The Ramanujan Journal

https://doi.org/10.1007/s11139-020-00330-6

An efficient determination of the coefficients in the

Chudnovskys’ series for 1/

Lorenz Milla1

Received: 20 June 2020 / Accepted: 21 August 2020

© Springer Science+Business Media, LLC, part of Springer Nature 2021

Abstract
In 1914, Srinivasa Ramanujan published several hypergeometric series for 1/π . One
of these series was used by Bill Gosper in 1985 in a world-record-computation of
π . Shortly after this, the Chudnovskys found a faster series for 1/π based on the
largest Heegner number and the Borweins proved Ramanujan’s series. Lately, the
Chudnovskys’ series has often been used in practice to calculate digits of π , it reads:
√ ∞
6403203  (6n)! 13591409 + 545140134n
=  n .
12π
n=0
(3n)! (n!)3 −6403203

In this paper, we calculate the coefficients in two of Ramanujan’s series for 1/π and
those in the Chudnovskys’ series. For our calculation, we don’t require special software
packages, but only the Fourier expansions of the Eisenstein series with a precision of
≈ 20 decimals. We also prove the exactness of our calculations by proving that the
values of certain non-holomorphic modular functions are algebraic integers. Our proof
uses the division values of the Weierstraß ℘ function.

Keywords Eisenstein series · Non-holomorphic modular functions · Weierstraß ℘

function · Division values · Chudnovsky algorithm

Mathematics Subject Classification 11F03 · 30D60

B Lorenz Milla
[email protected]

1 Heidelberg, Germany

123
 L. Milla

1 Definitions

As usual we write q = e2πiτ for τ in the complex upper half-plane H. Let

∞
 n qn
E 2 (τ ) = 1 − 24 ,
1 − qn
n=1
∞
n3 q n
E 4 (τ ) = 1 + 240 ,
1 − qn
n=1
∞
 n5 q n
and E 6 (τ ) = 1 − 504
1 − qn
n=1

be the normalized Eisenstein series and denote by

∞

πiτ
η(τ ) = e 12 (1 − q n )
n=1
1728E 4 (τ )3 E 4 (τ )3
and j(τ ) = =
E 4 (τ )3 − E 6 (τ )2 η(τ )24

the Dedekind η function and Klein’s j-invariant. Further, we define

E 4 (τ ) ∗ 3
s2 (τ ) = E (τ ), with E 2∗ (τ ) = E 2 (τ ) − . (1)
E 6 (τ ) 2 π Im(τ )

E 2∗ transforms like a modular form of weight 2 and is a polynomial in 1/ Im(τ ) with

holomorphic coefficients, thus E 2∗ is called an “almost holomorphic modular form
of weight 2”. This implies that s2 transforms like a modular function, i.e., that it is
invariant under transformations from the modular group .

2 Introduction

Our paper starts with the following equation which holds for all τ ∈ H with | j(τ )| >
1728 and Im(τ ) > 1:

 1/2 ∞

1 j(τ ) (6n)! 6n + 1 − s2 (τ )
= . (2)
2π Im(τ ) j(τ ) − 1728 (3n)!(n!)3 6 j(τ )n
n=0

This equation is due to the Chudnovskys [4, Eq. (1.4)], details can be found in [1,3,6].
In order to derive formulae for calculating π from Eq. (2), we have to calculate the
exact values of j(τ ) and of s2 (τ ). Two examples were found by Ramanujan [7, Eq.

123
An efficient determination of the coefficients...

√ √
(33, 34)] in 1914, corresponding to τ = i 3 and τ = i 7:
√ ∞
2 × 303  (6n)! 1 + 11n
=  n , (3)
72π
n=0
(3n)! (n!)3 2 × 303
√ ∞
2553  (6n)! 8 + 133n
=  n . (4)
162π
n=0
(3n)! (n!)3 2553

In 1988, the Chudnovskys [4, Eq. (1.5)] gave the

√fastest of these formulae, based on
the largest Heegner number 163 and τ = (1 + i 163)/2:
√ ∞
6403203  (6n)! 13591409 + 545140134n
=  n . (5)
12π
n=0
(3n)! (n!)3 −6403203

As Berndt and Chan write in [1]: “The Chudnovskys and the Borweins have cal-
culated s2 (τ ) for several values of τ ; but we are uncertain if these calculations are
theoretically grounded.” Later on, they calculate s2 (τ ) using the minpoly-command
in MAPLE.
In this paper, we will prove a method how to determine the values of s2 (τ ) efficiently,
i.e., without special software packages, requiring only the Fourier expansions of the
Eisenstein series with a precision of ≈ 20 decimals.
It is known (see [8]) that the modular j-function has integral values for all quadratic
irrationalities τ where Q(τ ) has class number 1. In [5, Thm. A1], Masser proved
that for all quadratic irrationalities τ which are not equivalent to i under modular
transformations, s2 (τ ) ∈ Q( j(τ )). Thus for all τ which are not equivalent to i, for
which the class number of Q(τ ) is 1, we have

j(τ ) ∈ Z and s2 (τ ) ∈ Q. (6)

The problem is to find a suitable denominator of the rational s2 (τ ) = a/b (as then
we easily find s2 (τ ) by approximating bs2 (τ )), and the main result of the paper is as
follows:
Theorem 1 Suppose τ is a CM-point which is not equivalent to i under modular
transformations. Suppose its minimal quadratic equation is A + Bτ + Cτ 2 = 0, the
discriminant D √= B 2 − 4 AC < 0 and the class number of Q(τ ) is 1. Take c ∈ Z such
that b = (AC) cD( j(τ ) − 1728) ∈ Z. Then b is a denominator of s2 (τ ).
2

The numerical results can be found in Table 1. In fact, this principal theorem is
valid for general quadratic irrationality τ , without restrictions on the class number of
Q(τ ): the number b so constructed makes bs2 (τ ) an algebraic integer.

3 Division values and complex multiplication

In this section, we prove Lemmas 1 and 2 using the division values of the Weierstraß ℘
function. These lemmas along with (6) are the main tools for proving Theorem 1.

123
 Table 1 Determination of the values of s2 (τ )

123
CM-point Quadratic form Discriminant Singular moduli Intermediate values c, b, and a from Thm. 1 Values of
τ Cτ 2 + Bτ + A = 0 D j(τ ) c b a s2 (τ ) = a/b

i τ2 + 1 = 0 −4 123 −1 0 – –
√
i 2 τ2 + 2 = 0 −8 203 −1 896 320 5/14
√
i 3 τ2 + 3 = 0 − 12 2 × 303 −1 7128 3240 5/11
√
i 4 τ2 + 4 = 0 − 16 663 −2 48384 25344 11/21
√
i 7 τ2 + 7 = 0 − 28 2553 −1 1055754 674730) 85/133
√
1+i 3
τ2 − τ + 1 = 0 −3 0 1 72 0 0
2√
1+i 7
τ2 − τ + 2 = 0 −7 −153 1 756 180 5/21
2√
1 + i 11
τ2 − τ + 3 = 0 − 11 −323 1 5544 2304 32/77
2√
1 + i 19
τ2 − τ + 5 = 0 − 19 −963 1 102600 57600 32/57
2√
1 + i 27
τ2 − τ + 7 = 0 − 27 −3 × 1603 1 892584 564480 160/253
2√
1 + i 43
τ 2 − τ + 11 = 0 − 43 −9603 1 23600808 16727040 640/903
2√
1 + i 67
τ 2 − τ + 17 = 0 − 67 −52803 1 907582536 695819520 33440/43617
2√
1 + i 163
τ 2 − τ + 41 = 0 − 163 −6403203 1 10996566783048 9351571368960 77265280/90856689
2
L. Milla
An efficient determination of the coefficients...

Definition 1 Given a lattice L ⊂ C, the Weierstraß ℘ function is defined by

  
1 1 1
℘ (z; L) = + − 2 .
z 2 (z − ω)2 ω
ω∈L
ω=0

Given any complex number c = 0, the “c-division points” are those u ∈ C with
cu ∈ L and u ∈
/ L. The “c-division values” are the values ℘ (u; L) at the c-division
points.

Definition 2 Those τ for which the elliptic curve associated with L τ = Z + Zτ has
complex multiplication are called “CM-points”. They are those τ for which there is
a ∈ C − Z such that a · L τ ⊆ L τ . This yields the set of CM-points:


(A, B, C) ∈ Z3 ;

CM = τ ∈ H A + Bτ + Cτ 2 = 0; .
gcd(A, B, C) = 1

For each τ ∈ C M, we call D = B 2 − 4 AC the “discriminant of τ ”.

Lemma 1 For all CM-points τ , the “singular moduli” j(τ ) and the values of
E 4 (τ )/η8 (τ ) and E 6 (τ )/η12 (τ ) are algebraic integers.

Proof From the definition of j(τ ) we deduce

 3  2
E 4 (τ ) E 6 (τ )
j(τ ) = and j(τ ) − 1728 = . (7)
η(τ )8 η(τ )12

Since j(τ ) is an algebraic integer for all CM-points τ (see [8, Thm. II.6.1]), it follows
that E 4 (τ )/η8 (τ ) and E 6 (τ )/η12 (τ ) are also algebraic integers. 

Lemma 2 If a CM-point τ has minimal quadratic equation

√ A + Bτ + Cτ 2 = 0 and
2 2 ∗
discriminant D = B − 4 AC < 0; then (AC) D E 2 (τ )/η4 (τ ) is an algebraic
integer.

Proof Suppose we are given a lattice L = Zω1 + Zω2 with complex multiplication
and period ratio τ = ω2 /ω1 . Since E 2∗ (i) = 0 and E 2∗ (eπi/3 ) = 0 (see [5, Lem. 3.2]),
it remains to prove the Lemma for τ which are not equivalent to i or to eπi/3 under
modular transformations. In [5, Lem. A3], Masser proves

−2g2 S
s2 (τ ) = ,
3g3 C(2 A + Bτ )

-division values of ℘ (z; L) and g2 and g3 are the invariants

where S is the sum of the Cτ√
of L. Using 2 A + Bτ = − Dτ yields
√ 3g3 S
D s2 (τ ) = .
2g2 Cτ

123
 L. Milla

Using g2 = 43 π 4 ω1−4 E 4 (τ ) and g3 = 8 6 −6

27 π ω1 E 6 (τ ) and the definition of s2 yields

π2 √ S
2
D E 2∗ (τ ) = . (8)
3ω1 Cτ

Cassels proved in [2, Eq. (11) and §2.(ii)] that for any
 AC-division
 point v of a lattice
L, the term AC℘ (v; L) is an algebraic integer of Z 41 g2 ; 41 g3 . Since all Cτ -division
 
points in S are also AC-division points, AC S is an algebraic integer of Z 41 g2 ; 41 g3 .
But if we rescale the given lattice to ω1 = √π η(τ )2 ,
3

1 E 4 (τ ) 1 E 6 (τ )
g2 = 3 8 and g3 = 2 12
4 η (τ ) 4 η (τ )

are algebraic integers (see Lemma 1), thus AC S is an algebraic integer.

Multiplying (8) with (AC)2 , using τ τ̄ = A/C and ω1 = √π η(τ )2 we obtain
3

√ E ∗ (τ )
(AC)2 D 2 4 = ACS · C τ̄ .
η(τ )

As both ACS and C τ̄ are algebraic integers, we have proven Lemma 2. 

An elaboration of Cassels’ and Masser’s proof can be found in [6, App. A & B].

4 Proof of Theorem 1

Let τ be any CM-point. Using the definition (1) of s2 (τ ) and (7) yields

a := bs2 (τ ) = (AC)2 cD( j(τ ) − 1728)s2 (τ )
  
√ E 6 (τ ) E 4 (τ ) ∗
= (AC)2 cD ± 12 E 2 (τ )
η (τ ) E 6 (τ )
√ ∗
√ E 4 (τ ) (AC) D E 2 (τ )
2
=± c 8 .
η (τ ) η4 (τ )
     
see Lemma 1 see Lemma 2

Then Lemmas 1 and 2 show that a is an algebraic integer.

From now on we assume (as in Theorem 1) that Q(τ ) has class number 1 and that
τ is not equivalent to i under modular transformations (the latter has to be assumed
since s2 (i) is undefined).
Then (6) implies that j(τ ) ∈ Z and we can choose c ∈ Z such that

b := (AC)2 cD( j(τ ) − 1728) ∈ Z.

We know from Eq. (6) that s2 (τ ) ∈ Q and thus a := bs2 (τ ) ∈ Q for these τ . Since a
is an algebraic integer, we have proven a ∈ Z for these τ . 

123
An efficient determination of the coefficients...

5 Conclusion

Since j(τ ) ∈ Z and we have proven bs2 (τ ) ∈ Z, we can now determine the coefficients
using the Fourier series of the Eisenstein series with limited precision – until the error
is less than ±0.5. For the numerical results see Table 1.
√ at the values of s2 (τ ) we observe that the chosen denominator b =
Looking
(AC)2 cD( j(τ ) − 1728) was larger than needed by at least a factor of (AC)2 . This
is not surprising, since s2 (τ ), j(τ ) and D are invariant under modular
√ transformations,
but (AC)2 isn’t. Using this invariance, one could prove that 2 cD( j(τ ) − 1728) is
also a suitable denominator.
We conclude our paper with√ the coefficients in the Chudnovskys’ series as computed
in Table 1 with τ = (1 + i 163)/2:

1 − s2 (τ ) 13591409
j(τ ) = −6403203 and = .
6 545140134
Acknowledgements For their invaluable help I am grateful to Henri Cohen, Michael Griffin, Jesús Guillera
and the anonymous referee.

References
1. Berndt, B., Chan, H.: Eisenstein series and approximations to π . Illinois J. Math. 45, 75–90 (2001)
2. Cassels, J.: A note on the division values of ℘ (u). Math. Proc. Camb. Philos. Soc. 45, 167–172 (1949)
3. Chen, I., Glebov, G.: On Chudnovsky–Ramanujan type formulae. Ramanujan J. 46, 677–712 (2018)
4. Chudnovsky, D., Chudnovsky, G.: Approximation and complex multiplication according to Ramanujan.
In: Andrews, G., Askey, R., Berndt, B., Ramanathan, K., Rankin, R. (eds.) Ramanujan Revisited, pp.
375–472. Academic Press, Boston (1988)
5. Masser, D.: Elliptic Functions and Transcendence. Springer, Berlin (1975)
6. Milla, L.: A detailed proof of the Chudnovsky formula with means of basic complex analysis (2019).
arXiv:1809.00533
7. Ramanujan, S.: Modular equations and approximations to π . Q. J. Math. 45, 350–372 (1914)
8. Silverman, J.: Advanced Topics in the Arithmetic of Elliptic Curves. Springer, New York (1994)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

123