TWO ENTRIES ON WELL-POISED BILATERAL

HYPERGEOMETRIC SERIES IN RAMANUJAN’S LOST NOTEBOOK

BRUCE C. BERNDT1 and WENCHANG CHU

1. Introduction
On page 200 of his lost notebook, in the pagination of [4], Ramanujan offers two
results on certain bilateral hypergeometric series. As we shall see, the second follows
from a theorem of J. Dougall [2]. The first gives a formula for the derivative of a
quotient of two certain bilateral hypergeometric series. Ramanujan’s formula needs
to be slightly corrected, but what is remarkable is that such a formula exists! This
is one of those instances where we can undauntedly claim that if Ramanujan had not
discovered the formula, no one else, at least in the foreseeable future, would have done
so.
We first state the second formula, which requires modest deciphering, because of
Ramanujan’s use of ellipses to denote missing terms. It will be used in the proof of
Ramanujan’s first formula on page 200.
Entry 1.1 (p. 200). Let α, β, γ, δ, and ξ be complex numbers such that Re(α + β +
γ + δ) > 3. Then
∞
X ξ + 2n
n=−∞
Γ(α + ξ + n)Γ(β − ξ − n)Γ(γ + ξ + n)Γ(δ − ξ − n)Γ(α − n)Γ(β + n)Γ(γ − n)Γ(δ + n)
sin(πξ)Γ(α + β + γ + δ − 3)
= .
πΓ(α + γ + ξ − 1)Γ(β + δ − ξ − 1)Γ(α + β − 1)Γ(β + γ − 1)Γ(γ + δ − 1)Γ(δ + α − 1)
(1.1)
We state a corrected version of Ramanujan’s more interesting formula. At the end
of the paper, we indicate the mistakes in Ramanujan’s original formula.
Entry 1.2 (Corrected, p. 200). Define, for any real numbers θ and s and for any
complex numbers α, β, γ, and δ, such that Re(α + β + γ + δ) > 3,
∞
X e(n+s)iθ
ϕs (θ) := . (1.2)
n=−∞
Γ(α + s + n)Γ(β − s − n)Γ(γ + s + n)Γ(δ − s − n)

1
Research partially supported by grant MDA904-00-1-0015 from the National Security Agency.
1
2 BRUCE C. BERNDT AND WENCHANG CHU

Then

α+β+γ+δ−4 i(π−θ)(α−β+γ−δ+2s+2t)/2
d ϕs (θ) i sin{π(s − t)} 2 sin( 12 θ) e
= . (1.3)
dθ ϕt (θ) πφ2t (θ)Γ(α + β − 1)Γ(β + γ − 1)Γ(γ + δ − 1)Γ(δ + α − 1)

2. Background on Bilateral Series

For every integer n, define
Γ(a + n)
(a)n := . (2.1)
Γ(a)
The bilateral hypergeometric series p Hp is defined for complex parameters a1 , a2 , . . . , ap
and b1 , b2 , . . . , bp by
  ∞
a1 a2 , . . . , a p ; X (a1 )n (a2 )n · · · (ap )n n
H
p p z := z .
b1 , b2 , . . . , bp ; (b 1 )n (b 2 )n · · · (b p )n
n=−∞

With the use of D’Alembert’s ratio test, it can be checked that p Hp converges only for
|z| = 1, provided that [5, p. 181, eq. (6.1.1.6)]
Re(b1 + b2 + · · · + bp − a1 − a2 − · · · − ap ) > 1. (2.2)
The series p Hp is said to be well-poised if
a1 + b1 = a2 + b2 = · · · = ap + bp .
In 1907, J. Dougall [2] proved that a well-poised series 5 H5 could be evaluated at z = 1.
In order to state this evaluation, define
 
a1 , a2 , . . . , am Γ(a1 )Γ(a2 ) · · · Γ(am )
Γ := .
b1 , b2 , . . . bn Γ(b1 )Γ(b2 ) · · · Γ(bn )
Then Dougall’s formula [5, p. 182, eq. (6.1.2.5)] is given by
1 + 21 a,
 
b, c, d, e;
5 H5 1 1
2
a, 1 + a − b, 1 + a − c, 1 + a − d, 1 + a − e;

1 − b, 1 − c, 1 − d, 1 − e, 1 + a − b, 1 + a − c, 1 + a − d,
=Γ
1 + a, 1 − a, 1 + a − b − c, 1 + a − b − d, 1 + a − b − e,

1 + a − e, 1 + 2a − b − c − d − e
, (2.3)
1 + a − c − d, 1 + a − c − e, 1 + a − d − e
where, for convergence, by (2.2),
1 + Re(2a − b − c − d − e) > 0. (2.4)
We need one further result, namely, the bilateral binomial theorem discovered by
M. E. Horn [3]. If a and b are real numbers and z is a complex number with |z| = 1,
then
(1 − z)b−a−1 Γ(1 − a)Γ(b)
 
a;
H
1 1 z = . (2.5)
b; (−z)c−1 Γ(b − a)
It would seem that Ramanujan had discovered (2.5), but we are unaware of any mention
of it by Ramanujan in his papers or notebooks. We remark that the bilateral binomial
 BILATERAL HYPERGEOMETRIC SERIES IN RAMANUJAN’S LOST NOTEBOOK 3

theorem can also be recovered from another bilateral hypergeometric series identity [1,
p. 110, Thm. 2.8.2] due to Dougall [2], namely,
 
a, b; Γ(1 − a)Γ(1 − b)Γ(c)Γ(d)Γ(c + d − a − b − 1)
2 H2 1 = , (2.6)
c, d; Γ(c − a)Γ(c − b)Γ(d − a)Γ(d − b)
where Re(c+d−a−b) > 1 for convergence. In fact, in the identity above, first replacing
b by dz and secondly letting d → +∞, we derive (2.5) in view of Stirling’s asymptotic
formula for the Γ-function.
In the sequel, we very often use the classical reciprocal formula
π
Γ(z)Γ(1 − z) = . (2.7)
sin(πz)

3. Proof of Entry 1.1

We show that (2.3) leads to a proof of Entry 1.1.

Proof. Let S denote the series on the left-hand side of (1.1). Define
sin{π(β − ξ)} sin{π(δ − ξ)} sin{πα} sin{πγ}
Ω := . (3.1)
π4
Using (2.7) and (3.1), we see that we can write S in the form
∞
X (ξ + 2n)Γ(1 + ξ + n − β)Γ(1 + ξ + n − δ)Γ(1 + n − α)Γ(1 + n − γ)
S = Ωξ
n=−∞
ξΓ(α + ξ + n)Γ(γ + ξ + n)Γ(β + n)Γ(δ + n)
Γ(1 + ξ − β)Γ(1 + ξ − δ)Γ(1 − α)Γ(1 − γ)
= Ωξ
Γ(α + ξ)Γ(γ + ξ)Γ(β)Γ(δ)
∞
X (1 + 1 ξ)n (1 − α)n (1 + ξ − β)n (1 − γ)n (1 + ξ − δ)n
2
× . (3.2)
n=−∞
( 12 ξ)n (α + ξ)n (β)n (γ + ξ)n (δ)n

Note that the series (3.2) is well-poised, and so we can invoke (2.3) with a = ξ, b = 1−α,
c = 1 + ξ − β, d = 1 − γ, and e = 1 + ξ − δ. Thus, for Re(α + β + γ + δ) > 3 for
convergence, we deduce that
Γ(1 + ξ − β)Γ(1 + ξ − δ)Γ(1 − α)Γ(1 − γ)
S = Ωξ
Γ(α + ξ)Γ(γ + ξ)Γ(β)Γ(δ)
Γ(α)Γ(β − ξ)Γ(γ)Γ(δ − ξ)Γ(α + ξ)Γ(β)Γ(γ + ξ)Γ(δ)Γ(α + β + γ + δ − 3)
×
Γ(1 + ξ)Γ(1 − ξ)Γ(α + β − 1)Γ(α + γ + ξ − 1)Γ(α + δ − 1)Γ(β + γ − 1)
1
×
Γ(β + δ − ξ − 1)Γ(γ + δ − 1)
sin(πξ)Γ(α + β + γ + δ − 3)
= ,
πΓ(α + γ + ξ − 1)Γ(β + δ − ξ − 1)Γ(α + β − 1)Γ(β + γ − 1)Γ(γ + δ − 1)Γ(δ + α − 1)
where we applied (2.7) five times, used the value of Ω from (3.1), and simplified. 
4 BRUCE C. BERNDT AND WENCHANG CHU

4. Proof of Entry 1.2

We first replace the functions in Entry 1.2 by another pair with which it is easier to
work. With four applications of (2.7), we see that we can write ϕs (θ) in the form

esiθ Hs (θ)
ϕs (θ) = (4.1)
Γ(α + s)Γ(β − s)Γ(γ + s)Γ(δ − s)

where
 
1 − β + s, 1 − δ + s; iθ
Hs (θ) := 2 H2 e . (4.2)
α + s, γ + s;

Thus, we prove an analogue with ϕs and ϕt replaced by Hs and Ht , respectively. At

the end of our proof, we convert our result to (1.2).
For brevity, we introduce the notation

(1 − β + s)n (1 − δ + s)n
hsin := .
(α + s)n (γ + s)n

In particular, we can then write

  ∞
1 − β + s, 1 − δ + s; iθ X
Hs (θ) = 2 H2 e = hsin einθ .
α + s, γ + s;
n=−∞

Proof. By the quotient rule for derivatives,

d n Hs (θ) (s−t)iθ o ∆
e = 2tiθ 2 , (4.3)
dθ Ht (θ) e Ht (θ)

where
d  siθ d  tiθ
∆ = etiθ Ht (θ) e Hs (θ) − esiθ Hs (θ)

e Ht (θ) . (4.4)
dθ dθ
Using the notation above and in the previous paragraph and setting k = m + n in
the second equality below, we find that
∞
X
∆=i (s − t + n − m) hsin htim e(s+t+n+m)iθ
m,n=−∞
X ∞
=i (s − t − k + 2n) hsin htik−n e(s+t+k)iθ
k,n=−∞
∞ ∞
X
(s+t+k)iθ
X s − t − k + 2n
=i (s − t − k) htik e hsin hk + ti−n . (4.5)
k=−∞ n=−∞
s−t−k
 BILATERAL HYPERGEOMETRIC SERIES IN RAMANUJAN’S LOST NOTEBOOK 5

Observe that the inner sum above is a well-poised 5 H5 , requiring that Re(α+β+γ+δ) >
3 for convergence. Thus, we can use (2.3) to obtain the evaluation
1 + 21 (s − t − k), 1 − α − t − k, 1 − β + s, 1 − γ − t − k, 1 − δ + s;
 
5 H5 1 1
2
(s − t − k), α + s, β − t − k, γ + s, δ − t − k;
 
α + t + k, γ + t + k, β − t − k, δ−t−k
=Γ
1 + s − t − k, 1 − s + t + k, α + γ + s + t + k − 1, β + δ − s − t − k − 1
 
α + s, β − s, γ + s, δ − s, α+β+γ+δ−3
×Γ . (4.6)
α + β − 1, β + γ − 1, γ + δ − 1, δ+α−1
Using the evaluation (4.6) in (4.5) and simplifying the expressions involving gamma
functions and rising factorials, we find that
 
α + t, β − t, γ + t, δ−t
∆ = iΓ
s − t, 1 − s + t, α + γ + s + t − 1, β + δ − s − t − 1
 
α + s, β − s, γ + s, δ − s, α+β+γ+δ−3
×Γ
α + β − 1, β + γ − 1, γ + δ − 1, δ+α−1
∞
X (s + t − β − δ + 2)k
× ei(s+t)θ eikθ . (4.7)
k=−∞
(s + t + α + γ − 1)k

We next apply Horn’s bilateral binomial theorem (2.5) with a = s + t − β − δ + 2

and b = s + t + α + γ − 1. Thus,
∞  
X (s + t − β − δ + 2)k ikθ s + t − β − δ + 2; iθ
e = 1 H1 e
(s + t + α + γ − 1)k s + t + α + γ − 1;
k=−∞

2−α−γ−s−t Γ(α + γ + s + t − 1)Γ(β + δ − s − t − 1)
= −eiθ (1 − eiθ )α+β+γ+δ−4 .
Γ(α + β + γ + δ − 3)
(4.8)
Now substitute (4.8) into (4.7), use (2.7), and cancel common gamma function factors
to arrive at

2−α−γ−s−t
∆ = e(s+t)iθ −eiθ (1 − eiθ )α+β+γ+δ−4
 
α + s, β − s, γ + s, δ − s, α + t, β − t, γ + t, δ−t
×Γ
s − t, 1 − s + t, α + β − 1, β + γ − 1, γ + δ − 1, δ + α − 1
i
α+β+γ+δ−4 i(π−θ)(α−β+γ−δ+2s+2t)/2
= sin{π(s − t)} 2 sin 2θ e
π  
α + s, β − s, γ + s, δ − s, α + t, β − t, γ + t, δ−t
×Γ . (4.9)
α + β − 1, β + γ − 1, γ + δ − 1, δ+α−1
Lastly, substituting the expression just displayed into (4.3) and then reformulating the
result according to the relation (4.1) between ϕt (θ) and Ht (θ), we derive the identity

α+β+γ+δ−4 i(π−θ)(α−β+γ−δ+2s+2t)/2
d n ϕs (θ) o i sin{π(s − t)} 2 sin 2θ e
= 2
,
dθ ϕt (θ) πϕt (θ)Γ(α + β − 1)Γ(β + γ − 1)Γ(γ + δ − 1)Γ(δ + α − 1)
6 BRUCE C. BERNDT AND WENCHANG CHU

which is (1.3). The proof is thus complete. 

Let φt (θ) = e−tiθ ϕt (θ). We end our paper with Ramanujan’s rendition of Entry 1.2,
given by
 α+β+γ+δ−4 i(α−β+γ−δ+2s−2t){(π−θ)/2+π[θ/(2π)]}
d n φs (θ) o i sin{π(s − t)} 2 sin 2θ  e
= 2
.
dθ φt (θ) πφt (θ)Γ(α + β − 1)Γ(β + γ − 1)Γ(γ + δ − 1)Γ(δ + α − 1)
(4.10)
Note that Ramanujan’s function φs (θ) does not have the factor esiθ in ϕs (θ), defined
in (1.2). The second major difference between the two formulas is in the exponent of
e on the right-hand sides. One would guess that [x] in Ramanujan’s exponent denotes
the greatest integer less than or equal to x. The powers of 2 sin( 12 θ) in both (1.3)
and (4.10) are the same, except that Ramanujan has absolute values around 2 sin( 12 θ),
which could reflect the assumption that 0 < θ < 2π. However, if that were the case,
then Ramanujan would not need to have recorded [θ/(2π)] in the exponent of e. In
conclusion, except for multiplicative expressions of absolute value equal to 1, the other
parts of the formulas (4.10) and (1.3) are identical.
References
[1] G. E. Andrews, R. Askey, and R. Roy, Special Functions, Cambridge University Press, Cambridge,
2000.
[2] J. Dougall, On Vandermonde’s theorem and some more general expansions, Proc. Edinburgh
Math. Soc. 25 (1907), 114–132.
[3] M. E. Horn, Bilateral binomial theorem, SIAM Problem 03-001 (2003).
[4] S. Ramanujan, The Lost Notebook and Other Unpublished Papers, Narosa, New Delhi, 1988.
[5] L. J. Slater, Generalized Hypergeometric Functions, Cambridge University Press, Cambridge,
1966.

Department of Mathematics, University of Illinois, 1409 West Green Street, Ur-

bana, IL 61801, USA
E-mail address: berndt@math.uiuc.edu

Dipartimento di Matematica, Università degli Studi di Lecce, Lecce-Arnesano P. O. Box

193, 73100 Lecce, Italia
E-mail address: chu.wenchang@unile.it