Noname manuscript No.

(will be inserted by the editor)

From Wallis and Forsyth to Ramanujan

Paul Levrie · Amrik Singh Nimbran

the date of receipt and acceptance should be inserted later

Abstract We show how a couple of Ramanujan’s series for 1/π can be de-
duced directly from Forsyth’s series and from Wallis’s product formula for π.
The same method is used to obtain Bauer’s alternating series.
Mathematics Subject Classification (2000) 11Y60 · 33D15 · 05A10
Keywords Series for 1/π, Wallis’s product formula, Ramanujan’s series

1 Introduction

In 1883 the Cambridge mathematician A.R. Forsyth derived the following

formula for π1 using Legendre polynomials [5]:

∞ 2n 2


X
n 4
= . (1)
n=0
(2n − 1)2 24n π

J.W.L. Glaisher [6] used elliptic functions to prove in 1905 a number of

similar formulas including the following one:

∞ 2n 2


X
n 4
= . (2)
n=0
(n + 1)24n π

Paul Levrie
Faculty of Applied Engineering, University of Antwerp, Groenenborgerlaan 171, 2020
Antwerpen; Department of Computer Science, KU Leuven, P.O. Box 2402, B-3001 Heverlee,
Belgium
E-mail: paul.levrie@cs.kuleuven.be
Amrik Singh Nimbran
B3-304, Palm Grove Heights, Ardee City, Gurgaon, Haryana, INDIA 122003
E-mail: amrikn622@gmail.com
2

And there is also this series:

∞ 2n 2


X
n 2
− = . (3)
n=0
(2n − 1)24n π

Both (2) and (3) are equivalent with Euler’s version of Wallis’s famous product
formula for π derivable from the infinite product expansion of the sine function
[14]:
∞ 2n 2


X
n 1 12 32 12 32 52 12 32 52 72
=1+ + + + + ···
n=0
(n + 1)24n 2·22 2·42 6 2·42 62 8 2·42 62 82 10
2 2 2 2 2 2 2 2 2
3·3 1 3 1 3 5 1 3 5 7
= + + + + ···
2·4 2·4 6 2·4 6 8 2·42 62 82 10
2 2 2

3·3 5·5 12 32 52 12 32 52 72
= · + + + ···
2·4 4·6 2·4 6 8 2·42 62 82 10
2 2
2 2 2 2
3·3 5·5 7·7 1 3 5 7
= · · + + ···
2·4 4·6 6·8 2·42 62 82 10
3·3 5·5 7·7 4
= · · ··· =
2·4 4·6 6·8 π
and
∞ 2n 2


X
n 1 12 3 12 32 5 12 32 52 7
− =1− − − − − ···
n=0
(2n − 1)24n 22 22 42 22 42 62 22 42 62 82
2 2 2 2 2 2
1·3 1 3 1 3 5 1 3 5 7
= − 2 2 − 2 2 2 − 2 2 2 2 − ···
2·2 2 4 2 4 6 2 4 6 8
1·3 3·5 12 32 5 12 32 52 7
= · − − 2 2 2 2 − ···
2·2 4·4 22 42 62 2 4 6 8
1·3 3·5 5·7 12 32 52 7
= · · − − ···
2·2 4·4 6·6 22 42 62 82
1·3 3·5 5·7 2
= · · ··· = .
2·2 4·4 6·6 π
The following series is due to G. Bauer (1859) [10]:
∞ 2n 3


n (4n + 1) 2
X
n
(−1) = . (4)
n=0
26n π

All the previous series converge very slowly. The last one can be found in
S. Ramanujan’s first letter to G. Hardy, dated January 31, 1913. In a paper
published in 1914 [16] Ramanujan lists some more rapidly converging series
for 1/π which he found using modular equations. These are two of them:

∞ 2n 3


X (6n + 1) n 4
= , (5)
n=0
28n π
 3

∞ 2n 3


X (42n + 5) n 16
= . (6)
n=0
212n π
In this paper we show how Ramanujan’s series (5) and a number of other
series for π1 can be derived from Forsyth’s in a straightforward way. The same
method can also be applied to Glaisher’s series (2) and to (3), leading to a
proof of (4) and (6).
Series for π1 have received much attention lately, see for instance the survey
paper by Baruah, Berndt and Chan [2] and the work of Guillera [7, 8] who uses
the Wilf-Zeilberger method to find series of this form. Chu [3] and Liu [11,
12] have obtained similar results by manipulating hypergeometric series using
Dougall’s and Gauss’s summation formulae.

2 A first recurrence

Note that all the series in the introduction contain the central binomial coef-
ficients which can be rewritten using the Pochhammer symbol:
1 1
 


2n 2n 2 n 2n 2 n
=2 =2 .
n n! (1)n
Using this in the general term of (1) we find that:
2n 2 1 2 1 2 1 2





n 1 2 n 1 2 n−1 1 2 n−1
= = = .
(2n − 1)2 24n (2n − 1)2 (1)2n 4 (1)2n 4 (2)2n−1
Hence Forsyth’s series can be written in the following form:
∞ 1 2


4 1X 2 n
=1+ . (7)
π 4 n=0 (2)2n

This last series is a special case of the more general series

∞
(a)2n
 
X a a 1
= 3 F2 ;1
n=0
(a + b)2n a+b a+b
1
with a = 2 and b = 32 . In the rest of the paper we will use the abbreviation:
∞
X (a)2n
s(a, b) = . (8)
n=0
(a + b)2n

This expression satisfies the following recurrence relation:

Theorem 1 (+0, +1–scheme)
2a + 3b − 2 b3
s(a, b) = + · s(a, b + 1). (9)
2(2b − 1) 2(2b − 1)(a + b)2
4

Proof Manipulation of the general term in the series s(a, b + 1) leads to the
required result. We start by writing:
∞ ∞
X (a)2n 1 X (a)2n (a + b + n − (a + n))2
= .
n=0
(a + b + 1)2n b2 n=0 (a + b + 1)2n

If we work out the numerator, we get the following 3 series:

∞ ∞
X (a)2n (a + b + n)2 2
X (a)2n
2
= (a + b) 2
= (a + b)2 s(a, b)
n=0
(a + b + 1) n n=0
(a + b)n
∞ ∞
X (a)2n (a + n)2 2
X (a)2n+1
2
= (a + b) 2 = (a + b)2 (s(a, b) − 1)
n=0
(a + b + 1) n n=0
(a + b)n+1
∞ ∞
X (a)2n (a + b + n)(a + n) 2
X (a)2n (a + n)
2
= (a + b) .
n=0
(a + b + 1)n n=0
(a + b)2n (a + b + n)

We now deal with this last series. Using a similar trick as before, we get:
∞ ∞
X (a)2n (a + n) 1 X (a)2n (a + n)(a + b + n − (a + n))
=
n=0
(a + b)2n (a + b + n) b n=0 (a + b)2n (a + b + n)
∞ ∞
!
1 X (a)2n (a + n) X (a)2n+1
= −
b n=0
(a + b)2n n=0
(a + b)2n (a + b + n)
∞ ∞
!
1 X (a)2n (a + n) X (a)2n+1
= a+ −
b n=1
(a + b)2n n=0
(a + b)2n (a + b + n)
∞
1 X (a)2n+1 (a + n + 1)
= a+
b n=0
(a + b)2n+1
∞
!
X (a)2n+1
−
n=0
(a + b)2n (a + b + n)
∞
a 1 X (a)2n+1 (a + n + 1 − (a + b + n))
= +
b b n=0 (a + b)2n+1
∞
a 1 − b X (a)2n+1 a 1−b
= + = + (s(a, b) − 1).
b b n=0 (a + b)2n+1 b b

We bring everything together:

(a + b)2
  
a 1−b
s(a, b + 1) = s(a, b) + s(a, b) − 1 − 2 + (s(a, b) − 1) .
b2 b b
Rearranging leads to (9).
Note that this theorem can be found in a slightly less general form in Knopp
[9, p. 261-262]. It is a special case of Kummer’s transformation of series.
Using Theorem 1 we are now able to rewrite s(a, b):
 5

Theorem 2
∞
X 2a + 3(b + n) − 2 (b)3n
s(a, b) = · 2n . (10)
n=0
2(2(b + n) − 1) 2 (b − 21 )n (a + b)2n

Proof From (9) it follows that:

2a + 3(b + n) − 2 (b + n)3
s(a, b + n) = + 2 · s(a, b + n + 1).
2(2(b + n) − 1) 2 (b + n − 12 )(a + b + n)2

Iterating this formula starting from n = 0 proves the result.

Ramanujan’s formula (5) is an immediate consequence of Theorem 2:

Corollary 1
∞ 2n 3


X (6n + 1) n 4
= .
n=0
28n π

Proof Forsyth’s formula is equivalent with (7):

16
− 4 = s( 12 , 23 ).
π
We rewrite the right-hand side using Theorem 2:
∞
X 2 + 3(3 + 2n) − 4 ( 3 )3
s( 12 , 32 ) = · 2n 2 n 2
n=0
2(2(3 + 2n) − 2) 2 (1)n (2)n
∞
X 6n + 7 8( 21 )3n+1
= · 2n 3
n=0
8 2 (1)n+1
∞
3
X (6n + 1) 2n
n
=4 .
n=1
28n

This leads to:

∞ 2n 3 ∞ 2n 3



16 X (6n + 1) n 4 X (6n + 1) n
=4+4 ⇒ = .
π n=1
28n π n=0 28n

3 Other recurrences

Note that the following recurrence relation:

a2
s(a, b) = 1 + s(a + 1, b) (11)
(a + b)2
is related to the series (8) in the same way that (9) is related to (10). By
combining (9) and (11) we get new recurrences.
6

Theorem 3 (+1, +1–scheme)

(2a + 3b − 2)(a + b)2 + b3 a2 b3

s(a, b) = 2
+ s(a+1, b+1).
2(2b − 1)(a + b) 2(2b − 1)(a + b)2 (a + b + 1)2

The corresponding series is:

∞
X (5n + 2a + 3b − 2)(a + b + 2n)2 + (b + n)3 (a)2n (b)3n
s(a, b) = · .
n=0
2(2(b + n) − 1)(a + b + 2n)2 22n (b − 12 )n (a + b)22n

Proof We replace s(a, b + 1) in the right-hand side of (9) by the corresponding

formula from (11):

2a + 3b − 2 b3
s(a, b) = + · s(a, b + 1)
2(2b − 1) 2(2b − 1)(a + b)2
b3 a2
 
2a + 3b − 2
= + · 1+ s(a + 1, b + 1) .
2(2b − 1) 2(2b − 1)(a + b)2 (a + b + 1)2

The series follows immediately from this recurrence.

If we take a = 21 , b = 23 , we get the following series:

∞ 2n 3


4 X 168n3 − 36n2 + 6n + 1 n
= . (12)
π n=0 (2n − 1)2 212n

By combining the two recurrences (9) and (11) we can derive other series. For
instance, if we use (11) again in the recurrence of Theorem 3, we get this series
(+2, +1–scheme):
∞ 2n 4n 2



4 X 18560n5 − 20736n4 + 8160n3 − 992n2 + 30n + 9 n n
= .
π n=0 (4n − 1)2 (4n − 3)2 216n

And this is the series we obtain if we use (9) again in the recurrence of Theorem
3 (+1, +2–scheme):
∞ 4n 4n 2



4 X 186368n5 − 128000n4 + 22304n3 − 152n2 + 2n + 1 2n n
= .
π n=0 (2n − 1)(4n − 1)3 220n

4 Recurrences related to Glaisher’s and Wallis’s series

Glaisher’s series (2) can be written in the following form:

∞ 3 2


4 1X 2 n
=1+ .
π 4 n=0 (2)n (2)n+1
 7

The series on the right-hand side is the special case a = 32 , b = 1

2 of this series:
∞
X (a)2n
t(a, b) = .
n=0
(a + b)n (a + b)n+1

In a similar way as in the previous sections we can prove the two recurrences
equivalent with (9) and (11) for t:
1 a2
t(a, b) = + t(a + 1, b),
a + b (a + b)2
a + 2ab + b + 3b2 b(b + 1)2
t(a, b) = + t(a, b + 1).
2(2b + 1)b(a + b) 2(2b + 1)(a + b)2
3 1
Using only the last recurrence (+0, +1–scheme) with a = 2 and b = 2 we get
the following new series:
∞ 2n 3


4 X 12n2 + 4n + 1 n
= .
π n=0 (2n − 1)2 (n + 1) 28n

With the +1, +1-scheme we again obtain the series (12).

The series (3) can be rewritten like this:
∞ 3 3



2 1 X 2 n−1 2 n
=1− .
π 8 n=0 (2)2n

where the series on the right is a special case of

∞
X (a)n−1 (a)n
u(a, b) =
n=0
(a + b)2n

with corresponding recurrences:

1 a2
u(a, b) = + u(a + 1, b),
a − 1 (a + b)2
a + 2ab + 3b2 − 1 b(b + 1)2
u(a, b) = + u(a, b + 1).
2(2b + 1)b(a − 1) 2(2b + 1)(a + b)2
The +0, +1–scheme leads in this case to the series:
∞ 2n 3


2 X 12n2 − 1 n
=− .
π n=0
(2n − 1)2 28n

With the +1, +1 –scheme we find this series:

∞ 2n 3


2 X 168n3 + 20n2 − 2n − 1 n
=− . (13)
π n=0
(2n − 1)2 212n

With these series we can prove (6):

8

Corollary 2
∞ 2n 3


X (42n + 5) n 16
= .
n=0
212n π

Proof If we add (12) and (13), we get a new series:

∞ 2n 3


3 X −28n2 + 4n + 1 n
= .
π n=0 (2n − 1)2 212n

If we substract (13) from 6 times the previous series, the sum of the new series
is 16
π , and the polynomial in the numerator is given by:

6(−28n2 + 4n + 1) + 168n3 + 20n2 − 2n − 1 = (42n + 5)(2n − 1)2 .

Hence the resulting series is Ramanujan’s series (6).
Note that (3) can also be written in this form:
∞ 1

3

2 1X 2 n 2 n
=1− (14)
π 4 n=0 (2)2n

and hence is a special case (a = 21 , b = 32 ) of this series:

∞  
X (a)n (b)n a b 1
v(a, b) = = 3 F2 ;1 .
n=0
(a + b)2n a+b a+b

The corresponding recurrences are:

Theorem 4
a + 2b − 1 b3
v(a, b) = − 2
· v(a, b + 1),
a+b−1 (a + b) (a + b − 1)
2a + b − 1 a3
v(a, b) = − 2
· v(a + 1, b).
a+b−1 (a + b) (a + b − 1)
Proof We prove the first one. The second one follows by symmetry. Note that
the identity we want to prove can be rewritten like this:
(a + b)2
 
a+b−1
v(a, b + 1) = 1− (v(a, b) − 1)
b2 b
or
∞
(a + b)2
 
X (a)n (b + 1)n a+b−1
= 1− (v(a, b) − 1) .
n=0
(a + b + 1)2n b2 b
We use the definition of the pochhammer symbol to rewrite the left-hand side
and at the same time we add a factor:
∞ ∞
X (a)n (b + 1)n (a + b)2 X (a)n (b)n+1
= (a + b + n − (a + n)).
n=0
(a + b + 1)2n b2 n=0
(a + b)2n+1
 9

After simplifying and using the definition of v, what we have to prove becomes:
∞ ∞ ∞
X (a)n (b)n+1 X (a)n+1 (b)n+1 a + b − 1 X (a)n+1 (b)n+1
(a+b+n)− = 1− .
n=0
(a + b)2n+1 n=0
(a + b)2n+1 b n=0
(a + b)2n+1

Note that two sums cancel out, and we are left with:
∞ ∞
X (a)n (b)n+1 a − 1 X (a)n+1 (b)n+1
(a + b + n) = 1 − .
n=0
(a + b)2n+1 b n=0 (a + b)2n+1

We multiply by b and rewrite the left-hand side:

∞ ∞
X (a)n (b)n+1 X (a)n+1 (b)n+1
2 (a + b + n)(a + b + n − (a + n)) = b − (a − 1)
n=0
(a + b)n+1 n=0
(a + b)2n+1
∞ ∞
X (a)n (b)n+1 X (a)n+1 (b)n+1
⇔ − (a + b + n)
n=0
(a + b)2n n=0
(a + b)2n+1
∞
X (a)n+1 (b)n+1
= b − (a − 1)
n=0
(a + b)2n+1
∞ ∞ ∞
X (a)n (b)n+1 X (a)n+1 (b)n+1 X (a)n+1 (b)n+1
⇔ − 2 (b + n) = b + .
n=0
(a + b)n2
n=0
(a + b)n+1 n=0
(a + b)2n+1

If we change the index of summation in the first term on the left, the b at the
right cancels out:
∞ ∞ ∞
X (a)n+1 (b)n+1 (b + n + 1) X (a)n+1 (b)n+1 (b + n) X (a)n+1 (b)n+1
− = .
n=0
(a + b)2n+1 n=0
(a + b)2n+1 n=0
(a + b)2n+1

It is now easy to see that both sides are equal.

Using the first recurrence in the same way as in Theorem 2, we get the following
result:

Theorem 5
∞
X a + 2(b + n) − 1 (b)3n
v(a, b) = (−1)n . (15)
n=0
a + b + n − 1 (a + b)2n (a + b − 1)n

An immediate consequence is (4):

Corollary 3
∞ 2n 3


n (4n + 1) 2
X
n
(−1) = .
n=0
26n π
10

Proof We rewrite the right-hand side of (14) using Theorem 5:

∞
X 1 + 2(3 + 2n) − 2 ( 32 )3n
v( 21 , 32 ) = (−1)n ·
n=0
2n + 2 (1)n (2)2n
∞
X 4n + 5 8( 21 )3n+1
= (−1)n ·
n=0
2 (1)3n+1
∞ 2n 3


n−1 (4n + 1) n
X
=4 (−1) .
n=1
26n

This leads to:

∞ 2n 3 ∞ 2n 3



2 X (4n + 1) X
n (4n + 1)
=1− (−1)n−1 n
= (−1) n
.
π n=1
26n n=0
26n
Other consequences of Theorem 5 are:
Corollary 4
√ ∞ ∞
3 3 X ( 1 )3 3 X ( 1 )3
= (−1)n (6n + 1) 3 3n = (−1)n (12n + 1) 6 3n
2π n=0
n! π n=0 n!
√ ∞ 1 3
p √ ∞
2 2 X ( ) 4 2− 2 X ( 1 )3
= (−1)n (8n + 1) 4 3n = (−1)n (16n + 1) 8 3n
π n=0
n! π n=0
n!
p √ ∞ 1
√ ∞
5 5− 5 X ( )3 5( 5 − 1) X ( 1 )3
√ = (−1)n (10n + 1) 5 3n = (−1)n (20n + 1) 10 3n
2 2π n=0
n! 2π n=0
n!
Proof These series follow from the Wallis-type products for π which can be
found in [1] (see also [13], formulas (10), (11) and (19)):
∞
sin(πm/k) Y nk + k − m nk + k + m
=
πm/k n=0
nk + k nk + k

for m, k positive integers with m < k. The choice m = 1, k = 2 leads to Wallis’s

product. If we take m = 1, k = 3, the product takes this form:
√
3 3 2 · 4 5 · 7 8 · 10
= · · ··· .
2π 3·3 6·6 9·9
As we did in the introduction with Wallis’s product, we can rewrite this prod-
uct as a series:
√
3 3 1 2·4 2·4·5·7
= 1 − 2 − 2 2 − 2 2 2 − ···
2π 3 3 ·6 3 ·6 ·9
∞
1 X ( 23 )n ( 43 )n
=1− 2
3 n=0 (n + 1)!2
1 2 4
=1− v( , ).
32 3 3
 11

Using Theorem 5, we get the first formula. The second formula is found by
taking m = 1, k = 4, the third one by taking m = 1, k = 5, and so on.

Concluding remarks.
1. The method used above to convert a product to a series can be applied
directly to Euler’s product formula for the sine-function:
∞ 
x2
Y 
sin πx = πx 1− 2 .
m=1
m

The result is the following series:

∞
sin πx X (−x)n (x)n
= . (16)
πx n=0
n!2

which converges (by Raabe’s test) for all x 6= 0.

2. Applying Theorem 5 to (16) results in this series:
∞
sin πx X 2n + x (x)3n
= (−1)n .
πx n=0
x n!3

It can be found in Dougall’s paper [4, p. 124 formula (16)]. All series in
Corollary 4 are special cases of this general formula.

References

1. I. Ben-Ari, D. Hay, A. Roitershtein, On Wallis-type products and Pólya’s urn schemes,

Amer. Math. Monthly, 121, 5, 422–432 (2014).
2. N. D. Baruah, B. C. Berndt and H. H. Chan, Ramanujan’s series for 1/π: A Survey,
Amer. Math. Monthly, 116, 7, 567-587 (2009).
3. W. Chu, Dougall’s bilateral 2 H2 -series and Ramanujan-like π-formulae, Math. Comp.
80, 2223–2251 (2011).
4. J. Dougall, On Vandermonde’s theorem and some more general expansions, Proc. Ed-
inburgh Math. Soc., 25, 114–132 (1907).
5. A. R. Forsyth, A Series for π1 , Messenger of Mathematics, XII, 142–143 (1883).
6. J. W. L. Glaisher, On series for π1 and π12 , Quarterly Journal of Pure and Applied
Mathematics, XXXVII, 173–198 (1905-06).
7. J. Guillera, Series de Ramanujan: Generalizaciones y conjeturas. Ph.D. Thesis, Univer-
sity of Zaragoza, Spain (2007).
8. J. Guillera, Accelerating Dougall’s 5 F4 sum and the WZ-algoritm https://arxiv.org/
pdf/1611.04385.pdf (2016). Accessed 14 March 2017.
9. K. Knopp, Theory and Application of Infinite Series, Blackie, London, 2nd English ed.,
4th reprint (1954).
10. P. Levrie, Using Fourier-Legendre expansions to derive series for 1/π and 1/π 2 , Ra-
manujan J. 22, no. 2, 221–230 (2010).
11. Z.-G. Liu, A summation formula and Ramanujan type series, J. Math. Anal. Appl. 389
(2), 1059–1065 (2012).
12. Z.-G. Liu, Gauss summation and Ramanujan-type series for 1/π, Int. J. Number Theory
8 (2), 289–297 (2012).
13. A. S. Nimbran, Generalized Wallis-Euler Products and New Infinite Products for π,
Mathematics Student, Vol. 83, Nos. 1–4, 155–64 (2014).
12

14. A. S. Nimbran, Deriving Forsyth-Glaisher type series for π1 and Catalan’s constant by
an elementary method, Mathematics Student, Vol. 84, Nos. 1–2, 69–86 (2015).
15. Hessami Pilehrood Kh. and Hessami Pilehrood T., Generating function identities for
ζ(2n + 2), ζ(2n + 3) via the WZ-method, Electron. J. Combinatorics 15, #R35 (2008).
16. S. Ramanujan, Modular equations and approximations to π1 , Quarterly Journal of
Pure and Applied Mathematics, XLV, 350–372 (1914). Available at http://ramanujan.
sirinudi.org/Volumes/published/ram06.pdf