Lecture Notes For An Introductory

Minicourse on q-Series

George Gasper
(September 19, 1995 version)

These lecture notes were written for a mini-course that was designed to introduce
arXiv:math/9509223v1 [math.CA] 25 Sep 1995

students and researchers to q-series, which are also called basic hypergeometric series be-
cause of the parameter q that is used as a base in series that are “over, above or beyond”
the geometric series. We start by considering q-extensions (also called q-analogues) of the
binomial theorem, the exponential and gamma functions, and of the beta function and
beta integral, and then progress on to the derivations of rather general summation, trans-
formation, and expansion formulas, integral representations, and applications. Our main
emphasis is on methods that can be used to derive formulas, rather than to just verify pre-
viously derived formulas. Since the best way to learn mathematics is to do mathematics, in
order to enhance the learning process and enable the reader to practice with the discussed
methods and formulas, we have provided several carefully selected exercises at the end of
each section. We strongly encourage you to obtain a deeper understanding of q-series by
looking at these exercises and doing at least three of them in each section. Solutions to
several of the exercises are given in the Gasper and Rahman [1990a] “Basic Hypergeo-
metric Series” book (which we will refer to as BHS) along with additional exercises and
material on this subject, and references to applications to affine root systems (Macdonald
identities), Lie algebras and groups, number theory, orthogonal polynomials, physics (such
as representations of quantum groups and Baxter’s work on the hard hexagon model),
statistics, etc. In particular, for applications to orthogonal polynomials we recommend the
Askey and Wilson [1985] A.M.S. Memoirs. For applications to number theory, physics and
related fields, we recommend the Andrews [1986] and Berndt [1993] lecture notes and the
Fine [1988] book. To add a historical perspective, we have followed the method in BHS of
referring to papers and books in the References at the end of these notes by placing the
year of publication in square brackets immediately after the author’s name.

1 The q-binomial theorem and related formulas

1.1 The binomial theorem. One of the most elementary summation formulas for
power series is the sum of the geometric series
∞
X
z n = (1 − z)−1 , (1.1.1)
n=0

where z is a real or complex number and |z| < 1. By applying Taylor’s theorem to the
function f (z) = (1 − z)−a , which is an analytic function of z for |z| < 1, and observing

1991 Mathematics Subject Classification. Primary 33D15, 33D20, 33D65; Secondary

33D05, 33D45, 33D60, 33D90.
This work was supported in part by the National Science Foundation under grant
DMS-9401452.

1
by mathematical induction that f (n) (z) = (a)n (1 − z)−a−n |z=0 = (a)n , where (a)n is the
shifted factorial defined by

Γ(a + n)
(a)0 = 1, (a)n = a(a + 1) · · · (a + n − 1) = , n = 1, 2, . . . , (1.1.2)
Γ(a)

one can extend (1.1.1) to

∞
X (a)n n
z = (1 − z)−a , |z| < 1. (1.1.3)
n=0
n!

This formula is usually called the binomial theorem because, when a = −m is a negative
integer and z = −x/y, it reduces to the binomial theorem for the m-th power of the
binomial x + y:
m  
m
X m n m−n
(x + y) = x y , m = 0, 1, 2, . . . . (1.1.4)
n=0
n

1.2 The q-binomial theorem. Let 0 < q < 1. Since, by l’Hôpital’s rule,

1 − qa
lim =a (1.2.1)
q→1− 1−q

and hence
(1 − q a )(1 − q a+1 ) · · · (1 − q a+n−1 ) (a)n
lim 2 n
= , (1.2.2)
q→1− (1 − q)(1 − q ) · · · (1 − q ) n!
it is natural to consider what happens when the coefficient of each z n in (1.1.3) is replaced
by the ratio displayed on the left side of (1.2.2) or, more generally, by

(a; q)n
,
(q; q)n

where (a; q)n is the q-shifted factorial defined by


1, n = 0,
(a; q)n = n−1 (1.2.3).
(1 − a)(1 − aq) · · · (1 − aq ), n = 1, 2, . . . .

Hence, let us set

∞
X (a; q)n n
f (a, z) = z , (1.2.4)
n=0
(q; q)n

where a, q, z are real or complex numbers such that |z| < 1 and, unless stated otherwise,
it is assumed that |q| < 1. The case when |q| > 1 will be considered later. Note that, by
the ratio test, since |q| < 1 the series in (1.2.4) converges for |z| < 1 to a function, which

2
we have denoted by f (a, z). One way to find a formula for f (a, z) analogous to that for
the sum of the series in (1.1.3) is to first observe that, since 1 − a = (1 − aq n ) − a(1 − q n ),
∞
X (a; q)n n
f (a, z) = 1 + z
n=1
(q; q)n
∞
X (aq; q)n−1
=1+ [(1 − aq n ) − a(1 − q n )]z n
n=1
(q; q)n
∞ ∞
X (aq; q)n n X (aq; q)n−1 n
=1+ z −a z
n=1
(q; q)n n=1
(q; q)n−1
= f (aq, z) − azf (aq, z) = (1 − az)f (aq, z). (1.2.5)

By iterating this functional equation n − 1 times we get that

f (a, z) = (az; q)n f (aq n , z), n = 1, 2, . . . ,

which on letting n → ∞ and using q n → 0 gives

f (a, z) = (az; q)∞ f (0, z) (1.2.6)

with (a; q)∞ defined by

∞
Y
(a; q)∞ = lim (a; q)n = (1 − aq k ), |q| < 1. (1.2.7)
n→∞
k=0

Since the above infinite product diverges when a 6= 0 and |q| ≥ 1, whenever (a; q)∞ appears
in a formula, it is usually assumed that |q| < 1. Now set a = q in (1.2.6) to obtain

f (q, z) (1 − z)−1 1
f (0, z) = = = ,
(qz; q)∞ (qz; q)∞ (z; q)∞

which, combined with (1.2.6) and (1.2.4), shows that

∞
X (a; q)n n (az; q)∞
z = , |z| < 1, |q| < 1. (1.2.8)
n=0
(q; q)n (z; q)∞

This summation formula was derived by Cauchy [1843] and Heine [1847]. It is called the
q-binomial theorem, because it is a q-analogue of the binomial theorem in the sense that
∞ ∞
X (q a ; q)n n X (a)n n
lim z = z = (1 − z)−a , |z| < 1. (1.2.9)
q→1−
n=0
(q; q) n n=0
n!

Notice that (1.2.8) and (1.2.9) yield

(q a z; q)∞
lim− = (1 − z)−a , |z| < 1, a real, (1.2.10)
q→1 (z; q)∞

3
which, by analytic continuation, holds for z in the complex plane cut along the positive
real axis from 1 to ∞, with (1 − z)−a positive when z is real and less than 1. In the special
case a = q −m , m = 0, 1, 2, . . . , (1.2.8) gives
m
X (q −m ; q)n n
z = (zq −m ; q)m , (1.2.11)
n=0
(q; q) n

where, by analytic continuation, z can be any complex number.

Heine’s proof of the q-binomial theorem, which is presented in Heine [1878], Bailey
[1935, p. 66], Slater [1966, p. 92], and in §1.3 of BHS along with some motivation from
Askey [1980], consists of using series manipulations to derive two difference equations that
are then used to derive the functional equation

(1 − z)f (a, z) = (1 − az)f (a, qz). (1.2.12)

Iterating this equation n − 1 times and letting n → ∞ gives

(az; q)n
f (a, z) = f (a, q n z)
(z; q)n
(az; q)∞ (az; q)∞
= f (a, 0) = (1.2.13)
(z; q)∞ (z; q)∞

and completes the proof. Since it is easy to use series manipulations to show that the
series in (1.2.4) satisfies the functional equation (1.2.12), once (1.2.12) has been discovered
it can be used to give a short verification type proof of the q-binomial theorem.
Another derivation of the q-binomial theorem can be given by calculating the coeffi-
(n)
cients cn = ga (0)/n!, n = 0, 1, 2, . . . , in the Taylor series expansion of the function
∞
(az; q)∞ X
ga (z) = = cn z n , (1.2.14)
(z; q)∞ n=0

which is an analytic function of z when |z| < 1 and |q| < 1. Clearly c0 = ga (0) = 1. One
may show that c1 = ga′ (0) = (1 − a)/(1 − q) by taking the logarithmic derivative of (az;q)∞
(z;q)∞
and then setting z = 0. But, unfortunately, the succeeding higher order derivatives of ga (z)
become more and more difficult to calculate, and so one is forced to abandon this approach
and to search for another way to calculate all of the cn coefficients. One simple method
is to notice that from the definition of ga (z) as the quotient of two infinite products, it
immediately follows that ga (z) satisfies the functional equation

(1 − z) ga (z) = (1 − az) ga (qz), (1.2.15)

which is of course the same as the functional equation (1.2.12) satisfied by f (a, z). In a
verification type proof of the q-binomial theorem, (1.2.15) provides substantial motivation

4
for showing, as in Heine’s proof, that the sum of the q-binomial series f (a, z) satisfies the
functional equation (1.2.12).
To calculate the cn coefficients, we first use (1.2.15) to find that

∞
X ∞
X ∞
X ∞
X
n n+1 n n
cn z − cn z = cn q z − a cn q n z n+1 ,
n=0 n=0 n=0 n=0

or, equivalently,
∞
X ∞
X
n
1+ (cn − cn−1 ) z = 1 + (cn q n − acn−1 q n−1 ) z n ,
n=1 n=1

which implies that

cn − cn−1 = cn q n − acn−1 q n−1
and hence
1 − aq n−1
cn = cn−1 (1.2.16)
1 − qn
for n = 1, 2, . . . . Iterating the recurrence relation (1.2.16) n − 1 times gives

(a; q)n (a; q)n

cn = c0 = , n = 0, 1, 2, . . . , (1.2.17)
(q; q)n (q; q)n

which concludes our third derivation of the q-binomial theorem (1.2.8). For a combinatorial
proof using a bijection between two classes of partitions, see Andrews [1969].

1.3 Related formulas. One immediate consequence the q-binomial theorem is the
product formula
∞ ∞ ∞
X (a; q)j X (b; q)k X (ab; q)n n
zj (az)k = z , |z| < 1, |q| < 1, (1.3.1)
j=0
(q; q)j (q; q)k n=0
(q; q) n
k=0

which is a q-analogue of (1 − z)−a (1 − z)−b = (1 − z)−a−b . By setting j = n − k in the

product on the left side of (1.3.1) and comparing the coefficients of z n on both sides of the
equation, we get
n
(ab; q)n X (a; q)n−k (b; q)k k
= a , (1.3.2)
(q; q)n (q; q)n−k (q; q)k
k=0

which gives a q-analogue of (1.1.4) in the form

n h i
X n
(ab; q)n = (a; q)n−k (b; q)k ak , (1.3.3)
k q
k=0

5
where the q-binomial coefficient is defined by
hni (q; q)n
= , k = 0, 1, . . . , n. (1.3.4)
k q (q; q)k (q; q)n−k

If we let
∞
X zn
eq (z) = , |z| < 1, (1.3.5)
n=0
(q; q) n

then the case a = 0 of (1.2.8) gives

1
eq (z) = , |z| < 1, |q| < 1. (1.3.6)
(z; q)∞

The function eq (z) is a q-analogue of the exponential function ez , since

lim eq (z(1 − q)) = ez . (1.3.7)

q→1−

Another q-analogue of ez can be obtained from (1.2.8) by replacing z with −z/a and then
letting a → ∞ to find that the q-exponential function defined by
∞
X q n(n−1)/2 n
Eq (z) = z , |z| < ∞, |q| < 1, (1.3.8)
n=0
(q; q)n

equals (−z; q)∞ for all complex values of z and satisfies the limit relation

lim Eq (z(1 − q)) = ez , |z| < ∞. (1.3.9)

q→1−

Hence, eq (z)Eq (−z) = 1 when |z| < 1 and |q| < 1.

In Exercise 1.1 below you will be asked to verify the inversion identity

n
(a; q)n = (a−1 ; q −1 )n (−a)n q ( 2 ) (1.3.10)

for n = 0, 1, 2, . . . . This identity enables us to convert a q-series formula containing sums

of quotients of products of q-shifted factorials in base q to a similar formula with base
q −1 . In particular, it follows from (1.2.8) and (1.3.10) that if |q| > 1, then the q-binomial
theorem takes the form
∞
X (a; q)n n (z/q; q −1 )∞
z = , |az/q| < 1, |q| > 1. (1.3.11)
n=0
(q; q) n (az/q; q −1 )
∞

Exercises 1

6
1.1 Verify the inversion identity (1.3.10) and show for nonnegative integers k and n that

(a; q)∞
(i) (a; q)n = ,
(aq n ; q)∞
(ii) (a; q)n+k = (a; q)n (aq n ; q)k ,
(a; q)k (aq k ; q)n
(iii) (aq n ; q)k = ,
(a; q)n
(a; q)k (qa−1 ; q)n −nk
(iv) (aq −n ; q)k = q ,
(a−1 q 1−k ; q)n
1 1
1 − aq 2n (qa 2 ; q)n (−qa 2 ; q)n
(v) = 1 1 .
1−a (a 2 ; q)n (−a 2 ; q)n

1.2 Use series manipulations to verify that the function f (a, z) defined in (1.2.4) satisfies
the functional equation (1.2.12).
1.3 Show that if we define eq (z) for |q| > 1 by
∞
X zn
eq (z) = , |z| < ∞, |q| > 1,
n=0
(q; q) n

then it follows from (1.3.8) and (1.3.10) that eq −1 (z) = Eq (−qz) when |z| < ∞ and
|q| < 1.
1.4 Derive (1.3.11) by using one (or more) of the methods used in §1.2 to derive (1.2.8).
1.5 Investigate what happens if, as in the third derivation of the q-binomial theorem
presented in §1.2, you try to calculate the cn coefficients in (1.2.14) by starting with
the functional equation ga (z) = (1 − az) gaq (z), which corresponds to the functional
equation derived in (1.2.5), instead of with the more complicated functional equation
(1.2.15).
1.6 Extend the definition of the q-binomial coefficient (1.3.4) to

(q β+1 ; q)∞ (q α−β+1 ; q)∞

 
α
=
β q (q; q)∞ (q α+1 ; q)∞

for complex α and β when |q| < 1. Show that

hαi (q −α ; q)k k
(−q α ) q −(2) ,
k
(i) =
k q (q; q)k
(q α+1 ; q)k
 
k+α
(ii) = ,
k q (q; q)k
     
α+1 hαi
k α hαi α
(iii) = q + = + q α+1−k ,
k q k q k−1 q k q k−1 q

7
 where k and n are nonnegative integers.
1.7 Use Ex. 1.6 (iii) and induction to prove that if x and y are indeterminates such that
xy = qyx, q commutes with x and y, and the associative law holds, then
n h i n h i
X n X n
(x + y)n = y k xn−k = xk y n−k .
k q k q −1
k=0 k=0

See, e.g., Koornwinder[1989].

2 Ramanujan’s 1 ψ1 summation formula, Jacobi’s triple

product identity and theta functions

2.1 Ramanujan’s 1 ψ1 summation formula. Since

(a; q)∞
(a; q)n = , |q| < 1, n = 0, 1, 2, . . . , (2.1.1)
(aq n ; q)∞

we may extend the definition (1.2.3) of (a; q)n to

(a; q)∞
(a; q)α = , |q| < 1, (2.1.2)
(aq α ; q)∞

for any complex number α, where the principal value of q α is (usually) taken when q 6= 0.
In particular, if 0 < |q| < 1 and n = 0, 1, 2, . . . , then

(a; q)∞ 1 (−q/a)n (n2 )

(a; q)−n = = = q (2.1.3)
(aq −n ; q)∞ (aq −n ; q)n (q/a; q)n

and
1
= (q n+k ; q)−k = 0, n ≥ 1, k = −n, −n − 1, . . . . (2.1.4)
(q n ; q) k

Let 0 < |q| < 1, 0 < |z| < 1, and n = 1, 2, . . . . Then the q-binomial theorem (1.2.8)
gives
∞ ∞
(az; q)∞ X (a; q)k k X (a; q)j+n−1 j+n−1
= z = z
(z; q)∞ (q; q)k j=−∞
(q; q)j+n−1
k=−∞
∞ ∞
X (a; q)n−1 (aq n−1 ; q)j j+n−1 (a; q)n−1 n−1 X (aq n−1 ; q)j j
= z = z z (2.1.5)
(q; q)n−1 (q n ; q)j (q; q)n−1 (q n ; q)j
j=−∞ j=−∞

by a shift in the index of summation. After replacing a by aq 1−n and then setting q n = b
it follows from the left and right sides of (2.1.5) that
∞
X (a; q)j j (q, b/a, az, q/az; q)∞
z = (2.1.6)
j=−∞
(b; q)j (b, q/a, z, b/az; q)∞

8
for b = q n when 0 < |q| < 1, 0 < |z| < 1, and n = 1, 2, . . . , where we used the compact
notation
(a1 , a2 , . . . , ak ; q)∞ = (a1 ; q)∞ (a2 ; q)∞ · · · (ak ; q)∞ . (2.1.7)
By applying (2.1.3) to the terms of the series in (2.1.6) with negative j, we get
∞ ∞ ∞
X (a; q)j j X (a; q)j j X (q/b; q)j b
j
z = z + , (2.1.8)
j=−∞
(b; q) j j=0
(b; q) j j=1
(q/a; q) j az

from which it is clear that the bilateral series on the left side of (2.1.8) converges in the
annulus |b/a| < |z| < 1 when |q| < 1 and a 6= 0. Since both sides of (2.1.8) are analytic
functions of b when |b| < min(1, |az|) and |z| < 1, and since (2.1.6) holds for b = q n when
0 < |q| < 1, n = 1, 2, . . . , it follows by analytic continuation that we have derived the
summation formula
∞
X (a; q)j j (q, b/a, az, q/az; q)∞
z = , |b/a| < |z| < 1, |q| < 1, (2.1.9)
j=−∞
(b; q)j (b, q/a, z, b/az; q)∞

which is an extension of the q-binomial theorem. This formula was stated without proof in
Ramanujan’s second notebook (see Hardy [1940] and Berndt [1993]). The first published
proofs were given by Hahn [1949] and M. Jackson [1950]. The above proof is essentially
the reverse order of Ismail’s [1977] proof, which first reduces the proof of (2.1.9) to the case
when b = q n , where n is a positive integer, and then proves this case by using a shift in the
index of summation to obtain a series that is summable by the q-binomial theorem. In the
next section, when we introduce the r ψs notation for general bilateral basic hypergeometric
series, it will be seen that (2.1.9) gives the sum of a 1 ψ1 series. Hence, (2.1.9) is sometimes
called Ramanujan’s 1 ψ1 summation formula.
In BHS a proof due to Andrews and Askey [1978] is given, which first considers the
sum of the series in (2.1.9) as a function of b, say f (b), and then shows that this function
satisfies the functional equation
1 − b/a
f (b) = f (bq). (2.1.10)
(1 − b)(1 − b/az)
Iterating (2.1.10) n − 1 times gives
(b/a; q)n
f (b) = f (bq n ). (2.1.11)
(b, b/az; q)n
Since f (b) is analytic for |b| < min(1, |az|), we may let n → ∞ in (2.1.11) to obtain
(b/a; q)∞
f (b) = f (0). (2.1.12)
(b, b/az; q)∞
To calculate f (0), it suffices to set b = q in (2.1.12) and observe that
(q, q/az; q)∞ (q, q/az, az; q)∞
f (0) = f (q) = , (2.1.13)
(q/a; q)∞ (q/a, z; q)∞

9
since
∞
X (a; q)n n (az; q)∞
f (q) = z =
n=0
(q; q)n (z; q)∞

by the q-binomial theorem. Combining (2.1.12) and (2.1.13) gives (2.1.9).

Just as in the third derivation given in §1.2 for the q-binomial theorem, one can also
derive (2.1.9) by using a functional equation and a recurrence relation to calculate the
coefficients cn in the Laurent expansion
∞
(az, q/az; q)∞ X
= cn z n . (2.1.14)
(z, b/az; q)∞ n=−∞

See, e.g., Venkatachaliengar [1988], whose proof is reproduced in Berndt [1993] (also see
Exercise 2.4).

2.2 Jacobi’s triple product identity and theta functions. If we set b = 0 in

(2.1.9), replace q and z by q 2 and −qz/a, respectively, and then let a → ∞, we obtain
Jacobi’s [1829] triple product identity
∞
X 2
q k z k = q 2 , −qz, −q/z; q 2


∞
. (2.2.1)
k=−∞

In BHS this formula is derived by using Heine’s [1847] summation formula for a 2 φ1 (c/ab)
series, which we will consider in §3.
Jacobi’s triple product identity has many important applications. In particular, it can
be used to express the theta functions (Whittaker and Watson [1965, Chapter 21])
∞
X 2
ϑ1 (x) = 2 (−1)n q (n+1/2) sin(2n + 1)x, (2.2.2)
n=0
∞
X 2
ϑ2 (x) = 2 q (n+1/2) cos(2n + 1)x, (2.2.3)
n=0
∞
X 2
ϑ3 (x) = 1 + 2 q n cos 2nx, (2.2.4)
n=1
∞
X 2
ϑ4 (x) = 1 + 2 (−1)n q n cos 2nx (2.2.5)
n=1

in terms of infinite products. To derive the infinite product representations, replace q by

q 2 in (2.2.1) and then set z equal to qe2ix , −qe2ix , −e2ix , e2ix , respectively, to obtain
∞
Y
1/4
ϑ1 (x) = 2q sin x (1 − q 2n )(1 − 2q 2n cos 2x + q 4n ), (2.2.6)
n=1

10
 ∞
Y
ϑ2 (x) = 2q 1/4 cos x (1 − q 2n )(1 + 2q 2n cos 2x + q 4n ), (2.2.7)
n=1
∞
Y
ϑ3 (x) = (1 − q 2n )(1 + 2q 2n−1 cos 2x + q 4n−2 ), (2.2.8)
n=1
and
∞
Y
ϑ4 (x) = (1 − q 2n )(1 − 2q 2n−1 cos 2x + q 4n−2 ). (2.2.9)
n=1

For other applications of (2.2.1), including a proof of the quintuple product identity
∞
X
(−1)n q n(3n−1)/2 z 3n (1 + zq n ) = (q, −z, −q/z; q)∞ (qz 2 , q/z 2 ; q 2 )∞ , z 6= 0,
n=−∞
(2.2.10)
see Berndt [1993] and the Notes to Exercise 5.6 in BHS.

Exercises 2

2.1 Verify the identities (2.1.3) and (2.1.4), and show that

(a; q)n
(i) (aq k ; q)n−k = ,
(a; q)k
(a; q)n (aq n ; q)k
(ii) (aq 2k ; q)n−k = ,
(a; q)2k
(a; q)k (qa−1 ; q)n −nk
(iii) (aq −n ; q)k = q ,
(a−1 q 1−k ; q)n
(q; q)n k
(iv) (q −n ; q)k = (−1)k q (2)−nk ,
(q; q)n−k
(a; q)n k
−1 k ( 2 )−nk
(v) (a; q)n−k = (−qa ) q ,
(a−1 q 1−n ; q)k

when k and n are integers and both sides of the identity are well-defined.
2.2 Use series manipulations to verify that the function f (b) defined in §2.1 satisfies the
functional equation (2.1.10).
2.3 Apply the inversion identity (1.3.10) to the q-shifted factorials in the left side of (2.1.9)
to derive a bilateral extension of the |q| > 1 case of the q-binomial given in (1.3.11).
2.4 Show that the function
(az, q/az; q)∞
h(z) =
(z, b/az; q)∞
satisfies the functional equation (b − aqz) h(qz) = q(1 − z) h(z) when |b/aq| < |z| < 1.
Use this equation to derive a recurrence relation for the coefficients cn in (2.1.14). As

11
 in the derivation of (1.2.17), use the recurrence relation to calculate these coefficients,
and hence derive Ramanujan’s 1 ψ1 summation formula (2.1.9). See Venkatachaliengar
[1988] and Berndt [1993].
2.5 Investigate what happens if, as in Ex. 2.4, you try to generalize Ramanujan’s 1 ψ1 sum-
mation formula by using a functional equation and a recurrence relation to calculate
the coefficients in the Laurent expansion
∞
(az, b/z; q)∞ X
= dn z n , |c| < |z| < 1.
(z, c/z; q)∞ n=−∞

2.6 Use functional equations in a and b to prove that

∞
(−tq b , −q a+1 /t; q)∞ dt (q, q a+b ; q)∞
Z
= − log q
0 (−t, −q/t; q)∞ t (q a , q b ; q)∞

when 0 < q < 1, Re a > 0 and Re b > 0. See Askey and Roy [1986], and Gasper
[1987].
2.7 Extend the integral formula in Ex. 2.6 by evaluating the more general integral
∞
(−tq b , −q a+1 /t; q)∞
Z
tc−1 dt
0 (−t, −q/t; q)∞

when 0 < q < 1, Re (a + c) > 0 and Re (b − c) > 0. See Ramanujan [1915], Askey and
Roy [1986], and Gasper [1987].

3 Basic hypergeometric series, q-gamma and q-beta functions, q-integrals,

and some important summation and transformation formulas

3.1 Basic hypergeometric series. So far, because of the relatively simple q-series
considered, we have not needed to introduce a compact notation for q-series containing
several parameters. Recall that the Gauss [1813] hypergeometric series is (formally) defined
by
  X ∞
a, b (a)n (b)n n
F (a, b; c; z) ≡ 2 F1 (a, b; c; z) ≡ 2 F1 ;z = z , (3.1.1)
c n!(c) n
n=0

where it is assumed that c 6= 0, −1, −2, ..., so that no zero factors appear in the denom-
inators of the terms of the series. Gauss’ series converges absolutely for |z| < 1, and for
|z| = 1 when Re(c − a − b) > 0. Heine [1846, 1847, 1878] introduced the series

φ(α, β, γ, q, z) = 2 φ1 (q α , q β ; q γ ; q, z) (3.1.2)

with   X ∞
a, b (a; q)n (b; q)n n
2 φ1 (a, b; c; q, z) ≡ 2 φ1 ; q, z = z , (3.1.3)
c (q; q) n (c; q) n
n=0

12
where it is assumed that γ 6= −m and c 6= q −m for m = 0, 1, . . . . Heine’s series converges
absolutely for |z| < 1 when |q| < 1, and it is a q-analogue of Gauss’ series because, by
using (1.2.1) and taking a formal termwise limit,
α
lim 2 φ1 (q , q β ; q γ ; q, z) = 2 F1 (α, β; γ; z). (3.1.4)
q→1−

In view of the base q, Heine’s series is also called the basic hypergeometric series or q-
hypergeometric series. We prefer to use the 2 φ1 (a, b; c; q, z) notation instead of Heine’s
φ(α, β, γ, q, z) notation, because when 0 < |q| < 1 the α, β, or γ → ∞ limit cases of
Heine’s series correspond to setting a, b, or c, respectively, equal to zero in (3.1.3).
The (generalized) hypergeometric series with r numerator parameters a1 , . . . , ar and
s denominator parameters b1 , . . . , bs is defined (formally) by
 
a 1 , a 2 , . . . , ar
r Fs (a1 , a2 , . . . , ar ; b1 , . . . , bs ; z) ≡ r Fs ;z
b 1 , . . . , bs
∞
X (a1 )n (a2 )n · · · (ar )n n
= z , (3.1.5)
n=0
n!(b1 )n · · · (bs )n

and an r φs basic hypergeometric series is defined by

 
a 1 , a 2 , . . . , ar
r φs (a1 , a2 , . . . , ar ; b1 , . . . , bs ; q, z) ≡ r φs ; q, z
b 1 , . . . , bs
∞ (3.1.6)
(a1 , a2 , . . . , ar ; q)n h i1+s−r
(−1)n q ( 2 )
X n
= zn
n=0
(q, b 1 , . . . , b s ; q) n

where n2 = n(n − 1)/2 and, analogous to (2.1.7), we employed the compact notation

(a1 , a2 , . . . , ar ; q)n = (a1 ; q)n (a2 ; q)n · · · (ar ; q)n . As in (3.1.4), r φs is a q-analogue of r Fs
and we have (formally)

q a1 , q a2 , . . . , q ar
   
1+s−r a 1 , a 2 , . . . , ar
lim r φs ; q, (q − 1) z = r Fs ;z . (3.1.7)
q→1− q1b , . . . , q bs b 1 , . . . , bs

In terms of these notations, the binomial theorem (1.1.3) and the q-binomial theorem
(1.2.8) may be written in the forms

2 F1 (a, c; c; z) = 1 F0 (a; —; z) = (1 − z)−a , |z| < 1, (3.1.8)

and
(az; q)∞
2 φ1 (a, c; c; q, z) = 1 φ0 (a; —; q, z) = , |z| < 1, |q| < 1, (3.1.9)
(z; q)∞
where a dash is used to indicate the absence of either numerator (when r = 0) or denom-
inator (when s = 0) parameters. Many other important special cases are considered in
BHS, its references, and in what follows.

13
 It is assumed in (3.1.5) and (3.1.6) that the parameters b1 , . . . , bs are such that the
denominator factors in the terms of the series are never zero, and in (3.1.6) it is assumed
that q 6= 0 when r > s + 1. From

(−m)n = q −m ; q n
= 0, n = m + 1, m + 2, . . . , (3.1.10)

we see that an r Fs series terminates if one of its numerator parameters is zero or a negative
integer, and an r φs series terminates if one of its numerator parameters is of the form q −m
with m = 0, 1, 2, . . ., and q 6= 0. Unless stated otherwise, when working with nonterminat-
ing basic hypergeometric series, we will (for simplicity) assume that |q| < 1 and that the
parameters and variables are such that the series converge absolutely. As in our derivation
of (1.3.11), if |q| > 1 then we can use (1.3.10) to perform an inversion with respect to the
base to transform the series (3.1.6) into a series in base q −1 , with |q −1 | < 1 (see Exercise
3.2).
The ratio test shows that an r Fs series converges absolutely for all z if r ≤ s, and for
|z| < 1 if r = s + 1. By Raabe’s test this series also converges absolutely for |z| = 1 if
r = s + 1 and Re [b1 + · · · + bs − (a1 + · · · + ar )] > 0. If r > s + 1 and z 6= 0 or r = s + 1
and |z| > 1, then this series diverges, unless it terminates. Similarly, if 0 < |q| < 1, then
the r φs series converges absolutely for all z if r ≤ s and for |z| < 1 if r = s + 1. It
converges absolutely if |q| > 1 and |z| < |b1 b2 · · · bs |/|a1 a2 · · · ar |. It diverges for z 6= 0
if 0 < |q| < 1 and r > s + 1, and if |q| > 1 and |z| > |b1 b2 · · · bs |/|a1 a2 · · · ar |, unless it
terminates. The r Fs and r φs notations are also used for the sums of these series inside the
circle of convergence and for their analytic continuations (called hypergeometric functions
and basic hypergeometric functions, respectively) outside the circle of convergence.
The series (3.1.6) has the property that if we replace z by z/ar and let ar → ∞ (called
a confluence process), then we obtain a series that is of the form (3.1.6) with r replaced by
r − 1. This is not the case for the r φs series defined without the factors [(−1)n q ( 2 ) ]1+s−r
n

in Bailey [1935] and Slater [1966]. It is because we need a notation that includes such limit
cases as special cases that we have chosen to use the definition (3.1.6). Also, there is no
loss in generality because the Bailey and Slater series can be obtained from the r = s + 1
case of (3.1.6) by choosing s sufficiently large and setting some of the parameters equal to
zero.
Notice that if we denote the terms of the series (3.1.5) and (3.1.6) which contain z n
by un and vn , respectively, then uun+1
n
is a rational function of n and vn+1 vn
is a rational
function of q . Conversely, any rational function of n can be written in the form of uun+1
n
n
,
n vn+1
and any rational function of q can be written in the form of vn . Hence, we have the
P∞ P∞
characterization that if n=0 un and n=0 vn are series with u0 = v0 = 1 such that
un+1 /un is a rational function of n and vn+1 /vn is a rational function of q n , then these
series are of the forms (3.1.5) and (3.1.6), respectively. This characterization provides
another reason why we defined r φs in (3.1.6) with the [(−1)n q ( 2 ) ]1+s−r factor.
n

Generalizing the bilateral q-series that we considered in §2.1, we define the bilateral

14
basic hypergeometric series in base q with r numerator and s denominator parameters by
 
a 1 , a 2 , . . . , ar
r ψs (z) ≡ r ψs ; q, z
b 1 , b 2 , . . . , bs
∞ (3.1.11)
X (a1 , a2 , . . . , ar ; q)n (s−r)n (s−r)( 2 ) n
n
= (−1) q z ,
n=−∞
(b 1 , b 2 , . . . , b s ; q) n

where it is assumed that q, z and the parameters are such that each term of the series
is well-defined (i.e., the denominator factors are never zero, q 6= 0 if s < r, and z 6= 0 if
negative powers of Pz∞occur). A bilateral basic hypergeometric series may be characterized
as being a series n=−∞ vn such that v0 = 1 and vn+1 /vn is a rational function of q n .
Employing the 1 ψ1 notation, Ramanujan’s summation formula (2.1.9) may be stated
in the form
(q, b/a, az, q/az; q)∞
1 ψ1 (a; b; q, z) = , |b/a| < |z| < 1, |q| < 1. (3.1.12)
(b, q/a, z, b/az; q)∞

As in our derivation of (2.1.8), to determine when r ψs series converge we first apply

(1.3.10) to the terms with negative n to obtain the decomposition
∞
(a1 , a2 , . . . , ar ; q)n
(−1)(s−r)n q (s−r)( 2 ) z n
X n
r ψs (z) =
n=0
(b1 , b2 , . . . , bs ; q)n
∞  n (3.1.13)
X (q/b1 , q/b2 , . . . , q/bs ; q)n b1 · · · bs
+ .
n=1
(q/a 1 , q/a 2 , . . . , q/a r ; q) n a 1 · · · a r z

Set R = |(b1 · · · bs )/(a1 · · · ar )|. From (3.1.13) it is obvious that if r > s and 0 < |q| < 1,
then the first series on the right side of (3.1.13) diverges for z 6= 0; if r > s and |q| > 1,
then the first series converges for |z| < R and the second series converges for all z 6= 0.
When r < s and |q| < 1 the first series converges for all z, but the second series converges
only when |z| > R. If r < s and |q| > 1, the second series diverges for all z 6= 0. If r = s
and |q| < 1, the first series converges when |z| < 1 and the second when |z| > R; while
if |q| > 1, then the second series converges when |z| > 1 and the first when |z| < R. In
particular, if r = s and |q| < 1, which is the most important case, the region of convergence
of the series r ψr (z) is the annulus |(b1 · · · br )/(a1 · · · ar )| < |z| < 1.

3.2 q-gamma and q-beta functions. Analogous to Gauss’ infinite product repre-
sentation for the gamma function
∞
Y
Γ(z) = z −1
[(1 + 1/k)z (1 + z/k)−1 ], (3.2.1)
k=1

the q-gamma function Γq (z) is defined as in Thomae [1869] and Jackson [1904] by

(q; q)∞
Γq (z) = (1 − q)1−z , 0 < q < 1. (3.2.2)
(q z ; q)∞

15
When z = n + 1 is a positive integer, this definition reduces to

Γq (n + 1) = 1(1 + q)(1 + q + q 2 ) · · · (1 + q + · · · + q n−1 ), (3.2.3)

which tends to n! as q → 1− . Thus Γq (n + 1) → Γ(n + 1) = n! as q → 1− . One can

extend the definition of Γq (z) to |q| < 1 by using the principal values of q z and (1 − q)1−z
in (3.2.2). In BHS it is shown that

lim Γq (z) = Γ(z) (3.2.4)

q→1−

and that Γq (z) satisfies the functional equation

1 − qz
f (z + 1) = f (z), f (1) = 1, (3.2.5)
1−q

which is a q-analogue of the well-known functional equation for the gamma function

Γ(z + 1) = z Γ(z), Γ(1) = 1. (3.2.6)

Analogous to the definition of the the beta function

Γ(x)Γ(y)
B(x, y) = , (3.2.7)
Γ(x + y)

the q-beta function is defined by

Γq (x)Γq (y)
Bq (x, y) = , (3.2.8)
Γq (x + y)

which tends to B(x, y) as q → 1− . From (3.2.2) and (1.2.8),

(q, q x+y ; q)∞

Bq (x, y) = (1 − q)
(q x , q y ; q)∞
∞
(q; q)∞ X (q y ; q)n
= (1 − q) q nx
(q y ; q)∞ n=0
(q; q)n
∞
X (q n+1 ; q)∞ nx
= (1 − q) n+y ; q)
q , Re x, Re y > 0. (3.2.9)
n=0
(q ∞

This expansion will be utilized in §3.3 to derive a q-integral representation for Bq (x, y).

3.3 q-integrals. Thomae [1869, 1870] defined the q-integral on the interval [0, 1] by
Z 1 ∞
X
f (t) dq t = (1 − q) f (q n )q n . (3.3.1)
0 n=0

16
The right side of (3.3.1) corresponds to using a Riemann sum with partition points tn =
q n , n = 0, 1, 2, . . . . Jackson [1910b] extended this to the interval [a, b] via
Z b Z b Z a
f (t) dq t = f (t) dq t − f (t) dq t, (3.3.2)
a 0 0

where Z a ∞
X
f (t) dq t = a(1 − q) f (aq n ) q n . (3.3.3)
0 n=0

He also defined an integral on (0, ∞) by

Z ∞ ∞
X
f (t) dq t = (1 − q) f (q n )q n . (3.3.4)
0 n=−∞

On the interval (−∞, ∞) the bilateral q-integral is defined by

Z ∞ ∞
X
f (t) dq t = (1 − q) [f (q n ) + f (−q n )] q n . (3.3.5)
−∞ n=−∞

When f is continuous on [0, a], it can be shown that

Z a Z a
lim f (t) dq t = f (t) dt (3.3.6)
q→1 0 0

and that a similar limit holds for (3.3.4) and (3.3.5) for suitable functions. By employing
the q-integral definition (3.3.1), the series expansion (3.2.9) for the q-beta function can be
rewritten in the form
Z 1
(tq; q)∞
Bq (x, y) = tx−1 y dq t, Re x > 0, y 6= 0, −1, −2, . . . , (3.3.7)
0 (tq ; q)∞

which, as q → 1− , tends to the beta function integral

Z 1
B(x, y) = tx−1 (1 − t)y−1 dt, Re x, Re y > 0. (3.3.8)
0

We can derive a q-integral representation for 2 φ1 (q a , q b ; q c ; q, z) by using (3.3.7) to get

1
(q b ; q)n Γq (c) (tq; q)∞
Z
c
= tb+n−1 dq t (3.3.9)
(q ; q)n Γq (b)Γq (c − b) 0 (tq c−b ; q)∞

for n ≥ 0, Re b > 0, c − b 6= 0, −1, −2, . . . , which leads to

∞
Γq (c) (q a ; q)n n 1 b+n−1 (tq; q)∞
X Z
a b c
2 φ1 (q , q ; q ; q, z) = z t dq t
Γq (b)Γq (c − b) n=0 (q; q)n 0 (tq c−b ; q)∞

17
 1
Γq (c) (tq; q)∞
Z
= tb−1 a
1 φ0 (q ; —; q, tz) dq t (3.3.10)
Γq (b)Γq (c − b) 0 (tq c−b ; q)∞
when |z| < 1, Re b > 0, and c − b 6= 0, −1, −2, . . . . Employing (3.1.9) to sum the 1 φ0
series in (3.3.10) gives Thomae’s [1869] q-integral
Z 1 a
a b c Γq (c) b−1 (tzq , tq; q)∞
φ
2 1 (q , q ; q ; q, z) = t dq t, (3.3.11)
Γq (b)Γq (c − b) 0 (tz, tq c−b ; q)∞

where |z| < 1, Re b > 0, and c − b 6= 0, −1, −2, . . . . This is a q-analogue of Euler’s integral
representation
Z 1
Γ(c)
2 F1 (a, b; c; z) = tb−1 (1 − t)c−b−1 (1 − tz)−a dt, (3.3.12)
Γ(b)Γ(c − b) 0
where | arg(1 − z)| < π and Re c > Re b > 0.

3.4 Heine’s 2 φ1 transformation and summation formulas. By employing

Thomae’s definition (3.3.1) of a q-integral to rewrite the integral in (3.3.11) as a series, we
obtain Heine’s [1847], [1878] 2 φ1 transformation formula

a b c (q b , q a z; q)∞ c−b
2 φ1 (q , q ; q ; q, z) = 2 φ1 (q , z; q a z; q, q b ), (3.4.1)
(q c , z; q)∞

or, equivalently, on replacing q a , q b , q c by a, b, c, respectively,

(b, az; q)∞
2 φ1 (a, b; c; q, z) = 2 φ1 (c/b, z; az; q, b), (3.4.2)
(c, z; q)∞
provided |z| < 1 and |b| < 1. Thomae [1869] had derived his integral representation (3.3.11)
by rewriting Heine’s formula (3.4.2) in the form (3.3.11), the reverse of our derivation of
(3.4.2). Formula (3.4.2) is derived in BHS by using the q-binomial theorem and series
manipulations of double series.
One particularly attractive feature of (3.4.2) is that it can be iterated to produce the
following chain of transformation formulas
(b, az; q)∞
2 φ1 (a, b; c; q, z) = 2 φ1 (c/b, z; az; q, b) (3.4.3)
(c, z; q)∞
(c/b, bz; q)∞
= 2 φ1 (abz/c, b; bz; q, c/b) (3.4.4)
(c, z; q)∞
(abz/c; q)∞
= 2 φ1 (c/a, c/b; c; q, abz/c), (3.4.5)
(z; q)∞
provided the series converge. In particular, (3.4.5) and the left side of (3.4.3) yield Heine’s
transformation formula
(abz/c; q)∞
2 φ1 (a, b; c; q, z) = 2 φ1 (c/a, c/b; c; q, abz/c) (3.4.6)
(z; q)∞

18
when |z| < 1 and |abz/c| < 1, which is a q-analogue of Euler’s transformation formula for
2 F1 series

2 F1 (a, b; c; z) = (1 − z)c−a−b 2 F1 (c − a, c − b; c; z), |z| < 1. (3.4.7)

Formulas (3.4.2) – (3.4.6) and (3.3.11) can be used to analytically continue 2 φ1 (a, b; c; q, z)
functions to regions outside the unit disk |z| < 1.
Another important application of (3.4.2) is that in the case z = c/ab it reduces to

(b, c/b; q)∞

2 φ1 (a, b; c; q, c/ab) = 1 φ0 (c/ab; —; q, b), |c/ab| < 1, |b| < 1, (3.4.8)
(c, c/ab; q)∞

where the 1 φ0 can be summed via the q-binomial theorem to yield Heine’s [1847] summation
formula for a 2 φ1 series

(c/a, c/b; q)∞

2 φ1 (a, b; c; q, c/ab) = , (3.4.9)
(c, c/ab; q)∞

which, by analytic continuation, holds when |c/ab| < 1. This formula is Heine’s [1847]
q-analogue of Gauss’ [1813] famous summation formula

Γ(c)Γ(c − a − b)
F (a, b; c; 1) = , Re (c − a − b) > 0. (3.4.10)
Γ(c − a)Γ(c − b)

In §1.6 of BHS it is shown that the Jacobi triple product identity (2.2.1) can be easily
derived from Heine’s summation formula (3.4.9).
For the terminating case when b = q −n , (3.4.9) reduces to

(c/a; q)n
2 φ1 (a, q
−n
; c; q, cq n /a) = , (3.4.11)
(c; q)n

which, by using the inversion identity (1.3.10) or by changing the order of summation, is
equivalent to
−n (c/a; q)n n
2 φ1 (a, q ; c; q, q) = a . (3.4.12)
(c; q)n
Formulas (3.4.11) and (3.4.12) are equivalent to (1.3.3), and they are q-analogues of the
Chu–Vandermonde summation formula
(c − a)n
F (a, −n; c; 1) = , n = 0, 1, . . . . (3.4.13)
(c)n

In §1.5 of BHS it is shown that (3.4.11) and (3.4.12) can be used to derive Sears’
[1951b] transformation formula
−n
2 φ1 (q , b; c; q, z)
 n
(c/b; q)n bz
= 3 φ2 (q
−n
, q/z, c−1 q 1−n ; bc−1 q 1−n , 0; q, q) (3.4.14)
(c; q)n q

19
and Jackson’s [1910a] transformation formula
∞
(az; q)∞ X (a, c/b; q)k k
2 φ1 (a, b; c; q, z) = (−bz)k q (2)
(z; q)∞ (q, c, az; q)k
k=0
(az; q)∞
= 2 φ2 (a, c/b; c, az; q, bz), (3.4.15)
(z; q)∞

which is a q-analogue of the Pfaff-Kummer transformation formula

2 F1 (a, b; c; z) = (1 − z)−a 2 F1 (a, c − b; c; z/(z − 1)). (3.4.16)

3.5 q-analogue of the Pfaff–Saalschütz summation formula. Observe that,

since
∞
(abz/c; q)∞ X (ab/c; q)k k
= z
(z; q)∞ (q; q)k
k=0

by the q-binomial theorem, the right side of (3.4.6) equals

∞ X∞  m
X (ab/c; q)k (c/a, c/b; q)m ab
z k+m ,
m=0
(q; q)k (q, c; q)m c
k=0

and hence, by equating the coefficients of z n on both sides of (3.4.6),

n
X (q −n , c/a, c/b; q)j j (a, b; q)n
1−n /ab; q)
q = .
j=0
(q, c, cq j (c, ab/c; q) n

After replacing a, b by c/a, c/b, respectively, this yields Jackson’s [1910a] summation for-
mula for a terminating 3 φ2 series

(c/a, c/b; q)n

3 φ2 (a, b, q
−n
; c, abc−1 q 1−n ; q, q) = , n = 0, 1, . . . . (3.5.1)
(c, c/ab; q)n

By replacing a, b, c in (3.5.1) by q a , q b , q c , respectively, and letting q → 1 one obtains the

sum of a terminating 3 φ2 series

(c − a)n (c − b)n
3 F2 (a, b, −n; c, 1 + a + b − c − n; 1) = , n = 0, 1, . . . , (3.5.2)
(c)n (c − a − b)n

which was discovered by Pfaff [1797] and rediscovered by Saalschütz [1890]. Since (3.5.2)
is usually called the Saalschütz formula or the Pfaff-Saalschütz formula, Jackson’s formula
(3.5.1) is usually called the q-Saalschütz or the q-Pfaff-Saalschütz formula. It should be
noted that letting a → ∞ in (3.5.1) gives (3.4.11), while letting a → 0 gives (3.4.12).

20
 The Pfaff-Saalschütz 3 F2 (1) series is said to be a balanced series (or Saalschützian)
because z = 1 and the sum of its denominator parameters equals the sum of its numerator
parameters plus 1. Analogously, the q-Pfaff-Saalschütz 3 φ2 (q) series is also called balanced
because z = q and the product of its denominator parameters equals the product of its
numerator parameters times q. In general, as defined in BHS, an r+1 Fr series is called k-
balanced if b1 +b2 +· · ·+br = k +a1 +a2 +· · ·+ar+1 and z = 1; a 1-balanced series is called
balanced (or Saalschützian). Analogously, an r+1 φr series is called k-balanced if b1 b2 · · · br =
q k a1 a2 · · · ar+1 and z = q, and a 1-balanced series is called balanced (or Saalschützian). In
general an r+1 Fr (z) series is called balanced if b1 +b2 +· · ·+br = 1+a1 +a2 +· · ·+ar+1 and
z = 1, and an r+1 φr (z) series is called balanced if b1 b2 · · · br = qa1 a2 · · · ar+1 and z = q.
We will encounter several balanced series in §4.

Exercises 3

3.1 Verify the convergence conditions stated for r Fs , r φs , and r ψs series in §3.1.
3.2 Use (1.3.10) to derive the (formal) inversion formula
∞ n
a 1 , . . . , ar (a−1
  −1 −1

1 , . . . , ar ; q )n a1 · · · ar z
X
r φs ; q, z = .
b 1 , . . . , bs n=0
−1 −1
(q −1 , b1 , . . . , bs ; q −1 )n b1 · · · bs q

3.3 Show that if z is replaced by z/ar in the series (3.1.6), then on letting ar → ∞ one
obtains a series that is of the form (3.1.6) with r replaced by r − 1. Show that this is
not the case if the r φs series in (3.1.6) is defined without the factors [(−1)n q ( 2 ) ]1+s−r .
n

3.4 Prove the limit relation (3.2.4) and the functional equation (3.2.5).
3.5 Derive the transformation formulas (3.4.14) and (3.4.15). Show that (3.4.15) is a
q-analogue of (3.4.16).
3.6 Derive the generating function
∞
X Hn (x|q) n 1
t = iθ
, |t| < 1,
n=0
(q; q)n (te , te−iθ ; q)∞

for the continuous q-Hermite polynomials defined by

n
X (q; q)n
Hn (x|q) = ei(n−2k)θ ,
(q; q)k (q; q)n−k
k=0

where x = cos θ. See Rogers [1894] and Askey and Ismail [1983].
3.7 Derive the generating function
∞
X (βteiθ , βte−iθ ; q)∞
Cn (x; β|q) tn = , |t| < 1,
n=0
(teiθ , te−iθ ; q)∞

21
 for the continuous q-ultraspherical polynomials defined by
n
X (β; q)k (β; q)n−k
Cn (x; β|q) = ei(n−2k)θ ,
(q; q)k (q; q)n−k
k=0

where x = cos θ. See Rogers [1895] and Askey and Ismail [1983].
3.8 The little q-Jacobi polynomials are defined in Andrews and Askey [1977] by

pn (x; a, b; q) = 2 φ1 (q −n , abq n+1 ; aq; q, qx).

Prove that they satisfy the orthogonality relation

∞
X (bq; q)j
(aq)j pn (q j ; a, b; q) pm(q j ; a, b; q)
j=0
(q; q) j

 0, if m 6= n,
n 2
= (q, bq; q) n (1 − abq)(aq) (abq ; q) ∞
, if m = n.
(aq, abq; q)n(1 − abq 2n+1 ) (aq; q)∞


3.9 Show that the little q-Jacobi polynomials defined in Ex. 3.8 satisfy the connection
coefficient formula
n
X
pn (x; c, d; q) = ak,n pk (x; a, b; q)
k=0

with
(q −n , aq, cdq n+1 ; q)k
 k−n
, cdq n+k+1 , aq k+1

q
ak,n = 3 φ2 ; q, q .
(q, cq, abq k+1 ; q)k cq k+1 , abq 2k+2

4 Summation, transformation, and expansion formulas,

integral representations, and applications

4.1 Finite differences, bibasic and very-well-poised series. Let n, m = 0, ±1, ±2, . . . .
Another method to derive summation formulas is to use the fact that if a finite dif-
ference operator ∆ is defined for any sequence {uk } (of real or complex numbers), by
∆uk = uk − uk−1 , then
Xn
∆uk = un − um−1 , (4.1.1)
k=m

where we employed the convention of defining

a + a
 m m+1 + · · · + an , m ≤ n,

Xn 

ak = 0, m = n + 1,

k=m 


−(an+1 + an+2 + · · · + am−1 ), m ≥ n + 2,
(4.1.2)

22
for any sequence {ak }. An excellent way to motivate this definition is to (formally) let
n
X
un = ak
k=−∞

and then observe that, by cancelling the ak ’s that appear in both series in the difference
n
X m
X
un − um−1 = ak − ak ,
k=−∞ k=−∞
Pn
we obtain (4.1.1) with k=m ak defined by (4.1.2).
If, as in Gasper [1989a], we let

(ap, bp; p)k (cq, aq/bc; q)k

sk = , (4.1.3)
(q, aq/b; q)k (ap/c, bcp; p)k

where q and p are independent bases, we obtain the factorization

∆sk = sk − sk−1
(ap, bp; p)k−1(cq, aq/bc; q)k−1
=
(q, aq/b; q)k (ap/c, bcp; p)k
· (1 − apk )(1 − bpk )(1 − cq k )(1 − aq k /bc)


−(1 − q k )(1 − aq k /b)(1 − apk /c)(1 − bcpk )

(1 − apk q k )(1 − bpk /q k ) (a, b; p)k (c, a/bc; q)k
= qk , (4.1.4)
(1 − a)(1 − b) (q, aq/b; q)k (ap/c, bcp; p)k

which is equivalent to the easily checked factorization

(1 − a)(1 − b)(1 − c)(1 − ad2 /bc) − (1 − d)(1 − ad/b)(1 − ad/c)(1 − bc/d)

= (1 − ad)(1 − b/d)(1 − ad/bc)(d − c). (4.1.5)

Since sk = 0 when k ≤ −1, (4.1.4) gives the indefinite bibasic summation formula
n
X (1 − apk q k )(1 − bpk q −k ) (a, b; p)k (c, a/bc; q)k
qk
(1 − a)(1 − b) (q, aq/b; q)k (ap/c, bcp; p)k
k=0
(ap, bp; p)n(cq, aq/bc; q)n
= (4.1.6)
(q, aq/b; q)n(ap/c, bcp; p)n

for n = 0, 1, . . . . In particular, since q 1−n ; q n = 0 unless n = 0, when p = q and

c = q −n , n = 0, 1, . . . , (4.1.6) reduces to the 6 φ5 summation formula

" 1 1
#
a, qa 2 , −qa 2 , b, aq n /b, q −n
6 φ5 1 1 ; q, q = δn,0 , (4.1.7)
a 2 , −a 2 , aq/b, bq 1−n, aq n+1

23
where 
1, m = n,
δm,n =
0, m =
6 n,
is the Kronecker delta function. In §4.2 we will need to use the 4 φ3 summation formula
1 1
a, qa 2 , −qa 2 , q −n
 
n
4 φ3 1 1 ; q, q = δn,0 (4.1.8)
a 2 , −a 2 , aq n+1

when n = 0, 1, . . . , which is the b → 0 and the b → ∞ limit cases of (4.1.7). The

above derivation of (4.1.8) is substantially simpler than that in §2.3 of BHS, which used
the q-Pfaff–Saalschütz formula (3.5.1) and the Bailey [1941] and Daum [1942] summation
formula
(−q; q)∞ (aq, aq 2 /b2 ; q 2 )∞
2 φ1 (a, b; aq/b; q, −q/b) = . (4.1.9)
(aq/b, −q/b; q)∞
Formula (4.1.9) is a q-analogue of Kummer’s formula

Γ(1 + a − b)Γ(1 + 21 a)
2 F1 (a, b; 1 + a − b; −1) = . (4.1.10)
Γ(1 + a)Γ(1 + 21 a − b)

The series in (4.1.7) and (4.1.8) are the form

 
a1 , a2 , . . . , ar+1
r+1 φr ; q, z (4.1.11)
b 1 , . . . , br

in which the the parameters satisfy the relations

qa1 = a2 b1 = a3 b2 = · · · = ar+1 br . (4.1.12)

Such series are called well-poised series, and they are called very-well-poised if, in addition
1 1
a2 = qa12 , a3 = −qa12 , which is the case for (4.1.7) and (4.1.8). Since very-well-poised series
appear quite offen, to simplify some of the displays containing very-well-poised r+1 φr series,
we will frequently replace
1 1
" #
a1 , qa1 , −qa1 , a4 , . . . , ar+1
2 2

r+1 φr 1 1 ; q, z (4.1.13)
a1 , −a12 , qa1 /a4 , . . . , qa1 /ar+1
2

with the more compact notation

r+1 Wr (a1 ; a4 , a5 , . . . , ar+1 ; q, z) . (4.1.14)

Observe that the expression on the right side of (4.1.6) is balanced and well-poised
since

(ap)(bp)(cq)(aq/bc) = q(aq/b)(ap/c)(bcp)
and

24
 (ap)q = (bp)(aq/b) = (cq)(ap/c) = (aq/bc)(bcp).

Also, the part of the series on the left side of (4.1.6) containing the q-shifted factorials is
split-poised in the sense that aq = b(aq/b) and c(ap/c) = (a/bc)(bcp) = ap. It is the first
property and previous special cases, such as Wm. Gosper’s b → 0 limit case of (4.1.6)
n
X 1 − apk q k (a; p)k (c; q)k −k (ap; p)n(cq; q)n −n
c = c , (4.1.15)
1−a (q; q)k (ap/c; p)k (q; q)n (ap/c; p)n
k=0

that led the author to consider the sequence sk in (4.1.3) and obtain the factorization
(4.1.4). See the discussion in Gasper [1989a, p. 259] .
This and the importance of the applications of (4.1.6) considered in Gasper [1989a]
led Gasper and Rahman [1990b] to generalize (4.1.3) to

(ap, bp; p)k (cq, ad2 q/bc; q)k

sk = (4.1.16)
(dq, adq/b; q)k (adp/c, bcp/d; p)k

for k = 0, ±1, ±2, . . ., and observe that

∆tk = tk − tk−1
(ap, bp; p)k−1 (cq, ad2 q/bc; q)k−1
=
(dq, adq/b; q)k (adp/c, bcp/d; p)k
· (1 − apk )(1 − bpk )(1 − cq k )(1 − ad2 q k /bc)


−(1 − dq k )(1 − adq k /b)(1 − adpk /c)(1 − bcpk /d)

d(1 − c/d)(1 − ad/bc)(1 − adpk q k )(1 − bpk /dq k )
=
(1 − a)(1 − b)(1 − c)(1 − ad2 /bc)
(a, b; p)k (c, ad2 /bc; q)k q k
· (4.1.17)
(dq, adq/b; q)k (adp/c, bcp/d; p)k

by (4.1.5), which, in view of (4.1.1), gives the following generalization of (4.1.6)

n
X (1 − adpk q k )(1 − bpk /dq k ) (a, b; p)k (c, ad2 /bc; q)k
qk
(1 − ad)(1 − b/d) (dq, adq/b; q)k (adp/c, bcp/d; p)k
k=−m

(1 − a)(1 − b)(1 − c)(1 − ad2 /bc)

=
d(1 − ad)(1 − b/d)(1 − c/d)(1 − ad/bc)
(ap, bp; p)n(cq, ad2 q/bc; q)n
 
(c/ad, d/bc; p)m+1(1/d, b/ad; q)m+1
· − ,
(dq, adq/b; q)n(adp/c, bcp/d; p)n (1/c, bc/ad2; q)m+1 (1/a, 1/b; p)m+1
(4.1.18)

where we used −m as the lower limit of summation instead of m. This has the advantage
that, by letting m → ∞ in (4.1.18), we immediately see that if |p| < 1 and |q| < 1, then

25
(4.1.18) tends to the bibasic summation formula
n
X (1 − adpk q k )(1 − bpk /dq k ) (a, b; p)k (c, ad2 /bc; q)k
qk
(1 − ad)(1 − b/d) (dq, adq/b; q)k (adp/c, bcp/d; p)k
k=−∞
(1 − a)(1 − b)(1 − c)(1 − ad2 /bc)
=
d(1 − ad)(1 − b/d)(1 − c/d)(1 − ad/bc)
(ap, bp; p)n(cq, ad2 q/bc; q)n
 
(c/ad, d/bc; p)∞(1/d, b/ad; q)∞
· −
(dq, adq/b; q)n(adp/c, bcp/d; p)n (1/c, bc/ad2; q)∞ (1/a, 1/b; p)∞
(4.1.19)
for integer n, and with n replaced by ∞ . Some generalizations of (4.1.18) are given in
Chu [1993] and Bhatnagar and Milne [1995].
Returning to (4.1.6), notice that when c = q −n it reduces to
n
X (1 − apk q k )(1 − bpk q −k ) (a, b; p)k (q −n , aq n /b; q)k
q k = δn,0 (4.1.20)
(1 − a)(1 − b) (q, aq/b; q)k (apq n , bpq −n ; p)k
k=0

when n = 0, 1, . . . . This formula was derived independently by Bressoud [1988], Gasper

[1989a], and Krattenthaler [1995]. If we replace n, a, b and k by n − m, apm q m , bpm q −m
and j − m, respectively, (4.1.20) gives the rather general orthogonality relation
n
X
anj bjm = δn,m (4.1.21)
j=m

with
(−1)n+j (1 − apj q j )(1 − bpj q −j )(apq n , bpq −n ; p)n−1
anj = ,
(q; q)n−j (apq n , bpq −n ; p)j (bq 1−2n /a; q)n−j

(apm q m , bpm q −m ; p)j−m  a 1+2m j−m 2(j−m

bjm = − q q 2 ).
(q, aq 1+2m /b; q)j−m b
Hence, the triangular matrix A = (anj ) is inverse to the triangular matrix B = (bjm ),
where j, m, n are nonnegative integers. Since inverse matrices commute, a calculation of
the jk th term of BA leads to the orthogonality relation
j−k
X (1 − apk q k )(1 − bpk q −k )(apk+1 q k+n , bpk+1 q −k−n ; p)j−k−1
n=0
(q; q)n (q; q)j−k−n (aq 2k+n /b; q)j−k−1
 a 
· 1 − q 2k+2n (−1)n q n(j−k−1)+( 2 ) = δj,k .
j−k−n
(4.1.22)
b
By replacing j, n, a, b by n + k, k, ap−k−1 q −k , bp−k−1 q k , respectively, we see that (4.1.22)
is equivalent to the bibasic summation formula
 n
b X (aq k , bq −k ; p)n−1 (1 − aq 2k /b)
 
a
(−1)k q ( 2) = δn,0
k
1− 1− k
(4.1.23)
p p (q; q)k (q; q)n−k (aq /b; q)n+1
k=0

26
when n = 0, 1, . . . . Al-Salam and Verma [1984] derived the b → 0 limit case of (4.1.23)
 n
a X (aq k ; p)n−1

n−k
1− (−1)k q ( 2 ) = δn,0 (4.1.24)
p (q; q)k (q; q)n−k
k=0

when n = 0, 1, . . . , by using the fact that the nth q-difference of a polynomial in q of

degree less than n is equal to zero. For applications to q-analogues of Lagrange inversion,
general expansion formulas, and to the positivity of certain sums (kernels), see Gessel and
Stanton [1986], Exercises 4.2 – 4.4, and Gasper [1989a, 1989b].

4.2 Expansion, summation and transformation formulas. Let k and n be

nonnegative integers. In order to derive rather general summation and transformation
formulas for r+1 φr series, it is efficient to first derive a rather general expansion formula
that follows by using, as in §2.2 of BHS, the q-Pfaff–Saalschütz formula (3.5.1) in the form
, aq k , aq/bc; aq/b, aq/c; q, q
−k


3 φ2 q

(c, q 1−k /b; q)k (b, c; q)k  aq k

= = (4.2.1)
(aq/b, cq −k /a; q)k (aq/b, aq/c; q)k bc
to obtain, for any sequence {vk },
n
X (b, c, q −n; q)k
vk
(q, aq/b, aq/c; q)k
k=0
n X k  k
X (aq/bc, aq k , q −k ; q)j (q −n ; q)k j bc
= q vk
(q, aq/b, aq/c; q)j (q; q)k aq
k=0 j=0

X aq/bc, aq i+j , q −i−j ; q j (q −n ; q)i+j  bc i+j

n n−j


X
= qj vi+j
j=0 i=0
(q, aq/b, aq/c; q)j (q; q)i+j aq
n aq/bc, aq j , q −n ; q j


(−1)j q −(2)
X j
=
j=0
(q, aq/b, aq/c; q)j
n−j
X q j−n , aq 2j ; q

 i+j
i −ij bc
· q vi+j . (4.2.2)
(q, aq j ; q)i aq
i=0
Setting
(a, a1 , . . . , ar ; q)k
vk = zk (4.2.3)
(b1 , b2 , . . . , br+1 ; q)k
in (4.2.2) yields the desired expansion formula
a, b, c, a1, a2 , . . . , ar , q −n
 
r+4 φr+3 ; q, z
aq/b, aq/c, b1, b2 , . . . , br , br+1
n j
(aq/bc, a1 , a2 , . . . , ar , q −n ; q)j

bcz
q −(2) (a; q)2j
X j
= −
j=0
(q, aq/b, aq/c, b1, . . . , br , br+1 ; q)j aq
 2j
aq , a1 q j , a2 q j , . . . , ar q j , q j−n

bcz
· r+2 φr+1 ; q, j+1 . (4.2.4)
b1 q j , b2 q j , . . . , br q j , br+1 q j aq

27
This formula enables us to reduce the problem of deriving a transformation formula for a
r+4 φr+3 series
in terms of a single series to that of summing the r+2 φr+1 series in (4.2.4) for some
values of the parameters.
1 1 1 1
By setting a1 = qa 2 , a2 = −qa 2 , b1 = a 2 , b2 = −a 2 , br+1 = aq n+1 and ak = bk , for
k = 3, 4, . . . , r in (4.2.4), we get
1 1
a, qa 2 , −qa 2 , b, c, q −n
 
6 φ5 1 1 ; q, z
a 2 , −a 2 , aq/b, aq/c, aq n+1
n 1 1  j
(aq/bc, qa 2 , −qa 2 , q −n ; q)j (a; q)2j bcz
q −(2)
X j
= 1 1 −
j=0
(q, a 2 , −a 2 , aq/b, aq/c, aq n+1; q)j aq
 2j j+1 1 1
aq , q a 2 , −q j+1 a 2 , q j−n

bcz
· 4 φ3 1 1 ; q, j+1 . (4.2.5)
q j a 2 , −q j a 2 , aq j+n+1 aq

If we now set z = aq n+1 /bc, then we can use (4.1.8) to sum the above 4 φ3 series; thus
deriving the summation formula

aq n+1
 1 1 
a, qa 2 , −qa 2 , b, c, q −n
6 φ5 1 1 ; q,
a 2 , −a 2 , aq/b, aq/c, aq n+1 bc
1 1
(aq/bc, qa 2 , −qa 2 , q −n ; q)n (a; q)2n
= 1 1 (−1)n q n(n+1)/2
(q, a , −a
2 2 , aq/b, aq/c, aq n+1; q) n
(aq, aq/bc; q)n
= , (4.2.6)
(aq/b, aq/c; q)n
which sums a terminating very-well-poised 6 φ5 series.
Similarly, from (4.2.4), we obtain
1 1
a2 q 2+n

a, qa 2 , −qa 2 , b, c, d, e, q −n

8 φ7 1 1 ; q,
a 2 , −a 2 , aq/b, aq/c, aq/d, aq/e, aq n+1 bcde
n 1 1 j
aq n+1

(aq/bc, qa 2 , −qa 2 , d, e, q −n; q)j (a; q)2j
q −(2)
X j
= 1 1 −
j=0
(q, a 2 , −a 2 , aq/b, aq/c, aq/d, aq/e, aq n+1; q)j de
1 1
aq 2j , q j+1 a 2 , −q j+1 a 2 , dq j , eq j , q j−n aq 1+n−j
 
· 6 φ5 j 1 1 ; q, (4.2.7)
q a 2 , −q j a 2 , aq j+1 /d, aq j+1 /e, aq j+n+1 de

in which we can employ (4.2.6) to sum the 6 φ5 series and derive Watson’s [1929] transfor-
mation formula for a terminating very-well-poised 8 φ7 series as a multiple of a terminating
balanced 4 φ3 series:
1 1
a, qa 2 , −qa 2 , b, c, d, e, q −n a2 q 2+n
 
8 φ7 1 1 ; q,
a 2 , −a 2 , aq/b, aq/c, aq/d, aq/e, aq n+1 bcde
 
(aq, aq/de; q)n q −n , d, e, aq/bc
= 4 φ3 ; q, q . (4.2.8)
(aq/d, aq/e; q)n aq/b, aq/c, deq −n/a

28
 If a2 q n+1 = bcde, the 4 φ3 series in (4.2.8) becomes a terminating balanced 3 φ2 se-
ries, which can be summed by the q-Pfaff–Saalschütz formula to derive Jackson’s [1921]
summation formula for a terminating very-well-poised 8 φ7 series
 1 1
a, qa 2 , −qa 2 , b, c, d, e, q −n

8 φ7 1 1 ; q, q
a 2 , −a 2 , aq/b, aq/c, aq/d, aq/e, aq n+1
(aq, aq/bc, aq/bd, aq/cd; q)n
= , (4.2.9)
(aq/b, aq/c, aq/d, aq/bcd; q)n

where a2 q n+1 = bcde. This formula is a q-analogue of Dougall’s [1907] 7 F6 summation

formula

a, 1 + 21 a, b, c, d, e, −n
 
7 F6 1 ;1
2 a, 1 + a − b, 1 + a − c, 1 + a − d, 1 + a − e, 1 + a + n
(1 + a)n (1 + a − b − c)n (1 + a − b − d)n (1 + a − c − d)n
= , (4.2.10)
(1 + a − b)n (1 + a − c)n (1 + a − d)n (1 + a − b − c − d)n

where the series is 2-balanced, i.e, 1 + 2a + n = b + c + d + e. The reason that this

series is 2-balanced instead of balanced is that the appropriate q-analogue of the term
1 1 1 1
(1+ 21 a)k /( 12 a)k = (a+2k)/a in the 7 F6 series is not (qa 2 ; q)k /(a 2 ; q)k = (1−a 2 q k )/(1−a 2 )
1 1 1 1
but (qa 2 , −qa 2 ; q)k /(a 2 , −a 2 ; q)k = (1 − aq 2k )/(1 − a), and this introduces an additional
q-factor in the ratio of the products of the numerator and denominator parameters. Krat-
tenthaler [1995] removed some of the mystery in the factorization (4.1.5) by observing that
it is equivalent to the n = 1 case of Jackson’s 8 φ7 sum (4.2.9).
Watson [1929] showed that the b, c, d, e → ∞ limit case of his transformation formula
(4.2.9) and Jacobi’s triple product identity can be used to give a relatively simple proof of
the famous Rogers-Ramanujan identities:

∞ 2
X qn (q 2 , q 3 , q 5 ; q 5 )∞
= , (4.2.11)
n=0
(q; q) n (q; q) ∞

∞
X q n(n+1) (q, q 4 , q 5 ; q 5 )∞
= , (4.2.12)
n=0
(q; q)n (q; q)∞

where |q| < 1. See Hardy [1940] for an early history of these identities.
An important limit case of Jackson’s summation formula (4.2.9) is the sum of a non-
terminating very-well-poised 6 φ5 series
 1 1 
a, qa 2 , −qa 2 , b, c, d aq
6 φ5 1 1 ; q,
a 2 , −a 2 , aq/b, aq/c, aq/d bcd
(aq, aq/bc, aq/bd, aq/cd; q)∞
= (4.2.13)
(aq/b, aq/c, aq/d, aq/bcd; q)∞

29
 1
with |aq/bcd| < 1, which follows by letting n → ∞ in (4.2.9). When d = a 2 this formula
reduces to
" 1 1
#
a, −qa 2 , b, c qa 2
4 φ3 1 ; q,
−a 2 , aq/b, aq/c bc
1 1
(aq, aq/bc, qa 2 /b, qa 2 /c; q)∞
= 1 1 , (4.2.14)
(aq/b, aq/c, qa 2 , qa 2 /bc; q)∞
1
where |qa 2 /bc| < 1, which is a q-analogue of Dixon’s [1903] formula for the sum of a
well-poised 3 F2 series

3 F2
[ a, b, c; 1 + a − b, 1 + a − c ; 1]
Γ(1 + 12 a)Γ(1 + a − b)Γ(1 + a − c)Γ(1 + 21 a − b − c)
= , (4.2.15)
Γ(1 + a)Γ(1 + 21 a − b)Γ(1 + 12 a − c)Γ(1 + a − b − c)

where Re (1 + 21 a − b − c) > 0.

4.3 Bailey’s transformation formulas and some integral representations.

In this section we conclude the some of the most important transformation formulas for
very-well-poised series. By rewriting Jackson’s formula (4.2.9) in the form
" 1 1
#
λ, qλ 2 , −qλ 2 , λb/a, λc/a, λd/a, aq m, q −m
8 φ7 1 1 ; q, q
λ 2 , −λ 2 , aq/b, aq/c, aq/d, λq 1−m/a, λq m+1
(b, c, d, λq; q)m
= (4.3.1)
(aq/b, aq/c, aq/d, a/λ; q)m
with λ = qa2 /bcd and proceeding as in (4.2.2), one obtains a sum of a terminating very-
well-poised 8 φ7 series that can be summed with Jackson’s formula, yielding Bailey’s [1929]
transformation formula between two terminating 10 W9 series
n+1


10 W9 a; b, c, d, e, f, λaq /ef, q −n ; q, q
(aq, aq/ef, λq/e, λq/f ; q)n
=
(aq/e, aq/f, λq/ef, λq; q)n
· 10 W9 λ; λb/a, λc/a, λd/a, e, f, λaq n+1/ef, q −n ; q, q


(4.3.2)
with λ = qa2 /bcd, where for compactness we employed the 10 W9 notation for very-well-
poised series.
Watson’s transformation formula (4.2.8) follows from (4.3.2) by letting b, c, or d → ∞.
By taking the limit n → ∞ of (4.3.2) we obtain a transformation formula for nontermi-
nating very-well-poised 8 φ7 series

8 W7 (a; b, c, d, e, f ; q, λq/ef )
(aq, aq/ef, λq/e, λq/f ; q)∞
=
(aq/e, aq/f, λq, λq/ef ; q)∞
· 8 W7 (λ; λb/a, λc/a, λd/a, e, f ; q, aq/ef ) (4.3.3)

30
with λ = qa2 /bcd, where, for convergence, max(|aq/ef |, |λq/ef ) < 1. Bailey iterated
(4.3.2) to obtain
3 n+2


10 W9 a; b, c, d, e, f, a q /bcdef, q −n ; q, q
(aq, aq/de, aq/df, aq/ef ; q)n
=
(aq/d, aq/e, aq/f, aq/def ; q)n
· 10 W9 def q −n−1 /a; aq/bc, d, e, f, bdef q −n−1/a2 , cdef q −n−1 /a2 , q −n ; q, q .

(4.3.4)
It is clear that the 10 W9 on the left side of (4.3.4) tends to the 8 φ7 series on the left
side of (4.3.3) as n → ∞. However, in trying to take the n → ∞ limit of the right side we
run into the problem that the terms near both ends of the series on the right side of (4.3.4)
are large compared to those in the middle for large n, which prevents us from directly
taking the term-by-term limit. Bailey overcame this difficulty by choosing n to be an odd
integer 2m + 1, dividing the series on the right into two halves, each containing m + 1
terms, reversing the order of the second series, and then taking the limit as m → ∞ to
derive the transformation formula
1 1
a2 q 2
 
a, qa 2 , −qa 2 , b, c, d, e, f
8 φ7 1 1 ; q,
a 2 , −a 2 , aq/b, aq/c, aq/d, aq/e, aq/f bcdef
aq/bc, d, e, f
 
(aq, aq/de, aq/df, aq/ef ; q)∞
= 4 φ3 ; q, q
(aq/d, aq/e, aq/f, aq/def ; q)∞ aq/b, aq/c, def /a
(aq, aq/bc, d, e, f, a2q 2 /bdef, a2 q 2 /cdef ; q)∞
+
(aq/b, aq/c, aq/d, aq/e, aq/f, a2q 2 /bcdef, def /aq; q)∞
aq/de, aq/df, aq/ef, a2q 2 /bcdef
 
· 4 φ3 ; q, q , (4.3.5)
a2 q 2 /bdef, a2 q 2 /cdef, aq 2 /def
provided |a2 q 2 /bcdef | < 1 when the 8 φ7 series on the left side does not terminate.
Al-Salam and Verma [1982] observed that (4.3.5) is equivalent to the q-integral formula
Z b
(qt/a, qt/b, ct, dt; q)∞
dq t
a (et, f t, gt, ht; q)∞
(q, bq/a, a/b, cd/eh, cd/f h, cd/gh, bc, bd; q)∞
= b(1 − q)
(ae, af, ag, be, bf, bg, bh, bcd/h; q)∞
· 8 W7 (bcd/hq; be, bf, bg, c/h, d/h; q, ah) , (4.3.6)
where cd = abef gh and |ah| < 1. Setting h = d in (4.3.6) and then replacing g by d gives
Sears’ [1951a] nonterminating extension of the q-Pfaff–Saalschütz formula
 
a, b, c (q/e, f /a, f /b, f /c; q)
3 φ2 ; q, q =
e, f (aq/e, bq/e, cq/e, f ; q)∞
(q/e, a, b, c, qf /e; q)∞
−
(e/q, aq/e, bq/e, cq/e, f ; q)∞
aq/e, bq/e, cq/e
 
· 3 φ2 ; q, q (4.3.7)
q 2 /e, qf /e

31
with ef = abcq in the equivalent q-integral form
Z b
(qt/a, qt/b, ct; q)∞
dq t
a (dt, et, f t; q)∞
(q, bq/a, a/b, c/d, c/e, c/f ; q)∞
= b(1 − q) , (4.3.8)
(ad, ae, af, bd, be, bf ; q)∞
where c = abdef.
Rahman [1984] employed this q-integral to give a rather simple proof of the Askey
and Wilson [1985] q-beta integral formula
Z 1 1 1
h(x; 1, −1, q 2 , −q 2 ) dx
√
−1 h(x; a, b, c, d) 1 − x2
2π(abcd; q)∞
= , (4.3.9)
(q, ab, ac, ad, bc, bd, cd; q)∞
where

h(x; a1 , a2 , . . . , am ) ≡ h(x; a1 , a2 , . . . , am ; q) = h(x; a1 )h(x; a2 ) · · · h(x; am ),

∞
Y
h(x; a) ≡ h(x; a; q) = (1 − 2axq n + a2 q 2n )
n=0

and
max (|a|, |b|, |c|, |d|, |q|) < 1.
This is the integral that Askey and Wilson [1985] used to derived their orthogonality
relation for the polynomials

pn (x) ≡ pn (x; a, b, c, d|q)

q −n , abcdq n−1 , aeiθ , ae−iθ
 
−n
= (ab, ac, ad; q)na 4 φ3 ; q, q , (4.3.10)
ab, ac, ad
which are now called the Askey-Wilson polynomials.
In §2.11 of BHS the q-integral representation (4.3.6) is applied to derive Bailey’s [1936]
3-term transformation formula for 8 W7 series
2 2
8 W7 (a, b, c, d, e, f ; q, a
q /bcdef )
(aq, aq/de, aq/df, aq/ef, eq/c, f q/c, b/a, bef /a; q)∞
=
(aq/d, aq/e, aq/f, aq/def, q/c, ef q/c, be/a, bf /a; q)∞
· 8 W7 (ef /c, aq/bc, aq/cd, ef /a, e, f ; q, bd/a)
b (aq, bq/a, bq/c, bq/d, bq/e, bq/f, d, e, f ; q)∞
+
a (aq/b, aq/c, aq/d, aq/e, aq/f, bd/a, be/a, bf /a, def /a; q)∞
(aq/bc, bdef /a2, a2 q/bdef ; q)∞
·
(aq/def, q/c, b2 q/a; q)∞
· 8 W7 (b2 /a, b, bc/a, bd/a, be/a, bf /a; q, a2q 2 /bcdef ), (4.3.11)

32
where |bd/a| < 1 and |a2 q 2 /bcdef | < 1, and it is pointed out that the special case when
qa2 = bcdef gives

8 W7 (a, b, c, d, e, f ; q, q)
b (aq, c, d, e, f, bq/a, bq/c, bq/d, bq/e, bq/f ; q)∞
−
a (aq/b, aq/c, aq/d, aq/e, aq/f, bc/a, bd/a, be/a, bf /a, b2q/a; q)∞
· 8 W7 (b2 /a, b, bc/a, bd/a, be/a, bf /a; q, q)
(aq, b/a, aq/cd, aq/ce, aq/cf, aq/de, aq/df, aq/ef ; q)∞
= , (4.3.12)
(aq/c, aq/d, aq/e, aq/f, bc/a, bd/a, be/a, bf /a; q)∞

where qa2 = bcdef . This nonterminating extension of Jackson’s summation formula

(4.2.9), which is called Bailey’s 8 φ7 summation formula, can be written in the equiva-
lent q-integral form
Z b 1 1
(qt/a, qt/b, t/a 2 , −t/a 2 , qt/c, qt/d, qt/e, qt/f ; q)∞
1 1 dq t
a (t, bt/a, qt/a 2 , −qt/a 2 , ct/a, dt/a, et/a, f t/a; q)∞
b(1 − q)(q, a/b, bq/a, aq/cd, aq/ce, aq/cf, aq/de, aq/df, aq/ef ; q)∞
= , (4.3.13)
(b, c, d, e, f, bc/a, bd/a, be/a, bf /a; q)∞

where qa2 = bcdef.

Also, in §2.12 of BHS the q-integral (4.3.6) was used to give a short derivation of
Bailey’s [1947] 4-term transformation formula for 10 W9 series

10 W9 (a, b, c, d, e, f, g, h; q, q)
(aq, b/a, c, d, e, f, g, h, bq/c, bq/d; q)∞
+ 2
(b q/a, a/b, aq/c, aq/d, aq/e, aq/f, aq/g, aq/h, bc/a, bd/a; q) ∞
(bq/e, bq/f, bq/g, bq/h; q)∞
·
(be/a, bf /a, bg/a, bh/a; q)∞
· 10 W9 (b2 /a, b, bc/a, bd/a, be/a, bf /a, bg/a, bh/a; q, q)
(aq, b/a, λq/f, λq/g, λq/h, bf /λ, bg/λ, bh/λ; q)∞
=
(λq, b/λ, aq/f, aq/g, aq/h, bf /a, bg/a, bh/a; q)∞
· 10 W9 (λ, b, λc/a, λd/a, λe/a, f, g, h; q, q)
(aq, b/a, f, g, h, bq/f, bq/g, bq/h, λc/a, λd/a; q)∞
+ 2
(b q/λ, λ/b, aq/c, aq/d, aq/e, aq/f, aq/g, aq/h, bc/a, bd/a; q)∞
(λe/a, abq/λc, abq/λd, abq/λe; q)∞
·
(be/a, bf /a, bg/a, bh/a; q)∞
· 10 W9 (b2 /λ, b, bc/a, bd/a, be/a, bf /λ, bg/λ, bh/λ; q, q) (4.3.14)

and it was observed that (4.3.14) can be written in terms of the q-integrals in the compact
form
Z b 1 1
(qt/a, qt/b, ta− 2 , −ta− 2 , qt/c, qt/d, qt/e, qt/f, qt/g, qt/h; q)∞
−1 −1
dq t
a (t, bt/a, qta 2 , −qta 2 , ct/a, dt/a, et/a, f t/a, gt/a, ht/a; q)∞

33
 a (b/a, aq/b, λc/a, λd/a, λe/a, bf /λ, bg/λ, bh/λ; q)∞
=
λ (b/λ, λq/b, c, d, e, bf /a, bg/a, bh/a; q)∞
Z b 1 1
(qt/λ, qt/b, tλ− 2 , −tλ− 2 , aqt/cλ, aqt/dλ, aqt/eλ, qt/f, qt/g, qt/h; q)∞
· 1 1 dq t,
λ (t, bt/λ, qtλ− 2 , −qtλ− 2 , ct/a, dt/a, et/a, f t/λ, gt/λ, ht/λ; q)∞
(4.3.15)

where λ = qa2 /cde and a3 q 2 = bcdef gh.

Exercises 4

4.1 Prove that

n
X (1 − adpk q k )(1 − bpk /dq k ) (a, b; p)k (q −n , ad2 q n /b; q)k
qk
(1 − ad)(1 − b/d) (dq, adq/b; q)k (adpq n , bp/dq n ; p)k
k=0
(1 − d)(1 − ad/b)(1 − adq n )(1 − dq n /b)
= , n = 0, 1, . . . .
(1 − ad)(1 − d/b)(1 − dq n )(1 − adq n /b)

4.2 Use (4.1.24) to show that Euler’s transformation formula

∞ ∞
X
n
X xk (k)
an bn x = (−1)k f (x)∆k a0 ,
n=0
k!
k=0

where
f (x) = b0 + b1 x + b2 x2 + · · ·
and
k  
k
X
jk
∆ a0 = (−1) ak−j ,
j
j=0

has the bibasic extension

∞ ∞ k
X
n
X
k
X
k (1 − apn q n )wn An
An Bn (xw) = (apq ; p)k−1 x
n=0 n=0
(q; q)k−n (apq k ; p)n
k=0
∞
(apk q k ; p)j j
Bj+k (−x)j q (2) .
X
·
(q; q)j
j=0

See Al-Salam and Verma [1984].

4.3 Use (4.1.6) to derive the Gasper [1989a] bibasic expansion formula
∞ ∞
(xw)n (1 − γpn q n )(1 − σpn q n ) n
(−x)n q n+( 2 )
X X
An B n =
n=0
(q; q)n n=0 (q; q)n

34
 ∞
X 1 − γσ −1 q 2n+2k
· Bn+k xk
(q; q)k (γpq n+k , σpq −n−k ; p)n
k=0
n
X (q −n ; q)j (γσ −1 q n+j+1 ; q)n+k−j−1
·
j=0
(q; q)j

· (γpq j , σpq −j ; p)n−1 Aj Cj,n+k−j wj q n(j−n−k) ,

where Aj , Bj , Cj,k are complex numbers such that the series converge absolutely and
Cj,0 = 1, for j = 0, 1, . . . .
4.4 Show that if p = q, then the σ → ∞ limit case of Ex. 4.3 gives the expansion formula
 
aR , cT
r+t φs+u ; q, xw
b S , dU
∞
(cT , eK , σ, γq j+1/σ; q)j  x j j
[(−1)j q (2) ]u+m−t−k
X
= j
j=0
(q, dU , fM , γq ; q)j σ
 2j p p
γq /σ, q j+1 γ/σ, −q j+1 γ/σ, σ −1 ,
· t+k+4 φu+m+3 p p
q j γ/σ, −q j γ/σ, γq 2j+1 , dU q j ,
cT q j , eK q j

j(u+m−t−k)
; q, xq
fM q j
q −j , γq j , aR , fM
 
· r+m+2 φs+k+2 ; q, wq ,
γq j+1 /σ, q 1−j /σ, bS , eK

where we employed the contracted notation of representing a1 , . . . , ar by

aR , etc.
4.5 Derive formulas (4.3.2) and (4.3.4).
4.6 Show that the q-integral formulas (4.3.6) and (4.3.8) are equivalent to formulas (4.3.5)
and (4.3.7), respectively.
4.7 Use the q-integral representation (4.3.6) to derive formulas (4.3.11) and (4.3.14).

References

Al-Salam, W.A. and Verma, A. [1982] Some remarks on q-beta integral, Proc. Amer.
Math. Soc., 85, 360–362.
Al-Salam, W.A. and Verma, A. [1984] On quadratic transformations of basic series,
SIAM J. Math. Anal., 15, 414–420.
Andrews, G.E. [1969] On a calculus of partition functions, Pacific J. Math., 31, 555–
562.

35
 Andrews, G.E. [1986] q-Series: Their Development and Application in Analysis, Num-
ber Theory, Combinatorics, Physics, and Computer Algebra, CBMS Regional Confer-
ence Lecture Series, 66, Amer. Math. Soc., Providence, R. I.
Andrews, G.E. and Askey, R. [1977] Enumeration of partitions: the role of Eulerian
series and q-orthogonal polynomials, Higher Combinatorics (M. Aigner, ed.), Reidel,
Boston, Mass., pp. 3–26.
Andrews, G.E. and Askey, R. [1978] A simple proof of Ramanujan’s summation of the
1 ψ1 , Aequationes Math., 18, 333–337.

Askey, R. [1980] Ramanujan’s extensions of the gamma and beta functions, Amer.
Math. Monthly, 87, 346–359.
Askey, R. and Ismail, M.E.H. [1983] A generalization of ultraspherical polynomials,
Studies in Pure Mathematics (P. Erdős, ed.), Birkhäuser, Boston, Mass., pp. 55–78.
Askey, R. and Roy, R. [1986] More q-beta integrals, Rocky Mtn. J. Math., 16, 365–372.
Askey, R. and Wilson, J.A. [1985] Some basic hypergeometric polynomials that gener-
alize Jacobi polynomials, Memoirs Amer. Math. Soc., 319.
Bailey, W.N. [1929] An identity involving Heine’s basic hypergeometric series, J. Lon-
don Math. Soc., 4, 254–257.
Bailey, W.N. [1935] Generalized Hypergeometric Series, Cambridge University Press,
Cambridge, reprinted by Stechert-Hafner, New York, 1964.
Bailey, W.N. [1936] Series of hypergeometric type which are infinite in both directions,
Quart. J. Math. (Oxford), 7, 105–115.
Bailey, W.N. [1941] A note on certain q-identities, Quart. J. Math. (Oxford), 12,
173–175.
Bailey, W.N. [1947] Well-poised basic hypergeometric series, Quart. J. Math. (Ox-
ford), 18, 157–166.
Berndt, B.C. [1993] Ramanujan’s theory of theta-functions, Theta Functions From the
Classical to the Modern (M. Ram Murty, ed.), CRM Proceedings & Lecture Notes, 1,
Amer. Math. Soc., Providence, R.I, pp. 1–63.
Bhatnagar, G. and Milne, S.C. [1995] Generalized bibasic hypergeometric series and
their U (n) extensions, to appear.
Bressoud, D.M. [1988] The Bailey Lattice: an introduction,
Ramanujan Revisited (G. E. Andrews et al., eds.), Academic Press, New York, pp.
57–67.
Cauchy, A.-L. [1843] Mémoire sur les fonctions dont plusieurs valeurs sont liées entre
elles par une équation linéaire, et sur diverses transformations de produits composés
d’un nombre indéfini de facteurs, C. R. Acad. Sci. Paris, T. XVII, p. 523, Oeuvres
de Cauchy, 1re série, T. VIII, Gauthier-Villars, Paris, 1893, pp. 42–50.

36
Chu, W.C. [1993] Inversion techniques and combinatorial identities, Bullettino U.M.I.,
7, 737–760.
Daum, J.A. [1942] The basic analog of Kummer’s theorem, Bull. Amer. Math. Soc.,
48, 711–713.
Dixon, A.C. [1903] Summation of a certain series, Proc. London Math. Soc. (1), 35,
285–289.
Dougall, J. [1907] On Vandermonde’s theorem and some more general expansions,
Proc. Edin. Math. Soc., 25, 114–132.
Fine, N.J. [1988] Basic Hypergeometric Series and Applications, Mathematical Surveys
and Monographs, Vol. 27, Amer. Math. Soc., Providence, R. I.
Gasper, G. [1987] Solution to problem #6497 (q-Analogues of a gamma function iden-
tity, by R. Askey), Amer. Math. Monthly, 94, 199–201.
Gasper, G. [1989a] Summation, transformation, and expansion formulas for bibasic
series, Trans. Amer. Math. Soc., 312, 257–277.
Gasper, G. [1989b] q-Extensions of Clausen’s formula and of the inequalities used by
de Branges in his proof of the Bieberbach, Robertson, and Millin conjectures, SIAM
J. Math. Anal., 20, 1019–1034.
Gasper, G. and Rahman, M. [1990a] Basic Hypergeometric Series, Encyclopedia of
Mathematics and Its Applications, 35, Cambridge University Press, Cambridge and
New York.
Gasper, G. and Rahman, M. [1990b] An indefinite bibasic summation formula and
some quadratic, cubic, and quartic summation and transformation formulas, Canad.
J. Math., 42, 1–27.
Gauss, C.F. [1813] Disquisitiones generales circa seriem infinitam ..., Comm. soc.
reg. sci. Gött. rec., Vol. II; reprinted in Werke 3 (1876), pp. 123–162.
Gessel, I. and Stanton, D. [1986] Another family of q-Lagrange inversion formulas,
Rocky Mtn. J. Math., 16, 373–384.
Hahn, W. [1949] Über Polynome, die gleichzeitig zwei verschiedenen Orthogonalsyste-
men angehören, Math. Nachr., 2, 263-278.
Hardy, G.H. [1940] Ramanujan, Cambridge University Press, Cambridge; reprinted
by Chelsea, New York, 1978.
Heine, E. [1846] Über die Reihe ..., J. reine angew. Math., 32, 210–212.
Heine, E. [1847] Untersuchungen über die Reihe ..., J. reine angew. Math., 34, 285–
328.
Heine, E. [1878] Handbuch der Kugelfunctionen, Theorie und Anwendungen, Vol. 1,
Reimer, Berlin.

37
Ismail, M.E.H. [1977] A simple proof of Ramanujan’s 1 ψ1 sum, Proc. Amer. Math.
Soc., 63, 185–186.
Jackson, F.H. [1904] A generalization of the functions Γ(n) and xn , Proc. Roy. Soc.
London, 74, 64–72.
Jackson, F.H. [1910a] Transformations of q-series, Messenger of Math., 39, 145–153.
Jackson, F.H. [1910b] On q-definite integrals, Quart. J. Pure and Appl. Math., 41,
193-203.
Jackson, F.H. [1921] Summation of q-hypergeometric series, Messenger of Math., 50,
101–112.
Jackson, M. [1950] On Lerch’s transcendant and the basic bilateral hypergeometric
series 2 ψ2 , J. London Math. Soc., 25, 189–196.
Jacobi, C.G.J. [1829] Fundamenta Nova Theoriae Functionum Ellipticarum, Regiomonti.
Sumptibus fratrum Bornträger; reprinted in Gesammelte Werke 1 (1881), 49–239,
Reimer, Berlin.
Koornwinder, T.H. [1989] Representations of the twisted SU (2) quantum group and
some q-hypergeometric orthogonal polynomials, Proc. Kon. Nederl. Akad. Wetensch.
Series A, 92, 97–117.
Krattenthaler, C. [1995] A new matrix inverse, Proc. Amer. Math. Soc., to appear.
Pfaff, J.F. [1797] Observationes analyticae ad L. Euler Institutiones Calculi Integralis,
Vol. IV, Supplem. II et IV, Historia de 1793, Nova acta acad. sci. Petropolitanae, 11
(1797), pp. 38–57.
Rahman, M. [1984] A simple evaluation of Askey and Wilson’s q-beta integral, Proc.
Amer. Math. Soc., 92, 413–417.
Ramanujan, S. [1915] Some definite integrals, Messenger of Math., 44, 10–18.
Rogers, L.J. [1894] Second memoir on the expansion of certain infinite products, Proc.
London Math. Soc., 25, 318–343.
Rogers, L.J. [1895] Third memoir on the expansion of certain infinite products, Proc.
London Math. Soc., 26, 15–32.
Saalschütz, L. [1890] Eine Summationsformel, Zeitschr. Math. Phys., 35, 186–188.
Sears, D.B. [1951a] Transformations of basic hypergeometric functions of special type,
Proc. London Math. Soc. (2), 52, 467–483.
Sears, D.B. [1951b] On the transformation theory of basic hypergeometric functions,
Proc. London Math. Soc. (2), 53, 158–180.
Slater, L.J. [1966] Generalized Hypergeometric Functions, Cambridge University Press,
Cambridge.

38
Thomae, J. [1869] Beiträge zur Theorie der durch die Heinesche Reihe ..., J. reine
angew. Math., 70, 258–281.
Thomae, J. [1870] Les séries Heinéennes supérieures, ou les séries de la forme ...,
Annali di Matematica Pura ed Applicata, 4, 105–138.
Venkatachaliengar, K. [1988] Development of Elliptic Functions According to Ramanu-
jan, Tech. Rep. no. 2, Madurai Kamaraj University, Madurai.
Watson, G.N. [1929] A new proof of the Rogers-Ramanujan identities, J. London Math.
Soc., 4, 4–9.
Whittaker, E.T. and Watson, G.N. [1965] A Course of Modern Analysis, 4th edition,
Cambridge University Press, Cambridge.

Department of Mathematics, Northwestern University, Evanston, IL 60208, USA

E-mail address: [email protected]

39