Relations between e and π: Nilakantha’s series

and Stirling’s formula

arXiv:2206.07174v3 [math.HO] 5 Sep 2022

V. Yu. Irkhin
September 7, 2022

Abstract
Approximate relations between e and π are reviewed, some new
connections being established. Nilakantha’s series expansion for π
is transformed to accelerate its convergence. Its comparison with the
standard inverse-factorial expansion for e is performed to demonstrate
similarity in several first terms. This comparison clarifies the origin of
the approximate coincidence e+2π ≈ 9. Using Stirling’s series enables
us to illustrate the relations π 4 + π 5 ≈ e6 and π 9 /e8 ≈ 10. The role
of Archimede’s approximation π = 22/7 is discussed.

The fundamental mathematical constants e and π are both transcenden-

tal, and their simple combinations are probably transcendental too, except
for famous Euler’s identity eiπ = −1.
However, there exist very close numerical coincidences in some combina-
tions of e and π, which seem to be accidental. Several examples can be found
in [1, 2]:
π 2 = 0.9996(4e − 1) ≈ 4e − 1 (1)
163(π − e) = 68.99966... (2)
(π 4 + π 5 )/e6 = 0.999999956... , (3)
π 9 /e8 = 9.9998... ≈ 10, (4)
eπ − π = 19.999..., (5)
π 2 (π − e)3/2 /e = 0.9999869..., (6)
For some of such approximate relations, complicated mathematical justi-
fications were found. For example, “Ramanujan’s constant”
√
eπ 163
= 262 537 412 640 768 743.999 999 999 999 25 . . . ≈ 640 3203 + 744 (7)
is very close to integer value owing to that 163 is a Heegner number [3].

1
1 Nilakantha’s series
First we focus on the simplest relation

e + 2π = 9.001..., (8)

or 2(π −3) ≈ 3 −e. We demonstrate a reason of this coincidence by using the

series expansions. For e we use the standard inverse-factorial series which
may be represented as
∞
X 1 1 1 1 1
e= =3− + + + + ... (9)
n=0
n! 3 24 120 720

and is rapidly convergent.

As for π, the situation is more complicated. The simple Gregory-Leibniz
expansion
π 1 1 1
= 1 − + − + ..., (10)
4 3 5 7
converges very slowly. We apply Nilakantha’s series which can be obtained
from (3) after acceleration of the convergence:
∞  
X
n+1 1 1 4
π = 3+ (−1) + −
n=1
n + 1 n 2n + 1
∞
X 1
= 3+ (−1)n+1
n=1
n(2n + 1)(n + 1)
∞
X 4
= 3+ (−1)n+1 . (11)
n=1
(2n + 1)3 − (2n + 1)

The series (11) was discovered in India already in the 15th century and
converges rather rapidly [1].
After grouping of pairs of opposite-sign terms, (11) can be transformed
as
∞
1 X 3
2π = 6 + −
3 n=1 n(n + 1)(4n + 1)(4n + 3)
1 3 1 1
= 6+ − − − − ... (12)
3 70 198 780
so that the convergence becomes still faster.
One can see that the expansions (9) and (12) demonstrate a close simi-
larity of several first terms, which justifies and clarifies the coincidence (8).

2
 The structure of Nilakantha’s series also reproduces roughly the relation
(1). We have from (12)
 2
1 2 1 1 1 1 1
(π + 1) ≈ 3+ − − +
4 4 6 48 200 4
1 1 1 1 1
≈3− + − − 0.01 ≈ 3 − + (13)
3 12 32 3 24
which corresponds to the expansion of e (9).
The relation (8) enables us to present combinations of e and π in terms
of one constant and therefore can be useful to clarify other above relations
between e and π. So, the relation (1) is transformed as

π 2 + 8π = 35 (14)

which gives √
π= 51 − 4 ≈ 3.1414... (15)
Expanding, we obtain
p
π = 7 1 + 2/49 − 4 ≈ 22/7 ≈ 3.143, (16)

i.e., Archimede’s number.

On substituting (8) into (2) we have

512 29
π= = = 3.1411..., (17)
163 163
which corresponds to Stoschek’s approximation using powers of two and the
number 163 that is the largest Heegner number [2].

2 Stirling’s formula
Another way to relate e and π is using Stirling’s series [5] which gives a good
accuracy even for n = 1:
√  n n  1 1

n! ≃ 2πn 1+ + +··· (18)
e 12n 288n2

so that for any n

nn
 
n
√ 1 1
e ≃ 2πn 1+ + +··· (19)
n! 12n 288n2

3
 Eq.(19) enables one to generate various approximations for e and π. For
example, we obtain
n+1/2 1
en+1 1 + 12(n+1) +···

1
e= n = 1+ 1 . (20)
e n 1 + 12n +···

For
√ half-integer n in (19), using the value of Euler’s Gamma function Γ(1/2) =
π we derive
√ (2n + 1)n+1
 
n+1/2 1
e = 2 1+ +··· (21)
(2n + 1)!! 6(2n + 1)
which yields a good approximation even for n = 0, e = 49/18 ≈ 2.72.
To demonstrate the usefulness of the relation (19) we substitute e from
(19) with n = 1 into (3). Expanding up to 1/n2 we obtain

π 2 + π = 8(1 + 1/2 + 1/8 + · · · ) ≈ 13. (22)

Then we derive again Archimede’s number:

p
π = −1/2 + (7/2) 1 + 4/49 ≈ 3 + 1/7 (23)

Otherwise, we can substitute (1) into (22) to obtain

4e + π = 14. (24)

Solving the system of linear equations (8), (24) we derive

π = 3 + 1/7, e = 3 − 2/7. (25)

Combining (8) and (24), we can generate approximate equations with de-
creasing accuracy:

3π − 2e = 4, 3e − π = 5, 5e − 4π = 1...

and also for any combinations nπ + me with integer n and m satisfying

n − 2m = 7k.
Another way to derive (22) from (3) is to take into account e3 = 20.08 ≈
20 and the well-known relation π 3 ≈ 31 which follows from the series
∞
6
X 1
π = 960 ≈ 960. (26)
n=0
(2n + 1)6

Then we have
π 2 + π ≈ 400/31 ≈ 13.

4
 Although the accuracy of (15), (23) is somewhat worse than that of orig-
inal relations (1), (3), they demonstrate close connection with ancient inves-
tigations of the problem of calculating π.
Now we are also able to clarify the relation (4). Using (19) for both n = 1
and n = 2 we derive
 4  2
8 3 1 1
e ≃ 64π 1 + +··· 1+ +···
12 24
3
≈ 64π 3 = 96π 3 . (27)
2
Using (26) wee see the coincidence of (27) and (4).
The coincidence (5) can be related to the continued-fraction representa-
tion of Gelfond’s constant, which also starts from Archimede’s approxima-
tion [6]:
eπ = [23; 7, 9, 1, 1, 591...] ≈ 20 + 3 + 1/7 + ... (28)
In more detail, we can use e3 ≈ 20 to write down

eπ = e3 eπ−3 ≈ 20(1 + 1/7 + 0.01) ≈ 20(1 + π/20) = 20 + π. (29)

The relation (6) is most difficult. To obtain an estimation, we take into

account (8) and rewrite (6) as

27π 8 (π − 3)3 /(π 2 e)2 ≈ 1. (30)

Further on, we use Archimede’s π − 3 ≈ 1/7, 73 ≈ 350, π 2 e ≈ 27 (which

follows from the geometric mean corresponding to (8)), and

π 8 = 9450 ζ(8) ≈ 9450 = 350 · 27

with ζ(x) the Riemann zeta function. Then Eq.(30) is satisfied.

Although the subject of the present work can seem somewhat artificial,
we note that combinations of e and π occur in physical problems, e.g., in the
exact solution of the Kondo model [7].

References
[1] E. W. Weisstein, Almost Integer, MathWorld–A Wolfram Web Resource,
https://mathworld.wolfram.com/AlmostInteger.html

[2] E. W. Weisstein, Pi Approximations, MathWorld–A Wolfram Web Re-

source, https://mathworld.wolfram.com/PiApproximations.html

5
[3] J. D. Barrow, The Constants of Nature. London: Jonathan Cape, 2002.

[4] R. Roy, The Discovery of the Series Formula for π by Leibniz, Gregory
and Nilakantha. Mathematics Magazine, Vol. 63, No. 5, pp. 291-306
(1990).

[5] Olver, F. W. J.; Olde Daalhuis, A. B.; Lozier, D. W.; Schneider, B.

I.; Boisvert, R. F.; Clark, C. W.; Miller, B. R.; Saunders, B. V., 5.11
Gamma function properties: Asymptotic Expansions, NIST Digital Li-
brary of Mathematical Functions, Release 1.0.13 of 2016-09-16.

[6] The On-Line Encyclopedia of Integer Sequences, A058287.

[7] A.M. Tsvelick, P.B. Wiegmann, Exact results in the theory of magnetic
alloys, Adv. Phys. 32, 453 (1983).