the non-homogenous equation (4.4.1) may be given by
(x, ) = c
1
(x)
1
(x, ) +c
2
(x)
2
(x, ), (4.4.5)
where c
1
(x), c
2
(x) are solutions of the rst order q-dierence equations
D
q,x
c
1
(x) =
q
()
2
(qx, )f(qx), D
q,x
c
2
(x) =
q
()
1
(qx, )f(qx). (4.4.6)
Dene the q-geometric set A
f
by
A
f
:=
_
x [0, a] :
_
x
0
i
(qt, )f(qt) d
q
t exists, i = 1, 2
_
. (4.4.7)
A
f
is a q-geometric set containing 0, aq
m
, m N because f L
2
q
(0, a). From Holder
inequality we obtain
_
x
0
[
i
(qt, )f(qt)[ d
q
t
__
x
0
[
i
(qt, )[
2
d
q
t
_
1/2
__
x
0
[f(t)[
2
d
q
t
_
1/2
, x A
f
.
Since f,
i
L
2
q
(0, a), i = 1, 2 and () ,= 0, then D
q
c
i
(), i = 1, 2, are qintegrable on
[0, x] for all x A
f
. Hence direct computations lead to the following appropriate solutions
of (4.4.6)
c
1
(x) = c
1
(0) +
q
()
_
x
0
2
(qt, )f(qt) d
q
t, x A
f
(4.4.8)
76
c
2
(x) = c
2
(a) +
q
()
_
a
x
1
(qt, )f(qt) d
q
t, x A
f
. (4.4.9)
That is the general solution of (4.4.1) is given by
(x, ) = c
1
1
(x, ) +c
2
2
(x, ) +
q
()
1
(x, )
_
x
0
2
(qt, )f(qt) d
q
t
+
q
()
2
(x, )
_
a
x
1
(qt, )f(qt) d
q
t, (4.4.10)
where x A
f
, and c
1
, c
2
are arbitrary constants. Now, we determine c
1
, c
2
for which
(x, ) satises (4.3.7)(4.3.8). It easy to see that
(0, ) = c
1
1
(0, ) +
_
c
2
+
q
()
_
a
0
1
(qt, )f(qt) d
q
t
_
2
(0, ),
D
q
1(0, ) = lim
n
xA
f
(xq
n
, ) (0, )
xq
n
= c
1
D
q
1
1
(0, ) +
_
c
2
+
q
()
_
a
0
1
(qt, )f(qt) d
q
t
_
D
q
1
2
(0, ).
The boundary condition a
11
(0, ) +a
12
D
q
1(0, ) = 0 implies that
_
c
2
+
q
()
_
a
0
1
(qt, )f(qt) d
q
t
_
W
q
(
1
,
2
)(0) = 0. (4.4.11)
Therefore,
c
2
=
q
()
_
a
0
1
(qt, )f(qt) d
q
t. (4.4.12)
Hence,
(x, ) = c
1
1
(x, ) +
q
()
_
x
0
_
1
(x, )
2
(qt, )
2
(x, )
1
(qt, )
_
f(qt) d
q
t. (4.4.13)
Now we compute (a, ) and D
q
1(a, ). Indeed, from the denition of the q-integration
(1.1.31) and relation (4.4.10)
(a, ) = c
1
1
(a, ) +
q
()
_
a
0
_
1
(a, )
2
(qt, )
2
(a, )
1
(qt, )
_
f(qt) d
q
t
= c
1
1
(a, ) +
q
()
_
q
1
a
0
_
1
(a, )
2
(qt, )
2
(a, )
1
(qt, )
_
f(qt) d
q
t
77
and
D
q
1(a, ) = D
q
1
1
(a, )
_
c
1
+
q
()
_
q
1
a
0
2
(qt, )f(qt) d
q
t
_
q
()
D
q
1
2
(a, )
_
q
1
a
0
1
(qt, )f(qt) d
q
t.
The boundary condition a
21
2
(a, ) +a
22
D
q
1
2
(a, ) = 0 implies
_
c
1
+
q
()
_
q
1
a
0
2
(qt, )f(qt) d
q
t
_
W
q
(
1
,
2
)(a) = 0. (4.4.14)
Hence
c
1
=
q
()
_
q
1
a
0
2
(qt, )f(qt) d
q
. (4.4.15)
So for x A
f
(x, ) =
q
()
2
(x, )
_
x
0
1
(qt, )f(qt) d
q
t
q
()
1
(x, )
_
q
1
a
x
2
(qt, )f(qt) d
q
t
=
1
()
2
(x, )
_
qx
0
1
(t, )f(t) d
q
t
1
()
1
(x, )
_
a
qx
2
(t, )f(t) d
q
t
=
1
()
2
(x, )
_
x
0
1
(t, )f(t) d
q
t
1
()
1
(x, )
_
a
x
2
(t, )f(t) d
q
t.
proving (4.4.2)(4.4.3). Conversely, by direct computations, if (x, ) is given by (4.4.2),
then it is a solution of (4.4.1) and satises the boundary conditions (4.3.7)(4.3.8). To prove
the uniqueness, suppose that there exists another function,
G(x, t, ), such that
(x, ) =
_
a
0
G(x, t, ) f(t) d
q
t, (4.4.16)
is a solution of (4.4.1) which satises (4.3.7)(4.3.8). For convenience, let
G(x, t, ) =
_
_
G
1
(x, t, ), 0 t x,
G
2
(x, t, ), x t a.
,
G(x, t, ) =
_
G
1
(x, t, ), 0 t x,
G
2
(x, t, ), x t a.
By subtraction, we see that
_
a
0
_
G(x, t, )
G(x, t, )
_
f(t) d
q
t = 0, x 0, aq
m
, m N , (4.4.17)
78
holds for all functions f(t) L
2
q
(0, a). Let us take
f(t) := G(x, t, )
G(x, t, ), x aq
m
, m N .
Then
_
a
0
G(aq
m
, t, )
G(aq
m
, t, )
2
d
q
t
=
_
aq
m
0
G
1
(aq
m
, t, )
G
1
(aq
m
, t, )
2
d
q
t +
_
a
aq
m
G
2
(aq
m
, t, )
G
2
(aq
m
, t, )
2
d
q
t
= a(1 q)
n=0
q
n
G(aq
m
, aq
n
, )
G(aq
m
, aq
n
, )
2
= 0 (4.4.18)
Therefore, from (4.4.18) we conclude that
G(aq
m
, aq
n
, ) =
G(aq
m
, aq
n
, ), m, n N.
If f is q-regular at zero, then A
f
[0, a] and (4.4.2) will be dened for all x [0, a].
Theorem 4.4.2. Basic Greens function has the following properties
(i) G(x, t, ) is continuous at the point (0, 0).
(ii) G(x, t, ) = G(t, x, ).
(iii) For each xed t (0, qa], as a function of x, G(x, t, ) satises the q-dierence equation
(4.3.6) in the intervals [0, t), (t, a] and it also satises the boundary conditions (4.3.7)
(4.3.8).
(iv) Let
0
be a zero of (). Then
0
can be a simple pole of the function G(x, t, ), and
in this case
G(x, t, ) =
0
(x)
0
(t)
0
+
G(x, t, ), (4.4.19)
where
G(x, t, ) is analytic function of in a neighborhood of
0
and
0
is a normalized
eigenfunction corresponding to
0
.
79
Proof. (i) Follows from the continuity of
1
(, ),
2
(, ) at zero for each xed C and
(ii) is easily checked. Now, we prove (iii). Let t (0, qa] be xed. If x [0, t], then
G(x, t, ) =
1
()
1
(x, )
2
(t, ).
So,
G(x, t, ) =
1
()
2
(t, )
1
(x, ) =
()
2
(t, )
1
(x, ) = G(x, t, ).
Similarly if x [t, a]. From (4.3.26) and (4.4.3), we have
a
11
G(0, t, ) +a
12
D
q
1G(0, t, ) =
1
()
2
(t, )
_
a
11
1
(0, ) +a
12
D
q
1
1
(0, )
_
= 0,
a
21
G(a, t, ) +a
22
D
q
1G(a, t, ) =
1
()
1
(t, )
_
a
21
1
(a, ) +a
22
D
q
1
1
(a, )
_
= 0.
(iv) Let
0
be a pole of G(x, t, ), and R(x, t) be the residue of G(x, t, ) at =
0
. From
(4.3.28) and (4.3.30), we obtain
R(x, t) = lim
0
(
0
)G(x, t, ) = k
1
0
1
(x,
0
)
1
(t,
0
) lim
0
0
()
=
1
(x,
0
)
1
(t,
0
)
_
a
0
[
1
(u, )[
2
d
q
u
=
0
(x,
0
)
1
(t,
0
).
4.5 Eigenfunction expansions
In this section, the existence of a countable sequence of eigenvalues of with no nite limit
points will be proved by using the spectral theorem of compact self adjoint operators in
Hilbert spaces, see e.g. [24]. Moreover it will be proved that the corresponding eigenfunctions
form an orthonormal basis of L
2
q
(0, a). We dene the operator L : T
L
L
2
q
(0, a) to be
Ly = y for all y T
L
, where T
L
is the subspace of L
2
q
(0, a) consisting of those complex
valued functions y that satises (4.3.7)(4.3.8) such that D
q
y is q-regular at zero and D
2
q
y
lies in L
2
q
(0, a). Thus L is the dierence operator generated by the dierence expression
and the boundary conditions (4.3.7)(4.3.8). By L(y) = y, we mean that (y) = y and
80
y satises (4.3.7)(4.3.8). The operator L has the same eigenvalues of the basic Sturm-
Liouville problem (4.3.6)(4.3.8). We assume without any loss of generality that = 0 is
not an eigenvalue. Thus ker L = 0. From the previous section the solution of the problem
(Ly)(x) = f(x), f L
2
q
(0, a), (4.5.1)
is given uniquely in L
2
q
(0, a) by
y(x) =
_
a
0
G(x, t)f(t) d
q
t, (4.5.2)
where
G(x, t) = G(x, t, 0) =
_
_
c
1
(t)
2
(x), 0 t x
c
1
(x)
2
(t), x t a,
, c :=
1
W
q
(
1
,
2
)
. (4.5.3)
Replacing f by y in (4.5.1). Then the eigenvalue problem
(Ly)(x) = y(x), (4.5.4)
is equivalent to the following basic Fredholm integral equation of the second kind
y(x) =
_
a
0
G(x, t)y(t) d
q
t, in L
2
q
(0, a). (4.5.5)
Let ( be the integral operator
( : L
2
q
(0, a) L
2
q
(0, a), ((f)(x) =
_
a
0
G(x, t)f(t) d
q
t. (4.5.6)
We prove that
(L()f = f, f L
2
q
(0, a). (4.5.7)
We show rst that y = (f T
L
. From (4.5.5) and (4.5.6)
y(x) = ((f)(x) =
2
(x)y
1
(x) +
1
(x)y
2
(x), (4.5.8)
81
where
y
1
(x) = c
_
x
0
1
(t)f(t) d
q
t, y
2
(x) = c
_
a
x
2
(t)f(t) d
q
t.
Thus, for all x A
f
, cf. (4.4.7),
D
q
y(x) = D
q
2
(x)y
1
(qx) +D
q
1
(x)y
2
(qx), (4.5.9)
D
2
q
y(x) = q(qx)(y)(qx) qf(qx) L
2
q
(0, a). (4.5.10)
Since D
q
i
(x, ), y
i
(x), i = 1, 2, are q-regular at zero, then so is D
q
y and
D
q
1y(0) = D
q
y(0) = lim
n
xA
f
y(xq
n
) y(0)
xq
n
= D
q
1
2
(0)y
2
(0),
y
1
(0) = 0, and y
2
(a) = 0, then
a
11
y(0) +a
12
D
q
1y(0) =
_
a
11
1
(0) +a
12
D
q
1
1
(0)
_
y
2
(0) = 0,
and
a
21
y(a) +a
22
D
q
1y(a) =
_
a
21
2
(a) +a
22
D
q
1
2
(a)
_
y
1
(a) = 0.
Thus y T
L
. It follows from (4.5.10) that Ly = (L()(f) = f. Hence we have established
(4.5.7). Also, we can see that
((L)(y) = y, y T
L
. (4.5.11)
Indeed, replacing f in (4.5.7) by Ly, we get Ly = L(Ly. Thus y = (Ly since L is assumed
to be injective. It follows from (4.5.7) and (4.5.11) that ker ( = 0 and is an eigenfunction
of ( with eigenvalue if and only if is an eigenfunction of L with eigenvalue 1/.
Theorem 4.5.1. The operator ( is compact and self adjoint.
Proof. Let f, h, L
2
q
(0, a). Since G(x, t) is a real valued function dened on [0, a] [0, a]
and G(x, t) = G(t, x), then for f, h L
2
q
(0, a),
((f), h) =
_
a
0
((f)(x)h(x) d
q
x =
_
a
0
_
a
0
G(x, t)f(t)h(x) d
q
t d
q
x
=
_
a
0
f(t)
_
_
a
0
G(t, x)h(x) d
q
x
_
d
q
t = f, ((h)) ,
82
i.e. ( is self adjoint. Let
ij
(x, t) =
i
(x)
j
(t)
i,j=1
be an orthonormal basis of L
2
q
_
(0, a) (0, a)
_
. Consequently G =
i,j=1
G,
ij
_
ij
. For
n Z
+
set G
n
=
n
i,j=1
G,
ij
_
ij
, and let (
n
be the nite rank integral operator dened
on L
2
q
(0, a) by
(
n
(f)(x) :=
_
a
0
G
n
(x, t)f(t) d
q
t, in L
2
q
(0, a). (4.5.12)
Obviously (
n
is compact for all n N. From the Cauchy-Schwarz inequality
|(( (
n
)(f)| =
__
a
0
[(( (
n
)(f)(x)[
2
d
q
t
_
1/2
=
__
a
0
_
a
0
(GG
n
)(x, t)(f)(t) d
q
t
2
d
q
x
_
1/2
__
a
0
_
a
0
[(GG
n
)(x, t)[
2
d
q
t d
q
x
_
1/2
__
a
0
[f(x)[
2
d
q
x
_
1/2
= |GG
n
|
2
|f| ,
then
|( (
n
| |GG
n
|
2
0 as n .
This completes the proof.
Corollary 4.5.2. The eigenvalues of the operator L form an innite sequence
k
k=1
of
real numbers which can be ordered so that
[
1
[ [
2
[ . . . [
n
[ . . . as n .
The set of all normalized eigenfunctions of L form an orthonormal basis for L
2
q
(0, a).
Proof. Since ( is a compact self adjoint operator on L
2
q
(0, a), then ( has an innite
sequence of non zero real eigenvalues
n
n=1
, R,
n
0 as n . Let
n
n=1
denote an orthonormal set of eigenfunctions corresponding to
n
n=1
. From the spectral
83
theorem of compact self adjoint operators, we have,
((f) =
n=0
f,
n
_
n
. (4.5.13)
Since the eigenvalues
n
n=1
of the operator L are the reciprocal of those of (, then
[
n
[ =
1
[
n
[
as n . (4.5.14)
Let y T
L
. Then, y = G(f), for some f L
2
q
(0, a). Consequently,
y =
n=0
n
f,
n
)
n
=
n=0
n
ly,
n
)
n
=
n=0
n
y,
n
)
n
=
n=0
y,
n
)
n
.
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 4.5.1. Consider the q-Sturm-Liouville boundary value problem
1
q
D
q
1D
q
y(x) = y(x), (4.5.15)
with the q-Dirichlet conditions
U
1
(y) = y(0) = 0, U
2
(y) = y(1) = 0. (4.5.16)
A fundamental set of solutions of (4.5.15) is
1
(x, ) = cos(
x; q),
2
(x, ) =
sin(
x; q)
. (4.5.17)
Now, the eigenvalues of problem (4.5.15) are the zeros of the determinant
() =
U
1
(
1
) U
2
(
1
)
U
1
(
2
) U
2
(
2
)
=
2
(1, ) =
sin(
; q)
. (4.5.18)
84
Hence, the eigenvalues
m
m=1
are the zeros of sin(
m
=
q
2m
(1 q)
2
_
1 +O(q
m
)
_
, m 1, (4.5.19)
for suciently large m and the corresponding set of eigenfunctions
_
sin(
m
x;q)
m
_
m=1
is an
orthogonal basis of L
2
q
(0, 1). In the previous notations
1
(x, ) =
sin(
x; q)
, (4.5.20)
and
2
(x, ) =
sin(
; q)
cos(
x; q) + cos(
; q)
sin(
x; q)
. (4.5.21)
So, if is not an eigenvalue, Greens function is given by
G(x, t, ) =
sin(
t; q)
sin(
; q)
_
cos(
x; q)
sin(
; q)
cos(
; q)
sin(
x; q)
_
, (4.5.22)
for 0 t x, and
G(x, t, ) =
sin(
x; q)
sin(
; q)
_
cos(
t; q)
sin(
; q)
cos(
; q)
sin(
t; q)
_
, (4.5.23)
for x t 1. Since = 0 is not an eigenvalue, then Greens function G(x, t) is nothing
but
G(x, t) = G(x, t, 0) =
_
_
t(1 x), 0 t x,
x(1 t), x t 1.
Hence the boundary value problem (4.5.15)(4.5.16) is equivalent to the basic Fredholm in-
tegral equation
y(x) =
_
1
0
G(x, t)y(t) d
q
t. (4.5.24)
Example 4.5.2. Consider the equation (4.5.15) with the q-Neumann boundary conditions
U
1
(y) = D
q
1y(0) = 0, U
2
(y) = D
q
1y(1) = 0. (4.5.25)
In this case
1
(x, ) = cos(
x; q), and
2
(x, ) = cos(
q
1/2
; q) cos(
x; q) +
q sin(
q
1/2
; q) sin(
x; q).
85
Since () =
qsin(
q
1/2
; q). Then
0
= 0 and for suciently large m, the eigenvalue
are given by
m
=
q
2m+1
(1 q)
2
(1 +O(q
m
)) , m 1 (4.5.26)
Therefore,
_
1, cos(
m
x; q)
_
m=1
is an orthogonal basis of L
2
q
(0, 1). If is not an eigen-
value, then the Greens function G(x, t, ) is dened for x, t [0, a] [0, a] by
G(x, t, ) =
cos(
t; q)
qsin(
q
1/2
; q)
_
cos(
q
1/2
; q) cos(
x; q)
+
q sin(
q
1/2
; q) sin(
x; q)
_
,
and
G(x, t, ) =
cos(
x; q)
qsin(
q
1/2
; q)
_
cos(
q
1/2
; q) cos(
t; q)
+
q sin(
q
1/2
; q) sin(
t; q)
_
, x t 1.
The operator L associated with problem (4.5.15), (4.5.25) is not invertible since zero is an
eigenvalue.
Example 4.5.3. Consider (4.5.15) with the following boundary conditions
U
1
(y) = y(0) = 0, U
2
(y) = y(1) +D
q
1y(1) = 0. (4.5.27)
Then
() =
2
(1, ) +D
q
1
2
(1, )
=
sin(
; q)
+ cos(
q
1/2
; q).
(4.5.28)
The eigenvalues
m
m=1
of this boundary value problem are the solutions of the equation
sin(
; q)
= cos(
q
1/2
; q) (4.5.29)
86
and the corresponding eigenfunctions are
_
sin(
m
x;q)
m
_
m=1
. The functions
1
(x, ) and
2
(x, ) are
1
(x, ) =
sin(
x; q)
2
(x, ) =
_
cos(
q
1/2
; q) +
sin(
; q)
_
cos(
_
x; q)
_
q sin(
q
1/2
; q) + cos(
; q)
_
sin(
x; q)
.
If is not an eigenvalue, then the Greens function G(x, t, ) is dened to be
G(x, t, ) =
1
sin(
; q) +
cos(
q
1/2
; q)
_
_
sin(
t; q)
2
(x, ), 0 t x,
sin(
x; q)
2
(t, ), x t 1.
and
G(x, t) =
1
2
_
_
t(2 x), 0 t x,
x(2 t), x t 1.
Therefore the boundary value problem (4.5.15), (4.5.27) is equivalent to the basic Fredholm
integral equation
y(x) =
_
1
0
G(x, t)y(t) d
q
t.
Remark 4.5.1. Let r be a function dened on [0, q
1
a] and let w be a function dened on
[0, a] such that D
q
1r(0) exists and w is positive on 0, aq
n
, n N. The SturmLiouville
problem (4.3.6)(4.3.8) for 0 x a < , C is
M(y) :=
1
q
D
q
1
_
r(x)D
q
y(x)
_
+(x)y(x) = w(x)y(x), (4.5.30)
U
1
(y) := a
11
y(0) +a
12
D
q
1y(0) = 0, (4.5.31)
U
2
(y) := a
21
y(a) +a
22
D
q
1y(a) = 0, (4.5.32)
87
where the functions and the constants a
ij
, i, j 1, 2 as in Section 4.3. In this case
we will have the Lagrange identity
_
a
0
_
(My)(x)z(x) y(x)(Mz)(x)
_
w(x) d
q
x
= r(aq
1
)[y, z](a) lim
n
r(aq
n1
)[y, z](aq
n
).
(4.5.33)
If y, z, and r are q-regular at zero, then the Lagrange identity (4.5.33) becomes
_
a
0
_
(My)(x)z(x) y(x)(Mz)(x)
_
w(x) d
q
x = r(aq
1
)[y, z](a) r(0)[y, z](0). (4.5.34)
The problem (4.5.30)(4.5.32) is formally self adjoint in L
2
q
_
(0, a), w
_
, where L
2
q
_
(0, a), w
_
is the space with the inner product
f, g)
w
=
_
a
0
f(x)g(x)w(x) d
q
x, f, g L
2
q
_
(0, a); w
_
. (4.5.35)
Let
1
(, ),
2
(, ) be the linearly independent solutions of (4.5.30) subject to the initial
conditions
D
j1
q
i
(0, ) =
ij
, i, j = 1, 2, C. (4.5.36)
The Greens function of the problem (4.5.30)(4.5.32)
1
q
D
q
1
_
r(x)D
q
y(x)
_
+w(x) +(x) y(x) = f(x), f L
2
q
_
(0, a), w
_
, (4.5.37)
is given by
G(x, t, ) =
1
()
_
2
(x, )
1
(t, )
r(t)
, 0 t x,
1
(x, )
2
(t, )
r(t)
, x < t a,
(4.5.38)
where
i
(x, ) are linearly independent solutions of (4.5.30) in [0, a] such that
U
1
1
(, ) = 0, U
2
2
(, ) = 0. (4.5.39)
We dene the operator / : T
M
L
2
q
_
(0, a), w
_
by /y = My for all y T
M
, where
T
M
is the subspace of L
2
q
_
(0, a), w
_
consisting of those complex valued functions that
88
satisfy (4.5.31)(4.5.32) such that
_
rD
q
y
_
() is q-regular at zero and D
q
1
_
(rD
q
y)()
_
L
2
q
_
(0, a), w
_
. An analysis similar work to Section 4.5 shows that the operator / has an
innite sequence of eigenvalues with no nite limit points and the eigenfunctions form an
orthogonal basis of L
2
q
_
(0, a), w
_
.
4.6 Asymptotics of eigenvalues and eigenfunctions
In this section we investigate the asymptotic behavior of the eigenvalues and the eigenfunc-
tions of the basic SturmLiouville problem (4.3.6)(4.3.8). The results of this section are
contained in [19]. The trigonometric identity
sin z cos w cos z sin w = sin(z w), (4.6.1)
where z, w are complex numbers, plays an important role in deriving the asymptotics of the
eigenvalues of the classical SturmLiouville problem (4.1.1)(4.1.3). Unfortunately, there is
no known q-analogue of such a rule. For For z in C and k in Z
+
, we dene the a function
S
k
(z; q) by
S
k
(z; q) := sin(z; q) cos(q
k
z; q) cos(z; q) sin(q
k
z; q). (4.6.2)
We have the following useful relation which plays a key role in our investigation.
Lemma 4.6.1. For any z C and k N, k 1, we have
S
k
(z; q) =
k1
n=0
(1)
n
q
n
2
+n
z
2n+1
q
(2n + 2)
(q
kn
; q)
2n+1
. (4.6.3)
Proof. Since the power series expansions of the basic cosine and sine functions dened
respectively in (1.1.16) and (1.1.17) are absolutely convergent, then we obtain
S
k
(z; q) =
n=0
(1)
n
q
n
2
+n
z
2n+1
q
(2n + 2)
2n+1
j=0
(1)
j
_
2n + 1
j
_
q
q
j(j1)/2
q
(kn)j
. (4.6.4)
89
Applying (1.1.8) with m = 2n + 1 and u = q
kn
yields
2n+1
j=0
(1)
j
_
2n + 1
j
_
q
q
j(j1)/2
q
(kn)j
= (q
kn
; q)
2n+1
. (4.6.5)
Moreover (q
kn
; q)
2n+1
= 0 whenever n k. This completes the proof.
Remark 4.6.1. Identity (4.6.3) is a q-analogue of the result of [115, p.259, Eq. (2)] when
= 1/2.
Lemma 4.6.2. For any k Z
+
we obtain
[S
k
(z; q)[ sinh([z[; q) q
1/4
E([z[; q). (4.6.6)
Proof. From (4.6.3) we nd that
[S
k
(z; q)[
k1
n=0
q
n
2
+n
[z[
2n+1
q
(2n + 2)
n=0
q
n
2
+n
[z[
2n+1
q
(2n + 2)
= sinh([z; q[). (4.6.7)
Moreover from (1.1.20), we have
E([z[; q) = cosh([z[; q) +q
1/4
sinh([z[; q). (4.6.8)
Consequently
sinh([z[; q) q
1/4
E([z[; q). (4.6.9)
Thus (4.6.6) follows from combining (4.6.7) and (4.6.9).
Lemma 4.6.3. For k Z
+
, x R, and z C, if [zx[
q
k
1 q
, then for any l N
S
k
(zx; q)
l+k
m=l+1
_
1 +
[z[
2
x
2
x
2
m
_
q
k
2
(1 q)
2k1
(q; q)
l+k
m=l+1
x
2
m
, (4.6.10)
where S
k
(; q) is dened in (4.6.3).
90
Proof. From (4.6.3) we obtain
S
k
(zx; q)
l+k
m=l+1
_
1 +
[z[
2
x
2
x
2
m
_
k1
r=0
q
r
2
+r
([zx[)
2r+1
_
q
kr
; q
_
2r+1
q
(2r + 2)
l+k
m=l+1
_
1 +
[z[
2
x
2
x
2
m
_
_
l+k
m=l+1
x
2
m
_
k1
r=0
q
r
2
+r
([zx[)
2r2k+1
q
(2r + 2)
=
_
l+k
m=l+1
x
2
m
_
k1
j=0
q
(kj)(kj1)
([zx[)
2j1
q
(2k 2j)
. (4.6.11)
If [zx[
q
k
1q
, then [zx[
2j1
q
2jk
q
k
(1 q)
2j+1
. Therefore
k1
j=0
q
k
2
k2kj+j(j+1)
([zx[)
2j1
q
(2k 2j)
q
k
2
k1
j=0
q
j(j+1)
(1 q)
2j+1
q
(2k 2j)
q
k
2
(1 q)
2k
k1
j=0
q
j(j+1)
(q; q)
2k2j1
q
k
2
(1 q)
2k1
(q; q)
.
(4.6.12)
The lemma follows by combining (4.6.11) and (4.6.12).
For a positive integer k let f
k
be a function dened on C by
f
k
(zx; q) := S
k
(z; q)
m=1
_
1 +
q
2k
[z[
2
x
2
m
_
_
1 +
[z[
2
x
2
m
_
. (4.6.13)
Theorem 4.6.4. There exists a positive constant h such that for all n N, if [xz[
q
n
1 q
,
where x R, and z C, then
[f
k
(zx; q)[ h, k = 1, 2, . . . , n. (4.6.14)
Proof. If x = 0, then f
k
(zx; q)[
x=0
= S
k
(0; q) = 0, for all k N and the lemma holds. In
the following we study the case x ,= 0. From relation (1.2.50) there exists a positive integer
91
m
0
and a positive constant A such that
1
q
2k
x
2
m+k
x
2
m
Aq
m
, k N, for all m m
0
. (4.6.15)
Choosing m
0
suciently large, from (1.2.20) we deduce that there exists a positive constant
B such that
q
m1/2
(1 q)x
m
1
Bq
m
, for all m m
0
. (4.6.16)
Set
h(m
0
) :=
_
B; q
_
2
_
Aq
m
0
+1
; q
_
(1 q)(q; q)
m
0
m=1
(1 +x
2
m
). (4.6.17)
In the following we prove that for any n N, h(m
0
) is an upper bound of the functions
f
k
([z[x; q), k = 1, 2, . . . , n in the region [zx[
q
n
1q
. Notice that for z C, x R, we have
1 +
q
2k
[z[
2
x
2
x
2
m
_
1 +
[z[
2
x
2
x
2
m+k
_
+[zx[
2
1
x
2
m+k
q
2k
x
2
m
. (4.6.18)
Thus from (4.6.15) we obtain for m m
0
,
1 +
q
2k
[z[
2
x
2
x
2
m
1 +
|z|
2
x
2
x
2
m+k
1 +
[zx[
2
1 +
|z|
2
x
2
x
2
m+k
1
x
2
m+k
q
2k
x
2
m
1 +
1
q
2k
x
2
m+k
x
2
m
_
1 +Aq
m
_
.
(4.6.19)
Hence by (4.6.17)
m=m
0
+1
1 +
q
2k
|z|
2
x
2
x
2
m
1 +
|z|
2
x
2
x
2
m+k
m=m
0
+1
(1 +Aq
m
) =
_
Aq
m
0
+1
; q
_
. (4.6.20)
Consequently
m=m
0
+1
1 +
q
2k
[z[
2
x
2
x
2
m
1 +
[z[
2
x
2
x
2
m
_
Aq
m
0
+1
; q
_
m=m
0
+1
1 +
[z[
2
x
2
x
2
m+k
1 +
[z[
2
x
2
x
2
m
=
_
Aq
m
0
+1
; q
_
m
0
+k
m=m
0
+1
_
1 +
|z|
2
x
2
x
2
m
_
.
(4.6.21)
92
For k N, x R, and z C let
g
k
(zx; q) := S
k
(zx; q)
m=m
0
+1
_
1 +
q
2k
[z[
2
x
2
x
2
m
_
_
1 +
[z[
2
x
2
x
2
m
_
. (4.6.22)
Thus
f
k
(zx; q) =
_
m
0
m=1
_
1 +
q
2k
|z|
2
x
2
x
2
m
_
_
1 +
|z|
2
x
2
x
2
m
_
_
g
k
(zx; q), k N, (4.6.23)
From (4.6.21) and Lemma 4.6.3 we conclude that
[g
k
(zx; q)[
_
Aq
m
0
+1
; q
_
S
k
(zx; q)
m
0
+k
m=m
0
+1
_
1 +
|z|
2
x
2
x
2
m
_
_
Aq
m
0
+1
; q
_
q
k
2
(1 q)
2k1
(q; q)
_
m
0
+k
m=m
0
+1
x
2
m
_
.
(4.6.24)
Indeed, since [zx[
q
n
1 q
, then [zx[
q
k
1 q
, k = 1, 2, . . . , n, then from Lemma 1.2.4 we
obtain
m
0
m=1
1 +
q
2k
[z[
2
x
2
x
2
m
1 +
[z[
2
x
2
x
2
m
=
m
0
m=1
x
2
m
+q
2k
[z[
2
x
2
x
2
m
+[z[
2
x
2
q
2km
0
m
0
m=1
(1 +x
2
m
), k = 1, 2, . . . , n. (4.6.25)
Combining (4.6.23), (4.6.24) and (4.6.25), yields
[f
k
(zx; q)[
_
Aq
m
0
+1
; q
_
q
k
2
+2m
0
k
(1 q)
2k1
(q; q)
m
0
+k
m=m
0
+1
x
2
m
m
0
m=1
(1 +x
2
m
), (4.6.26)
k = 1, 2, . . . , n, [zx[
q
n
1 q
. From (4.6.16) we have
0 x
2
m
q
2m+1
(1 q)
2
(1 +Bq
m
)
2
. (4.6.27)
Therefore
m
0
+k
m=m
0
+1
x
2
m
q
2m
0
kk
2
(1 q)
2k
m
0
+k
m=m
0
+1
(1 +Bq
m
)
2
= q
2m
0
kk
2
(1 q)
2k
_
m=0
(1 +Bq
m
)
_
2
= q
2m
0
kk
2
(1 q)
2k
(B; q)
2
.
(4.6.28)
93
Thus for [zx[
q
n
1 q
, formula (4.6.14) is satised. Clearly the constant h(m
0
) dened on
(4.6.17) is independent of n and x. This completes the proof of the theorem.
Lemma 4.6.5. The functions
1
(x, ) = cos(sx; q)+
q
s
_
x
0
cos(sx; q) sin(sqt; q) sin(sx; q) cos(sqt; q) (qt)
1
(qt, ) d
q
t,
(4.6.29)
2
(x, ) =
sin(sx; q)
s
+
q
s
_
x
0
cos(sx; q) sin(sqt; q) sin(sx; q) cos(sqt; q) (qt)
2
(qt, ) d
q
t,
(4.6.30)
form a fundamental set of solutions of (4.3.6) subject to the initial conditions
D
j1
q
i
(0, ) =
ij
, i, j = 1, 2, (4.6.31)
where s :=
1
q
D
q
1D
q
y(x) = y(x), x R, (4.6.32)
by the identity
0
(x, ) = u
1
(x)
1
(x, ) +u
2
(x)
2
(x, ),
where the functions u
1
and u
2
are solutions of the rst order q-dierence equations
D
q
u
i
(x) =
W
q,i
(
1
,
2
)(qx, )
W
q
(
1
,
2
)(qx, )
(qx)(x, ), i = 1, 2, (4.6.33)
where W
q,i
(x) is the determinant obtained from W
q
(x) by replacing the i-th column by the
column
_
0
1
_
. The functions
1
(x, ) = cos(sx; q),
2
(x, ) =
sin(sx; q)
s
,
94
form a fundamental set of solutions of (4.6.32) with W
q
(
1
,
2
)(x, ) 1. Then,
W
q,1
(
1
,
2
)(x, ) =
2
(x, ) and W
q,2
(
1
,
2
)(x, ) =
1
(x, ).
Thus, the functions
u
1
(x, ) :=
_
x
0
2
(qt, )(qt)(qt, ) d
q
t, and u
2
(x, ) =
_
x
0
1
(qt, )(qt)(qt, ) d
q
t,
are solutions of (4.6.33). Since the general solution of (4.3.6) is given by
(x, ) = c
1
1
(x, ) +c
2
2
(x, ) +
0
(x, ), (4.6.34)
where c
1
, c
2
are not both zero, then one can easily verify that (4.6.29) and (4.6.30) are a
fundamental set of solutions satisfying (4.6.31).
Computing the q
1
-derivatives of
1
(, ) and
2
(, ) we obtain the following lemma.
Lemma 4.6.6. For x (0, a] and ,= 0, we have
D
q
1
1
(x, ) :=
qs sin(sq
1/2
x; q)+
q
_
xq
1
0
_
cos(sq
1/2
x; q) cos(sqt; q) +
q sin(sq
1/2
x; q) sin(sqt; q)
_
(qt)
1
(qt, ) d
q
t,
(4.6.35)
D
q
1
2
(x, ) := cos(sq
1/2
x; q)+
q
_
xq
1
0
_
cos(sq
1/2
x; q) cos(sqt; q) +
q sin(sq
1/2
x; q) sin(sqt; q)
_
(qt)
2
(qt, ) d
q
t.
(4.6.36)
Theorem 4.6.7. For x [0, a], C, when [[ , we have
1
(x, ) = cos(sx; q) +O
_
E([s[x; q)
[s[
_
, (4.6.37)
2
(x, ) =
sin(sx; q)
s
+O
_
E([s[x; q)
[s[
2
_
, (4.6.38)
where for each x (0, a] the O-terms are uniform on xq
n
, n N. Moreover, if is
bounded on [0, a], the O-terms (4.6.37)(4.6.38) will be uniform for all x [0, a].
95
Proof. Let x (0, a] and n N be xed. Consider the annulus
q
n+1
1 q
[sx[
q
n
1 q
.
Then [sx[q
j
(1 q)
1
, for all j n. Hence from (4.6.3), we obtain for [sx[
q
n
1q
[S
k
(sxq
j
; q)[
k1
j=0
q
j
2
+j
(1 q)
2j1
q
(2j + 2)
j=0
q
j
2
+j
(1 q)
2j1
q
(2j + 2)
= sinh((1 q)
1
; q),
(4.6.39)
j n. Similarly for j n,
[ cos(sxq
j
; q)[ cosh((1 q)
1
; q), [sx[
q
n
1 q
. (4.6.40)
From (4.6.3) and (4.6.29) we deduce
1
(xq
j
, ) = cos(sxq
j
; q)
q
s
k=0
xq
k+j
(1 q)S
k+1
(sxq
j
; q)(xq
k+j+1
)
1
(xq
k+j+1
, )
= cos(sxq
j
; q)
1
s
k=1
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, ).
(4.6.41)
Set
1
(xq
j
, ) :=
1
(xq
j
, )
cosh([sx[q
j
; q)
=
1
(xq
j
, )
r=1
_
1 +
q
2j
[xs[
2
x
2
r
_
, j N. (4.6.42)
Multiplying both sides of the second equation of (4.6.41) by 1
_
cosh([sx[q
j
; q) and noting
that [ cos(z; q)[ cosh([z[; q), we obtain for j n
1
(xq
j
, )
1 +
sinh((1 q)
1
; q)
[s[
sup
jn
1
(xq
j
, )
k=1
xq
k+j
(1 q)
r=1
_
x
2
r
+[sx[
2
q
2j+2k
_
_
x
2
r
+[sx[
2
q
2j
_ [(xq
k+j
[
1 +
sinh((1 q)
1
; q)
[s[
sup
jn
1
(xq
j
, )
_
x
0
[(t)[ d
q
t,
(4.6.43)
Thus
sup
jn
[
1
(xq
j
, )[
_
1
q sinh((1 q)
1
; q)
[s[
_
x
0
[(t)[ d
q
t
_
1. (4.6.44)
96
We can choose n large enough such that
:=
_
1
sinh((1 q)
1
; q)
[s[
_
qx
0
[(t)[ d
q
t
_
> 0.
Therefore
[
1
(xq
j
, )[
1
, for all j n. (4.6.45)
Set :=
r=1
_
1 +
(1q)
2
x
2
r
_
. Then
[
1
(xq
j
, )[
r=1
_
1 +
q
2j
[xs[
2
x
2
r
_
r=1
_
1 +
(1 q)
2
x
2
r
_
= , (4.6.46)
for all j n. For j < n, we have
1
(xq
j
, ) = cos(sxq
j
; q)
1
s
nj1
k=1
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, )
1
s
k=nj
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, ). (4.6.47)
But from (4.6.46) and (4.6.10) we obtain
k=nj
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, )
__
x
0
[(t)[ d
q
t
_
sinh([sx[q
j
; q).
(4.6.48)
That is
1
s
k=nj
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, )
= O
_
1
[s[
sinh([sx[q
j
; q)
_
. (4.6.49)
Substituting in (4.6.47) yields
1
(xq
j
, ) = cos(sxq
j
; q)
1
s
nj1
k=1
xq
k+j
(1 q)S
k
(sxq
j
; q)(xq
k+j
)
1
(xq
k+j
, )
+O
_
1
[s[
sinh([sx[q
j
; q)
_
.
(4.6.50)
97
Multiplying both sides of (4.6.47) by 1
_
cosh([sx[q
j
; q) and using (4.6.42), we obtain
1
(xq
j
, )
1
+
1
[s[
max
1knj1
1
(xq
k+j
, )
nj1
k=1
xq
k+j
(1 q)
f
k
(sxq
j
; q)(xq
k+j
)
+O
_
1
[s[
sinh([sx[q
j
; q)
cosh([sx[q
j
; q)
_
,
(4.6.51)
where f
k
(; q) is dened in (4.6.13). Since [sx[
q
n+1
1q
, then for j = 1, 2, . . . , n1, we have
[sxq
j
[
q
(nj1)
1q
. Hence, from Theorem 4.6.4, there exists a constant h > 0 such that for
all n N and j 1, 2, . . . , n 1, if [sxq
j
[
q
(nj1)
1q
, then
[f
k
(sxq
j
; q)[ h, k = 1, 2, . . . , n j 1. (4.6.52)
Consequently for j = 1, 2, . . . , n 1 we get
[
1
(xq
j
, )[ 1
+ max
1<j<n
[
1
(xq
j
, )[
h
[s[
n
k=j+1
xq
k
(1 q)[(xq
k
)[ +O
_
sinh([sx[q
j
; q)
[s[ cosh([sx[q
j
; q)
_
1 +h max
1<j<n
[
1
(xq
j
, )[
_
x
0
[(t)[ d
q
t +O(
sinh([sx[q
j
; q)
[s[ cosh([sx[q
j
; q)
).
(4.6.53)
From Lemma 1.3.8 we have
[
1
(xq
j
, )[ 1 +h max
1<j<n
[
1
(xq
j
, )[
_
x
0
[(t)[ d
q
t +O
_
1
[s[
_
, (4.6.54)
j = 1, 2, . . . , n 1. Hence
max
1j<n
[
1
(xq
j
, )[
_
1
h
[s[
_
x
0
(qt) d
q
t
_
1 +O
_
1
[s[
_
. (4.6.55)
Again, we choose n large enough such that
_
1
h
[s[
_
x
0
[(qt)[ d
q
t
_
> 0. (4.6.56)
Then there exists a constant > 0 such that
[
1
(xq
j
, )[
r=1
_
1 +
q
2j
[xs[
2
x
2
r
_
, j = 1, 2 . . . , n 1. (4.6.57)
98
Using (4.6.57), (4.6.45), and (4.6.13) we obtain
1
s
n1
k=1
xq
k
(1 q)S
k
(sx; q)(xq
k
)
1
(xq
k
, )
r=1
_
1 +
[xs[
2
x
2
r
_
1
s
n1
k=1
xq
k
(1 q)f
k
([sx[; q)(xq
k
)
h
[s[
r=1
_
1 +
[xs[
2
x
2
r
_
_
qx
0
[(qt)[ d
q
t = O
_
E([sx[; q)
[s[
_
,
(4.6.58)
and
q
s
k=n
xq
k
(1 q)S
k+1
(sx; q)(xq
k+1
)
1
(xq
k+1
, )
sinh([sx[; q)
_
qx
0
[(t)[ d
q
t = O
_
E([sx[; q)
[s[
_
.
(4.6.59)
Thus
1
s
k=1
xq
k
(1 q)S
k
(sx; q)(xq
k
)
1
(xq
k
, )
max ,
_
_
qx
0
[(t)[ d
q
t
_
E([sx[; q)
[s[
.
(4.6.60)
One can verify that if we replace x by xq
m
, m Z
+
in the left-hand side of (4.6.60) we
obtain the same right-hand side of (4.6.60). That is the O-term in (4.6.37) is uniform on
xq
m
, m N. If is bounded on [0, a], and M := sup
x[0,a]
[(x)[ then we can redene
the constants to be
:= 1
Ma sinh((1 q)
1
; q)
[s[
, (4.6.61)
and (4.6.60) would be
1
s
k=1
xq
k
(1 q)S
k
(sx; q)(xq
k
)
1
(xq
k
, )
Ma max ,
E([sx[; q)
[s[
. (4.6.62)
Thus the O-term is uniform for x [0, a]. This proves (4.6.37) and (4.6.38) is similarly
proved.
99
Corollary 4.6.8. As [[ , we have
1
(x, ) = cos(sx; q) +O
_
[s[
1
exp
_
(log [sx[)
2
log q
_
_
, (4.6.63)
2
(x, ) =
sin(sx; q)
s
+O
_
[s[
2
exp
_
(log [sx[)
2
log q
_
_
. (4.6.64)
Proof. From (1.3.41) and (1.3.54) we obtain
M(r; sinh([z[; q)), M(r; cos([z[; q)) = O
_
exp
_
(log [z[)
2
log q
_
_
as z . Consequently from (1.1.21)
M
_
r; E(z; q)
_
= O
_
exp
_
(log [z[)
2
log q
_
_
as z .
Thus for each xed x [0, a]
E([sx[; q) = O
_
exp
_
(log [sx[)
2
log q
_
_
as [[ . (4.6.65)
Thus the big O-terms in (4.6.37) and (4.6.38) can be replaced by the big O-terms in (4.6.63)
and (4.6.64) respectively.
Theorem 4.6.9. Let s =
qs sin(sq
1/2
x; q) +O
_
exp
_
_
log [sxq
1/2
[
_
2
log q
_
_
, (4.6.66)
D
q
1
2
(x, ) = cos(sq
1/2
x; q) +O
_
[s[
1
exp
_
_
log [sxq
1/2
[
_
2
log q
_
_
, (4.6.67)
where
Proof. From (4.6.63), we conclude that as [[ ,
1
(x, ) = cos(sx; q) +g(x, ), [g(x, )[ = O
_
[s[
1
exp
_
(log [sx[)
2
log q
_
_
. (4.6.68)
100
Hence there exists a constant R > 0 such that
[g(t, )[
R
[s[
exp
_
(log [s[t)
2
log q
_
, t = xq
n
, n = 1, 0, s C 0 . (4.6.69)
Consequently
D
q
1g(x, )
g(xq
1
, ) g(x, )
xq
1
(1 q)
R
[sx[(1 q)
_
exp
_
(log [s[q
1
x)
2
log q
_
+ exp
_
(log [sx[)
2
log q
__
=
q
3/4
R
1 q
exp
_
(log [s[q
1/2
x)
2
log q
_
_
1 +
1
[s[
2
x
2
q
1
_
= O
_
exp
_
(log [s[q
1/2
x)
2
log q
_
_
, (4.6.70)
as s . Since D
q
1 cos(sx; q) =
qs sin(
qs sin(sq
1/2
a; q) +O
_
exp
_
(log [s[q
1/2
a)
2
log q
__
_
+a
11
a
21
_
sin(sa; q)
s
+O
_
[s[
2
exp
_
(log [s[a)
2
log q
__
_
+a
11
a
22
_
q cos(sq
1/2
a; q) +O
_
[s[
1
exp
_
(log [s[q
1/2
a)
2
log q
__
_
a
12
a
21
_
cos(sa; q) +O
_
[s[
1
exp
_
(log [s[a)
2
log q
__
_
.
(4.6.71)
Proof. From the denition of (), cf. (4.3.24), we obtain
() = U
1
(
1
)U
2
(
2
) U
1
(
2
)U
2
(
1
). (4.6.72)
Since
D
j1
q
i
(0, ) =
ij
, i, j = 1, 2, (4.6.73)
101
then from (4.3.7) and (4.3.8), we obtain
U
1
(
1
) = a
11
, U
2
(
2
) = a
12
, (4.6.74)
U
2
(
1
) = a
21
1
(a, ) +a
22
D
1
q
1
(a, ), (4.6.75)
U
2
(
2
) = a
21
2
(a, ) +a
22
D
1
q
2
(a, ). (4.6.76)
Thus
() = a
11
a
21
2
(a, ) +a
11
a
22
D
q
1
2
(a, ) a
12
a
22
1
(a, ) a
12
a
22
D
q
1
1
(a, ).
Substituting from (4.6.63)(4.6.64) and (4.6.66)(4.6.67) when x = a in the previous equa-
tion, we obtain (4.6.71).
It follows from (4.6.71) that () as . Consequently () has at most a
nite number of negative eigenvalues because
Theorem 4.6.11. () has an innite number of positive zeros and may have a nite
number of negative zeros. The positive zeros
m
are given asymptotically as m by
m
=
_
_
q
a
2
y
2
m
_
1 +O(q
m
2
)
_
, a
12
a
22
,= 0;
q
a
2
x
2
m
_
1 +O(q
m
2
)
_
, a
12
= 0;
1
a
2
x
2
m
_
1 +O(q
m
2
)
_
, a
22
= 0.
(4.6.77)
Proof. We shall consider the case of a
12
a
22
,= 0 in detail, the proof of the other cases are
similar.
If a
12
a
22
,= 0, then
() = a
12
a
22
qs sin(sq
1/2
a; q) +O
_
exp
_
(log [s[q
1/2
a)
2
log q
_
_
. (4.6.78)
Set
() = F() +R(), (4.6.79)
102
where
F() = a
12
a
22
qs sin(sq
1/2
a; q), R() = O
_
exp
_
(log [s[q
1/2
a)
2
log q
_
_
, (4.6.80)
Therefore there exists a constant c > 0 such that as [[
[R()[ c exp
_
(log [s[q
1/2
a)
2
log q
_
, for all C.
Recall that A
s
m
, m 1, is the annulus dened in (1.3.52) in terms of the positive ze-
ros y
m
m=1
of sin(z; q) and the sequences c
m
m=1
and d
m
m=1
dened respectively in
(1.3.49) and (1.3.50). Thus from (1.3.53) if saq
1/2
A
s
m
, m 1, then
log [ sin(sq
1/2
a; q)[ =
_
log [s[q
1/2
a
_
2
log q
+ log
1
qa
2
s
2
y
2
m
+O(1) (4.6.81)
as m . So there exists l
0
N, and
1
> 0 such that
log [ sin(sq
1/2
a; q)[ +
_
log [s[q
1/2
a
_
2
log q
log
1
a
2
q
1
s
2
y
2
m
1
. (4.6.82)
Then
log [R()[ log c
_
log [s[q
1/2
a
_
2
log q
log [ sin(sq
1/2
a; q)[ log
1
s
2
a
2
q
1
y
2
m
+
1
+ log c
log [F()[ log [s[ log
1
s
2
a
2
q
1
y
2
m
+
2
,
(4.6.83)
where
2
:=
1
+ log c + log
q[a
12
a
22
[. If z A
s
m
, then
log
1
s
2
a
2
q
1
y
2
m
log
_
1 q
2d
m
_
, [saq
1/2
[ = y
m
q
d
m
,
log
1
s
2
a
2
q
1
y
2
m
log
_
q
2c
m
1
_
, [saq
1/2
[ = y
m
q
c
m
.
Since the sequences c
m
m=1
, d
m
m=1
are bounded and positive, then there exists a con-
stant
3
> 0 such that
3
:= inf
m1
_
log(1 q
2d
m
), log(q
2c
m
1)
_
.
103
Therefore
log
1
s
2
a
2
q
y
2
m
3
.
We choose l
0
suciently large such that log [s[
3
+
2
< 0. That is
[R()[ [F()[, A
s
m
, m l
0
.
An application of Rouches theorem shows that () and F() have the same number of
zeros in A
s
m
, m l
0
. As sin(sq
1/2
a; q) has only two real simple symmetric zeros in A
s
m
,
then () has exactly two zeros there. In the following we show that these zeros are real,
simple and symmetric. If is a zero of (), then [F()[ = [R()[ and we deduce from
(4.6.83) that if [[ is suciently large, s :=
, > 0, then
log
1
a
2
q
1
y
2
m
log [s[ +
2
. (4.6.84)
Consequently either
log
1
saq
1/2
y
m
log [s
1/2
[ +
2
2
, (4.6.85)
or
log
1 +
saq
1/2
y
m
log [s
1/2
[ +
2
2
. (4.6.86)
Thus if (4.6.85) holds, then
1
saq
1/2
y
m
= O
_
[s[
1/2
_
. (4.6.87)
This implies that s q
1/2
a
1
y
m
. More precisely, from (1.2.21), we have
s = q
1/2
a
1
y
m
_
1 +O(y
m
)
1/2
_
(4.6.88)
Similarly we have if (4.6.86) holds that
s = q
1/2
a
1
y
m
_
1 +O(y
m
)
1/2
_
. (4.6.89)
Thus s are the zeros of () in A
s
m
, s is suciently large. On the other hand () has
a nite number of negative zeros, because if < 0, then the left hand side of (4.6.71) does
not tends to zero when [w[ is large enough.
104
Corollary 4.6.12. As m
_
m
=
_
_
q
m+1/2
a(1 q)
_
1 +O(q
m/2
)
_
, a
12
,= 0;
q
m+1
a(1 q)
_
1 +O(q
m/2
)
_
, a
12
= 0;
(4.6.90)
Proof. The proof follows by substituting in (4.6.77) with the asymptotic behavior of x
m
, y
m
given in equations (1.2.20), (1.2.21) respectively.
The following lemma will be needed in the sequel.
Lemma 4.6.13. Let f : C C be a function. If there exists a constant > 0 such that
[f(aq
n
z)[ E(aq
n
[z[; q), n N, a > 0,
then
_
a
0
[f(tz)[ d
q
t = O
_
exp
_
(log [z[a)
2
log q
_
_
as [z[ . (4.6.91)
Proof. From the denition of the q-integration (1.1.31) we obtain
_
a
0
[f(tz)[ d
q
t =
n=0
aq
n
(1 q)[f(aq
n
z)[
n=0
aq
n
(1 q)E(aq
n
[z[; q)
aM (ar; E(a[z[; q)) = O
_
exp
_
(log [z[a)
2
log q
_
_
,
(4.6.92)
as [z[ .
Theorem 4.6.14. The corresponding normalized eigenfunctions
m
()
m=1
have the
asymptotics
m
(x) =
_
_
cos(
m
x; q)
_ _
a
0
cos
2
(
m
x; q) d
q
x
_
1/2
_
1 +O(q
m
)
_
, a
12
,= 0;
sin(
m
x; q)
_ _
a
0
sin
2
(
m
x; q) d
q
x
_
1/2
_
1 +O(q
m
)
_
, a
12
= 0,
(4.6.93)
as m .
105
Proof. Since
m
(x) satises (4.3.6) with =
m
, there are numbers c
1
and c
2
independent
of x such that
m
(x) = c
1
1
(x,
m
) +c
2
2
(x,
m
). (4.6.94)
Since
m
(x) also satises the boundary condition (4.3.7), we have
c
1
a
11
+c
2
a
12
= 0. (4.6.95)
Since [c
1
[ +[c
2
[ > 0, we can write c
1
= k
m
a
12
and c
2
= k
m
a
11
, where k
m
,= 0. Hence
k
1
m
m
(x) = a
12
1
(x,
m
) a
11
2
(x,
m
). (4.6.96)
If a
12
,= 0, Theorem 4.6.7 with =
m
gives
k
1
m
m
(x) = a
12
cos(
_
m
x; q) +O
_
E(
m
x; q)
m
_
(4.6.97)
as m , where the O-term is uniform on xq
m
, m N. Thus
K
2
m
2
m
(x) = a
2
12
cos
2
(
_
m
x; q) +O
_
E
2
(
m
x; q)
m
_
(4.6.98)
But
cos(
m
x; q)
2
E
2
(
m
x; q), for all m Z
+
and x R. Then
_
a
0
cos
2
(
_
m
x; q) d
q
x
= O
_
exp
_
2(log a
m
)
2
log q
_
_
, (4.6.99)
as m . Applying Lemma 4.6.13 we obtain
K
2
m
= k
2
m
_
a
0
2
m
(x) d
q
x
= a
2
12
_
a
0
cos
2
(
_
m
x; q) d
q
x +O
_
exp
_
2(log a
m
)
2
log q
_
_
_
m
_
= a
2
12
__
a
0
cos
2
(
_
n
x; q) d
q
x
_
_
1 +O(
1
m
)
_
.
106
Hence from (4.6.96)
m
(x) =
cos(
m
x; q)
_
_
a
0
cos
2
(
m
x; q) d
q
x
_
1/2
_
1 +O(
1
m
)
_
1/2
+O(
1
m
)
=
cos(
m
x; q)
_
_
a
0
cos
2
(
m
x; q) d
q
x
_
1/2
+O(
1
m
)
=
cos(
m
x; q)
_
_
a
0
cos
2
(
m
x; q) d
q
x
_
1/2
_
1 +O(q
m
)
_
.
Similarly, we can prove the second formula of
m
when a
12
= 0.
Part III
Basic Fractional Calculus and
Equations
107
Chapter 5
Basic Fractional Calculus
In this chapter we study q-analogue of RiemannLiouville fractional calculus and fractional
dierence equations. We begin with a brief account of the history of fractional calculi and
some results which we are interested in deriving their q-analogues. We give also q-analogues
of the Gr unwaldLetinkov fractional derivative and the Caputo fractional derivative at the
end of this chapter.
5.1 Fractional calculus and equations
The following is a brief account about the history of the fractional calculus. This summary
which is taken from [30, 107] contains some important stages in the developments of the
concepts of fractional integrals and derivatives which we are interested in. The idea of
generalizing the notation of dierentiation d
k
/dx
k
, k N to non-integer values has begun
with the birth of dierential calculus itself. In 1695 L Hospital raised the question of the
meaning of d
n
y/dx
n
if n = 1/2, c.f. [30]. Euler (1738) was the rst who observed that the
derivative d
p
x
a
/dx
p
of the power function x
a
, a R, has a meaning for a non-integer p,
cf. [107]. S. F. Lacroix [81, 1819] derived the explicit formula
d
1/2
dx
1/2
x
a
=
(a + 1)
(a + 1/2)
x
a1/2
, a R. (5.1.1)
108
109
Formula (5.1.1) coincides with the RiemannLiouville fractional derivative of x
a
. See equa-
tion (5.2.12) below. J.B.J Fourier 1882 derived the integral representation
f(x) =
1
2
_
f(u) du
_
p
d
_
dt. (5.1.5)
J. L utzen [89, p. 314] indicated that Abel never solved the problem of fractional calculus.
He only showed that the solution, found by other means, could be written as a fractional
derivative. Throughout his series of papers published between 1823 and 1837, Liouville
has been considered as the founder of the theory of fractional calculus, see e.g. [30, 88,
107]. Liouvilles rst denition of fractional derivative is established for functions having
expansion of the form f(x) =
k=0
c
k
e
a
k
x
. For such functions Liouvilles denition is
D
f(x) =
k=0
c
k
a
k
e
a
k
x
, C. (5.1.6)
The restrictiveness of such a denition is clearly related to the convergence properties of
the series expansion of the dierentiated function. He also dened the formula
D
f(x) =
(1)
()
_
0
(x +t)t
1
dt, x R, ' > 0. (5.1.7)
110
This is now called the Liouvilles form of fractional integration where the factor (1)
is
omitted. It is worth mentioning that Liouville in 1823 gave the idea of dening the fractional
derivative as a limit of a dierence quotient
h
f/h
, where
h
f(x) is the righthanded
dierence of fractional order
h
f(x) =
j=0
(1)
j
_
j
_
f(x jh),
_
j
_
=
( 1) . . . ( j + 1)
j!
, (5.1.8)
which coincides with the classical r-th right handed dierence
r
h
f(x) for = r N.
However, Liouville gave no substantial development of this idea. He derived an example
involving Fouriers formula (5.1.3) for non integer p, based on this idea. He also evalu-
ated fractional derivative of some elementary functions via this approach. This idea was
more deeply considered by Gr unwald [56] and Letinkov [83] who established the so called
Gr unwaldLetinkov fractional derivative. Riemann in [106] dened the fractional integral
for R to be
1
()
_
x
0
(x t)
1
(t) dt, x > 0. (5.1.9)
which had become one of the main formulae of fractional integration together with Li-
ouvilles construction (5.1.7). Formula (5.1.9) is called the RiemannLiouville fractional
integral operator of a real-valued function (x) dened on (0, T], T > 0, as is seen in the
next section. Another direction in the history of fractional calculus is connected with the
Cauchy integral formula
f
(n)
(z) =
n!
2i
_
f(t)
(t z)
n+1
dt, (5.1.10)
for an analytic function f(z) in the complex plane. The direct extension of this formula to
non-integer values of n was rst made by Sonine. In 1917 Weyl, [116], dened fractional
integration for periodic functions as
K
()
n=
(in)
n
e
inx
,
n
=
1
2
_
2
0
e
int
(t) dt,
0
= 0. (5.1.11)
111
It is realized as a convolution
K
()
=
1
2
_
2
0
+
=
1
()
_
x
(t)
(x t)
1
dt, K
=
1
()
_
x
(t)
(x t)
1
dt (5.1.13)
0 < < 1, provided that the integrals in the last equation of the periodic function were
understood to be conventionally convergent ones over an innite interval. For this reason,
fractional integrals over innite intervals, especially the integral K
t
f(t) =
1
(n )
_
t
a
(t )
n1
f
(n)
() d, (5.1.14)
n 1 < < n. In the following section, we would exhibit some properties of the Riemann-
Liouville fractional derivative, M. Caputo fractional derivative, Gr unwald-Letinkov frac-
tional derivative, and Liouville fractional integral operator. These will be the main proper-
ties and relations of which we will derive the basic analogue.
112
5.2 Some fractional integral operators
This section is devoted to constructing the fractional calculi of which we will derive the
q-analogues. First we denote by /C
(n)
[a, b], n N, to the set of functions f which have
continuous derivatives up to order n 1 on [a, b] with f
(n1)
/C[a, b].
Lemma A. [79] The set /C
(n)
[a, b] consists of those and only those functions f, which are
represented in the form
f(x) =
n1
k=0
c
k
(x a)
k
+
1
n 1!
_
x
a
(x t)
n1
(t) dt, (5.2.1)
where L
1
(a, b), and the c
k
s, k = 0, 1, . . . , n 1, are arbitrary constants. Moreover
(x) = f
(n)
(x) a.e., and c
k
=
f
(k)
(a)
k!
, k = 1, 2, . . . , n.
The rst fractional integral operator we dene in this section is that of Riemann and
Liouville. It is connected to Abels integral equation
1
()
_
x
a
(x t)
1
(t) dt = f(x), x > a, > 0, f L
1
(a, b), (5.2.2)
cf. [107, p. 32].
Theorem B. The Abel integral equation (5.2.2) with 0 < < 1 has a unique solution in
L
1
(a, b) if and only if the function f
1
dened by
f
1
(x) :=
1
(1 )
_
x
a
(x t)
f(t) dt (5.2.3)
is absolutely continuous on (a, b) and f
1
(a) = 0. Moreover the unique solution is given
by
(x) =
1
(1 )
d
dx
_
x
a
(x t)
f(t) dt =
d
dx
f
1
(x). (5.2.4)
If f /C[a, b], then f
1
/C[a, b] and (5.2.4) becomes
(x) =
1
(1 )
_
f(a)
(x a)
+
_
x
a
f
(s)
(x s)
ds
_
. (5.2.5)
113
The following Cauchy formulae can be considered as the n-th primitive of a function
f L
1
(a, b).
_
x
a
_
x
n1
a
. . .
_
x
1
a
f(t) dtdx
1
. . . dx
n1
=
1
(n 1)!
_
x
a
(x t)
n1
f(t) dt, (5.2.6)
_
b
x
_
b
x
n1
. . .
_
b
x
1
f(t) dtdx
1
. . . dx
n1
=
1
(n 1)!
_
b
x
(t x)
n1
f(t) dt, (5.2.7)
n Z
+
. Since the right hand side of (5.2.6) and (5.2.7) exist also for non integer values
of n, the RiemannLiouville fractional integral can be considered as an extension of (5.2.6)
and (5.2.7) when we replace n by R
+
and n 1! by (). Indeed for (0, ) and
f L
1
(a, b) the fractional RiemannLiouville integral is dened to be
D
a+
f(x) :=
1
()
_
x
a
(x t)
1
f(t) dt, D
b
f(x) :=
1
()
_
b
x
(t x)
1
f(t) dt, (5.2.8)
x (a, b) with respect to x = a and x = b respectively. In some literature they are called
left-sided and right-sided RiemannLiouville fractional integrals respectively. It is known,
see e.g [100, 107], that if f L
1
(a, b), then both of D
a
+
f and D
b
f exist a.e. and they are
L
1
(a, b)-functions. Moreover for f L
1
(a, b), we have
lim
0+
D
a+
f(x) := D
k
D
(k)
a+
f(x) =
1
(k )
d
k
dx
k
_
x
a
(x t)
k1
f(t) dt, (5.2.10)
and
D
b
f(x) := (1)
k
D
k
D
(k)
b
f(x) =
(1)
k
(k )
d
k
dx
k
_
b
x
(t x)
k1
f(t) dt, (5.2.11)
114
x (a, b), k = [] + 1, respectively. The fractional derivatives D
a
+
f, D
f exist if f
L
1
(a, b) and D
(k)
a
+
f, D
(k)
b
a+
x
a
=
(a + 1)
(a + 1)
x
a
, a R. (5.2.12)
Now we state some properties of the RiemannLiouville fractional calculus which we will
derive their q-analogues in Section 5.4. We conne ourselves to the case of the leftsided
RiemannLiouville fractional calculus since we will study its basic analogue. Let , R
+
.
If f L
1
(a, b), then the semigroup property
D
a+
D
a+
f(x) = D
a+
D
a+
f(x) = D
(+)
a+
f(x), (5.2.13)
holds for almost all x [a, b]. If f(x) satises the conditions
f L
1
(a, b) and D
(k)
a+
f /C
(k)
[a, b], k = [] + 1, (5.2.14)
then D
j
a+
f L
1
(a, b), j = 0, 1, . . . , k. Besides, D
j
a+
f AC
(j)
[a, b], 1 j k 1.
Moreover, for functions satisfying the appropriate conditions
D
a+
D
a+
f(x) = f(x), a.e. (5.2.15)
D
a+
D
a+
f(x) = D
()
a+
f(x), a.e. if 0, (5.2.16)
D
a+
D
a+
f(x) = D
a+
f(x), a.e. if > 0. (5.2.17)
Also,
D
a+
D
a+
f(x) = f(x)
k
j=1
D
j
a+
f(t)
t=a
(x a)
j
(1 + j)
. (5.2.18)
If f L
1
(a, b) and, in addition, D
(n)
a+
f /C
(n)
[a, b], n = [] + 1, then the equality
D
a+
D
a+
f(x) = D
a+
f(x)
n
j=1
_
D
j
a+
f(t)
_
t=a
(x a)
j
(1 + j)
(5.2.19)
115
holds almost everywhere in (a, b) for any > 0. Finally, the identities
D
a+
D
a+
f(x) = f(x), and D
a+
D
a+
f(x) = D
a+
f(x), (5.2.20)
hold if f I
a+
(L
1
) and f I
a+
(L
1
), respectively, cf. [107], where by I
a+
(L
1
) we denote
the set of functions f represented by the left sided fractional integral of order , > 0, of
a summable function, i.e. f = I
a+
, L
1
(a, b). For proof of these and other properties,
see e.g. [40, 100, 107].
The other types of fractional derivatives we discuss here are those of Gr unwaldLetinkov
and Caputo. It is known that the nth derivative of a function f is dened for x (a, b) to
be
f
(n)
(x) = lim
h0
n
h
f(x)
h
n
, (5.2.21)
where
n
h
f(x) =
1
h
n
n
r=0
(1)
r
_
n
r
_
f(x rh). (5.2.22)
Starting from this formula, A.K. Gr unwald [56] and A.V. Letinkov [83] developed an ap-
proach to fractional dierentiation for which the formal denition of the fractional derivative
D
a
+
f(x) is the limit
D
a
+f(x) := lim
h0
h
f(x)
h
, R
+
, (5.2.23)
h
f(x) being the right-handed dierence of fractional order dened in (5.1.8).
Gr unwald [56] gave a formal proof while Letinkov s [83] gave a rigorous proof of the fact
that D
coincides with the RiemannLiouville fractional integral operator when > 0 and
f is continuous on [a, b], cf. [82]. The following theorem gives the conditions on which the
Gr unwaldLetinkov fractional derivative is nothing but the RiemannLiouville fractional
derivative.
Theorem C. [100] Assume that f /C
(n)
[a, b]. Then for every , 0 < < n, the
RiemannLiouville fractional derivative D
a
+
f(t) exists and coincides with the Gr unwald
Letinkov derivative D
a
+
f(t). i.e. If 0 m 1 m n, then for a < t < b the
116
following holds:
D
a
+
f(t) = D
a
+
f(t) =
m1
j=0
f
(j)
(a)(t a)
j
j!
+
1
(m)
_
t
a
(t )
m1
f
(m)
() d.
(5.2.24)
It is known that in applied problems we require denitions of fractional problems which
allow the involvement of initial conditions with physical meanings, i.e. conditions like e.g.
f
(j)
(a) = b
j
, j = 1, 2, . . . , k (5.2.25)
where the b
j
s and k Z
+
are given constants. The RiemannLiouville approach leads to
initial conditions containing the limit values of the RiemannLiouville fractional derivatives
at the lower end point t = a, for example
lim
ta
D
j
a
+
f(t) = c
j
, j = 1, 2, . . . , k, (5.2.26)
where the c
j
s and k Z
+
are given constants and is not an integer. It is frequently
stated that the physical meaning of initial conditions of the form (5.2.26) is unclear or
even non existent, see e.g. [107, p. 78]. The requirement for physical interpretation of such
initial conditions was most clearly formulated recently by Diethelm et al. [74]. In [62], N.
Heymans and I. Podlubny show that initial conditions of the form (5.2.26) for Riemann-
Liouville fractional dierential equations have physical meaning, and that the corresponding
quantities can be obtained from measurements. However, the problem of interpretation of
initial conditions still remained open. In [31], [32], M. Caputo gave a solution for this
problem when he dened the fractional derivative of order ,
C
a
D
t
f(t), by the formula
C
a
D
t
f(t) =
1
(n )
_
t
a
(t )
n1
f
(n)
() d, (5.2.27)
n1 < n, n Z
+
. The Caputo approach leads to initial conditions of the form (5.2.25)
which has a physical meaning. Recently, several initial value problems based on the Caputo
117
fractional operator have been studied, see e.g. [42, 43, 99, 101]. The relationship between the
Caputo fractional derivative and the RiemannLiouville Calculus is obviously
C
a
D
t
f(t) = D
(n)
a
+
f
(n)
(t). (5.2.28)
If the function f(t) has n + 1 continuous derivatives in [a, b], then cf. [100]
lim
n
C
a
D
t
f(t) = f
(n)
(t), t [a, b].
The main advantage of the Caputo fractional derivative is that
C
a
D
t
c = 0, where c is a
constant, while the Riemann-Liouville fractional derivative of a constant c is not equal to
zero. In fact, if n = [] + 1, then
D
a
+
c = D
n
I
n
c =
c
(n + 1)
D
n
(x a)
n
=
c
(1 )
(x a)
. (5.2.29)
Since we will establish a method for using the basic Laplace transform dened by [57,
1949] in solving basic fractional equations, we state before closing this section some results
concerning the Laplace transform.
Let f L
1
(0, b) for any b > 0. If I
n
f /C
(n)
[0, b] for any b > 0 such that I
n
f(x)
is of exponential order as x . That is there exist real constants K, T, p
0
R
+
, p
0
0,
such that
[I
n
f(x)[ Ke
p
0
x
if x > T. (5.2.30)
Then the Laplace transform of the RiemannLiouville fractional derivative D
a
+
f(x) exists,
cf. [117], and satises the following identity
_
0
e
pt
D
a
+
f(t) dt = p
F(p)
n1
k=0
p
k
D
a
+
f(t)[
t=0
, n 1 < n, (5.2.31)
for all 'p > p
0
, cf. [107,117]. Also the Laplace transform of the Caputo derivative
C
a
D
t
f(x)
exists if f(x) /C
(n)
[0, b] for any b > 0 and f(x) is of exponential order as x . In this
118
case we can nd p
0
0 such that the Laplace transform of the Caputo derivative satises
the identity
_
0
e
ptC
0
D
t
f(t) dt = p
F(p)
n1
k=0
p
k1
f
(k)
(0), (5.2.32)
for all 'p > p
0
. The Laplace transform method is frequently used for solving applied prob-
lems. So, it is clear that the Laplace transform of the Caputo derivative allows utilization of
initial values of classical integer-order derivative with known physical interpretations, while
the Laplace transform of the Riemann-Liouville fractional derivative allows utilization of
initial conditions of the form
D
j
f(t)[
t=0
= b
j
, j = 1, 2, . . . , [] + 1,
which can cause problems without their physical interpretation.
5.3 q-Notations and results
In this section, we introduce q-notations and results which we need throughout this part.
The following denition of q-translation
y
is given by M.E.H. Ismail [65]. The q-translation
is dened on monomials by
y
x
n
:= x
n
_
y/x; q
_
n
, (5.3.1)
and it is extendeed to polynomials as a linear operator. Thus
y
_
m
n=0
f
n
x
n
_
:=
m
n=0
f
n
x
n
_
y/x; q
_
n
. (5.3.2)
The q-translation is dened for x
a
, a R
+
to be
y
x
a
:= x
a
_
y/x; q
_
a
. (5.3.3)
Let L
1
q
(0, a) be the space of all functions f dened on [0, a] such that f L
1
q
(0, x) for all
x (qa, a].
119
Lemma 5.3.1. For positive numbers , and positive integer n, we have
n
k=0
_
n
k
_
q
q
k
(q
; q)
k
(q
; q)
nk
= (q
+
; q)
n
. (5.3.4)
Proof. We prove the lemma by induction. Clearly For n = 0, (5.3.4) is true. Assume that
(5.3.4) is true at n. Then, using
_
n + 1
k
_
q
=
_
n
k 1
_
q
+q
k
_
n
k
_
q
, (5.3.5)
cf. e.g [53, p. 25], we obtain
n+1
k=0
_
n + 1
k
_
q
q
k
(q
; q)
k
(q
; q)
n+1k
= (q
; q)
n+1
+
n
k=1
_
n
k 1
_
q
q
k
(q
; q)
k
(q
; q)
n+1k
+ q
(n+1)
(q
; q)
n+1
+
n
k=1
_
n
k
_
q
q
(+1)k
(q
; q)
k
(q
; q)
nk
=
n
k=0
_
n
k
_
q
q
(k+1)
(q
; q)
k+1
(q
; q)
nk
+
n
k=0
_
n
k
_
q
q
(+1)k
(q
; q)
k
(q
; q)
n+1k
=
_
q
(1 q
) + (1 q
)
_
(q
++1
; q)
n
= (q
+
; q)
n+1
.
This completes the proof.
Remark 5.3.1. There is another proof of the previous Lemma. This proof depends on Heines
q-analogue of Gauss summation formula
2
1
(q
n
, b; c; q, q) =
(c/b, q)
n
(c, q)
n
b
n
, (5.3.6)
see e.g. [53, p.14]. Indeed
n
k=0
_
n
k
_
q
q
k
(q
; q)
k
(q
; q)
nk
= (q
; q)
n 2
1
(q
n
, q
; q
1n
; q, q)
= (q
; q)
n
q
n
(q
1n
; q)
n
(q
1n
; q)
n
= (q
+
; q)
n
.
Corollary 5.3.2. For , > 0 and n Z
+
, we have
n
k=0
q
k
(1 q)(q
k+1
; q)
1
(q
nk+1
; q)
1
= B
q
(, )(q
n+1
; q)
+1
. (5.3.7)
120
Proof. Using equations (1.1.5) and (1.1.10) one can easily see that for any > 0, j N
(q
j+1
; q)
1
=
(q; q)
(q
; q)
j
(q; q)
j
(q
; q)
=
q
()(1 q)
1
(q
; q)
j
(q; q)
j
(5.3.8)
Thus from (5.3.4) we obtain
n
k=0
q
k
(1 q)(q
k+1
; q)
1
(q
nk+1
; q)
1
=
(1 q)
+1
q
()
q
()
(q; q)
n
n
k=0
q
k
_
n
k
_
q
(q
; q)
k
(q
; q)
nk
= (1 q)
+1
q
()
q
()
(q
+
; q)
n
(q; q)
n
. (5.3.9)
But from (5.3.8)
(q
+
; q)
n
(q; q)
n
=
(1 q)
1
q
( +)
(q
n+1
; q)
+1
. (5.3.10)
By combining (5.3.9) and (5.3.10) we obtain (5.3.7).
The following property is essential in our investigations and it is taken from [90]. This
property indicates that the nth q-derivative D
n
q
of a function f can be expressed in terms
of its values at the points q
j
x, j = 0, 1, . . . , n.
Property 5.3.3. Let f be a function dened in an interval I and assume that f has q-
derivatives up to order n in I, n in N. Then, (D
k
q
f)(x) can be expressed as
D
k
q
f(x) =
1
x
k
(1 q)
k
j=k
j=0
(1)
j
_
k
j
_
q
f(q
j
x)
q
j(j1)/2+j(kj)
, (5.3.11)
for every x in I 0, k = 1, 2, . . . , n.
Lemma 5.3.4. Let h(t, x) be a function dened on [0, a] [0, a], such that
D
j
q,x
h(t, x), j = 0, 1, . . . , k 1,
are q-integrable functions on [0, a], . If for some x (0, a] and k N
h(xq
r
, xq
j
) = 0, r = 0, 1, 2, . . . , j 1; j = 1, 2, . . . , k, (5.3.12)
121
then
D
k
q,x
_
x
0
h(t, x) d
q
t =
_
x
0
D
k
q,x
h(t, x) d
q
t, x (0, a]. (5.3.13)
Proof. From (5.3.11) we obtain
D
k
q,x
_
x
0
h(t, x) d
q
t =
j=k
j=0
(1)
j
_
k
j
_
q
q
j(j+1)
2
kj
x
k
(1 q)
k
_
xq
j
0
h(t, xq
j
) d
q
t. (5.3.14)
From the condition (5.3.12) we nd that
_
xq
j
0
h(t, xq
j
) d
q
t =
_
x
0
h(t, xq
j
) d
q
t, j = 1, 2, . . . , k. (5.3.15)
Hence
D
k
q,x
_
x
0
h(t, x) d
q
t =
j=k
j=0
(1)
j
_
k
j
_
q
q
j(j+1)
2
kj
x
k
(1 q)
k
_
x
0
h(t, xq
j
) d
q
t
=
_
x
0
_
_
j=k
j=0
(1)
j
_
k
j
_
q
q
j(j+1)
2
kj
x
k
(1 q)
k
h(t, xq
j
)
_
_
d
q
t
=
_
x
0
D
k
q,x
h(t, x) d
q
t. (5.3.16)
Lemma 5.3.5. For , > 0 and f L
1
q
(0, x), x > 0, we obtain
x
1
_
x
0
(qt/x; q)
1
t
1
_
t
0
(qu/t; q)
1
f(u) d
q
ud
q
t
= B
q
(, )x
+1
_
x
0
_
qt/x; q
_
+1
f(t) d
q
t.
(5.3.17)
Proof. Let , > 0 and f L
1
q
(0, x), x > 0. Then by the denition of the q-integration,
122
cf. (1.1.31) we obtain
x
1
_
x
0
(qt/x; q)
1
t
1
_
t
0
(qu/t; q)
1
f(u) d
q
ud
q
t (5.3.18)
= x
+1
n=0
q
n
(q
n+1
; q)
1
m=0
xq
n+m
(1 q)(q
m+1
; q)
1
f(xq
n+m
)
= x
+1
n=0
xq
n
(1 q)(q
n+1
; q)
1
(q
n+1
; q)
1
k=n
q
k
(1 q)(q
kn+1
; q)
1
f(xq
k
)
= x
+1
k=1
xq
k
(1 q)f(xq
k
)
k
n=0
q
n
(1 q)(q
n+1
; q)
1
(q
kn+1
; q)
1
. (5.3.19)
Thus from (5.3.7)
x
1
_
x
0
(qt/x; q)
1
t
1
_
t
0
(qu/t; q)
1
f(u) d
q
ud
q
t
= x
+1
B
q
_
,
_
k=1
xq
k
(1 q)(q
k+1
; q)
+1
f(xq
k
)
= x
+1
B
q
_
,
_
_
x
0
(qt/x; q)
+1
f(t) d
q
t.
(5.3.20)
Denition 5.3.1. A function f dened on [0, a] is called qabsolutely continuous if for
any > 0 and x, y (0, a] there exists n
0
N such that for any n n
0
, if
_
(xq
k
, yq
k
)
_
n
k=n
0
are pairwise disjoint and non-intersecting intervals, then we have
n
j=n
0
f(xq
j
f(yq
j
)
. (5.3.21)
The space of these functions will be denoted by /C
q
[0, a].
Lemma 5.3.6. If f /C
q
[0, a], then f is q-regular at zero.
Proof. Let f /C
q
[0, a]. Then given > 0 and t, v (0, a], there exists n
0
N such that
for any n n
0
if
_
(tq
k
, vq
k
)
_
n
k=n
0
are pairwise disjoint non-intersecting intervals, then
n
j=n
0
f(tq
j
f(vq
j
)
. (5.3.22)
123
Take v = qt in (5.3.22) and assume that t (qa, a]. Then
n
j=n
0
f(tq
j
) f(tq
j+1
)
, n n
0
. (5.3.23)
Thus for all m > n n
0
, t (qa, a], we obtain
[f(tq
n
) f(tq
m
)[ =
m1
j=n
f(tq
j
) f(tq
j+1
)
< . (5.3.24)
Consequently
l
t
:= lim
j
f(tq
j
) exists for all t (qa, a]. (5.3.25)
We prove that l
t
= l
v
for all t, v (qa, a]. Indeed from (5.3.25) we can nd j n
0
such
that
[l
t
f(tq
j
)[, [l
v
f(vq
j
)[ < . (5.3.26)
Then by (5.3.22)
[l
t
l
v
[ [l
t
f(tq
j
)[ +[f(tq
j
) f(vq
j
)[ +[f(vq
j
l
v
)[ < 3. (5.3.27)
Hence l
t
does not depend on t (qa, q] as required. Set l := l
t
, t (qa, a]. Then If we
redene f(0) to be l, then f is qregular at zero.
Remark 5.3.2. Form the previous theorem all the functions in /C
q
[0, a] are qregular at
zero. Therefore from now on if f /C
q
[0, a], then by f(0) we mean lim
n
x(qa,a]
f(xq
n
).
Theorem 5.3.7. Let f be a function in the space /C
q
[0, a]. If t xq
n
, n N, x (qa, a],
then
f /C
q
[0, a] f(t) = c +
_
t
0
(u) d
q
u, L
1
q
(0, x), (5.3.28)
where c = f(0) and (x) = D
q
f(x) for all x (0, a].
Proof. Assume that f /C
q
[0, a]. Fix x (qa, a] the series
n=0
f(xq
n
) f(xq
n+1
) is
absolutely convergent and lim
n
f(xq
n
) = c. Obviously if t = xq
n
, x (qa, a], n N, we
124
get
f(t) = c +
k=n
f(xq
k
) f(xq
k+1
) = c +
_
t
0
D
q
f(u) d
q
u. (5.3.29)
But
_
t
0
D
q
f(u) d
q
u
k=0
[f(xq
k
) f(xq
k+1
)[ < , t = xq
n
; n N. (5.3.30)
Hence D
q
f L
1
q
(0, x) for all x (qa, a]. To prove the converse if f satises the identity
in (5.3.28), then D
q
f(x) = (x) for all x ,= 0. Hence
n=0
xq
n
(1 q)[(D
q
f)(xq
n
)[ is
convergent. Hence from Lemma 1.1.1 and relation (1.1.34) there exists
i
(0, 1], i = 1, 2,
and n
0
N such that
[f(xq
n
) f(xq
n+1
)[ C[xq
n
[
1
1
, [f(yq
n
) f(yq
n+1
)[ C[yq
n
[
1
2
, n n
0
(5.3.31)
for some constant C > 0. Thus from (5.3.28) we have
f(xq
n
) f(yq
n
) =
k=n
f(xq
k
) f(xq
k+1
)
k=n
f(yq
k
) f(yq
k+1
). (5.3.32)
Then for all n n
0
we obtain
[f(xq
n
) f(yq
n
)[ C
_
x
1
1
(1 q
1
1
)
q
n(1
1
)
+
y
1
2
(1 q
1
2
)
q
n(1
2
)
_
, (5.3.33)
Thus
n=n
0
[f(xq
n
) f(yq
n
)[ C
_
x
1
1
(1 q
1
1
)
2
q
n
0
(1
1
)
+
y
1
2
(1 q
1
2
)
2
q
n
0
(1
2
)
_
. (5.3.34)
We choose n
0
large enough such that
C
_
x
1
1
(1 q
1
1
)
2
q
n
0
(1
1
)
+
y
1
2
(1 q
1
2
)
2
q
n
0
(1
2
)
_
< .
Hence f /C
q
[0, a].
Denition 5.3.2. Let /C
(n)
q
[0, a], n Z
+
, be the space of all functions f dened on [0, a]
such that f, D
q
f, . . ., D
n1
q
f are q-regular at zero and D
n1
q
f(x) /C
q
[0, a].
125
When n = 1, we call /C
(1)
q
[0, a] the space of q-absolutely continuous function and we
refer to it by /C
q
[0, a] for simplicity.
Lemma 5.3.8. A function f : [0, a] C lies in /C
(n)
q
[0, a] if and only if there exists a
function L
1
q
(0, x), qa < x a such that
f(t) =
n1
k=0
c
k
t
k
+
t
n1
q
(n)
_
t
0
_
qu/t; q
_
n1
(u) d
q
u, (5.3.35)
where t = xq
n
, x (qa, a], n N, c
k
=
D
k
q
f(0)
q
(k + 1)
and (x) = D
n
q
f(x) for all x ,= 0.
Proof. The sucient condition is easy to prove. We shall prove the necessary condition
by using the mathematical induction. The case of n = 1 is proved in Theorem 5.3.7. Now
assume that (5.3.35) holds at n = m. Let f(x) /C
(m+1)
q
[0, a]. That is, f, . . . , D
m
q
f
are q-regular at zero and D
m
q
f /C
1
q
[0, a]. Hence, D
q
f /C
(m)
q
[0, a]. Then, from the
hypothesis of the induction
D
q
f(t) =
m1
j=0
c
j
t
j
+
t
m1
q
(m)
_
t
0
_
qu/t; q
_
m1
(u) d
q
u, (5.3.36)
where t xq
n
, n N, x (qa, a],
c
j
= D
j
q
D
q
f(0)/
q
(j + 1) = D
j+1
q
f(0)/
q
(j + 1) and (x) = D
m+1
q
f(x)
for all x ,= 0. Now, replacing t with u and u with v in (5.3.36) and integrating from 0 to t.
Then applying Lemma 5.3.5 yields
f(t) =
m
k=0
D
k
q
f(0)
q
(k + 1)
t
k
+
t
m
q
(m+ 1)
_
t
0
_
qu/t; q
_
m
D
m+1
q
f(u) d
q
u, (5.3.37)
which completes the induction steps.
126
5.4 Basic RiemannLiouville fractional calculi
In 1966, W.A. Al-Salam [110] dened q-integral operator K
q
by the basic integral
K
q
(x) :=
q
1
2
(1)
q
()
_
x
t
1
_
x/t; q
_
1
(tq
1
) d
q
t,
K
0
q
(x) := (x),
(5.4.1)
where ,= 1, 2, . . .. Al-Salam dened (5.4.1) as a generalization of the following basic
Cauchy formula which is introduced in [12] for a positive integer n via
K
n
q
(x) =
_
x
_
x
n1
. . .
_
x
1
(t) d
q
td
q
x
1
. . . d
q
x
n1
=
q
1
2
N(N1)
q
(N)
_
x
t
N1
(x/t; q)
N1
(tq
1N
) d
q
t.
(5.4.2)
Using (1.1.33), we can write the equation (5.4.1) as
K
q
(x) = q
1
2
(1)
x
(1 q)
k=0
(1)
k
q
k(k1)
2
_
k
_
q
(xq
k
), (5.4.3)
which is valid for all . The fractional q-integral operator K
q
is a q-analogue of Liouville
fractional integral, K
q
K
q
(x) = K
+
q
(x), , > 0. (5.4.4)
Moreover, in [13], Al-Salam dened a two parameter basic fractional operator by
K
,
q
f(x) :=
q
q
()
_
x
_
x/t; q
_
1
t
1
f(tq
1
) d
q
t, (5.4.5)
,= 1, 2, . . .. This is a q-analogue of the Erdelyi and Sneddon fractional operator,
cf. [44, 46],
K
,
f(x) =
x
()
_
x
_
t x)
1
t
1
f(t) dt. (5.4.6)
By means of (1.1.33), (5.4.5) can be written as
K
,
q
f(x) = (1 q)
k=0
(1)
k
q
k
_
k
_
q
f(xq
k
), (5.4.7)
127
valid for all . Again Al-Salam gave the following semigroup identity for , and in R
+
.
K
,
q
K
+,
q
f(x) = K
,+
q
f(x). (5.4.8)
In 1968, R.P. Agarwal [10] dened another two parameter family fractional q-operator
corresponding to the basic integral (1.1.31). This operator is dened by
I
,
q
f(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
t
f(t) d
q
t, ,= 1, 2, . . . (5.4.9)
=
(1 q)
q
()
k=0
q
(+1)k
_
q
k+1
; q
_
1
f(xq
k
). (5.4.10)
valid for all . He also proved the following semigroup identity when , , and are positive
constants.
I
,
q
I
+,
q
f(x) = I
,+
q
= I
+,
q
I
,
q
f(x) (5.4.11)
= I
,
q
I
+,
q
f(x) = I
+,
q
I
,
q
f(x). (5.4.12)
It should be mentioned here that in most of the proofs of [10, 11, 13], the domain where
the fractional integrals and the related properties hold is not determined precisely. A very
restrictive condition might be found in [10, p.366]. A q-analogue of the Riemann-Liouville
fractional(RLF) integral operator is introduced in [11] by Al-Salam through
I
q
f(x) :=
x
1
q
()
_
x
0
_
qt/x; q
_
1
f(t) d
q
t, (5.4.13)
,= 1, 2, . . .. This RLF integral is an extension of the basic Cauchy formula, [12],
I
n
q
f(x) =
_
x
a
_
x
n1
a
. . .
_
x
1
a
f(t) d
q
td
q
x
1
. . . d
q
x
n1
=
x
n1
q
(n)
_
x
a
_
qt/x; q
_
n1
f(t) d
q
t.
(5.4.14)
This basic analogue of RLF integral is also given by Agarwal in [10]. In fact Agarwal dened
the q-fractional derivative to be,
D
q
f(x) := I
q
f(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
f(t) d
q
t, > 0. (5.4.15)
128
Using (1.1.31), (5.4.13) reduces to
I
q
f(x) = x
(1 q)
n=0
q
n
(q
; q)
n
(q; q)
n
f(xq
n
). (5.4.16)
In the following we will derive basic analogue of the results mentioned above concerning
RLF calculus. These results will be used in our investigations below. From now on we
assume that > 0.
The following Theorem taken from [14, p.491] is needed in the proof of Lemma 5.4.2
below.
Theorem 5.4.1. Suppose and are real. Then
lim
q1
(q
x; q)
(q
x; q)
= (1 x)
, (5.4.17)
uniformly on
x C : [x[ 1 , if , + 1,
and uniformly on compact subsets of x C : [x[ 1, x ,= 1 for other choices of and .
Lemma 5.4.2. Let f be a function dened on [0, a], a > 0. If f is Riemann integrable on
[0, x], then
lim
q1
q
f(x) = I
x
1
_
qt/x; q
_
1
= lim
q1
x
1
_
qt/x; q
_
_
q
t/x; q
_
= x
1
(1
t
x
)
1
= (x t)
1
(5.4.19)
uniformly for 0 < t < x. Hence given > 0 there exists > 0 such that for all t
_
q
k
x, k N
_
, we have
x
1
_
qt/x; q
_
1
(x t)
1
_
2
129
whenever 0 < 1 q < . That is
q
f(x)
1
q
()
_
x
0
(x t)
1
f(t) d
q
t
2
q
()
(5.4.20)
and since
q
f(x) I
f(x)
q
f(x)
1
q
()
_
x
0
(x t)
1
f(t) d
q
t
q
()
_
x
0
(x t)
1
f(t) d
q
t I
f(x)
,
(5.4.21)
then the lemma follows if f is Riemann integrable on [0, x].
Lemma 5.4.3. If
lim
n
q
n
f(xq
n
) = 0 for all x (qa, a], (5.4.22)
then the fractional q-integral of order exists.
Proof. We apply Lemma 1.1.1 to (5.4.16) and use that lim
n
(q
n+1
; q)
1
= 1.
Remark 5.4.1. From the previous lemma we conclude that I
q
f(x) exists if and only if
f L
1
q
(0, x) for all x (qa, a], i.e. f L
1
q
(0, a).
Lemma 5.4.4. Let f : [0, a] C be a function. If f L
1
q
(0, a), then I
q
f L
1
q
(0, a).
Proof. Let f be a function dened on [0, a] such that f L
1
q
(0, a). Then from (5.4.13) we
get for x (qa, a]
_
x
0
[I
q
f(t)[ d
q
t =
1
q
()
_
x
0
t
1
_
t
0
_
qu/t; q
_
1
f(u) d
q
u
d
q
t
q
()
_
x
0
t
1
_
t
0
_
qu/t; q
_
1
[f(u)[ d
q
ud
q
t
=
x
+1
(1 q)
2
q
()
n=0
q
n
m=0
q
n+m
_
q
m+1
; q
_
1
[f(xq
n+m
)[
=
x
+1
(1 q)
2
q
()
n=0
q
n
k=n
q
k
_
q
kn+1
; q
_
1
[f(xq
k
)[
=
x
+1
(1 q)
q
()
k=0
q
k
[f(xq
k
)[
k
n=0
(1 q)q
n
_
q
kn+1
; q
_
1
. (5.4.23)
130
But from (5.3.7) we obtain
k
n=0
(1 q)q
n
_
q
kn+1
; q
_
1
= B
q
(1, )x
+1
k=0
q
k
(1 q)
_
q
k+1
; q
_
[f(xq
k
)[
=
x
q
()
B
q
(1, )
_
x
0
_
qt/x; q
_
[f(t)[ d
q
t
q
( + 1)
_
x
0
[f(t)[ d
q
t (5.4.24)
Combining (5.4.23) and (5.4.24) we conclude that I
q
f L
1
q
(0, x), for all x (qa, a]. That
is I
q
f L
1
q
(0, a).
Similarly to property (5.2.9) we have the following Lemma
Lemma 5.4.5. Let f L
1
q
(0, a). Then for x (qa, a] and t xq
m
, m N
lim
0
+
I
q
f(t) = f(t). (5.4.25)
Proof. Let x (qa, a] and t xq
m
, m N be xed. From (5.4.16) we have
lim
0+
I
q
f(t) = lim
0+
t
(1 q)
n=0
q
n
(q
; q)
n
(q; q)
n
f(tq
n
). (5.4.26)
Since
q
n
(q
; q)
n
(q; q)
n
[f(tq
n
)[ q
n
[f(tq
n
)[, n N, (0, 1], (5.4.27)
then from Weierstrass M-test, the series in (5.4.26) is uniformly convergent for (0, 1).
Thus
lim
0+
n=0
q
n
(q
; q)
n
(q; q)
n
f(tq
n
) = f(t). (5.4.28)
Hence
lim
0+
I
q
f(t) = lim
0+
t
(1 q)
lim
0+
n=0
q
n
(q
; q)
n
(q; q)
n
f(tq
n
) = f(t). (5.4.29)
From now on we use the notation I
0
q
f(x) = f(x). The property
I
q
I
q
f(x) = I
q
I
q
f(x) = I
(+)
q
f(x), , 0 (5.4.30)
131
is a q-analogue of the semigroup property (5.2.13) and it is proved by R.P. Agarwal [10].
As in the case of the RLF derivative, the existence of the fractional q-Riemann-Liouville
derivative holds only for a more restrictive class of functions. For this reason we study a
q-analogue of Abels integral equation.
Theorem 5.4.6. The q-Abel integral equation
x
1
q
()
_
x
0
_
qt/x; q)
1
(t) d
q
t = f(x), 0 < < 1, x (0, a] (5.4.31)
has a unique solution (x) L
1
q
(0, a) if and only if
(x) = D
q,x
I
1
q
f(x), (5.4.32)
and
I
1
q
f(x) /C
q
[0, a] and I
1
q
f(0) = 0. (5.4.33)
Proof. Assume that (5.4.31) has a solution L
1
q
(0, a). Hence replacing x with t and
t with u in (5.4.31), and multiplying both sides of the equation with x
_
qt/x; q
_
, and
then make the q-integration from 0 to x. Then, we have
x
q
()
_
x
0
_
qt/x; q
_
_
t
0
t
1
_
qu/t; q
_
1
(u) d
q
ud
q
t
= x
_
x
0
_
qt/x; q
_
f(t) d
q
t.
(5.4.34)
But from (5.3.17),
x
q
()
_
x
0
_
qt/x; q
_
_
t
0
t
1
_
qu/t; q
_
1
(u) d
q
ud
q
t
=
B(1 , )
q
()
_
x
0
(t) d
q
t.
(5.4.35)
Consequently,
I
q
(x) = I
1
q
f(x). (5.4.36)
132
Applying the operator D
q
on the two sides of (5.4.36) yields (5.4.32). Since L
q
(0, a),
then by Remark 5.3.2
I
1
q
f(0) := lim
n
x(qa,a]
I
1
q
f(xq
n
)
= lim
n
x(qa,a]
I
q
(xq
n
) = lim
n
x(qa,a]
k=n
xq
k
(1 q)(xq
k
) = 0.
(5.4.37)
Moreover from Lemma 5.3.7 I
1
q
f /C
q
[0, a].
Conversely assume that f satises (5.4.33) and (x) := D
q,x
I
1
q
f(x). Then
L
1
q
(0, a). We shall prove that is a solution of (5.4.31). Indeed let g be the function
dened by
x
1
q
()
_
x
0
_
qt/x; q
_
1
(t) d
q
t = g(x), x (0, a]. (5.4.38)
It suces to show g(x) = f(x), x (0, a]. Hence is a solution of (5.4.38). From the
sucient part, should have the form (5.4.32) with g instead of f. Thus
(x) = D
q,x
I
1
q
f(x) = D
q,x
I
1
q
g(x)
= D
q,x
x
q
(1 )
_
x
0
_
qt/x; q
_
g(t) d
q
t,
(5.4.39)
for all x (0, a]. Hence
D
q,x
_
I
1
q
_
f g
_
(x)
_
= 0, x > 0. (5.4.40)
Solving equation (5.4.40) we obtain
I
1
q
_
f g
_
(x) = c(x), x > 0, (5.4.41)
where c is a q-periodic function. The function I
1
q
g(x) /C
q
[0, a] because it satises
equation (5.4.36) with g(x) on the right hand side. Also, we have I
1
q
g(0) = 0 because
(5.4.38) is solvable equation. Thus I
1
q
_
f g
_
(x) /C
q
[0, a]. So, I
1
q
_
f g
_
is q-regular
at zero such that I
1
q
_
f g
_
(0) = 0. Consequently, c 0. That is,
x
q
(1 )
_
x
0
_
qt/x; q
_
(f(t) g(t)) d
q
t = 0, for all x > 0. (5.4.42)
133
The previous is an equation of the form (5.4.31). The uniqueness of its solutions leads to
f(x) = g(x), for all x (0, a].
From Theorem 5.2.2, the solution should be the fractional derivative of order of f.
Starting from the previous theorem we dene the basic fractional derivative as follows.
Denition 5.4.1. Let f be a function dened on [0, a]. For > 0 the fractional q-derivative
of order is dened to be
D
q
f(x) := (x) = D
k
q
I
k
q
f(x), k = [] + 1, (5.4.43)
provided that
f(x) L
1
q
(0, a), I
k
q
f(x) /C
(k)
q
[0, a]. (5.4.44)
The following lemma gives the relationship between the fractional q-derivative of order
and I
q
.
Lemma 5.4.7. Let f L
1
q
(0, a) and > 0, k = [] + 1, be such that I
k
q
f /C
(k)
q
[0, a].
Then
D
q
f(x) = I
q
f(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
f(t) d
q
t, x (0, a]. (5.4.45)
Proof. From (5.4.43) we have
D
q
f(x) = D
k
q
I
k
q
f(x) = D
k
q
_
x
k1
q
(k )
_
x
0
_
qt/x; q
_
k1
f(t) d
q
t
_
(5.4.46)
Set
h(t, x) :=
x
k1
q
(k )
_
qt/x; q
_
k1
f(t). (5.4.47)
Hence
h(xq
r
, xq
j
) = 0, r = 0, 1, . . . , j 1, j = 1, 2, . . . , k. (5.4.48)
Thus applying Lemma 5.3.4 to (5.4.46) we obtain
D
q
f(x) =
_
x
0
D
k
q,x
x
k1
q
(k )
_
qt/x; q
_
k1
f(t) d
q
t
=
x
1
q
()
_
x
0
_
qt/x; q
_
1
f(t) d
q
t = I
q
f(t)
(5.4.49)
134
Now we investigate some properties of the fractional q-integral and q-derivative. Similar
to property (5.2.15) we have the following.
Lemma 5.4.8. If f L
1
q
(0, a), then
D
q
I
q
f(x) = f(x); > 0, x (0, a]. (5.4.50)
Proof. It is obvious that if = n, n N, then D
n
q
I
n
q
f(x) = f(x). For a non integer and
positive , k 1 < k, k Z
+
, we have by applying the semi group identity (5.4.30)
that
D
q
I
q
f(x) = D
k
q
I
k
q
I
q
f(x) = D
k
q
I
k
q
f(x) = f(x), x (0, a].
The converse of (5.2.18) is the following
Lemma 5.4.9. Let R
+
and k := [] + 1. If
f L
1
q
(0, a) such that I
k
q
f /C
k
q
(0, a), (5.4.51)
then
I
q
D
q
f(x) = f(x)
k
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
, x (0, a]. (5.4.52)
Proof. Set
h(t, x) = x
(qt/x; q)
q
f(t), x (0, a], 0 < t x. (5.4.53)
Hence h(x, qx) = 0 for all x (0, a]. So applying Lemma 5.3.4 yields
I
q
D
q
f(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
D
q
f(t) d
q
t
= D
q,x
_
x
q
( + 1)
_
x
0
_
qt/x; q
_
q
f(t) d
q
t
_
.
(5.4.54)
135
Now, applying the q-integration by parts (1.1.41) k times on the last q-integral of (5.4.54),
we get
x
q
( + 1)
_
x
0
_
qt/x; q
_
q
f(t) d
q
t =
x
1
q
()
_
x
0
_
qt/x; q
_
1
D
k
q
I
k
q
f(t) d
q
t
=
k
j=1
x
j+1
q
( j + 2)
D
j
q,x
f(x)
x=0
+
x
k
q
( k + 1)
_
x
0
_
qt/x; q
_
k
I
k
q
f(t) d
q
t
=
k
j=1
x
j+1
q
( j + 2)
D
j
q,x
f(x)
x=0
+I
k+1
q
I
k
q
f(x).
(5.4.55)
Applying the semigroup identity (5.4.30) on (5.4.55) we obtain
x
q
( + 1)
_
x
0
_
qt/x; q
_
q
f(t) d
q
t =
k
j=1
x
j+1
q
( j + 2)
D
j
q,x
f(x)
x=0
+I
q
f(x). (5.4.56)
Thus we deduce (5.4.52) by computing the q-derivative of the two sides of (5.4.56) for all
x (0, a].
From (5.4.52) the equality
I
q
D
q
f(x) = f(x), for all x (0, a] (5.4.57)
holds if and only if
D
j
q
f(0) = 0, j = 1, 2, . . . , k. (5.4.58)
Lemma 5.4.10. Let f satisfy the conditions
f L
1
q
(0, a) and I
k
q
f /C
(k)
q
[0, a], (5.4.59)
where > 0 and k := [] + 1. Then
D
j
q
f(x) L
1
q
(0, a), j = 0, 1, 2, . . . , k 1. (5.4.60)
Moreover
D
j
q
f(x) /C
(j)
q
[0, a], j = 1, 2, . . . , k 1. (5.4.61)
136
Proof. The proof of (5.4.60) follows immediately from (5.4.59) and that of (5.4.61) follows
by noting that
D
j
q
f(x) = D
kj
q
I
kj(j)
q
f(x)
= D
kj
q
I
k
q
f(x) /C
(j)
q
[0, a], 1 j k 1.
(5.4.62)
In the following three lemmas we discuss the results of combining q-fractional integral
and derivative of not necessarily equal orders.
Lemma 5.4.11. If f L
1
q
(0, a), then
D
q
I
q
f(x) = I
q
f(x), 0, x (0, a]. (5.4.63)
If in addition D
q
f(x) exists in (0, a], then
D
q
I
q
f(x) = D
q
f(x), > 0. (5.4.64)
Proof. First, assume that . Then, = + ( ) and from (5.4.30) we obtain
D
q
I
q
f(x) = D
q
I
q
I
q
f(x) = I
q
f(x).
Now let < and m := [] + 1 and n := [ ] + 1. Then n m. Using (5.4.43) and
(5.4.30) we have for x (0, a]
D
q
I
q
f(x) = D
m
q
I
m
q
I
q
f(x) = D
m
q
I
mn
q
I
n+
q
f(x)
= D
n
q
I
n+
q
f(x) = D
q
f(x).
(5.4.65)
Lemma 5.4.12. Let f L
1
q
(0, a) such that I
n
q
f /C
(n)
q
[0, a], where > 0 and
n := [] + 1. Then for any 0 we have
I
q
D
q
f(x) = D
+
q
f(x)
n
j=1
D
j
q
f(x)[
x=0
x
j
q
( j + 1)
, x (0, a]. (5.4.66)
137
Proof. If = 0, then
q
(j + 1) = 0 for all j N and the identity (5.4.66) holds for any
> 0. Therefore we shall assume that > 0. We distinguish between two cases. First if
, then from (5.4.30) and (5.4.52) we obtain
I
q
D
q
f(x) = I
q
_
I
q
D
q
_
f(x)
= I
q
_
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
_
= I
q
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
,
(5.4.67)
for all x (0, a]. If > , then from (5.4.63) and (5.4.52) we obtain
I
q
D
q
f(x) = D
q
_
I
q
D
q
f(x)
_
= D
q
_
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
_
= D
q
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
,
(5.4.68)
for all x (0, a].
Lemma 5.4.13. If f L
1
q
(0, a) and I
n
q
f(x) /C
(n)
q
[0, a], where > 0 and n := [] +1,
then
D
q
D
q
f(x) = D
+
q
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(0), x (0, a], (5.4.69)
provided that D
+
q
f(x) exists for any > 0.
Proof. Let x (0, a]. From (5.4.52) and (5.4.64) we obtain
D
q
D
q
f(x)=D
+
q
I
q
D
q
f(x)=D
+
q
_
_
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
_
_
= D
+
q
f(x)
n
j=1
x
j
q
( j + 1)
D
j
q
f(x)
x=0
.
(5.4.70)
138
Remark 5.4.2. Similar to (5.4.69), if I
k
q
/C
(k)
q
[0, a] and D
+
q
f(x) exists, where k :=
[] + 1, then
D
q
D
q
f(x) = D
+
q
f(x)
k
i=1
D
i
q
f(x)[
x=0
x
i
q
( i + 1)
, x (0, a]. (5.4.71)
If
D
j
q
f(x)
x=0
= 0, j = 1, 2, . . . , n, (5.4.72)
and
D
j
q
f(x)[
x=0
= 0, j = 1, 2, . . . , k, (5.4.73)
then we have
D
q
D
q
= D
q
D
q
= D
+
q
. (5.4.74)
Similar to property (5.2.24) we have the following
Lemma 5.4.14. Let > 0, n := [] + 1. If f /C
(n)
q
[0, a], then for any x (0, a], we
have
D
q
f(x) =
n1
j=0
x
+j
q
( +j + 1)
D
j
q
f(0) +
x
n1
q
(n )
_
x
0
_
qt/x; q
_
n1
D
n
q
f(t) d
q
t. (5.4.75)
Moreover, D
q
f(0) = 0 if and only if D
j
q
f(0) = 0, for j = 0, 1, . . . , n 1.
Proof. Since f /C
(n)
q
[0, a], then from lemma 5.3.8 we obtain
f(x) =
n1
k=0
D
k
q
f(0)
q
(k + 1)
x
k
+
x
n1
q
(n)
_
x
0
_
qt/x; q
_
n1
D
n
q
f(t) d
q
t, x (0, a]. (5.4.76)
Consequently,
f(x) =
n1
k=0
D
k
q
f(0)
q
(k + 1)
x
k
+I
n
q
D
n
q
f(x), x (0, a]. (5.4.77)
139
Applying the operator D
q
to both sides of (5.4.77) and using (5.4.63) yields
D
q
f(x) =
n1
k=0
D
k
q
f(0)
q
(k + 1)
x
k
+D
q
I
n
q
D
n
q
f(x)
=
n1
k=0
D
k
q
f(0)
q
(k + 1)
x
k
+I
n
q
D
n
q
f(x)
=
n1
k=0
D
k
q
f(0)
q
(k + 1)
x
k
+
x
n1
q
(n )
_
x
0
_
qt/x; q
_
n1
D
n
q
f(t) d
q
t,
(5.4.78)
which is (5.4.75). The rest of the proof arises from (5.4.75).
In the following we dene a q-analogue of the Gr unwaldLetinkov fractional derivative.
If f is a function dened on [0, a], a > 0, then, from Property 5.3.3, D
k
q
f(x), k N, x (0, a]
is given by
D
k
q
f(x) =
1
x
k
(1 q)
k
j=k
j=0
(1)
j
_
k
j
_
q
f(q
j
x)
q
j(j1)/2+j(kj)
. (5.4.79)
Since the right hand side of (5.4.79) has a meaning when we replace k by any real number
, we dene the fractional operator
D
q
f(x) =
1
x
(1 q)
j=0
(1)
j
_
j
_
q
f(q
j
x)
q
j(j1)/2+j(j)
. (5.4.80)
Clearly D
q
f(x) = D
k
q
f(x), if = k N because
_
k
j
_
q
= 0 for all j > k. The operator
(5.4.80) is a q-analogue of the Gr unwaldLetinkov fractional operator. From (1.1.4)
(1)
j
_
j
_
q
q
j(j+1)/2j
= q
j
(q
; q)
j
(q; q)
j
, j N. (5.4.81)
Then
D
q
f(x) = x
(1 q)
j=0
(1)
j
q
j(j+1)/2
q
j
_
j
_
q
f(q
j
x)
= x
(1 q)
j=0
q
j
(q
; q)
j
(q; q)
j
f(q
j
x).
(5.4.82)
Therefore, from (5.4.16), the fractional operator D
q
, R, is the same as Riemann
Liouville fractional operator D
q
.
Now we introduce a q-analogue of the Caputo fractional derivative dened in (5.2.27).
140
Denition 5.4.2. For > 0, the Caputo fractional q-derivative of order is dened to be
c
D
q
f(x) := I
k
q
D
k
q
f(x), k = [] + 1. (5.4.83)
Clearly
c
D
q
f(x) exists if f AC
(k)
q
[0, a]. Obviously
c
D
q
c = 0 for any constant c C.
We shall study existence and uniqueness of solutions of systems of fractional order of
basic Caputo derivative in the Section 7.2.
Chapter 6
Basic Mittag-Leer Functions
The two parameter Mittag-Leer functions play an important role in the theory of fractional
calculus and fractional dierential equations, see e.g. [24, 40, 100]. In fact they are solutions
of fractional integro dierential equations. In this chapter we derive a basic analogue of
Mittag-Leer function and study some of their properties. In particular the zeros of the
basic Mittag-Leer functions will be investigated.
6.1 Basic Mittag-Leer function
The special function of the from
E
(z) =
n=0
z
n
(
n
+ 1)
; > 0, (6.1.1)
and more general functions
E
,
(z) =
n=0
z
n
(
n
+)
; > 0, C, (6.1.2)
where z C, are known as Mittag-Leer functions, see e.g. [45, 65, 76]. The function E
(z)
which denes a family of exponential functions of one parameter was rst introduced by
Mittag-Leer [97] who in [97, 98] investigated some of its properties, while E
,
(z) was rst
introduced in [118]. It is an entire function of z of order and type 1. See [39, 40, 54, 100]
141
142
for properties and historical notes. The following group of identities follows directly from
the denition of E
,
(z).
E
1/2,1
(z) = cosh
z, E
1/2,2
(z) =
sinh
z
, (6.1.3)
E
1,1
(z) = e
z
, E
1,2
(z) =
e
z
1
z
, (6.1.4)
E
,
(z) =
1
()
+zE
,+1/
(z), (6.1.5)
E
,
(z) = E
,+1
(z) +
z
,+1
(z),
:=
d
dz
(6.1.6)
Furthermore, the termwise integration of the power expansion of the function E
,
(z) along
the interval (0, z) leads to the integral relation
_
z
0
E
,
(t
1
)t
1
dt = z
E
,+1
(z
1/
) (6.1.7)
and its generalization
1
()
_
z
0
(z t)
1
E
,
(t
1
)t
1
dt = z
+1
E
,+
(z
1/
), > 0, > 0. (6.1.8)
In particular, when > 0, we have
1
()
_
z
0
(z t)
1
cosh
t dt = z
E
1/2,+1
(z
2
), (6.1.9)
1
()
_
z
0
(z t)
1
sinh
dt = z
1+
E
1/2,2+
(z
2
), (6.1.10)
1
()
_
z
0
(z t)
1
e
t
dt = z
E
1,1+
(z), (6.1.11)
cf. [40]. Since we have two major q-exponential functions, namely, e
q
(z) and E
q
(z), then
we will dene two q-analogues of the Mittag-Leer functions. We dene
e
,
(z; q) :=
n=0
z
n
q
(n +)
, [z[ < (1 q)
,
E
,
(z; q) :=
n=0
q
n(n1)
2
z
n
q
(n +)
, z C,
(6.1.12)
143
where > 0, C. When = 1, e
,1
(z; q) and E
,1
(z; q) dene a family of q-exponential
functions of one parameter. Another one-parameter family of q-exponential functions is
dened by, cf. [22, 50, 65],
E
()
q
(z) :=
k=0
q
k
2
/2
(q; q)
k
z
k
, C. (6.1.13)
Similarly to formulae (6.1.3)(6.1.6), the following formulae follow at once from the deni-
tion of e
,
(z; q):
e
2,1
(z; q) = cosh
q
z(1 q), e
2,2
(z; q) =
sinh
q
z(1 q)
z
, (6.1.14)
e
1,1
(z; q) = e
q
(z(1 q)), e
1,2
(z; q) =
e
q
_
z(1 q)
_
1
z
, (6.1.15)
e
,
(z; q) =
1
q
()
+ze
+1,
(z; q), (6.1.16)
q
zD
q
e
,+1
(z
; q) =
1 q
1 q
e
,+1
(z
; q) +e
,
(z
; q). (6.1.17)
The following group of formulae follow at once from the denition of E
,
(z; q).
E
2,1
(z; q) = cosh(q
1/2
z; q), E
2,2
(z; q) =
sinh(q
1
z; q)
q
1
z
, (6.1.18)
E
1,1
(z; q) = E
q
(z(1 q)), E
1,2
(qz; q) =
1
z
(1 +E
q
(z(1 q))), (6.1.19)
E
,
(z; q) =
1
q
()
+zE
,+
(qz, q), > 0, (6.1.20)
E
,
(z
; q) =
1 q
1 q
E
,+1
(z
; q) +q
zD
q
E
,+1
(z
; q). (6.1.21)
Lemma 6.1.1. Let , , be positive numbers. Then
z
1
q
()
_
z
0
_
qt/z; q
_
1
E
,
(t
; q)t
1
d
q
t = z
+1
E
,+
(z
; q). (6.1.22)
Proof. Replace E
,
(t
q
()
_
z
0
_
qt/z; q
_
1
E
,
(t
; q)t
1
d
q
t
=
z
1
q
()
_
z
0
_
qt/z; q
_
1
t
1
_
n=0
q
n(n1)
_
2
_
t
_
n
q
(n +)
_
d
q
t
=
z
1
q
()
n=0
q
n(n1)
_
2
n
q
(n +)
_
z
0
_
qt/z; q
_
1
t
n+1
d
q
t. (6.1.23)
Substituting with u := t/z, z ,= 0, in the q-integral of (6.1.23) and using (1.1.12) yields
_
z
0
_
qt/z; q
_
1
t
n+1
d
q
t = z
n+
_
1
0
_
qu; q
_
1
u
n+1
d
q
u
= z
n+
B
q
_
, n +
_
.
(6.1.24)
Consequently
z
1
q
()
n=0
q
n(n1)
_
2
n
q
(n +)
_
z
0
_
qt/z; q
_
1
t
n+1
d
q
t
=
z
1
q
()
n=0
q
n(n1)
_
2
n
z
n+
q
(n +)
B
q
_
, n +
_
=
n=0
q
n(n1)
_
2
n
z
n++1
q
(n + +)
= z
+1
E
,+
(z
; q),
(6.1.25)
this completes the proof.
Corollary 6.1.2. Let , , be positive numbers. Then
_
z
0
E
,
(t
; q)t
1
d
q
t = z
E
,+1
(z
; q), (6.1.26)
z
1
q
()
_
z
0
_
qt/z; q
_
1
cosh(
t; q) d
q
t = z
E
2,+1
(z
2
q; q), (6.1.27)
z
1
q
()
_
z
0
_
qt/z; q
_
1
sinh(q
1
t; q)
d
q
t = qz
+1
E
2,2+
(q
2
z
2
; q), (6.1.28)
z
1
q
()
_
z
0
_
qt/z; q
_
1
E
q
(t(1 q)) d
q
t = z
E
1,+1
(z; q). (6.1.29)
Proof. Formula (6.1.26) follows from (6.1.22) by taking = 1. Formula (6.1.27) follows
from (6.1.22) by taking = 2, = 1, and from (6.1.18) by replacing z in E
1,2
(z; q) by t
2
.
145
As for formula (6.1.28) it follows from (6.1.22) by taking = = 2 and using (6.1.18) when
z in E
2,2
(z; q) is replaced by t
2
. Finally formula (6.1.29) follows from (6.1.22) by taking
= = 1 and using (6.1.19) when z is replaced by t.
Lemma 6.1.3. E
,
(z; q) is an entire function of order zero.
Proof. From (1.2.4) we have
_
E
,
(z; q)
_
= limsup
n
nlog n
log(1/[c
n
[)
, (6.1.30)
where c
n
is the coecient of z
n
in (6.1.12), see [27], i.e.
c
n
=
q
n(n1)
2
q
(n +)
, n N. (6.1.31)
Hence,
log
_
1
[c
n
[
_
=
2
n(n 1) log q + log [
q
(n +)[ .
From (1.1.2) and the denition of the qGamma function, cf. (1.1.10) we obtain for n N,
R
+
, and C
log [
q
(n +)[ = log
(q; q)
(q
n+
; q)
(1 q)
1n
= log
(q; q)
(q
n+
; q)
(1 q)
1n
= log(q; q)
(q
n+
; q)
,
(6.1.32)
and
log
(q
n+
; q)
= log
_
k=0
1 q
n++k
_
= log
_
lim
m
m
k=0
1 q
n++k
_
= lim
m
m
k=0
log
1 q
n++k
k=0
log
1 q
n++k
.
(6.1.33)
Since
log
1 q
n++k
log
_
1 +[q
n++k
[
_
[q
n++k
[ = q
n+k+
,
(6.1.34)
146
then
k=0
log
1 q
n++k
k=0
q
n+k+
=
q
n+
1 q
. (6.1.35)
Therefore lim
n
log [(q
n+
; q)
[
nlog n
= 0. Consequently by (6.1.32) lim
n
log [
q
(n +)[
nlog n
= 0.
Since lim
n
2
n(n 1) log q
nlog n
= , then lim
n
log(1/|c
n
|)
nlog n
= , i.e.
_
E
,
(z; q)
_
= 0.
Remark 6.1.1. From Theorem 1.2.1 we conclude that E
,
(z; q) has innitely many zeros.
In Section 6.3 we prove that for specic values of and , E
,
(z; q), 0 < q < 1, may have
only a nite number of complex zeros. Moreover if q satises an additional relation, then
the zeros of E
,
(z; q) are all real.
6.2 Basic Volterra integral equations
In this section we investigate the solutions of q-analogues of a Volterra integral equation of the
second kind. Actually we introduce basic analogues of the following theorem of [40, p.3].
Theorem 6.2.1. Let > 0 and C. If f L
1
(0, a), 0 < a < , then the integral
equation
u(x) = f(x) +
(1
_
)
_
x
0
(x t)
1/1
u(t)dt, x (0, a) (6.2.1)
has a unique solution in L
1
(0, a) given by the formula
u(x) = f(x) +
_
x
0
(x t)
1/1
E
_
(x t)
1/
; 1/
_
f(t) dt, x (0, a). (6.2.2)
The following theorem gives explicit solutions of a basic Volterra integral equation of
the second kind in terms of the translation operator dened by Ismail [65], cf. (5.3.2).
Theorem 6.2.2. Let , a > 0 and C be such that
[[a
(1 q)
< 1 (6.2.3)
147
If f L
1
q
(0, a), then the q-integral equation
u(x) = f(x) +
x
1
q
()
_
x
0
_
qt/x; q
_
1
u(t) d
q
t, x (0, a], (6.2.4)
has the unique solution
u(x) =
k=0
k
I
k
q
f(x) = f(x) +x
1
_
x
0
_
qt/x; q
_
1
t
e
,
(x
; q)f(t) d
q
t, (6.2.5)
x (0, a] in the space L
1
q
(0, a).
Proof. We prove the theorem in four steps.
i. First we prove the existence of a solution by using q-analogue of the method of successive
approximations. Dene a sequence u
m
m=0
recursively by
u
0
(x) = f(x),
u
m
(x) = f(x) +
x
1
q
()
_
x
0
_
qt/x; q
_
1
u
m1
(t) d
q
t, x (0, a], m 1.
(6.2.6)
We prove by mathematical induction that
u
m
(x) =
m
k=0
k
I
k
q
f(x), m N, x (0, a]. (6.2.7)
Indeed if u
m
satises the identity (6.2.7), then from (6.2.6) we obtain
u
m+1
(x) = f(x) +
x
1
q
()
_
x
0
_
qt/x; q
_
1
m
k=0
k
I
k
q
f(t) d
q
t
= f(x) +
m
k=0
k+1
x
1
q
()
_
x
0
_
qt/x; q
_
1
I
k
q
f(t)
= f(x) +
m
k=0
k+1
I
q
I
k
q
f(x), (6.2.8)
for all x (0, a]. Hence from the semigroup identity (5.4.30) we have, x (0, a],
u
m+1
(x) = f(x) +
m
k=0
k+1
I
k+
q
f(x) =
m+1
k=0
k
I
k
q
f(x). (6.2.9)
148
Thus (6.2.7) is true for all m N because it is true at m = 0. Now we prove that the series
k=0
k
I
k
q
f(x) is absolutely convergent for all x (0, a] and denes a function u(x) on
(0, a]. Take y (qa, a] and t yq
m
, m N. Since
k
I
k
q
f(t)
[[
k
t
k1
q
(k)
r=0
tq
r
(1 q)
_
q
r+1
; q
_
k1
[f(tq
r
)[
[[
k
y
k1
(q
; q)
q
(k)
_
y
0
[f(u)[ d
q
u,
(6.2.10)
k Z
+
, and the series
k=1
k
y
k1
q
(k)
= y
1
k=0
k
y
k
q
(k +)
= y
1
e
,
(y
; q),
is convergent for all y a and a satises the condition (6.2.3), then the series
k=0
k
I
k
q
f(t)
is uniformly convergent on yq
m
, m N, y (qa, a]. That is it is convergent on (0, a].
Thus the function u(x) is dened for x (0, a] by
u(x) :=
k=0
k
I
k
q
f(x). (6.2.11)
ii . Now we prove that u L
1
q
(0, a), i.e. u L
1
q
(0, y) for all y (qa, a]. Indeed from (6.2.11)
and the semigroup identity (5.4.30) we obtain
_
y
0
[u()[ d
q
= I
q
[u[(y)
k=0
[[
k
I
k+1
q
f(y)
. (6.2.12)
But from (6.2.10)
k=0
[[
k
I
k+1
q
f(y)
k=0
[[
k
_
y
0
I
k
q
f(t)
d
q
t
_
y
0
[f(u)[ d
q
u
(q
; q)
k=0
[[
k
y
k
[
q
(k + 1)[
=
_
y
0
[f(u)[ d
q
u
(q
; q)
e
,1
([[y
; q)
_
y
0
[f(u)[ d
q
u
(q
; q)
e
,1
([[a
; q),
(6.2.13)
149
for all y (qa, a]. Since a satises the condition (6.2.3) and f L
1
q
(0, y), for all y (qa, a],
then so is u.
iii. To prove the uniqueness assume that u and v are solutions of (6.2.4) in L
1
q
(0, a) valid
in (0, h], h a. Set
z(x) u(x) v(x), x (0, h].
Then, z(x) satises the functional equation
z(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
z(t) d
q
t. (6.2.14)
Since z L
1
q
(0, a), then z L
1
q
(0, h), i.e.
|z
0
|
w
:=
_
w
0
[z()[ d
q
< , w (qh, h]. (6.2.15)
Consequently from (6.2.14) we obtain
[z()[
[[
q
()(q
; q)
|z
0
|
w
1
, wq
m
, m N . (6.2.16)
Applying (6.2.16) to (6.2.14) to estimate the value of z(), we get
[z()[
[[
2
2
q
()(q
; q)
|z
0
|
w
21
B
q
(, ) =
[[
2
q
(2)(q
; q)
|z
0
|
w
21
.
Repeating the estimate n times, we obtain
[z()[
[[
n
q
(n)(q
; q)
|z
0
|
w
n1
, n Z
+
, wq
m
, m N , w (qh, h]. (6.2.17)
Since
|z
0
|
w
n=1
[[
n
q
(n)(q
; q)
n1
=
1
(q
; q)
|z
0
|
w
n=0
[[
n
q
(n +)
n
=
1
(q
; q)
|z
0
|
w
e
,
([[
; q),
(6.2.18)
then
lim
n
q
(n +)
n
= 0, wq
m
, m N, w (qh, h] .
150
Consequently,
z() = 0, wq
m
, m N, w (qh, h] .
That is z(x) = 0 for all x (0, h] and the uniqueness of the solution follows.
iv. Now we prove that u can be represented by the second identity in (6.2.5). Since
u(x) =
k=0
I
k
q
f(x) = f(x) +
k=1
k
x
k1
q
(k)
_
x
0
_
qt/x; q
_
k1
f(t) d
q
t (6.2.19)
= f(x) +x
1
k=0
k
x
k
q
(k +)
_
x
0
_
qt/x; q
_
k+1
f(t) d
q
t (6.2.20)
and
(qt/x; q)
k+1
= (qt/x; q)
1
(q
t/x; q)
k
,
then
u(x) = f(x) +x
1
_
x
0
_
k=0
k
x
k
q
(k)
_
q
t/x; q
_
k
_
_
qt/x; q
_
1
f(t) d
q
t
u(x) = f(x) +x
1
_
x
0
_
qt/x; q
_
1
t
e
,
(x
)f(t) d
q
t.
(6.2.21)
where the qtranslation operator
y
is dened in (5.3.3).
Theorem 6.2.3. Let , a > 0 and C. If f L
1
q
(0, a), then the q-integral equation
u(x) = f(x) +
x
1
q
()
_
x
0
_
qt/x; q
_
1
u(qt) d
q
t, (6.2.22)
has the unique solution
u(x) = f(x) +q
x
1
_
x
0
_
qt/x; q
_
1
j=0
j
q
j(j1)
2
q
(j +)
_
q
t/x; q
_
j1
f(q
j+1
t) d
q
t, (6.2.23)
x (0, a], in the space L
1
q
(0, a).
151
Proof. First we prove the existence of a solution of (6.2.23). Dene a sequence u
m
(x)
m=0
recursively by
u
0
(x) = f(x),
u
m
(x) = f(x) +
x
1
q
()
_
x
0
_
qt/x; q
_
1
u
m1
(qt) d
q
t, x (0, a], m 1.
(6.2.24)
We can prove by mathematical induction that
u
m
(x) =
m
j=0
j
q
j(j1)
2
I
j
q
f(q
j
x), m N. (6.2.25)
Indeed from (6.2.24) we obtain
u
m+1
(x) = f(x) +I
q
(f(qx))
+
x
1
q
()
_
x
0
(qt/x; q)
1
m
j=1
j
q
j(j1)
2
_
I
j
q
_
f(q
j
x)
_
(x)
x=qt
d
q
t
= f(x) +I
q
_
f(qx)
_
+
x
1
q
()
m
j=1
j
q
j(j+1)
2
1
q
(j)
_
x
0
(qt/x; q)
1
t
j1
_
qt
0
(u/t; q)
j1
f(q
j
u) d
q
ud
q
t.
But from Lemma 5.3.5
_
x
0
(qt/x; q)
1
t
j1
_
qt
0
(u/t; q)
j1
f(q
j
u) d
q
ud
q
t
= q
_
x
0
(qt/x; q)
1
t
j1
_
t
0
(q/t; q)
j1
f(q
j+1
) d
q
d
q
t
= qB
q
(, j)x
j
_
x
0
(qt/x; q)
(j+1)1
f(q
j+1
t) d
q
t.
(6.2.26)
Thus
u
m+1
(x) = f(x) +I
q
_
f(qx)
_
+
m
j=1
j+1
q
j(j+1)
2
I
(j+1)
q
f(q
j+1
x)
=
m+1
j=0
j
q
j(j1)
2
I
j
q
f(q
j
x).
(6.2.27)
This proves (6.2.25) because it holds at m = 0. Now we prove that the series
m
j=0
j
q
j(j1)
2
I
j
q
f(q
j
x)
152
is absolutely convergent for all x (0, a] and denes a function u(x) on (0, a]. Take y
(qa, a] and t yq
m
, m N. Since
(q
n+1
; q)
j1
(q
n+1
; q)
(q
n+
; q)
< (q
; q)
1
, n N, > 0, j 1, (6.2.28)
and f L
1
q
(0, a), then for all y (qa, a] and j 1 we obtain
I
j
q
_
f(q
j
y)
_
=
y
j1
q
(j)
_
y
0
_
qt/y; q
_
j1
f(q
j
t) d
q
t
(q
; q)
1
|f|
y
y
j1
q
(j)
. (6.2.29)
Hence
j=1
j
q
j(j1)
2
I
j
q
_
f(q
j
y)
_
|f|
y
(q
; q)
1
j=1
[[
j
q
j(j1)
2
y
j1
q
(j)
= |f|
y
(q
; q)
1
j=0
[[
j+1
q
j(j+1)
2
y
j+1
q
(j +)
= |f|
y
(q
; q)
1
[[y
1
E
,
(q
[[y
; q).
Since the basic Mittag-Leer function E
,
(q
[[y
k=0
k
I
k
q
f(t)
is uniformly convergent on yq
m
, m N, y (qa, a].
That is it is convergent on (0, a]. Thus the function u(x) is dened for x (0, a] by
u(x) :=
j=0
j
q
j(j1)
2
I
j
q
f(q
j
x). (6.2.30)
Then the series u(x) is the limit of (6.2.25) as m . The proof of u is the unique solution
in L
1
q
(0, a) is similar to that of Theorem 6.2.2 and it is omitted.
6.3 Zeros of basic Mittag-Leer functions
In this section we investigate the zeros of the basic Mittag-Leer functions E
,
(z; q) when
0 < q < 1, and when q satises an additional relation. The zeros will be investigated
for specic values for the parameters and . Actually we shall give an analogue of the
following theorem, cf. [40, p.7].
153
Theorem 6.3.1. Let
c
(z; ) = E
1/2
(
2
z; 1 +), R
+
. (6.3.1)
Then
1. If 0 < 2, then the zeros of
(z
2
; ) are simple, real and they are symmetric with
respect to the point z = 0.
2. If 0 < < 1, the all zeros of the function
(1)
k
=
_
k
2
,
k +
2
_
, k Z
, (6.3.2)
one zero in each interval and if = 0, then the zeros of this function are the endpoints
of these intervals.
3. If 1 < < 2, then the zeros of c
(z
2
; ) are situated in the intervals
(2)
k
=
_
_
_
_
k,
(k + 1)
_
, 1 k < ,
_
(k 1),
k
_
, < k 1,
(6.3.3)
one zero in each interval and if = 1, then they are the endpoints of these intervals.
Consider the function
c
,
(z; q) := E
2,1+
(
2
z; q), z C, R
+
, (6.3.4)
and recall that x
n
n=1
and 0, y
n
n=1
, x
n
, y
n
> 0, denote the zeros of cos(z; q) and
sin(z; q). Then we have the following theorem
Theorem 6.3.2. Let 0 < 2. If ,= 0 ,= 1. Then
1. For any q (0, 1) the function c
,
(z
2
; q) has at most a nite number of complex
zeros and it has an innite number of real, simple and symmetric zeros,
_
()
n
_
n=1
,
()
n
> 0, with the asymptotic behavior
()
n
=
_
_
q
1/2
x
n
_
1 +O(q
n
)
_
, 0 < < 1,
qy
n
_
1 +O(q
n
)
_
, 1 < < 2,
(6.3.5)
as n .
154
2. If q-satises the condition
q
1
(1 q)(1 q
+1
)(1 q
+2
) > 1, (0, 2), ,= 1, (6.3.6)
then the zeros of c
,
(z
2
; q) are real, simple and symmetric with respect to the point
z = 0 such that for n Z
+
()
n
_
_
_
q
n+3/2
_
(1 q
+1
)(1 q
+2
)
(1 q)
,
q
n+1/2
_
(1 q
+1
)(1 q
+2
)
(1 q)
_
, (0, 1),
_
q
n+5/2
_
(1 q
+1
)(1 q
+2
)
(1 q)
,
q
n+3/2
_
(1 q
+1
)(1 q
+2
)
(1 q)
_
, (1, 2).
(6.3.7)
3. If 0, 1, then c
,
(z
2
; q) has only real, simple and symmetric zeros such that
(0)
n
:=
q
1/2
x
n
,
(1)
n
:=
qy
n
, n = 1, 2, . . . . (6.3.8)
Proof. If 0, 1, then from (6.1.18)
c
0,
(z
2
; q) = cos(q
1/2
z; q), c
1,
(z
2
; q) =
sin(q
1
z; q)
q
1
z
. (6.3.9)
Since the zeros of cos(; q) and sin(; q) are real and simple, then the desired statement
follows for 0, 1. From formula (6.1.22) with = 2, =
2
we have
E
2,+
(
2
z
2
; q) =
z
q
()
_
z
0
_
qt/z; q
_
1
E
2,
_
2
t
2
; q
_
t
1
d
q
t. (6.3.10)
Substitute with u := t/z, z ,= 0, on the qintegral of (6.3.10). Then
_
z
0
_
qt/z; q
_
1
E
2,
_
2
t
2
; q
_
t
1
d
q
t = z
_
1
0
(qu; q)
1
E
2,
(
2
u
2
z
2
; q)t
1
d
q
t.
(6.3.11)
Thus (6.3.10) is nothing but
E
2,+
(
2
z
2
; q) =
_
1
0
(qu; q)
1
E
2,
(
2
u
2
z
2
; q)t
1
d
q
t. (6.3.12)
155
If 0 < < 1, substituting in (6.3.12) with = 1 and = yields
c
,
(z
2
; q) =
1
q
()
_
1
0
(qt; q)
1
cos(q
1/2
tz; q) d
q
t. (6.3.13)
If 1 < < 2, substituting in (6.3.12) with = 2 and = 1 yields
q
1
zc
,
(z
2
; q) =
1
q
( 1)
_
1
0
(qt; q)
2
sin(q
1
tz; q) d
q
t. (6.3.14)
For (0, 1) let f
(t) :=
(qt; q)
1
q
()
. (6.3.15)
Since
f
(q
k
)
f
(q
k+1
)
=
1 q
k+1
1 q
k++1
> 1, k N,
then
f
(1) > f
(q) > f
(q
2
) > . . . > f(q
n
) > . . . > f(0) = 1.
That is f
(t) d
q
t = B
q
( + 1, 1) =
1 q
1 q
+1
.
Hence applying Theorem 1.4.2 on (6.3.13) and Theorem 1.4.3 on (6.3.14) we conclude that
for any q (0, 1), c
,
(z
2
; q) has at most a nite number of non-real zeros, and it has an
innite number of real, simple and symmetric zeros such that
_
()
n
_
satises the asymptotic
properties (6.3.5). From (2.1.3) and (2.1.4)we have
c
k,1/2,f
1
=
1
(1 q
2k++1
)(1 q
2k++2
)
, 0 < < 1, (6.3.16)
b
k,1/2,f
2
=
1
(1 q
2k++1
)(1 q
2k++2
)
, 1 < < 2, (6.3.17)
for all k N. Consequently
c
1/2,f
1
= 1, C
1/2,f
1
=
1
(1 q
+1
)(1 q
+2
)
, (0, 1), (6.3.18)
156
b
1/2,f
2
= 1, C
1/2,f
2
=
1
(1 q
+1
)(1 q
+2
)
, (1, 2). (6.3.19)
Thus the condition (2.3.9) of Theorem 2.3.3 and condition (2.3.13) of Theorem 2.3.4 are
nothing but the condition (6.3.6). Therefore the proof of the point 2 of the theorem follows
from Theorems 2.3.3 and 2.3.4 of Section 2.3.
Chapter 7
Basic Fractional Dierence
Equations
As in the classical theory of ordinary fractional dierential equations, q-dierence equations
of fractional order are divided into linear, nonlinear, homogeneous, and inhomogeneous
equations with constant and variable coecients. This chapter is devoted to study certain
problems of fractional q-dierence equations based on the basic Riemann-Liouville fractional
derivative and the basic Caputo fractional derivative. We shall study questions concerning
the solvability of these equations in a certain space of functions. Also we shall solve an equa-
tion of fractional order with constant coecients by using qanalogue of Laplace transform.
A solution for a fractional integro-dierential equation q-Cauchy problem is given.
7.1 Basic RiemannLiouville fractional order sys-
tems
In this section we deal with two systems of basic RiemannLiouville fractional derivatives,
namely the system
D
q
y
i
(x) = f
i
(x, y
1
(x), . . . , y
n
(x)), i = 1, 2, . . . , n, (7.1.1)
157
158
and the system
D
q
y
i
(x) = f
i
(qx, y
1
(qx), . . . , y
n
(qx)), i = 1, 2, . . . , n, (7.1.2)
where 0 < < 1. System (7.1.1) together with the initial conditions
D
1
q
y
i
(x)[
x=0
= b
i
, b
j
R, i = 1, 2, . . . , n, (7.1.3)
is called an initial value problem. By a solution of the initial value problem (7.1.1), (7.1.3)
we mean a set of functions y
i
n
i=1
that satises (7.1.1) and the initial conditions (7.1.3)
in an interval I containing zero such that x
1
y
i
, i = 1, 2, . . . , n, are continuous at zero.
Similarly we can dene the solution of the initial value problem (7.1.2)(7.1.3). We begin
this section with the existence and uniqueness of solutions of the initial value problem
(7.1.1), (7.1.3).
Theorem 7.1.1. Let f
i
(x, y
1
, . . . , y
n
) be functions dened for x (0, a] and y
i
in a domain
G
i
, satisfying the following conditions
(i) there is a positive constant A such that, for x (0, a] and y
i
, y
i
G
i
, 1 i n, the
following Lipschiz condition is fullled
[f
i
(x, y
1
, . . . , y
n
) f
i
(x, y
1
, . . . , y
n
)[ A([y
1
y
1
[ +. . . +[y
n
y
n
[) . (7.1.4)
(ii) There exists M > 0 such that
[f
i
(x, y
1
, . . . , y
n
)[ M, y
i
G
i
, i = 1, . . . , n, x (0, a]. (7.1.5)
Let 0 < < 1 and a, K be such that
0 < a
1
(An)
1
(1 q)
and K
Ma
q
( + 1)
. (7.1.6)
Let D
i
(a, K) G
i
be the set of points y
i
G
i
satisfying the relation
[y
i
(x) b
i
x
i
1
q
(
i
)
[ < K, for all x (0, a]. (7.1.7)
159
Then, the initial value problem (7.1.1), (7.1.3) has a unique solution y
i
(x)
n
i=1
valid in
(0, a], y
i
D
i
(a, K), i = 1, 2, . . . , n.
Proof.
Existence. Dene sequences
i,m
m=1
, i = 1, . . . , n, by the following relationships
i,1
(x) =
b
i
x
1
q
()
,
i,m
(x) =
x
1
q
()
_
b
i
+
_
x
0
_
qt/x; q
_
1
f
i
_
t,
1,m1
(t), . . . ,
n,m1
(t) d
q
t
_
, (7.1.8)
where m 2. We will show that lim
m
i,m
(x) exists and gives the required solution
n
i=1
of the initial value problem (7.1.1)(7.1.3). We prove the existence in four steps.
i. We prove by induction on m that
i,m
D
i
(a, k), m Z
+
, i = 1, 2, . . . , n. (7.1.9)
Clearly,
i,1
D
i
(a, K) for i = 1, . . . , n. If we assume that
i,m
D
i
(a, k) for i = 1, . . . , n,
then by (7.1.5) we obtain
[f
i
(x,
1,m
(x), . . . ,
n,m
(x))[ M, x (0, a]. (7.1.10)
Thus by (1.1.12)
[
i,m+1
(x)
b
i
q
()
x
1
[
x
1
q
()
_
x
0
_
qt/x; q
_
1
[f
i
(t,
1,m
(t), . . . ,
n,m
(t))[ d
q
t
Mx
1
q
()
_
x
0
_
qt/x; q
_
1
d
q
t =
M
q
()
B
q
(, 1)x
=
Mx
q
( + 1)
Ma
q
( + 1)
K. (7.1.11)
So,
i,m+1
D
i
(a, K). This completes the induction steps and prove (7.1.9).
ii. We prove that x
1
i,m
is continuous at zero for all m N, i = 1, 2, . . . , n. From (7.1.9)
we conclude that
i,m
(x)
b
i
q
()
x
1
[
K, m Z
+
, x (0, a]. (7.1.12)
160
Consequently
x
1
i,m
(x)
b
i
q
()
[
Kx
1
, m Z
+
, x (0, a]. (7.1.13)
Therefore x
1
i,m
is continuous at zero for all m N, i = 1, 2, . . . , n.
iii. We prove by induction on m that
[
i,m+1
(x)
i,m
(x)[
MB
m1
x
(m)
q
(m + 1)
, B := An, m Z
+
. (7.1.14)
From (7.1.11), inequality (7.1.14) is true at m = 1. Assume that (7.1.14) is true at m = k,
that is,
[
i,k+1
(x)
i,k
(x)[
MB
k1
x
k
q
(k + 1)
, i = 1, 2, . . . , n. (7.1.15)
Consequently, we have
[
i,k+2
(x)
i,k+1
(x)[
x
1
q
()
_
x
0
_
qt/x; q
_
1
[f
i
(t,
1,k+1
(t), . . . ,
n,k+1
(t)) f
i
(t,
1,k
(t), . . . ,
n,k
(t))[ d
q
t
Ax
1
q
()
_
x
0
_
qt/x; q
_
1
n
j=1
[
j,k+1
(t)
j,k
(t)[ d
q
t
=
MB
k
x
1
q
()
q
(k + 1)
_
x
0
_
qt/x; q
_
1
t
k
d
q
t
=
MB
k
q
()
q
(k + 1)
B
q
(, k + 1)x
(k+1)
=
B
k
M
q
_
(k + 1) + 1
_x
(k+1)
.
That is (7.1.14) is true at m = k + 1 and hence it is true for all m 1.
iv. We prove that lim
m
i,m
exists for i = 1, 2, . . . , n, x (0, a] such that
i
n
i=1
,
i
(x) := lim
m
i,m
(x), i = 1, 2, . . . , n, x (0, a],
denes a solution of (7.1.1), (7.1.3). Consider the innite series
i,1
(x) +
m=1
i,m+1
(x)
i,m
(x). (7.1.16)
161
From (7.1.14) we obtain
m=1
[
i,m+1
(x)
i,m
(x)[
M
B
m=1
(Bx
)
m
q
(m + 1)
M
B
m=0
(Bx
)
m
q
(m + 1)
=
M
B
e
,
(Bx
; q)
M
B
e
,1
(Ba
; q).
(7.1.17)
But e
,1
(Ba
B < (1 q)
i,m
(x). Since x
1
i,m
is continuous at zero for all m N, i = 1, . . . , n, then so is
i
.
Also
i,m
D
i
(a, k) implies that
i
D
i
(a, k). The uniform convergence of the sequences
i,m
allows us to let m in the relationship (7.1.8), this gives that
i
(x) =
b
i
q
()
x
1
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
(t,
1
(t), . . . ,
n
(t)) d
q
t, x (0, a]. (7.1.18)
Consequently
I
1
q
i
(x) = b
i
+I
1
q
I
q
f
i
(x, y
1
(x), . . . , y
n
(x)) = b
i
+I
q
f
i
(x, y
1
(x), . . . , y
n
(x)). (7.1.19)
Since f
i
(x, y
1
(x), . . . , y
n
(x)), i = 1, 2, . . . , n, are bounded on (0, a], then I
1
q
i
(0) = b
i
,
i = 1, 2, . . . , n, i.e.
i
n
i=1
satises the initial conditions (7.1.3).
Uniqueness. To prove the uniqueness assume that
i
n
i=1
is another solution valid in an
interval (0, h], h a. Set
i
(x) :=
i
(x)
i
(x), x (0, h], i = 1, 2, . . . , n, (7.1.20)
and
g
i
(x) := f
i
(x,
1
(x), . . . ,
n
(x)) f
i
(x,
1
(x), . . . ,
n
(x)), j = 1, 2, . . . , n. (7.1.21)
Hence
D
i
(x) = g
i
(x), j = 1, 2, . . . , n, (7.1.22)
162
then by (5.4.52) we obtain
i
(x) =
x
1
q
()
_
x
0
_
qt/x; q
_
1
g
i
(t) d
q
t. (7.1.23)
Now x
1
i
is continuous at zero for i = 1, 2, . . . , n. Then for each x (qh, h] there exists
a constant C
x
> 0 such that
t
1
[
i
(t)[
C
x
q
()
, t xq
m
, m N . (7.1.24)
Fix x (qh, h] and t xq
m
, m N. Hence from (7.1.23) we obtain
[
i
(t)[
At
1
q
()
n
i=1
_
t
0
_
qu/t; q
_
1
i
(u) d
q
u
AnC
x
t
1
2
q
()
_
t
0
u
1
_
qu/t; q
_
1
d
q
u
=
AnC
x
t
21
2
q
()
B
q
(, ) =
BC
x
q
(2)
t
21
(7.1.25)
Repeating the estimates k times we obtain
[
i
(t)[ C
x
B
k+1
t
k+1
q
(k +)
, k Z
+
, i = 1, 2, . . . , n. (7.1.26)
Since
B
k
t
k
q
(k +)
is the general term of the series of the q-Mittag-Leer function e
,
(t
),
t
< (1 q)
,
then lim
k
B
k
t
k
q
(k +)
= 0. Therefore
i
(t) = 0, t xq
m
, m N , x (qh, h].
That is (x) = 0 for all x (0, h], proving the uniqueness.
As for the initial value problem (7.1.2)(7.1.3) we have the following theorem
Theorem 7.1.2. Let f
i
(x, y
1
, . . . , y
n
) be functions dened for x (0, a] and y
i
in a domain
G
i
, satisfying the following conditions
163
(i) there is a positive constant A such that, for x (0, a] and y
r
, y
i
G
i
, 1 r, i n, the
following Lipschiz condition is fullled
f
i
(x, y
1
, . . . , y
n
) f
i
(x, y
1
, . . . , y
n
)
A
_
[ y
1
y
1
[ +. . . +[ y
n
y
n
[
_
. (7.1.27)
(ii) There exists M > 0 such that
[f
i
(x, y
1
, . . . , y
n
)[ M, i = 1, . . . , n, x (0, a], and y
i
G
i
.
Let 0 < < 1 and K be a constant that satises K
Ma
q
(+1)
, and D
i
(a, K) G
i
be the
set of points y
i
G
i
which satisfy that
[y
i
(x) b
i
x
1
q
()
[ < K, for all x (0, a]
Then, the initial value problem (7.1.2)(7.1.3) has a unique solution y
i
n
i=1
valid in (0, a].
Proof. Consider the system of functions
i,m
m=1
, i = 1, . . . , n,
i,1
(x) =
b
i
x
1
q
()
,
i,m
(x) =
x
1
q
()
_
b
i
+
_
x
0
_
qt/x; q
_
1
f
i
_
qt,
1,m1
(qt), . . . ,
n,m1
(qt)
_
d
q
t,
(7.1.28)
where m 2. We will show that lim
m
i,m
(x) exists and give the required solution
n
i=1
of the initial value problem (7.1.2)(7.1.3). As in the proof of Theorem 7.1.1 we
can prove by induction on m that
i,m
D
i
(a, k), m Z
+
and
[
i,m+1
(x)
i,m
(x)[
MB
m1
q
m(m1)
_
2
x
m
q
(m + 1)
, B := An, m 1, (7.1.29)
for i = 1, . . . , n. According to the estimate (7.1.29), for 0 < x a, the absolute value of
the terms of the series
i,1
(x) +
m=1
i,m+1
(x)
i,m
(x). (7.1.30)
is less than the corresponding terms of the convergent numeric series
M
B
m=0
B
m
q
m(m1)/2
a
(m)
q
(m + 1)
=
M
B
E
,1
(Ba
; q).
164
Therefore the series (7.1.30) is uniformly convergent on (0, a] to a function
i
, where
i
(x) =
lim
m
i,m
(x). Since x
1
i,m
is continuous at zero for all m Z
+
, i = 1, . . . , n, then so
is
i
. Also
i,m
D
i
(a, k) implies that
i
D
i
(a, k). The uniform convergence of the
sequences
_
i,m(x)
_
, x (0, a], allows us to let m in the relationship (7.1.28), this
gives that
i
(x) =
b
i
q
()
x
1
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
(t,
1
(qt), . . . ,
n
(qt)) d
q
t.
Consequently
I
1
q
i
(x) = b
i
+I
1
q
I
q
f
i
(x, y
1
(qx), . . . , y
n
(qx)) = b
i
+I
q
f
i
(x, y
1
(qx), . . . , y
n
(qx)). (7.1.31)
Thus I
1
q
i
(0) = b
i
, i = 1, 2, . . . , n, i.e.
i
n
i=1
satises the initial conditions (7.1.3).
The proof of the uniqueness of the solution is similar to that of Theorem 7.1.1 so it is
omitted.
Remark 7.1.1. When = 1, the initial value problems (7.1.1), (7.1.3) and (7.1.2)(7.1.3)
are nothing but the initial value problems (3.1.5), (3.2.1) and (3.1.6), (3.2.1), respectively.
See Section 3.2. Also cf. [9, 90].
Example 7.1.1. Using the method of successive approximation, one can prove that the
solution of the initial value problem
D
q
y
1
(x) = y
2
(x), D
q
y
2
(x) = y
1
(x), (7.1.32)
D
1
q
y
1
(0) = 0, D
1
q
y
2
(0) = 1, (7.1.33)
is
y
1
(x) = x
21
e
2,2
(x
2
; q), y
2
(x) = x
1
e
2,
(x
2
; q), [x[ < (1 q)
. (7.1.34)
If we take = 1, we get
y
1
(x) = sin
q
x(1 q), and y
2
(x) = cos
q
x(1 q). (7.1.35)
165
Example 7.1.2. Using the successive approximation method, one can prove that the solu-
tion of the initial value problem
D
q
y
1
(x) = y
2
(x), D
q
y
2
(x) = q
12
y
1
(q
x), (7.1.36)
D
1
q
y
1
(0) = 0, D
1
q
y
2
(0) = 1, (7.1.37)
is
y
1
(x) = x
21
E
2,2
(x
2
; q), y
2
(x) = x
1
E
2,
(x
2
; q), x R. (7.1.38)
If we take = 1, we get
y
1
(x) = q
1
sin(q
1
x; q), and y
2
(x) = cos(q
1/2
x; q), x R. (7.1.39)
7.2 Basic Caputo fractional order systems
In this section we treat two systems of basic Caputo fractional derivatives, namely the
system
c
D
q
y
i
(x) = f
i
(x, y
1
, . . . , y
n
), i = 1, 2, . . . , n, (7.2.1)
and the system
c
D
q
y
i
(x) = f
i
(qx, y
1
(qx), . . . , y
n
(qx)), i = 1, 2, . . . , n, (7.2.2)
where 0 < < 1. We call system (7.2.1) together with initial conditions of the form
y
i
(0) = b
i
, b
j
R, i = 1, 2, . . . , n, (7.2.3)
an initial value problem. Similarly for the system (7.2.2) with the initial conditions (7.2.3).
By a solution of the initial value problem (7.2.1), (7.2.3) we mean a set of functions y
i
n
i=1
that satises (7.2.1) and the initial conditions (7.2.3) in an interval I containing zero such
that y
i
(x), i = 1, 2, . . . , n, are continuous at zero. Similarly we can dene the solution
of the initial value problem (7.2.2)(7.2.3). Now consider the existence and uniqueness of
solutions of the initial value problem (7.2.1), (7.2.3).
166
Theorem 7.2.1. Let f
i
(x, y
1
, . . . , y
n
) be functions dened for x [0, a] and y
i
in a domain
G
i
, satisfying the following conditions
(i) there is a positive constant A such that, for x (0, a] and y
i
, y
i
G
i
, 1 i n, the
following Lipschiz condition is fullled
[f
i
(x, y
1
, . . . , y
n
) f
i
(x, y
1
, . . . , y
n
)[ A([y
1
y
1
[ +. . . +[y
n
y
n
[) . (7.2.4)
(ii) There exists M > 0 such that
[f
i
(t, y
1
, . . . , y
n
)[ M, i = 1, . . . , n, t [0, a], and y
i
G
i
.
For 0 < 1, let a, K be such that
0 < a
1
(An)
1
(1 q)
and K
Ma
q
( + 1)
Let D
i
(a, K) G
i
be the set of points y
i
G
i
which satisfy that
[y
i
(x) b
i
x
1
q
()
[ < K, for all x [0, a]
Then, the initial value problem (7.2.1), (7.2.3) has a unique solution y
i
(x)
n
i=1
valid in
(0, a].
Proof. Dene sequences
i,m
m=1
, i = 1. . . . , n, by the following relationships
i,1
(x) = b
i
,
i,m
(x) = b
i
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
_
t,
i,m1
(t), . . . ,
n,m1
(t) d
q
t,
(7.2.5)
where m 2. We will show that lim
m
i,m
(x) exists and give the required solution
i
n
i=1
of the initial value problem (7.2.1), (7.2.3). First we prove by induction on m that
i,m
D
i
(a, k), i = 1, . . . , n, m Z
+
. (7.2.6)
Clearly,
i,1
D
i
(a, K) for i = 1, . . . , n. If we assume that
i,m
D
i
(a, k) for i = 1, . . . , n,
then
[f
i
(x,
1,m
(x), . . . ,
n,m
(x))[ M, x [0, a].
167
Then
[
i,m+1
(x) b
i
[
x
1
q
()
_
x
0
_
qt/x; q
_
1
[f
i
(t,
1,m
(t), . . . ,
n,m
(t))[ d
q
t
Mx
1
q
()
_
x
0
_
qt/x; q
_
1
d
q
t =
M
q
()
B
q
(, 1)x
=
Mx
q
( + 1)
(7.2.7)
Thus
i,m
is continuous at zero for all m Z
+
and i = 1, 2, . . . , n. Moreover from (7.2.7)
we obtain
[
i,m+1
(x) b
i
[
Ma
q
( + 1)
K. (7.2.8)
So,
i,m
D
i
(a, K), for all m N, i = 1, 2, . . . , n. Second, we prove by induction on m
that
[
i,m+1
(x)
i,m
(x)[
MB
m1
x
m
q
(m + 1)
, B := An, m 1. (7.2.9)
From (7.2.7), inequality (7.2.9) is true at m = 1. Assume that (7.2.9) is true at m = k,
that is,
[
i,k+1
(x)
i,k
(x)[
MB
k1
x
k
q
(k + 1)
, i = 1, 2, . . . , n. (7.2.10)
Consequently,
[
i,k+2
(x)
i,k+1
(x)[
x
1
q
()
_
x
0
_
qt/x; q
_
1
[f
i
(t,
1,k+1
(t), . . . ,
n,k+1
(t)) f
i
(t,
1,k
(t), . . . ,
n,k
(t))[ d
q
t
Ax
1
q
()
_
x
0
_
qt/x; q
_
1
n
j=1
[
j,k+1
(t)
j,k
(t)[ d
q
t
MB
k
x
1
q
()
q
(k + 1)
q
k
_
x
0
_
qt/x; q
_
1
t
k
d
q
t
=
MB
k
q
()
q
(k + 1)
B
q
(, k + 1)x
(k+1)
=
B
k
M
q
_
(k + 1) + 1
_x
(k+1)
.
Now, consider the innite series
i,1
(x) +
m=1
i,m+1
(x)
i,m
(x). (7.2.11)
168
According to the estimate (7.2.9), for 0 < x a, the absolute value of its terms is less than
the corresponding terms of the convergent numeric series
M
B
m=0
B
m
a
m
q
(m + 1)
=
M
B
e
,1
(Ba
; q).
This means that the series (7.2.11) is uniformly convergent on (0, a] to a function
i
, and
i
(x) = lim
m
i,m
(x). Since
i,m
is continuous at zero for all m N, i = 1, . . . , n, then so
is
i
. Also
i,m
D
i
(a, k) implies that
i
(x) D
i
(a, k). The uniform convergence of the
sequences
i,m
allows us to let m in identity (7.2.5), this gives that
i
(x) = b
i
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
(t,
1
(t), . . . ,
n
(t)) d
q
t.
Obviously the initial conditions (7.2.3) holds. The proof of the uniqueness of the solution
is similar to of that of Theorem 7.1.1 so it is omitted.
Similar to Theorem 7.2.1 we have the following theorem
Theorem 7.2.2. Let f
i
(x, y
1
, . . . , y
n
) be functions dened for x [0, a] and y
i
in a domain
G
i
, satisfying the following conditions
(i) for xed x [0, a], and y
i
, y
i
in G
i
[f
i
(x, y
1
, . . . , y
n
) f
i
(x, y
1
, . . . , y
n
)[ A([y
1
y
1
[ +. . . +[y
n
y
n
[) ,
where A > 0 and i = 1, 2, . . . , n.
(ii) There exists M > 0 such that
[f
i
(t, y
1
, . . . , y
n
)[ M, i = 1, . . . , n, t [0, a], and y
i
G
i
.
Let K be a constant that satises K
Ma
q
(
i
+ 1)
, and D
i
(a, K) G
i
be the set of points
y
i
G
i
which satisfy that
[y
i
(x) b
i
[ < K, for all x (0, a].
Then, the initial value problem (7.2.2)(7.2.3) has a unique solution y
i
n
i=1
valid in (0, a].
169
Proof. Consider the system of successive functions
i,m
m=1
, i = 1, . . . , n,
i,1
(x) = b
i
,
i,m
(x) = b
i
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
_
qt,
i,m1
(qt), . . . ,
n,m1
(qt) d
q
t,
(7.2.12)
where m 2. We will show that lim
m
i,m
(x) exists and give the required solution
n
i=1
of the initial value problem (7.2.2)(7.2.3). As in the proof of Theorem 7.2.1 we
can prove by induction on m that
i,m
D
i
(a, k) for i = 1, . . . , n, m Z
+
and
[
i,m+1
(x)
i,m
(x)[
MB
m1
q
m(m1)
_
2
x
m
q
(m + 1)
, B := An, m 1. (7.2.13)
According to the estimate (7.2.13), for 0 x a, the absolute value of the terms of the
series
i,1
(x) +
m=1
i,m+1
(x)
i,m
(x). (7.2.14)
is less than the corresponding terms of the convergent numeric series
M
B
m=0
B
m
q
m(m1)/2
a
m
q
(m + 1)
=
M
B
E
,1
(Ba
; q).
This means that the series (7.2.14) is uniformly convergent on (0, a] to a function
i
, where
i
(x) = lim
m
i,m
(x). Since
i,m
L
1
q
(0, a), for all m N, i = 1, . . . , n, then so is
i
. Also
i,m
D
i
(a, k) implies that
i
(x) D
i
(a, k). The uniform convergence of the sequences
i,m
(x), x (0, a], allows us to let m in the relationship (7.2.12), this gives that
i
(x) = b
i
+
x
1
q
()
_
x
0
_
qt/x; q
_
1
f
i
(t,
1
(t), . . . ,
n
(t)) d
q
t.
The proof of the uniqueness of the solution is similar to that of Theorem 7.1.1 so it is
omitted.
Example 7.2.1. Using the method of successive approximations, one can easily see that
the solution of the initial value problem
c
D
q
y
1
(x) = y
2
(x),
c
D
q
y
2
(x) = y
1
(x), (7.2.15)
y
1
(0) = 0, y
2
(0) = 1, (7.2.16)
170
is
y
1
(x) = x
e
2,2+1
(x
2
; q), y
2
(x) = e
2,1
(x
2
; q), [x[ < (1 q)
. (7.2.17)
If we take = 1, we get
y
1
(x) = sin
q
x(1 q), y
2
(x) = cos
q
x(1 q), [x[ < (1 q)
1
. (7.2.18)
Example 7.2.2. Using the successive approximation method, one can prove that the solu-
tion of the initial value problem
c
D
q
y
1
(x) = y
2
(x),
c
D
q
y
2
(x) = q
y
1
(q
x), (7.2.19)
y
1
(0) = 0, y
2
(0) = 1, (7.2.20)
is
y
1
(x) = x
E
2,+1
(x
2
; q), y
2
(x) = E
2,1
(x
2
; q), x R. (7.2.21)
If we take = 1, we get
y
1
(x) = q sin(q
1
x; q), and y
2
(x) = cos(q
1/2
x; q), x R. (7.2.22)
7.3 Solutions via a q-Laplace transform
In 1949, W. Hahn [57] dened two q-analogues of the classical Laplace transform through
the functional transformations
q
L
s
f(x) = (s) =
1
1 q
_
s
1
0
E
q
(qsx) f(x) d
q
x, (7.3.1)
and
q
L
s
f(x) = (s) =
1
1 q
_
0
e
q
(sx)f(x) d
q
x, (7.3.2)
'(s) > 0. In 1961, Abdi [1] studied certain properties of these analogues. In [2] he used
these analogues to solve linear q-dierence equations with constant coecients and certain
171
allied equations. The object of this section is to use (7.3.1) to solve fractional Riemann
Liouville q-dierence equations with constant coecients. First we establish some properties
of the operator
q
L
s
of (7.3.1). For extensive study of the properties of the Laplace transforms
(7.3.1) and (7.3.2), cf. e.g. [13, 5759, 9296]. Using the denition of the qintegration
(1.1.31), the q-Laplace transform (7.3.1) is nothing but
q
L
s
f(x) =
(q; q)
j=0
q
j
(q; q)
j
f(s
1
q
j
). (7.3.3)
As an example if
r
(x) =
x
r
q
(r+1)
, r > 1, then
q
L
s
r
(x) =
(1 q)
r
s
r+1
. (7.3.4)
In [57, equation (9.5)] Hahn dened the convolution of two functions F, G to be
(F G)(x) =
t
1 q
_
1
0
F(tx)G[t tqx] d
q
t, (7.3.5)
where G[x y], for
G(x) :=
n=0
a
n
x
n
,
is dened to be
G[x y] :=
n=0
a
n
[x y]
n
, with [x y]
n
:= x
n
_
y
x
; q
_
n
. (7.3.6)
There is a misprint in (7.3.5) and its correction is
(F G)(x) :=
x
1 q
_
1
0
F(tx)G[x tqx] d
q
t. (7.3.7)
Obviously using the denition of q-integration (F G) is nothing but
(F G)(x) =
1
1 q
_
x
0
F(t)G[x qt] d
q
t. (7.3.8)
From (5.3.2), we use that
G[x qt] =
qt
G(x). (7.3.9)
172
This motivated Ismail [65] to dene the convolution of two functions F, G through
_
F G
_
=
1
1 q
_
x
0
F(t)
qt
G(x) d
q
t. (7.3.10)
It is proved by W. Hahn [57] that
q
L
s
(F G) =
q
L
s
F
q
L
s
G. (7.3.11)
Abdi [1] showed that if (s) =
q
L
s
F(x), then
q
L
s
F(ax) =
1
a
(
s
a
), a ,= 0, (7.3.12)
and
q
L
s
D
n
q
F(x) =
_
s
1 q
_
n
(s)
n
m=1
D
nm
q
F(0)
s
m1
(1 q)
m
, n Z
+
. (7.3.13)
We shall compute the q-Laplace transform of the fractional q-integral and q-derivative in
the following lemma.
Lemma 7.3.1. If F L
1
q
(0, a) and (s) :=
q
L
s
F(x), then
q
L
s
I
q
F(x) =
(1 q)
1
s
q
F(x) =
s
(1 q)
+1
(s)
n
m=1
D
m
q
F(0)
s
m1
(1 q)
m
. (7.3.15)
Proof. Since
I
q
F(x) =
x
1
q
()
F(x) =
(x) F(x),
then from (7.3.4) and (7.3.11) we get (7.3.14). Now we prove (7.3.15). From (5.4.43),
(7.3.13), and (7.3.14) we get
q
L
s
_
D
q
F(x)
_
=
q
L
s
_
D
n
q
I
n
q
F(x)
_
=
s
n
(1 q)
n
q
L
s
I
n
q
F(x)
n
m=1
D
nm
q
I
n
q
F(0)
s
m1
(1 q)
m
=
s
(1 q)
+1
(s)
n
m=1
D
m
q
F(0)
s
m1
(1 q)
m
,
(7.3.16)
proving the lemma.
173
Remark 7.3.1. It is worth mentioning that if we take = n in (7.3.15) we do not obtain
(7.3.13). This is because
q
L
s
I
n
q
f(x) =
(1 q)
n1
s
n
, n 1 < < n.
Thus if we let n on the second identity of (7.3.16) we obtain 1
_
(1 q), and we obtain
formula (7.3.13) with (1 q)
n+1
instead of (1 q)
n
.
Lemma 7.3.2. If
q
L
s
f(x) = (s), then the Laplace transform of the Caputo fractional
q-derivative is given by
q
L
s
c
D
q
f(x) =
s
(1 q)
+1
_
(s)
n1
r=0
D
r
q
f(0)
(1 q)
r1
s
r+1
_
. (7.3.17)
Proof. Since
c
D
q
f(x) = D
n
q
f(x)
n1
(x), then by (7.3.4) and (7.3.13) we obtain
q
L
s
c
D
q
f(x) =
q
L
s
n1
(x)
q
L
s
D
n
q
f(x) =
(1 q)
n1
s
n
q
L
s
(D
n
q
f(x))
=
(1 q)
n1
s
n
_
(
s
1 q
)
n
(s)
n
m=1
D
nm
q
f(0)
s
m1
(1 q)
m
_
=
_
s
(1 q)
(s)
n1
r=0
D
r
q
f(0)
s
r+1
(1 q)
r++1
_
. (7.3.18)
From now on by f
(k)
(u) we mean
d
k
du
k
f(u). The next lemma will be needed in the proof
of the main result of this section.
Lemma 7.3.3. Let , , a R
+
and k N. Then
q
L
s
_
t
k+1
e
(k)
,
_
at
; q
_
_
=
p
1 q
k!
_
p
a)
k+1
(7.3.19)
Proof. From (6.1.12) we obtain
e
(k)
,
(u; q) =
n=0
n(n 1) . . . (n k + 1)
q
(n +)
u
nk
, [u[ < (1 q)
. (7.3.20)
174
Thus for [at
[ < (1 q)
we have
t
k+1
e
(k)
,
_
at
_
=
n=0
n(n 1) . . . (n k + 1)
q
(n +)
(a)
nk
t
k+1
. (7.3.21)
Consequently
q
L
s
_
t
k+1
e
(k)
,
_
at
; q
_
_
=
n=0
n(n 1) . . . (n k + 1)(a)
nk
(1 q)
n+1
s
n+
=
1
1 q
p
k
n=0
n(n 1) . . . (n k + 1)(ap
)
nk
=
1
1 q
p
k
d
k
d
k
n=0
n
,
(7.3.22)
where := ap
1
. But
p
k
1 q
d
k
d
k
n=0
n
=
p
k
1 q
d
k
d
k
1
1
=
p
k
1 q
k!
(1 )
k+1
=
p
k
1 q
k!
_
1 ap
_
k+1
=
p
1 q
k!
_
p
a)
k+1
.
(7.3.23)
Theorem 7.3.4. Let a
i
C, a
n
,= 0, and 1 >
n
>
n1
> . . . >
1
>
0
> 0. Let
F(s) :=
q
L
s
f exists for [s[ R
0
, R
0
> 0. Consider the n term fractional q-dierence
equation
a
n
D
n
q
y(x) +a
n1
D
n1
q
y(x) +. . . +a
1
D
1
q
y(x) +a
0
D
0
q
y(x) = f(x), x > 0 (7.3.24)
If
(x) =
_
1
a
n
m=0
(1)
m
m!
k
0
+...+k
n2
=m
k
0
0,...,k
n2
0
_
m
k
0
, k
1
, . . . , k
n2
_
n2
i=0
_
a
i
a
n
_
k
i
x
(
n
n1
)m+
n
n2
i=0
(
i
n
)k
i
1
e
(m)
n1
,
n
n2
i=0
(
i
n
)k
i
(
a
n1
a
n
x
n1
)
_
,
(7.3.25)
175
[x(1 q)[ <
_
a
n
a
n1
_
1/(
n
n1
)
, then the function y() dened by
y(x) =
n
j=0
a
j
I
1
j
q
y(0)(x) + (1 q)(x) f(x)
=
n
j=0
a
j
I
1
j
q
y(0)(x) +
_
x
0
f(t)
qt
(x) d
q
t,
(7.3.26)
is a solution of (7.3.24) valid in [x(1 q)[ <
_
a
n
a
n1
_
1/(
n
n1
)
.
Proof. Assume that
q
L
s
y(x) = Y (s) and
q
L
s
f(x) = F(s) for [s[ R
0
> 0. From (7.3.15),
the q-Laplace transform of (7.3.24) is
_
_
n
j=0
a
j
s
j
(1 q)
j
+1
_
_
Y (s)
1
1 q
n
j=0
a
j
I
1
j
q
y(0) = F(s). (7.3.27)
Set
p :=
s
1 q
, and g
n
(p) :=
1
n
j=0
a
j
p
j
. (7.3.28)
We can choose R R
0
such that for [s[ R,
n
j=0
a
j
p
j
,= 0 and
[
n2
j=0
a
j
p
j
[ < [a
n1
p
n1
+a
n
p
n
[. (7.3.29)
Consequently for [s[ R
Y (s) = cg
n
(p) + (1 q)F(s)g
n
(p), c :=
n
j=0
a
j
I
(1
j
)
q
y(0). (7.3.30)
Using (7.3.29) we get
g
n
(p) =
1
a
n
p
n
+a
n1
p
n1
+. . . +a
0
p
0
=
_
a
n
p
n
+a
n1
p
n1
_
1
_
1 +
n2
j=0
a
j
p
j
a
n
p
n
+a
n1
p
n1
_
=
m=0
(1)
m
_
_
n2
j=0
a
j
p
j
_
_
m
_
a
n
p
n
+a
n1
p
n1
_
m+1
=
m=0
(1)
m
a
n
p
n1
_
n2
j=0
a
j
a
n
p
n1
_
m
_
a
n1
a
n
+p
n1
_
m+1
.
176
Applying the multinomial theorem on
_
n2
j=0
a
j
a
n
p
n1
_
m
we obtain
_
n2
j=0
a
j
a
n
p
n1
_
m
=
k
0
+...+k
n2
=m
k
0
0,...,k
n2
0
_
m
k
0
, k
1
, . . . , k
n2
_
n2
j=0
_
a
j
a
n
_
k
j
p
(
j
n1
)k
j
=
k
0
+...+k
n2
=m
k
0
0,...,k
n2
0
_
m
k
0
, k
1
, . . . , k
n2
_
_
_
n2
j=0
_
a
j
a
n
_
k
j
_
_
p
n2
j=0
(
j
n1
)k
j
.
(7.3.31)
Thus
g
n
(p) =
m=0
(1)
m
a
n
k
0
+...+k
n2
=m
k
0
0,...,k
n2
0
_
m
k
0
, k
1
, . . . , k
n2
_
n2
j=0
_
a
j
a
n
_
k
j
p
n1
+
n2
j=0
(
j
n
)k
j
_
a
n1
a
n
+p
n1
_
m+1
,
(7.3.32)
Since the last series in (7.3.32) is absolutely convergent for [s[ R, then using the addition
theorem [1, p.393] and (7.3.19) we obtain g
n
(p) =
q
L
s
(x), [s[ R. Consequently from
(7.3.30)
y(x) = c(x) + (1 q)f(x) (x),
which is nothing but (7.3.26). This completes the proof of the theorem.
7.4 A q-Cauchy problem
In this section we investigate the existence and uniqueness of solutions of a q-Cauchy prob-
lem dened in terms of a set of real parameters
j
3
j=0
and a set L
j
3
j=0
of q-integro
dierence operators of fractional order. This problem is a q-analogue of a Cauchy problem
given by Djrbashian [40].
Let the set of parameters
j
3
j=0
satisfy the conditions
1
2
<
0
,
3
1, 0
1
,
2
1,
3
j=0
j
= 3. (7.4.1)
Then obviously
1 <
0
+
3
2, 1
1
+
2
< 2. (7.4.2)
177
We shall also consider the two special cases
0
=
2
=
3
= 1 when
1
= 0,
0
=
1
=
3
= 1 when
2
= 0.
(7.4.3)
We associate a given set of parameters
j
3
j=0
satisfying (7.4.1) with a set L
j
3
j=0
of
q-integro-dierence operators of fractional order setting
L
0
y(x) := D
(1
0
)
q
y(x),
L
1
y(x) := D
(1
1
)
q
D
q
L
0
y(x) = D
(1
1
)
q
y(x)D
0
q
y(x),
L
2
y(x) := D
(1
2
)
q
D
q
L
1
y(x) = D
(1
2
)
q
D
1
q
D
0
q
y(x),
L
1/2
y(x) := L
3
y(x) = D
(1
3
)
q
D
q
L
2
y(x) = D
(1
3
)
q
D
2
q
D
1
q
D
0
q
y(x).
(7.4.4)
In the following we introduce the domain of the denitions of the operators L
j
3
j=0
.
Denition 7.4.1. Let /C
{}
[0, a], 0 < a < , be the set of functions y(x) satisfying the
conditions
1. y(x) L
1
q
(0, a).
2. L
j
y(x) /
q
C[0, a], j = 0, 1, 2.
Obviously if L
j
y(x) /C[0, a], then D
q
L
j
y(x) L
1
q
(0, a), j = 0, 1, 2, and L
1/2
y(x)
L
1
q
(0, a). Let y(x) /C
x=0
, k = 0, 1, 2, (7.4.5)
and
k
=
k
j=0
j
, k = 0, 1, 2. (7.4.6)
One can easily verify that
j1
+
j2
2, j
1
,= j
2
, (7.4.7)
for any pair of numbers of the set
j
3
j=0
.
178
The q-Cauchy Problem which we are interested in is
L
1/2
y(x) +y(qx) = 0, 0 < x < a, (7.4.8)
L
k
y(x)[
x=0
= a
k
, k = 0, 1, 2, (7.4.9)
where a
k
2
k=0
and are arbitrary complex numbers.
The following lemmas are essential in our investigation.
Lemma 7.4.1. If y /C
{}
[0, a], then
D
2
q
L
1/2
y(x) = y(x)
2
k=0
m
k
(y)
x
k
1
q
(
k
)
, for all x (0, a]. (7.4.10)
Proof. Since L
1/2
y(x) L
1
q
(0, a), using (5.4.30) we obtain the following equality for all
x (0, a]
D
2
q
L
1/2
y(x) = D
2
q
D
(1
3
)
q
D
q
L
2
y(x) = D
(2
3
)
q
_
D
1
q
D
q
L
2
y(x)
_
= D
(2
3
)
q
L
2
y(x) m
2
(y) = D
(2
3
)
q
L
2
y(x) m
2
(y)
x
2
3
q
(3
3
)
.
(7.4.11)
Further, we obtain
D
(2
3
)
q
L
2
y(x) = D
(2
3
)
q
_
D
(1
2
)
q
D
q,x
L
1
y(x)
_
= D
(2
2
3
)
q
_
D
1
q
D
q
L
1
y(x)
_
= D
(2
2
3
)
q
L
1
y(x) m
1
(y)
= D
(2
2
3
)
q
L
1
y(x) m
1
(y)
x
2
2
q
(3
2
3
)
. (7.4.12)
Similarly,
D
(2
2
3
)
q
L
1
y(x) = D
0
q
D
q,x
L
0
y(x)
= D
0
q
D
q,x
_
D
(1
0
)
q
y(x)
_
= D
0
q
D
0
q
y(x).
(7.4.13)
But
D
0
q
D
0
q
y(x) = y(x) D
(1
0
)
q
y(x)
x=0
.
179
Therefore,
D
(2
2
3
)
q
L
1
y(x) = y(x) m
0
(y)
x
0
1
q
(
0
)
, x (0, a).
This formula together with (7.4.11) and (7.4.12) implies the desired representation (7.4.10).
Lemma 7.4.2. If y(x) /C
k=0
m
k
(y)
x
k
3
q
(
k
2)
. (7.4.14)
Proof. Applying the operation D
2
q
to both sides of identity (7.4.10), we arrive at formula
(7.4.14).
Lemma 7.4.3. The q-Cauchy problem (7.4.8)(7.4.9) may have only a unique solution in
the class /C
[0, a].
Proof. If there exist two solutions y
j
(x) /C
(x, ) = x
1
E
2,
(q
1
x
2
; q) =
k=0
(1)
k
q
k
2
+(2)k
()
k
x
2k+1
q
(2k +)
, (7.4.17)
which obviously satises the conditions
y
(x, )
_
_
L
1
q
(0, a) when, > 0,
L
2
q
(0, a) when, > 1/2,
(7.4.18)
180
where by L
2
q
(0, a) we mean the set of all functions f dened on (0, a] such that f L
2
q
(0, x)
for all x (qa, a].
Lemma 7.4.4. If
r
=
r
j=0
j
, r = 0, 1, 2 (7.4.19)
then the functions y
r
(, )
2
r=0
is a fundamental set of solution of the q-Cauchy problem
L
1/2
y(x) +y(x) = 0, x (0, a], (7.4.20)
L
k
y(x)[
x=0
=
k,r
_
_
1 when k = r,
0 when k ,= r,
k = 0, 1, 2. (7.4.21)
in the space /C
{}
[0, a].
Proof. We consider three cases
1. If =
0
=
0
, then
L
0
y
0
(x; ) =
k=0
()
k
q
k
2
+(
0
2)k
x
2k
q
(1 + 2k)
,
L
1
y
0
(x; ) =
k=0
()
k
q
(k+1)
2
+(
0
2)(k+1)
x
2k+2
1
q
(3
1
+ 2k)
,
L
2
y
0
(x; ) =
k=0
()
k
q
(k+1)
2
+(
0
2)(k+1)
x
2k+2
1
q
(3
1
2
+ 2k)
.
Since
1
+
2
< 2, and 3
1
3
=
0
=
0
, we obtain
L
1/2
y(x) = L
3
y(x) =
k=0
()
k
q
(k+1)
2
+(
0
2)(k+1)
x
2k+2
1
q
(3
1
3
+ 2k)
=
k=0
()
k
q
k
2
+(
0
2)k
(qx)
2k+
0
1
q
(2k +
0
)
= y(qx).
That is, y
0
(x; ) is a solution of the Problem (7.4.20)(7.4.21).
181
2. If =
1
=
0
+
1
, then
L
0
y
1
(x; ) =
k=0
()
k
q
k
2
+(
1
2)k
x
2k+
1
q
(2k +
1
+ 1)
,
L
1
y
1
(x; ) =
k=0
()
k
q
k
2
+(
1
2)k
x
2k
q
(2k + 1)
,
L
2
y
1
(x; ) =
k=0
q
(k+1)
2
+(
1
2)(k+1)
x
2k
2
+2
q
(3
2
+ 2k)
,
L
3
y
1
(x; ) = L
1/2
y
1
(x; ) =
k=0
()
k
q
k
2
+(
1
2)k
(qx)
2k+2
2
q
(3
2
3
+ 2k)
.
(7.4.22)
But in this case, 3
2
3
=
0
+
1
=
1
according to (7.4.1). Therefore the last identity
may be written in the form
L
1/2
y
1
(x; ) +y
1
(x; ) = 0, x (0, a]. (7.4.23)
By formulae (7.4.22), y
1
(x; ) /C
{}
[0, a] and, in addition, this function satises (7.4.20)
with initial condition (7.4.21) with r = 1. This completes the proof of the Lemma.
3. if =
2
=
0
+
1
+
2
, then
L
0
y
2
(x; ) =
k=0
()
k
q
k
2
+(
2
2)k
x
2k+
1
+
2
q
(2k +
1
+
2
+ 1)
,
L
1
y
2
(x; ) =
k=0
()
k
q
k
2
+(
2
2)k
x
2k+
2
q
(2k +
2
+ 1)
,
L
2
y
2
(x; ) =
k=0
()
k
q
k
2
+(
2
2)k
x
2k
q
(2k + 1)
,
L
3
y
2
(x; ) = L
1/2
y
2
(x; ) =
k=0
()
k
q
k
2
+(
2
2)k
(qx)
2k+2
3
q
(3
3
+ 2k)
.
(7.4.24)
But
2
= 3
3
according to (7.4.1) and the last identity may be written in the form
L
1/2
y
2
(x; ) +y
2
(x; ) = 0, x (0, a). (7.4.25)
By formulae (7.4.24), y
2
(x; ) /C
[0, a] and y
2
(x; ) satises (7.4.20) with initial con-
dition (7.4.21) with r = 1.
182
Now we are ready to prove the following theorem
Theorem 7.4.5. Let a > 0 be an arbitrary number. Then the function
Y (x; ) =
2
j=0
a
j
y
j
(x; ) L
2
q
(0, a) (7.4.26)
is the unique solution of the cauchy type problem (7.4.8)(7.4.9) in the class /C
[0, a].
Proof. By Lemma 7.4.4,
L
1/2
Y (x; ) +Y (qx; ) =
2
j=0
a
j
_
L
1/2
y
j
(x; ) +Y (qx; )
_
= 0, x (0, a]
and
L
k
Y (x; )
x=0
=
2
j=0
a
j
L
j
(x; )
x=0
=
2
j=0
a
j
k,j
= a
k
, k = 0, 1, 2.
The uniqueness of the solution is proved in Lemma 7.4.3.
Finally, we present the forms taken by the operators L
j
3
j=0
, the corresponding Cauchy
type problems and their solutions in the special cases (7.4.3). In the case when
1
= 0 and
0
=
1
=
3
= 1,
L
0
y(x) = y(x), L
1
y(x) = D
1
q
D
q
y(x) = y(x) y(0),
L
2
y(x) = D
q
y(x), L
1/2
y(x) = L
3
y(x) = D
2
q
y(x),
(7.4.27)
and we have also
0
= 1,
1
= 1,
2
= 2. In the case when
2
= 0 and
0
=
1
=
3
= 1,
L
0
y(x) = y(x), L
1
y(x) = D
q
y(x),
L
2
y(x) = D
1
q
D
2
q
y(x) = D
q
y(x) D
q
y(0), L
1/2
y(x) = L
3
y(x) = D
2
q
y(x),
(7.4.28)
and we also have
0
= 1,
1
= 2,
2
= 2. In both these cases we arrive at the same Cauchy
problem
1
q
D
q
1D
q
y(x) +y(x) = 0, x (0, a), y(0) = , D
q
y(0) =
183
whose solution is
Y (x; ) = E
2,1
(x
2
; q) +E
2,2
(qx
2
; q)x
= cos
_
q
x; q
_
+
_
q
sin
_
q
x; q
_
.
184
Table 7.1:
I
q
x
1
x
+1
q
()
q
(+)
, ' > 0
e
q
(x) x
e
1,+1
(x; q)
E
q
(x) x
E
1,+1
(x; q)
x
1
e
q
(x)
x
+1
(1q)
(q
+
;q)
(q
;q)
1
(0, q
; q
+
, q, x)
x
1
E
q
(x)
x
+1
(1q)
(q
+
;q)
(q
;q)
1
(q
; q
+
, q, x)
cos
q
x x
e
2,+1
(
2
x
2
), ' > 0
sin
q
x x
+1
e
2,+1
(
2
x
2
), ' > 0
Cos
q
x
x
q
(+1)
1
2
(q
2
; q
, q
+1
; q
2
,
2
x
2
)
Sin
q
x
x
+1
q
(+2)
xq
2
(1 q)
1
2
(q
2
; q
, q
+1
; q
2
,
2
x
2
)
cos(x; q) x
E
2,+1
(
2
qx
2
; q)
sin(x; q) x
+1
E
2,+2
(
2
q
2
x
2
; q)
Bibliography
[1] W.H. Abdi. On q-Laplace transforms. Proc. Nat. Acad. Sci. (India), 29(A):389408,
1961.
[2] W.H. Abdi. On certain q-dierence equations and q-Laplace transforms. Proc. Nat.
Acad. Sc. (India), 28:115, 1962.
[3] W.H. Abdi. Certain inversion and representation formula for q-Laplace transforms.
Math. Zeitschr., 83:238249, 1964.
[4] N.H. Abel. Auosung einer Mechanischen aufgabe. J. F ur Reine und Angew. Math.,
1:153157, 1826.
[5] N.H. Abel. Solutions de quelques proble`emes `a Iaide dintegrales denies. Gesam-
melte mathematische Werke, 1:1127, 1881. First Publ. in Mag. Naturvidenkaberne,
Aurgang, 1, no.2, Christiania, 1823.
[6] L.D. Abreu and J. Bustoz. On the completeness of sets of q-Bessel functions. In:
Theory and Applications of Special Functions, (eds. M.E.H. Ismail and E. Koelink).
Springer, New York, 2938, 2005.
[7] L.D. Abreu, J. Bustoz, and J.L. Caradoso. The roots of the third Jackson q-Bessel
functions. Internat. J. Math. Math. Sci., 67:42414248, 2003.
[8] M.H. Abu-Risha, M.H. Annaby, M.E.H. Ismail, and Z.S. Mansour. Linear q-dierence
equations. To appear in Z. f. Anal. Anwend.
185
186
[9] M.H. Abu-Risha, M.H. Annaby, M.E.H. Ismail, and Z.S. Mansour. Existence and
uniqueness theorems of q-dierence equations. Submitted, 2005.
[10] P.P. Agarwal. Certain fractional q-integrals and q-derivatives. Proc. Camb. Phil. Soc.,
66:365370, 1969.
[11] W.A. Al-Salam. Fractional q-integrations and q-dierentiations. Notices Amer. Math.
Soc., 13(243), 1966.
[12] W.A. Al-Salam. q-analogue of the Cauchy formula. Proc. Amer. Math. Soc., 17:616
621, 1966.
[13] W.A. Al-Salam. Some fractional q-integrals and q-derivatives. Proc. Edin. Math. Soc.,
15:135140, 1966.
[14] G.E. Andrews, R. Askey, and R. Roy. Special Functions. Cambridge University Press,
Cambridge, 1999.
[15] M.H. Annaby. q-type sampling theorems. Result. Math., 44:214225, 2003.
[16] M.H. Annaby and Z.S. Mansour. Basic SturmLiouville problems. J. Phys. A: Math.
Gen., 38:37753797, 2005. Corrigendum; J. Phys. A: Math. Gen. 39: 8747-8747, 2006.
[17] M.H. Annaby and Z.S. Mansour. On the zeros of basic nite Hankel transforms. To
appear in J. Math. Anal. Appl.
[18] M.H. Annaby and Z.S. Mansour. A basic analog of a theorem of polya. Submitted.
[19] M.H. Annaby and Z.S. Mansour. Asymptotic formulae for eigenvalues and eigenfunc-
tions of basic Sturm-Liouville problems. In preparation.
[20] M.H. Annaby and Z.S. Mansour. Basic fractional calculus. In preparation.
[21] R.A. Askey and J.A. Wilson. Some Basic Hypergeometric Orthogonal Polynomials
That Generalize Jacobi Polynomials, Mem. Amer. Math. Soc., 319, 1985.
187
[22] N.M. Atakishiyev. On a one-parameter family of q-exponential functions. J. Phys.
A: Math. Gen., 29:L223L227, 1996.
[23] F.V. Atkinson. Discrete and Continuous Boundary Problems. Academic Press, New
York, 1964.
[24] Y.M. Berezansky, Z.G. Shaftl, and G.F. US. Functional Analysis, Vol. I. Birkhauser,
Basel, 1996.
[25] W. Bergweiler and W.K. Hayman. Zeros of solutions of a functional equation. Comp.
Meth. Func. Theory, 3:5578, 2003.
[26] W. Bergweiler, K. Ishizaki, and N. Yanagihara. Growth of meromorphic solutions of
some functional equations. I. Aequationes Math., 63:140151, 2002.
[27] R.P. Boas. Entire Functions. Academic Press, New York, 1954.
[28] J. Bustoz and J.L. Cardoso. Basic analog of Fourier series on a q-linear grid. J.
Approx. Theory, 112:154157, 2001.
[29] J. Bustoz and S.K. Suslov. Basic analog of Fourier series on a q-quadratic grid.
Methods Appl. Anal., 5:138, 1998.
[30] P.L. Butzer and U. Westphal. An introduction to fractional calculus. In: Applications
of Fractional Calculus in Physics, R. Hilfer (ed). World Scientic, Singapore, 385,
2000.
[31] M. Caputo. Linear model of dissipation whose Q is almost frequency independentII.
Geophys. J. R. Astr. Soc., 13:529539, 1967.
[32] M. Caputo. Elasticit`ae Dissipazione. Bologna, 1969.
[33] R.D. Carmichael. Linear dierence equations and their analytic solutions. Trans.
Amer. Math. Soc., 12:99134, 1912.
188
[34] R.D. Carmichael. On the theory of linear dierence equations. Amer. J. Math.,
35:163182, 1913.
[35] Y. Chen, M.E.H. Ismail, and K.A. Muttalib. Asymptotics of basic Bessel functions
and q-Laguerre polynomials. J. Comp. Appl. Math., 54:263273, 1994.
[36] K. Chung, W. Chung, S. Nam, and H. Kang. New q-derivative and q-logarithm.
Internat. J. Theoret. Phys., 33 (10):20192029, 1993.
[37] E.A. Coddington and N. Levinson. Theory of Ordinary Dierential Equations.
McGraw-Hill, New York, 1955.
[38] A.B.O. Daalhuis. Asymptotic expansions for q-gamma, q-exponential and q-Bessel
functions. J. Math. Anal. Appl., 186:896913, 1994.
[39] M.M. Djrbashian. Integral Transforms and Representations of Functions in the Com-
plex Domain. Nauka, Moscow, 1962. in Russian.
[40] M.M. Djrbashian. Harmonic Anylysis and Boundary Value Problem in the Complex
Domain. Birkhauser, Basel, 1993.
[41] M.S.P. Eastham. Theory of Ordinary Dierential Equations. Van Nostrand, Reinhold,
1970.
[42] A.M.A. El-Sayed. Multivalued fractional dierential equations. Appl. Math. Comput.,
80:111, 1994.
[43] A.M.A. El-Sayed. Fractional order evolution equations. J. Fract. Calc., 7:89100,
1995.
[44] A. Erdelyi. On some functional transformations. Rend. Sem. Mat. Univ. Politec.
Torino, 10:217234, 1951.
[45] A. Erdelyi, W. Magnus, F. Oberhettinger, and F.G. Tricomi. Higer Transcendental
Functions, vol. III. McGraw-Hill, New York, 1955.
189
[46] A. Erdelyi and I.N. Sneddon. Fractional integration and dual integral equations.
Cand. J. Math., 14(4):547557, 1962.
[47] T. Ernst. The history of q-Calculus and a New Method. PhD thesis, Department of
Mathematics, Uppsala University, Sweden, 2001.
[48] H. Exton. q-Hypergeometric Functions and Applications. Ellis-Horwood, Chichester,
1983.
[49] H. Exton. Basic Sturm-Liouville theory. Rev. Tecn. Fac. Ingr. Univ. Zulia, 1:85100,
1992.
[50] R. Floreanini, J . LeTourneux, and L. Vinet. More on the qoscillator algebra. J.
Phys. A: Math. Gen., 28:L287L293, 1995.
[51] R. Floreanini and L. Vinet. A model for the continuous q-ultraspherical polynomials.
J. Math. Phys., 36:38003813, 1995.
[52] R. Floreanini and L. Vinet. More on the qoscillator algebra and qorthogonal
polynomials. J. Phys. A: Math. Gen., 28:L287L293, 1995.
[53] G. Gasper and M. Rahman. Basic Hypergeometric Series. Cambridge university
Press, Cambridge, 2004.
[54] R. Goreno, Yu. Luchko, and J. Loutchko. Computation of the Mittag-Leer function
e
,
(z) and its derivative. Fractional Calculus & Appl. Anal, 5:491-518, 2002.
[55] R.W. Gray and C.A. Nelson. A completeness relation for the q-analogue coherent
states by q-integration. J. Phys. A: Math. Gen., 23:L945L950, 1990.
[56] A.K. Gr unwald.
Uber begrenzte deriviatio und deren anwedung. Z. Math. Phys.,
12:441480, 1867.
[57] W. Hahn. Beitrage zur Theorie der Heineschen Reihen. Math. Nachr., 2:340379,
1949.
190
[58] W. Hahn.
Uber die Zerlegung einer Klasse von Polynomen in irreduzible faktoren.
Math. Nachr., 3:257294, 1950.
[59] W. Hahn.
Uber Uneigentliche Losungen linearer geometrischer Dierenzengleichun-
gen. Math. Annalen, 125:6781, 1952.
[60] W.K. Hayman. On the zeros of a q-Bessel function. Contemporary Mathematics,
American Mathematical Society, preprint.
[61] W.K. Hayman. Subharmonic Functions, volume II. Academic Press, London, 1989.
[62] N. Heymans and I. Podlubny. Physical interpretation of initial conditions for fractional
dierential equations with Riemann-Liouville fractional derivatives. Rheologica Acta,
Online rst, November 2005.
[63] M.E.H. Ismail. The zeros of basic Bessel functions, the functions J
+ax
(x) and asso-
ciated orthogonal polynomials. J. Math. Anal. Appl., 86:119, 1982.
[64] M.E.H. Ismail. On Jacksons third q-Bessel function. preprint, 1996.
[65] M.E.H Ismail. Classical and Quantum Orthogonal Polynomials in One Variable, Cam-
bridge University Press, Cambridge, 2005.
[66] M.E.H. Ismail and C. Zhang. Asymptotics of q-functionas and their zeros. submitted,
2006.
[67] F.H. Jackson. The applications of basic numbers to Bessels and Legendres functions.
Proc. Lond. Math. Soc. (2), 2:192220,1905.
[68] F.H. Jackson. The applications of basic numbers to Bessels and Legendres functions
(Second paper). Proc. Lond. Math. Soc. (2), 3:123,1905.
[69] F.H. Jackson. On generalized functions of Legendre and Bessel. Trans. Roy. Soc.
Edin., 41:128, 1903.
191
[70] F.H. Jackson. A basic-sine and cosine with sympolical solutions of certain dierential
equations. Proc. Edin. Math. Soc., 22:2834, 1903-1904.
[71] F.H. Jackson. A generalization of the function (n) and x
n
. Proc. Roy. Soc. London,
74:6472, 1904.
[72] F.H. Jackson. On q-denite integrals. Quart. J. Pure and Appl. Math., 41:193203,
1910.
[73] A. Jirari. Second-Order SturmLiouville Dierence Equations and Orthogonal Poly-
nomials. Mem. Amer. Math. Soc., 542, 1995.
[74] k. Diethelm, N.J. Ford, A.D. Freed, and Yu. Luchko. Algorithms for the fractional
calculus: A selection of numerical methods. Comput. Methods Appl. Mech. Engrg.,
194:743773, 2005.
[75] V. Kac and P. Cheung. Quantum Calculus. Springer-Verlag, New York, 2002.
[76] A.A. Kilbas, M. Saigo, and R.K Saxena. Generalized Mittag-Leer functions and
generalized fractional calculus operator. Integral Transforms Spec. Funct., 15(1):31
49, 2004.
[77] H.T. Koelink. Some basic Lommel polynomials. J. Approx. Theory, 96:45365, 1999.
[78] H.T. Koelink and R.F. Swarttouw. On the zeros of the HahnExton q-Bessel function
and associated q-Lommel polynomials. J. Math. Anal. Appl., 186:690710, 1994.
[79] A.N. Kolmogorov and S.V. Fomin. Introductory Real Analysis. Dover Publications,
Newyork, 1957.
[80] T.H. Koornwinder and R.F. Swarttouw. On a q-analogues of Fourier and Hankel
transforms. Trans. Amer. Math. Soc., 333:445461, 1992.
[81] S.F. Lacroix. Traite du calcul dierential et du calcul integral. Courcier, Paris,
3:409410, 1819.
192
[82] J.L. Lavoie, T.J. Osler, and R. Tremplay. Fractional derivative and special functions.
SIAM Rev., 18(2):240268, 1976.
[83] A.V. Letinkov. Theory of dierentiation of fractional order. Mat. Sb., 3:168, 1868.
[84] B.Ja. Levin. Distribution of Zeros of Entire Functions. Translations of Mathematical
Monographs. A.M.S. Providence, Rohdisland, 1980.
[85] B.M. Levitan and I.S. Sargsjan. Introduction to Spectral Theory: Selfadjoint Ordinary
Dierential Operators. Amer. Math. Soc., Providence, 1975.
[86] B.M. Levitan and I.S. Sergsjan. Sturm Liouville and Dirac Operators. Kluwer Aca-
demic, Dordrecht, 1991.
[87] J.E. Littelwood. On the asymptotic approximation to integral functions of order zero.
Proc. London Math. Soc., 5(2):361410, 1907.
[88] J. L utzen. Joseph Liouvilles contribution to the theory of integral equations. Historia
Math., 9:371391, 1982.
[89] J. L utzen. Joseph Liouville 18091882. Master of Pure and Applied Mathematics.
Springer, New York, 1990.
[90] Z.S. Mansour. q-Dierence Equations. M. Sc. Thesis, Faculty of Science, Cairo
University, 2002.
[91] V.A. Marchenko. Sturm-Liouville Operators and Applications. Birkhauser, Basel,
1986.
[92] R. Mishra. On certain generalized q-Laplace transforms and Stieltjes transforms. J.
Math. Sci., 5(2):2237, 1970.
[93] R. Mishra. A few inversion formula for generalized q-Laplace transforms. Indian J.
Math., 14(3), 1972.
193
[94] R. Mishra. Applications of fractional q-integration on the theory of a generalized
q-Stieltjes transforms. Bull. Cal. Math. Soc., 69:247254, 1977.
[95] R. Mishra. Application of generalized q-Laplace transforms of the solutions of certain
q-integral equations. Indian J. Math., 23:115120, 1981.
[96] R. Mishra. On a relation between a q-Laplace transform and two generalized q-Laplace
transforms. Proc. Nat. Acad. Sci. India, 53(A):379385, 1983.
[97] G.M. Mittag-Leer. Sur la nouvelle fonction e
(x). Acta
Math., 29:217234, 1905.
[119] A. Zygmund. Trigonometric Series, volume II. Cambridge University Press, 1959.
Index
C
2
q
(0, a), 63
E
q
(z), 5
E
,
(z; q), 143
E
(z), 141
J
(; q), 4
L
1
q
(0, a), 10
L
2
q
(0, a), 8, 63
L
2
q
_
(0, a) (0, a)
_
, 9
W
q
(y
1
, . . . , y
n
)(x), 58
Cos
q
z, 5
Sin
q
z, 5
cos(z; q), 10
cos
q
z, 5
/C
(n)
[a, b], 112
/C
q
[0, a], 122, 123, 125
/C
(n)
q
[0, a], 125
/C
{}
[0, a], 177
L
1
q
(0, a), 10, 118, 147
L
2
q
(0, a), 180
-geometric set, 6
, 11
sin(z; q), 10
sin
q
z, 5
e
q
(z), 5
e
,
(z; q), 143
q-Abel integral equation, 131
q-Bessel functions, 4
q-Beta function, 4
q-Cauchy problem, 56
q-Cauchy problem, 53, 176
q-Gamma function, 3
q-Hankel Transforms, 35
q-Hypergeometric series, 3
q-Laplace transform, 170
q-Wronskian, 58
q-analogues of Mittag-Leer functions, 142
q-binomial coecients, 3
q-dierence operator, 6
q-integration by parts, 8
q-moments of a function f, 36
q-regular at zero function, 7
q-shifted factorial, 2
q-successive approximations, 51
q-translation , 118
q-type Fredholm integral operator, 60
196
197
Abels integral equation, 113
NewtonPuiseux diagram, 12
A fundamental theorem of q-calculus, 7
Abels integral equation, 112
Basic Greens function, 74
Basic Sturm-Liouville problem, 68, 69, 71
Basic Volterra integral equations, 146
Caputo fractional derivative, 139
fractional derivative, 115
Gauss summation formula, 119
Gr unwaldLetinkov, 110
fractional derivative, 139
Greens function, 60
Jackson q-integration, 6
Jacobi triple product identity, 3
Linear q-dierence equations, 55
Mittag-Leer functions, 141
Multiple q-shifted factorial, 2
O, 11
Polya, 30, 49
RiemannLiouville
fractional integral operator, 110
zeros of E
,
(z; q), 152