On integral representations of

q-gamma and qbeta functions

Alberto De Sole , Victor G. Kac
arXiv:math/0302032v1 [math.QA] 4 Feb 2003

Department of Mathematics, MIT

77 Massachusetts Avenue, Cambridge, MA 02139, USA
Emails: [email protected]
[email protected]

Abstract
We study qintegral representations of the qgamma and the qbeta functions. This
study leads to a very interesting qconstant. As an application of these integral rep-
resentations, we obtain a simple conceptual proof of a family of identities for Jacobi
triple product, including Jacobis identity, and of Ramanujans formula for the bilateral
hypergeometric series.

1. Introduction
The qgamma function q (t), a qanalogue of Eulers gamma function, was
introduced by Thomae [9] and later by Jackson [4] as the infinite product
(1 q)t1
q
q (t) = , t>0, (1.1)
(1 q)t1
where q is a fixed real number 0 < q < 1. Here and further we use the following
notation:
n1
Y
(a + b)nq = (a + q j b) , if n Z+ , (1.2)
j=0
Y
(1 + a)
q = (1 + q j a) , (1.3)
j=0
(1 + a) q
(1 + a)tq = , if t C . (1.4)
(1 + q t a)q

Notice that, under our assumptions on q, the infinite product (1.3) is convergent.
Moreover, the definitions (1.2) and (1.4) are consistent.
Though the literature on the qgamma function and its applications is rather
extensive, [2], [3], [1], the authors usually avoided the use of its qintegral repre-
sentation. In fact, each time when a qintegral representation was discussed, it
was, as a rule, not quite right. The first correct integral representation of q (t)
that we know of is in reference [7]:
1
Z 1q
q (t) = xt1 Eqqx dq x . (1.5)
0

1
Here Eqx is one of the two qanalogues of the exponential function:

X xn
Eqx = q n(n1)/2 = (1 + (1 q)x)
q , (1.6)
n=0
[n]!

X xn 1
exq = = , (1.7)
n=0
[n]! (1 (1 q)x)
q

and the qintegral (introduced by Thomae [9] and Jackson [5]) is defined by
Z a
X
f (x)dq x = (1 q) aq j f (aq j ) . (1.8)
0 j=0

The qbeta function was more fortunate in this respect. Already in the
mentioned papers by Thomae and Jackson it was shown that the qbeta function
defined by the usual formula

q (s)q (t)
Bq (t, s) = , (1.9)
q (s + t)

has the qintegral representation, which is a qanalogue of Eulers formula:

Z 1
Bq (t, s) = xt1 (1 qx)qs1 dq x , t, s > 0 . (1.10)
0

Jackson [5] made an attempt to give a qanalogue of another Eulers integral

representation of the beta function:
Z
xt1
B(t, s) = dx . (1.11)
0 (1 + x)t+s

However, his definition is not quite right, since it is not quite equal to Bq (t, s),
as will be explained in Remark 4.4. A correct qanalogue of (1.11) is the famous
Ramanujans formula for the bilateral hypergeometric series, see [1, pp 502505]
(in fact, Ramanujans formula was known already to Kronecker).
In the present paper we give a qintegral representation of q (t) based on
the qexponential function exq , and give a qintegral representation of Bq (t, s)
which is a qanalogue of (1.11). Both representations are based on the following
remarkable function:
xt  1 t
K(x, t) = 1+ (1 + x)1t
q . (1.12)
1+x x q
This function is a qconstant in x, i.e.

K(qx, t) = K(x, t) ,

and for t an integer it is indeed independent on x, and is equal to q t(t1)/2 .

However, for t (0, 1) this function does depend on x, since for these t one has

lim K(x; t) = xt + xt1 .

q0

2
Our integral representations are as follows:
Z /A(1q)
q (t) = K(A, t) xt1 ex
q dq x , (1.13)
0
/A
xt1
Z
Bq (t, s) = K(A, t) dq x , (1.14)
0 (1 + x)t+s
q

where the improper integral, following [5] and [8], is defined by

Z /A X qn  qn 
f (x)dq x = (1 q) f . (1.15)
0 A A
nZ

Since K(A, t) depends on A, we conclude that integrals in both formulas do

depend on A. In his formulas Jackson used the factor q t(t1)/2 in place of
K(A, t), which is correct only for an integer t.
Note also that formula (1.5) for q (t) can be written via an improper integral
n
q
1q
too (since Eq = 0 for n 0):
Z /(1q)
q (t) = xt1 Eqqx dq x . (1.16)
0

We will refer to the book [6] for notation and basic facts on qcalculus.
Unfortunately the authors of the book didnt know about the reference [7] and
made the same mistake as their predecessors in the definition (1.16) (taking
in place of /(1 q), which gives a divergent series). However all their
arguments hold verbatim for the definition (1.5) (or (1.16)) and are used in the
present paper to derive (1.13) and (1.14).
In Sections 5 and 6 we will apply equation (1.14) to find an integral represen-
tation of the qbeta function which is manifestly symmetric under the exchange
of t and s, and to find a qanalogue of translation invariance of certain improper
integrals. Finally, in Section 7 we will show that equation (1.13) is equivalent to
a family of triple product identities, a special case of which is the Jacobi triple
product identity:

X
(1 q)
q (1 x)q (1 q/x)q = (1)n q n(n1)/2 xn , (1.17)
n=

equation (1.14) is equivalent to the famous Ramanujans identity, see [1, pp

501505]:

X (1 a)nq n (1 q)
q (1 b/a)q (1 ax)q (1 q/ax)q

x = , (1.18)
n=
(1 b)nq (1 b)
q (1 q/a)q (1 x)q (1 b/ax)q

and the symmetric integral representation of the qbeta function is equivalent

to the following identity:

X (1 a)nq (1 q/a)n
q (1 a)
q (1 q/a)q

(1 q)
q (1 bc/q)q
= .
n=
(1 b)nq (1 c)n
q (1 b) (1 b/a) (1 ac/q)
q (1 c)q q q
(1.19)

3
 One of the authors wishes to thank A. Varchenko for sending corrections to
the book [6], including the one mentioned above.

2. Notation and preliminary results

Throughout this paper we will assume q to be a fixed number between 0 and
1. We denote by Dq the qderivative of a function

f (qx) f (x)
(Dq f )(x) = .
(q 1)x

The Jackson definite integral of the function f is defined to be [5]

Z a X
f (x)dq x = (1 q) q n af (q n a) .
0 n0

Notice that the series on the righthand side is guaranteed to be convergent as

soon as the function f is such that, for some C > 0, > 1, |f (x)| < Cx in
a right neighborhood of x = 0. Jackson integral and qderivative are related by
the fundamental theorem of quantum calculus, [6, p 73]
Theorem 2.1. (a) If F is any anti qderivative of the function f , namely
Dq F = f , continuous at x = 0, then
Z a
f (x)dq x = F (a) F (0) .
0

(b) For any function f one has

Z x
Dq f (t)dq t = f (x) .
0

Remark 2.2. It is easy to check the qanalogue of the Leibniz rule:

Dq (f (x)g(x)) = g(x)Dq f (x) + f (qx)Dq g(x) .

An immediate consequence is the qanalogue of the rule of integration by parts:

Z a a Z a
g(x)Dq f (x)dq x = f (x)g(x) f (qx)Dq g(x)dq x .

0 0 0

One defines the Jackson integral in a generic interval [a, b] by [5]:

Z b Z b Z a
f (x)dq x = f (x)dq x f (x)dq x .
a 0 0

One also defines improper integrals in the following way [5], [8]:
Z /A X qn  qn 
f (x)dq x = (1 q) f . (2.1)
0 A A
nZ

4
Remark 2.3. Notice that in order the series on the righthand side to be conver-
gent, it suffices that the function f satisfies the conditions: |f (x)| < Cx , x
[0, ), for some C > 0, > 1, > 0, and |f (x)| < Dx , x [N, ), for
some D > 0, < 1, N > 0. In general though, even when these conditions
are satisfied, the value of the sum will be dependent on the constant A. In order
the integral to be independent of A, the anti qderivative of f needs to have
limits for x 0 and x +.
One has the following reciprocity relations:
Z A Z /A
1 1
f (x)dq x = f dq x ,
0 q/A x2 x
Z /A Z A
1 1
f (x)dq x = f dq x . (2.2)
0 0 x2 x
This is a special case of the following more general change of variable formula,
[6, p 107]. If u(x) = x , then
Z u(b) Z b
f (u)dq u = f (u(x))Dq1/ u(x)dq1/ x .
u(a) a

The qanalogue of an integer number n is

1 qn
[n] = = 1 + q + + q n1 .
1q
t
In general we will denote [t] = 1q
1q even for a non integer t.
The following lemma follows immediately from the definitions (1.2), (1.3)
and (1.4) of (a + b)tq . The proof is left as an exercise to the reader. See also [6,
pp 106-107].
Lemma 2.4. Let n Z+ and t, s, a, b, A, B R.
(1) Dq xt = [t]xt1
(2) Dq (Ax + b)nq = [n]A(Ax + b)qn1
(3) Dq (a + Bx)nq = [n]B(a + Bqx)qn1
(4) Dq (1 + Bx)tq = [t]B(1 + Bqx)t1
q
s s1 s
Ax Ax Ax
(5) Dq (1+Bx) t = [s]
(1+Bx)t+1
B([t] [s]) (1+Bx) t+1
q q q

(1+Ax)sq (1+Aqx)s1
q (1+Ax)sq
(6) Dq (1+Bx)tq = [s]A (1+Bqx)tq B[t] (1+Bx)t+1
q

(7) (1 + x)s+t
q = (1 + x)sq (1 + q s x)tq
1
(8) (1 + x)t
q = (1+qt x)tq

(1+x)s+t (1+qt x)sq

(9) (1 + q s x)tq = q
(1+x)sq = (1+x)sq (1 + x)tq

(x+q)n
(10) (1 + q n x)tq = q
(qt x+q)n (1 + x)tq
q

5
 The qanalogues of the exponential function are given by (1.6) and (1.7).
The equality between the series expansion and the infinite product expansion
of exq and Eqx (in the domain where both expansions converge) is proved by
taking the limit for n in the Gauss and Heines qbinomial formulas, [6,
pp 2932].
Remark 2.5. Notice that for q (0, 1) the series expansion of exq has radius of
convergence 1/(1 q). This corresponds to the fact that the infinite product
1/(1 (1 q)x) q has a pole at x = 1/(1 q). On the contrary, the series
expansion of Eqx converges for every x. Both product expansions (1.6) and (1.7)
converge for all x.
Lemma 2.6. The qexponential functions satisfy the following properties:
(a) Dq exq = exq , Dq Eqx = Eqqx .
(b) exq Eqx = Eqx ex
q = 1.

The proof is straightforward and it is left as an exercise to the reader. See also
[6, pp 2932].

3. Definition of qgamma and qbeta functions

The Eulers gamma and beta functions are defined as the following definite
integrals (s, t > 0):
Z
(t) = xt1 ex dx , (3.1)
0
Z 1
B(t, s) = xt1 (1 x)s1 dx , (3.2)
0

xt1
Z
= dx . (3.3)
0 (1 + x)t+s
Notice that (3.3) follows immediately from (3.2) after a change of variable x =
1/(1 + y). From the expression (3.2) it is clear that B(t, s) is symmetric in t and
s. Recall some of the main properties of the gamma and beta functions:

(t + 1) = t(t) , (1) = 1 , (3.4)

B(t, s) = (t)(s)/(t + s) . (3.5)

In this paper we are interested in the qanalogue of the gamma and beta
functions. They are defined in the following way.
Definition 3.1. (a) For t > 0, the qgamma function is defined to be
Z 1/(1q)
q (t) = xt1 Eqqx dq x . (3.6)
0

(b) For s, t > 0, the qbeta function is

Z 1
Bq (t, s) = xt1 (1 qx)qs1 dq x . (3.7)
0

6
 q (t) and Bq (t, s) are the correct qanalogues of the gamma and beta
functions, since they reduce to (t) and B(t, s) respectively in the limit q 1,
and they satisfy properties analogues to (3.4) and (3.5). This is stated in the
following
Theorem 3.2. (a) q (t) can be equivalently expressed as

(1 q)t1
q
q (t) = . (3.8)
(1 q)t1

In particular one has

q (t + 1) = [t]q (t) , t > 0 , q (1) = 1 .

(b) The qgamma and q-beta functions are related to each other by the follow-
ing two equations

Bq (t, )
q (t) = , (3.9)
(1 q)t
q (t)q (s)
Bq (t, s) = . (3.10)
q (t + s)

Proof. We reproduce here the proof of Kac and Cheung [6, pp 7679], because
similar arguments will be used to prove the results in the next section. If we
put s = in the definition of the qbeta function, use (1.6) and the change of
variable x = (1 q)y, we get
Z 1 qx
1q
Bq (t, ) = xt1 Eq dq x
0
1/(1q)
1 1
Z
= y t1 Eqqy dq y = q (t) ,
(1 q)t 0 (1 q)t

which proves (3.9). It follows from qintegration by parts and Lemma 2.4 (parts
(3) and (4)) that Bq (t, s) satisfies the following recurrence relations (t, s > 0):

[t]
Bq (t + 1, s) = Bq (t, s + 1) ,
[s]
Bq (t, s + 1) = Bq (t, s) q s Bq (t + 1, s) .

Putting these two conditions together we get

[s]
Bq (t, s + 1) = Bq (t, s) .
[s + t]
1
Since clearly Bq (t, 1) = [t] , we get, for t > 0 and any positive integer n,

[n 1] . . . [1] (1 q)qn1
Bq (t, n) = = (1 q)
[t + n 1] . . . [t] (1 q t )nq
(1 q)qn1 (1 q)t1
q
= (1 q) . (3.11)
(1 q)qt+n1

7
Taking the limit for n in this expression we get

Bq (t, ) = (1 q)(1 q)t1

q .

This together with (3.9) proves (3.8). We are left to prove (3.10). By comparing
(3.11) and (3.8) we have that (3.10) is true for any positive integer value of s. To
conclude that (3.10) holds for non integer values of s we will use the following
simple argument. If we substitute a = q s and b = q t in (3.10) we can write the
lefthand side as
X (1 q n )
q
(1 q) bn ,
(1 aq n1 )q
n0

and the righthand side as

(1 q)
q (1 ab)q

(1 q) .
(1 a)
q (1 b)q

Both these expressions can be viewed as formal power series in q with coefficients
rational functions in a and b. Since we already know that they coincide, for any
given b, for infinitely many values of a (of the form a = q n , with positive integer
n), it follows that they must be equal for every value of a and b. This concludes
the proof of the Theorem.

4. An equivalent definition of qgamma and q

beta functions
In the previous section the definition of q (t) was obtained from the integral
expression (3.1) of the Eulers gamma function, simply by replacing the integral
with a Jackson integral and the exponential function ex with its qanalogue
Eqqx . It is natural to ask what happens if we use the other qexponential
function. In other words, we want to study the following function (A > 0):
Z /A(1q)
q(A) (t) = xt1 ex
q dq x . (4.1)
0

Similarly, the function Bq (t, s) was obtained by taking the qanalogue of the
integral expression (3.2) of the Eulers beta function. We now want to study
the qanalogue of the integral expression (3.3). We thus define
/A
xt1
Z
q(A) (t, s) = dq x . (4.2)
0 (1 + x)t+s
q

(A) (A)
In this section we will show how the functions q (t) and q (t, s) are related
to the qgamma and qbeta function respectively.
We want to adapt the arguments in the proof of Theorem 3.2 to the functions
(A) (A)
q (t) and q (t, s). First, by taking the limit s in the definition of

8
 (A)
q (t, s), using the infinite product expansion of exq and making the change of
variables x = (1 q)y, we get
Z /A Z /A
xt1 x
q(A) (t, ) =
dq x = xt1 eq 1q dq x
0 (1 + x)q 0
Z /A(1q)
= (1 q)t y t1 ey t (A)
q dq y = (1 q) q (t) .
0

We therefore proved
1
q(A) (t) = (A) (t, ) . (4.3)
(1 q)t q
(A) (A)
We now want to find recursive relations for q (t) and q (t, s). By inte-
gration by parts we get
q(A) (t + 1) = q t [t]q(A) (t) .
Here we used the fact that xt ex
q tends to zero as x 0 and x + (The
(A)
second fact follows from Lemma 2.6 (b)). Since obviously q (1) = 1, we
conclude that for every positive integer n (and any value of A > 0),
q n(n1)/2 q(A) (n) = [n 1]! = q (n) . (4.4)
(A)
Let us now consider the function q (t, s). From integration by parts and the
results in Lemma 2.4 we get (t, s > 0)
Z /A
(A) 1 1
q (t + 1, s) = q t
(qx)t Dq dq x (4.5)
[t + s] 0 (1 + x)t+s
q
Z /A
1 1 [t]
= q t Dq xt dq x = q t (A) (t, s) .
[t + s] 0 (1 + x)t+s
q [t + s] q
For t = 1 we have
/A
1 1
Z
q(A) (1, s) = s+1 dq x = [s] . (4.6)
0 (1 + x)q
Formulas (4.5) and (4.6) imply (s > 0, n Z+ )
(1 q)qs1 (1 q)qn1
q n(n1)/2 q(A) (n, s) = (1 q) = Bq (n, s) . (4.7)
(1 q)qs+n1
Similarly we have
/A
1 1 xt+s
Z
q(A) (t, s + 1) = qs Dq dq x (4.8)
[t + s] 0 (qx)s
(1 + x)t+s
q
/A
1 xt+s 1 [s]
Z
= q s t+s Dq s dq x = (A) (t, s) .
[t + s] 0 (1 + x)q x [t + s] q
(A)
We now need to compute q (t, 1). By definition and Lemma 2.4 (5)
Z /A
xt1 1 /A xt
Z
q(A) (t, 1) = dq x = D q dq x . (4.9)
0 (1 + x)t+1
q [t] 0 (1 + x)tq

9
When using the fundamental theorem of qcalculus to compute the righthand
side, we have to be careful, since the limit for x + of the function F (x) =
xt
(1+x)tq does not exist. On the other hand, by definition of qderivative and
Jackson integral, we have
Z /A  1   qN 
Dq F (x)dq x = lim F lim F ,
0 N Aq N N A
where the limits on the righthand side are taken over the sequence of integer
numbers N . We then have from (4.9)
1 1 t 1
q(A) (t, 1) = lim (Aq N )t (1 + ) . (4.10)
[t] N Aq N q
If we denote by K(A; t) the limit in parenthesis in the righthand side of (4.10),
we can use Lemma 2.4 (10) to get
 q N t
K(A; t) = At lim q N t 1 +
N A q
 N
1
t +q
 1  A q
= At 1 + lim q N t  N
A q N qt
A +q
q
 1 t (1 + qA)N
q
= At 1 + lim
A q N (1 + q 1t A)N q
1  1 t
= At 1 + (1 + A)1t
q .
1+A A q
From (4.8) and (4.10) we conclude that for any t > 0 and positive integer n

(1 q)qn1 (1 q)t1
q
K(A; t)q(A) (t, n) = (1 q) = Bq (t, n) . (4.11)
(1 q)qn+t1
In the following lemma we enumerate some interesting properties of the
function
1  1 t
K(x; t) = xt 1 + (1 + x)1t
q .
1+x x q
Lemma 4.1. (a) In the limit q 1 and 0 we have

lim K(x; t) = 1 , x, t R
q1

lim K(x; t) = xt + xt1 , t (0, 1), x R .

q0

In particular K(x, t) is not constant in x.

(b) Viewed as a function of t, K(x; t) satisfies the following recurrence rela-
tion:
K(x; t + 1) = q t K(x; t) .
Since obviously K(x; 0) = K(x; 1) = 1, we have in particular that for any
positive integer n
K(x; n) = q n(n1)/2 .

10
 (c) As function of x, K(x; t) is a qconstant, namely

Dq K(x, t) = 0 , t, x R .

In other words K(q n x; t) = K(x; t) for every integer n.

Proof. The limit for q 1 of K(x; t) is obviously 1. In the limit q 0 we have,
for any > 0,
(1 + qx)q
(1 + x)
q = (1 + x) (1 + x) .
(1 + q x)
q
 
We therefore have, for t (0, 1): limq0 K(x; t) = xt 1 + x1 . For part (b), it
follows from the definition of K(x; t) and Lemma 2.4 that
1  1 t+1
K(x; t + 1) = xt+1 1 + (1 + x)t
q
1+x x q
 qt  1
= x 1+ K(x; t) = q t K(x; t) .
x (1 + q t x)
For part (c) it suffices to prove that K(qx; t) = K(x; t). By definition
1  1 t
K(qx; t) = (qx)t 1 + (1 + qx)1t
q .
1 + qx qx q
We can replace in the righthand side
 t 1 
1+ qx
t
1
1 + qx = qt
1 + x1 ,
q 1+ qx q
1+q1t x
(1 + qx)1t
q = 1+x (1 + x)1t
q .

The claim follows from the following trivial identity

 
1
q t 1 + qx (1 + q 1t x)
  =1.
1
(1 + qx) 1 + q1t x

This concludes the proof of the lemma.

Remark 4.2. The function K(x; t) is an interesting example of a function which
is not constant in x and with qderivative identically zero.
(A)
It follows from (4.7), (4.11) and Lemma 4.1 that the functions K(A; t)q (t, s)
and Bq (t, s) coincide for any A > 0 as soon as either t or s is a positive integer.
We want to prove that they actually coincide for every t, s > 0.
Theorem 4.3. For every A, t, s > 0 one has:

K(A; t)q(A) (t) = q (t) , (4.12)

K(A; t)q(A) (t, s) = Bq (t, s) . (4.13)

Remark 4.4. This result corrects and generalizes a similar statement of Jackson
[5]. There (4.13) is proved in the special case in which s + t is a positive integer,
But, due to a computational mistake, the factor K(A; t) is missing.

11
Proof. (4.12) is an immediate corollary of (3.9), (4.3) and (4.11). As in the proof
(A)
of Theorem 3.2, in order to prove (4.13) it suffices to prove that K(A; t)q (t, s)
can be written as formal power series in q with coefficients rational functions in
a = q s and b = q t . After performing a change of variable y = Ax, we get
Z /1
1  1 t y t1
K(A; t)q(A) (t, s) = 1+ (1 + A)1t
q t+s dq y . (4.14)
1+A A q

0 y
1+ A
q

Fix A > 0. After letting b = q t , we can rewrite the factor in front of the integral
as  
1 + A1 (1 + A)
1 q q
   ,
1+A 1+ b

qA
A 1+ b
q q
and this is manifestly a formal power series in q with coefficients rational func-
tions in b. We then only need to study the integral term in (4.14), which we
decompose as
Z 1 Z /1
y t1 y t1
 d
t+s q y +  t+s dq y (4.15)
0 y 1 y
1+ A 1+ A
q q

After letting a = q s and b = q t the first term in (4.15) can be written as

 n

X 1 + ab qA
(1 q) bn  n
 ,
n0 1 + qA

and this is a formal power series in q with coefficients rational functions in a

and b. Consider now the second term of (4.15). By relation (2.2) we can rewrite
it as Z q
xs1
 t+s dq x . (4.16)
0 xt+s 1 + 1
Ax
q

Recalling the definition of K(x; t), we have the identity

1 1 At+s
t+s = (1 + Ax)1ts
q .
1 + Ax K(Ax; t + s)

1
xt+s 1 + Ax q

The main observation is that, even though K(Ax; t + s) is not constant in x, by

Lemma 4.1 K(Aq n ; t + s) = K(A; t + s), n Z, therefore inside the Jackson
integral it can be treated as a constant. Using this fact, we can rewrite (4.16)
as Z q
At+s 1
xs1 (1 + Ax)1ts
q dq x . (4.17)
K(A; t + s) 0 1 + Ax
After letting a = q s and b = q t we can finally rewrite the first factor in (4.17)
as    
qA
(1 + A) 1 + abA 1 + ab
q q
    (4.18)
1
1+ A 1+A
q q

12
and the integral term in (4.17) as
X (1 + Aq n+2 )
q
(1 q) an+1   . (4.19)
Aqn+2
n0 1 + ab
q

Clearly both expression (4.18) and (4.19) are formal power series in q with
coefficients rational functions in a and b. This concludes the proof of the theo-
rem.

5. Application 1: an integral expression of the

qbeta function manifestly symmetric in t and s
Theorem 3.2 implies that Bq (t, s) is a symmetric function in t and s. This is
not obvious from its integral expression (3.7). We now want to use Theorem 4.3
to find an integral expression for Bq (t, s) which is manifestly symmetric under
the exchange of t and s. By Theorem 4.3 we have that, for any A > 0
Z /A
xt1
Bq (t, s) = K(A; t) dq x . (5.1)
0 (1 + x)t+s
q

By definition of K(x, t) and using the results of Lemma 2.4 we get, after simple
algebraic manipulations
1  1  q t
t
(1 + x)t+s
q = K ;t 1 + t (1 + q t x)sq . (5.2)
x x qx q
Since by Lemma 4.1 we have
1  qn
K ; t = K(A; t) , x= , nZ,
x A
when we substitute (5.2) back into (5.1) we get, after a change of variable
y = q t x,
Z /
1
Bq (t, s) =  dq y , > 0 . (5.3)
q
0 y 1 + y )tq (1 + y)sq

To conclude, we just notice that this integral expression of Bq (t, s) is manifestly

symmetric in t and s, since performing the change of variable x = yq (namely
applying the reciprocity relation (2.2) gives the same integral with replaced
by 1/ and s replaced by t.

6. Application 2: translation invariance of a cer-

tain type of improper integrals
One of the main failures of the Jackson integral is that there is no analogue
of the translation invariance identity
Z a Z a+c
f (x)dx = f (x c)dx ,
0 c

13
obviously true for classical integrals. By using Theorem 4.3 we are able to
write a q analogue of translation invariance for improper integrals of a special
class of function, namely of the form x /(1 + x)q . More precisely we want to
prove the following
Corollary 6.1. For > 0 and > + 1 we have
 
/ /1 x 1 x1
x q
Z Z
q
dq x = dq x . (6.1)
0 (1 + x)q q K(A, ) 1 x

Remark 6.2. In the classical limit q = 1, the righthand side is obtained from
the lefthand side by translating x x 1.
Proof. From the definition of Bq (t, s) we have
Z 1
Bq (t, s) = xs1 (1 qx)t1
q dq x
0
/1
1  q t1
Z
= 1 dq x
q xs+1 x q
 t1
/1 xt1 1 x1
1
Z
q
= dq x . (6.2)
qs 1 xt+s
The first identity was obtained by applying (2.2) and the second by a change of
variable y = x/q. From Theorem 4.3 we also have
Z /A
xt1
Bq (t, s) = K(A; t) dq x . (6.3)
0 (1 + x)t+s
q

Equation (6.1) is obtained by comparing (6.2) and (6.3), after letting = t

1, = t + s and using the fact that K(A; + 1) = q K(A; ).

7. Application 3: identities
If we rewrite equations (4.12), (4.13) and (5.3) using the definition of im-
proper integrals, we get some interesting identities involving qbilateral series.
After using the infinite product expansion (1.7) of the qexponential function
exq , the expression (3.8) of the qgamma function, the definition (1.15) of the
improper Jackson integral and simple algebraic manipulations, we can rewrite
equation (4.12) as

X
t
(1 q)
q (1 + q /A)q (1 + qA/q t )
q = (1 + qA) t
q (1 q )q q tn (1 + 1/A)nq .
n=
(7.1)
If we let x = q t /A in equation (7.1) we get

X
(1 q)
q (1 x)q (1 q/x)q = (1 + qA)q (1 + Ax)q (x)n An (1 + 1/A)nq .
n=
(7.2)

14
This is a 1parameter family of identities for the Jacobi triple product (1
q)
q (1 x)q (1 q/x)q , parametrized by A. Notice that

lim An (1 + 1/A)nq = q n(n1)/2 .

A0

This implies that, in the special case A = 0, equation (7.2) reduces to the famous
Jacobi triple product identity (1.17).
Lets consider now equation (4.13). After using the definition (1.15) of im-
proper qintegral, the expression (1.9) for the qbeta function and simple alge-
braic manipulations, we can rewrite it as

X (1 + 1/A)nq tn (1 q)
q (1 q
t+s
)q (1 + q t /A) t
q (1 + qA/q )q
q = .
n=
(1 + q t+s /A)nq (1 + q t+s /A) t s
q (1 + qA)q (1 q )q (1 q )q
(7.3)
Notice that, after letting a = 1/A, b = q t+s /A, x = q t , equation (7.3) is
equivalent to the famous Ramanujans identity (1.18). In other words, the proof
of Theorem 4.3 in Section 4 can be viewed as a new, more conceptual proof of
Ramanujans identity.
Finally we can rewrite equation (5.3) as

X (1 + 1/)nq (1 + q)n
q

n=
(1 + q s /)nq (1 + q t+1 )n
q
t+s
(1 + 1/)
q (1 + q)q

(1 q)
q (1 q )q
= s t+1 s t
. (7.4)
(1 + q /)q (1 + q )q (1 q )q (1 q )

q

After letting a = 1/, b = q s /, c = q t+1 , equation (7.4) reduces to

(1.19).

References
[1] G. E. Andrews, R. Askey and R. Roy Special functions, volume 71 of Encyclopedia
of Mathematics and its Applications, Cambridge University Press, Cambridge,
1999.

[2] H. Exton q-hypergeometric functions and applications, Ellis Horwood Series:

Mathematics and its Applications, Ellis Horwood Ltd., Chichester, 1983.

[3] G. Gasper and M. Rahman, Basic hypergeometric series, volume 35 of Encyclope-

dia of Mathematics and its Applications, Cambridge University Press, Cambridge,
1990.

[4] F. H. Jackson, A generalization of the functions (n) and xn , Proc. Roy. Soc.
London, (74) pp 64-72, 1904.

[5] F. H. Jackson, On qdefinite integrals, Quart. J. Pure and Appli. Math, (41) pp
193-203, 1910.

[6] V. Kac and P. Cheung, Quantum Calculus, Universitext, Springer-Verlag, New

York, 2002.

15
[7] H. T. Koelink and T. H. Koornwinder, q-special functions, a tutorial, in De-
formation theory and quantum groups with applications to mathematical physics
(Amherst, MA, 1990), volume 134 of Contemp. Math., pp 141142, Amer. Math.
Soc., Providence, RI, 1992.

[8] T. H. Koornwinder, Special functions and q-commuting variables, in Special func-

tions, q-series and related topics (Toronto, ON, 1995), volume 14 of Fields Inst.
Commun., pp 131166, Amer. Math. Soc., Providence, RI, 1997.

[9] J. Thomae, Beitrage zur Theorie der durch die Heinesche Reihe., J. reine angew.
Math, (70) pp 258281, 1869.

16