Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
c ISSN 1607-2510
Available free at mirror sites of http://www.math.nthu.edu.tw/∼amen/
Abstract
We present new sums involving binomial coefficients for π and various loga-
rithms and polylogarithms constants. These sums are a generalization of BBP
formulas first introduced by D. Bailey, P. Borwein and S. Plouffe in 1995. In this
paper, we describe how to find and prove such sums using the Beta function at
integer and rational arguments.
1 Introduction
In 1995, David Bailey, Peter Borwein and Simon Plouffe discovered a new seminal
formula for π
X∞
1 4 2 1 1
π= − − − . (1)
n=0
16n 8n + 1 8n + 4 8n + 5 8n + 6
This amazing formula allows extracting the n-th binary digit of π without knowledge
of previous digits [1]. Such sums are called BBP (Bailey-Borwein-Plouffe) formulas.
Many new BBP formulas for π, G (Catalan’s constant) or ζ(3) (Riemann Zeta function)
were subsequently discovered by Adamchik [2], Bellard [3], Broadhurst [4], Lupas [5]
and Huvent [6]. The main tool for finding BBP formulas is an algorithm designed
to detect linear combinations on the set of natural numbers N like PSLQ (from its
use of a partial-sum-of-squares (PS) vector and lower-diagonal-orthogonal (LQ) matrix
factorization) or LLL (Lenstra–Lenstra–Lovász) [7], [8]. To then prove such P formulas,
∞ zn
an integral is evaluated giving rise to logarithms or polylogarithms (Lip (z) = n=1 n p,
[9]).
In October 2001, Almkvist and colleagues presented new sums for π involving bi-
nomial coefficients such as
X∞
50n − 6
π=
3n n . (2)
n=0 n
2
∗ Mathematics Subject Classifications: 40C10
† 17rue Montera, 75012 Paris, email: boris@pi314.net
‡ Av. Cesareo Alierta, 31 esc. izda 4oVA, Zaragoza
237
238 Binomial Sums for π and Polylogarithmic Constants
They proved this formula using the so-called Beta method which is based on use of
R1
Beta function B(r, s) = 0 xr−1 (1 − x)s−1dx for positive integer values r and s [10].
In this paper, using an adapted Beta method, we show how to prove some more
general sums of the form
∞
X
1 b1 b2 bk bm−1
T = mn
+ +···+ + ···+ (3)
n=0
qn pn
mn + 1 mn + 2 mn + k mn + (m − 1)
where T is a constant often related to π and the natural logarithm ln(z). These sums
involve binomial coefficients and they are similar to BBP formulas. We therefore call
them “BBP binomial formulas”. One example of such formulas is
∞
1 X 1 59296 10326 3200 1352 792 552
π= − − − − +
16807 n=0 2n 7n
2n
7n + 1 7n + 2 7n + 3 7n + 4 7n + 5 7n + 6
(4)
that we discovered with LLL algorithm. An explicit proof of this formula is detailed in
this paper. Then we show how to generate an infinite set of similar formulas for π and
the natural logarithm ln(z). We also present outlines for proofs of formulas showing
that Beta method is not restrained to integer parameters.
Γ(r)Γ(s)
B(r, s) = (6)
Γ(r + s)
R∞
where Γ(z) = 0 tz−1e−t dt is the Gamma function. With parameters m, p, n ∈ N, we
use Γ(n) = (n − 1)! to obtain
Z 1
1
mn = (mn + 1) xpn (1 − x)(m−p)n dx. (7)
pn 0
B. Gourévitch and J. Guillera Goyanes 239
1
Summing Eq. (7) on n with (mn+1)( mn
on the left side, we obtain
pn )
X
∞ X∞ Z 1
1 1 1
mn
= xpn(1 − x)(m−p)n dx
n=0
qn pn
mn + 1 n=0 qn 0
Z 1X ∞ p n
x (1 − x)(m−p)
= dx
0 n=0 q
Z 1
q
= p (1 − x)(m−p)
dx. (8)
0 q − x
We can evaluate this integral by partial fraction decomposition. More generally, to
1
obtain the elements in Eq. (3) we write
mn + k
Z 1
1 1 (pn + 1)(pn + 2)...(pn + k − 1)
xpn+k−1(1−x)(m−p)n dx = mn+k = mn . (9)
0 pn+k−1 pn
(mn + 1)(mn + 2)...(mn + k)
(pn+1)(pn+2)...(pn+k−1)
For a given k, the partial fraction decomposition of (mn+1)(mn+2)...(mn+k) in n gives
(ak,j )j=1,..,k such that
Z 1
1 ak,1 ak,2 ak,k
xk−1xpn (1 − x)(m−p)n dx = mn
+ + ... + (10)
0 pn
mn + 1 mn + 2 mn + k
with ak,j ∈ Q. For values k ranging from 1 to m − 1, we linearly combine (ak,j )j=1,..,k
in order to find the m − 1 values bk of Eq. (3). This is equivalent to solving the system
a1,1 a2,1 ... am−1,1 b1
0 a2,2 ... am−1,2
V = b2 . (11)
.... 0 ... ... ....
0 0 ... am−1,k bm−1
1
X
Vector V contains the coefficients of the polynomial P (X) = V t .
.... of degree
X m−2
m − 2, such that
X∞
1 b1 b2 bm−1
+ + · · · +
n=0
qn mnpn
mn + 1 mn + 2 mn + (m − 1)
X 1
∞ Z
xpn(1 − x)(m−p)n
= P (x) dx
n=0 0
qn
Z 1
qP (x)
= p (1 − x)(m−p)
dx
0 q − x
when we use Eq. (10).
240 Binomial Sums for π and Polylogarithmic Constants
1
R1
REMARK 1. Almkvist et al. [10] use the relation = (3n + 1) 0 x2n(1 − x)n dx
(3n
n)
in order to prove the formula in Eq. (2). In general, they consider formulas of the form
X∞
S(n)
π=
mn n (12)
n=0 pn
q
where S ∈ Qm−2 [X] and q ∈ N. However, such forms of sums lead to the intricate
P∞ xp (1−x)m−p
n
evaluation of f(x) = n=0 (mn + 1)S(n) q . f(x) is indeed an integral
whose denominator is a power of the one considered in Eq. (8). Our method uses
1
(mn+1)(mn
on the left side of Eq. (7), which produces formulas of the general form of
pn )
Eq. 3.
3 Proof of Eq. 4
We write Eq. (4) as
∞
X
Un = π
n=0
Using Proposition 1, we are looking for the polynomial P (x) such that
X∞ Z 1 X∞ Z 1
x2n(1 − x)5n 1
Un = P (x) dx = P (x) dx
0 2n
0 x (1 − x)5
2
n=0 n=0 1−
2
Z 1
2
= P (x) 2 dx.
0 (x − 2x + 2) (x5 − 3x4 + 2x3 + x + 1)
R1 4
Because 0 2 dx = π, it is likely that P (x) = 2 x5 − 3x4 + 2x3 + x + 1 .
(x − 2x + 2)
Let us verify this conjecture.
To obtain the equivalent linear system in (11), and therefore vector V , we calculate
all coefficients (ak,i)i (columns of left matrix in Eq. (11)) with help from the compu-
tation of the integral in Eq. (10). For example, for k = 2 (2nd column of matrix), we
have
Z 1
1 (2n + 1)
x.x2n(1 − x)5n dx = B(5n + 1, 2n + 2) = 7n
0 2n
(7n + 1)(7n + 2)
1 5 1 3 1
= 7n − .
2n
7 (7n + 1) 7 (7n + 2)
Consequently, a2,1 = 57 and a2,2 = − 37 . The whole linear system is
59296
1 57 30
49
190
343
1235
2401
8151
16 807 16807
0 − 3
− 49 − 343 − 2401 − 16 807
30 255 2040 15 810 − 10326
7 16807
0 0 4 60 660 6380 − 3200
49 343 2401 16 807 V = 16807 .
0 0 0 13 260 3510 − 1352
343 2401 16 807 16807
0 0 0 0 99
− 2401 2475
− 16 − 792
807 16807
276 552
0 0 0 0 0 16 807 16807
B. Gourévitch and J. Guillera Goyanes 241
Its solution is V = 1 1 0 2 −3 1 . Finally, we have
1
X
P (X) = V t 3 4
.... = 2 1 + X + 2X − 3X + X
5
X5
p < m:
P∞
(−1)m b1 b2 bm−1
For alternating sums n=0 q n (mn ) mn+1 + mn+2 + ... + mn+(m−1) with equiva-
R1 pn
qP (x)
lent integral representation 0 q+xp (1−x)(m−p) dx, we get the following table:
Z 1
pn+ a mn+ dc Γ pn + ab + 1 Γ mn + dc + 1
x b (1 − x) dx =
0 Γ (p + m)n + ab + dc + 2
a
c
a
c
b pn+1 d mn+1 Γ b Γ d
= a c
a c
(13)
b + d (p+m)n+2 Γ b + d
Γ (x + n)
where (x)n = = x(x+1) · · ·(x+n−1) is Pochhammer symbol. Γ ab is often
Γ (x)
known for small integers values of a et b. All following formulas stem from summing
this relation on n for particular values of ab , dc . We now emphasize three particular
cases of interest.
B. Gourévitch and J. Guillera Goyanes 243
a
5.1 Case 1 : b
= − 12 , c
d
=0
We have
Z Z ∞
1
1 q 1 X xpn (1 − x)mn
1
√ dx = √ dx
0 x q − x (1 − x)m
p
0 x n=0 qn
∞ Z 1
X 1
= q−n xpn− 2 (1 − x)mn dx
n=0 0
X∞
pn + 12 Γ (mn + 1)
−n Γ
= q
n=0
Γ (p + m)n + 12 + 1
X∞
(2pn)!(mn + pn)!(mn)!
= (−1)n . (14)
n=0
qn (2mn + 2pn)!(pn)!
EXAMPLE 1. Case p = 3, m = 1. We have
Z 1 Z 1
1 1 3 2
2
−π = −2 √ dx = √ x − 2x + 2x − 2 dx
0 x(x + 1) 0 x 2 − x3(1 − x)
Z 1 ∞
1 X (−1)n x3n(1 − x)n
= √ x3 − 2x2 + 2x − 2 dx
0 x n=0
2n
X∞ Z
(−1)n 1 3 x3n
= n
x − 2x2 + 2x − 2 √ (1 − x)ndx (15)
n=0
2 0 x
a c
5.2 Case 2 : b
+ d
∈ Z (positive and negative integers)
This case allows us to obtain special values of Γ ab Γ dc (equal to π multiplied by
an algebraic number). In some cases, π is absent and we obtain amazing formulas for
algebraic numbers.
244 Binomial Sums for π and Polylogarithmic Constants
a
EXAMPLE 2. The following relations are based on Beta values for b = − 43 , dc = − 23
or = − 13 , dc = − 23 or ab = − 53 , dc = − 13
a
b
3n
√ Z 1 √ Z 1
3 4 2 3 1 2
n
= xn− 3 (1 − x)n− 3 dx − xn− 3 (1 − x)n− 3 dx
27n(3n − 1) 4π 0 2π 0
3n
√ Z 1 √ Z 1
3 5 1 3 1 2
n
n
= xn− 3 (1 − x)n− 3 dx − xn− 3 (1 − x)n− 3 dx
27 (3n − 2) 8π 0 4π 0
which gives the formula
∞
3n
8 X (−1)n n −1 1 √
3
n
+ = 2.
3 n=0 54 3n − 1 3n − 2
a c
5.3 Case 3 : b
+ d
∈ Q\N (rationals but not naturals)
Γ ab Γ dc
We obtain Pochhammer symbols in sums then factors in the outcome.
Γ ab + dc
EXAMPLE 3. Considering
Z 1
n− 14 n− 12 Γ n + 34 Γ n + 12
x (1 − x) dx =
0 Γ 2n + 54
Z 1
n+ 34 n− 12 Γ n + 74 Γ n + 12
x (1 − x) dx =
0 Γ 2n + 94
and the following integral
Z 1 X∞ Z 1
(−1)n n− 1 1 1
(x − 2) x 4 (1 − x)n− 2 dx = −2 √ √ dx = −π
2n
0 n=0 0
4
x 1 − x(x + 1)
we obtain finally
2
1
X∞
(−1)n
1
2 n
1
2 2n 11 1 √ Γ 4 √ 1
1
1
+ = 2 √ = 4 2K √
n=0
8n 4 n 4 2n
8n + 1 8n + 5 π 2
R π/2 dθ
where K(k) = 0
√ is the complete elliptic integral of the first kind, thus
1−k 2 sin 2 θ
making the connection with singular values of the elliptic integral.
are of order 3. A BBP binomial sum involves a constant of order p, as for BBP sums,
if the polynomial in n in the denominator is of degree p. For example, we have
X
∞
(−1)n 3
2n n 2 = −2 ln3 (2) + 3ζ(3). (17)
(2n + 1)
n=0 n 2 (n + 1)
X
∞ Z 1Z 1
1 q
2 = dxdy.
n mn q − ym xm (1 − x)(m−p)
n=0 q pn (mn + 1) 0 0
The evaluation is not often easy, there is no known proof of Eq. (17) to our knowl-
edge. Nevertheless, there is evidence for existence of BBP binomial formulas for higher
order constants.
Acknowledgements. The authors would like to thank Typhenn Brichieri-Colombi,
Professor Sui Sun Cheng and the anonymous reviewer for their helpful comments and
corrections.
References
[1] D. Bailey, P. Borwein and S. Plouffe, On The Rapid Computation of Various
Polylogarithmic Constants, Math. Comp., 66(1997), 903–913.
[5] A. Lupas, Some BBP functions, preprint, Sibiu University, Romania (2000).
[6] G. Huvent, Formules BBP, Seminar ANO, IRMA, Lille University (2001),
http://ano.univ-lille1.fr/seminaries/expo huvent01.pdf