A Compendium of BBP-Type Formulas for Mathematical Constants

David H. Bailey1
1 Mar 2004

Abstract
A 1996 paper by the author, Peter Borwein and Simon Plouffe showed that any mathematical constant given by an infinite series of a certain type has the property that its
n-th digit in a particular number base could be calculated directly, without needing to
compute any of the first n 1 digits, by means of a simple algorithm that does not require
multiple-precision arithmetic. Several such formulas were presented in that paper, including formulas for the constants and log 2. Since then, numerous other formulas of this
type have been found. This paper presents a compendium of currently known results of
this sort, both formal and experimental. Many of these results were found in the process
of compiling this collection and have not previously appeared in the literature. Several
conjectures suggested by these results are mentioned.

Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA, [email protected]. Baileys

work is supported by the Director, Office of Computational and Technology Research, Division of Mathematical, Information, and Computational Sciences of the U.S. Department of Energy, under contract
number DE-AC03-76SF00098.

1. Introduction
This is a collection of formulas for various mathematical constants that are of the
form similar to that first noted in the BBP paper [4]. That article presented the
following formula for (which was discovered using Fergusons PSLQ integer relation
finding algorithm [12, 5]):
=

X
1
k=0

16k

4
2
1
1

8k + 1 8k + 4 8k + 5 8k + 6

(1)

It was shown in [4] that this formula permits one to calculate the n-th hexadecimal or
binary digit of , without computing any of the first n 1 digits, by means of a simple
algorithm that does not require multiple-precision arithmetic. Further, as shown in [4],
several other well-known constants also have this individual digit-computation property.
One of these is log 2, based on the following centuries-old formula:
log 2 =

X
1
k=1

(2)

k2k

In general, any constant C that can be written in the form

C =

X
p(k)
k=0

q(k)bk

where p and q are integer polynomials, deg(p) < deg(q), and p(k)/q(k) is nonsingular for
nonnegative k, possesses this individual digit-computation property. Note that formula 1
can be written in this form, since the four fractions can be combined into one, yielding
=

X
1

47 + 151k + 120k 2
k
2
3
4
k=0 16 15 + 194k + 712k + 1024k + 512k

Since the publication of [4], other papers have presented formulas of this type for
various constants, including several constants that arise in quantum field theory [8, 9, 6].
More recently, interest in BBP-type formulas has been heightened by the observation that
the question of the statistical randomness of the digit expansions of these constants can be
reduced to the following hypothesis regarding the behavior of a particular class of chaotic
iterations [6]:
Hypothesis A (from the paper [6]). Denote by rn = p(n)/q(n) a rational-polynomial
function, i.e. p, q Z[X]. Assume further that 0 deg p < deg q, with rn nonsingular
for positive integers n. Choose an integer b 2 and initialize x0 = 0. Then the sequence
x = (x0 , x1 , x2 , . . .) determined by the iteration:
xn = (bxn1 + rn ) mod 1.
either has a finite attractor or is equidistributed in [0, 1).
Assuming this hypothesis, it is shown in [6] that any BBP-type constant is either
normal to base b (i.e., any n-long string digits appears in the base b expansion with
2

limiting frequency bn ), or else it is rational. No proof of Hypothesis A was presented

in [6], and indeed it is likely that Hypothesis A is rather difficult to prove. However,
it should be emphasized that even particular instances of Hypothesis A, if established,
would have interesting consequences. For example, if it could be established that the
specific iteration given by x0 = 0, and
xn = (2xn1 +

1
) mod 1
n

is equidistributed in [0, 1), then it would follow that log 2 is normal to base 2. In a similar
vein, if it could be established that the iteration given by x0 = 0 and
xn =

120n2 89n + 16
16xn1 +
512n4 1024n3 + 712n2 206n + 21

mod 1

is equidistributed in [0, 1), then it would follow that is normal to base 16 (and thus to
base 2 also).
One additional impetus for the study of BPP-type constants comes from a recent paper
by Lagarias [15], who demonstrates a connection to G-functions and to a conjecture of
Furstenberg from ergodic theory. Lagarias analysis suggests that there may be a special
signficance to constants that have BBP-type formulas in two or more bases say both
a base 2 and a base 3 formula.
This paper is a compendium of the growing set of BBP-type formulas that have been
found by various researchers. Part of these formulas are collected here from previously
published sources. In other cases, formulas whose existence has been demonstrated in the
literature are presented here explicitly for the first time. Still others are new, having been
found using the authors PSLQ program [5] in the course of this research.
The PSLQ integer relation algorithm [12] or one of its variants [5] can be used to find
formulas such as those listed in this paper as follows. Suppose, for example, that it is
conjectured that a given constant satisfies a BBP-type formula of the form

1
1X
=
r k=0 bk

a1
a2
an
+
+ +
,
s
s
(kn + 1)
(kn + 2)
(kn + n)s

where r and ak are unknown integers, for a specified selection of the parameters b, s and
P
n. Then one calculates the vector ( k0 1/(bk (kn + j)s ), 1 j n), as well as itself,
to very high precision and then gives this (n + 1)-long vector (including at the end)
to an integer relation finding program. If a solution vector (aj ) is found with sufficiently
high numerical fidelity, then

1
1 X
=
an+1 k=0 bk

a1
a2
an
+
+ +
s
s
(kn + 1)
(kn + 2)
(kn + n)s

(at least to the level of numeric precision used).

This compendium is not intended to be a comprehensive listing of all such formulas
ordinarily a formula is not listed here if
3

1. it is a telescoping sum.
2. it is a formal rewriting of another formula on the list.
3. it can be derived by a straightforward formal manipulation starting with another
formula in the list.
4. it is a linear combination of two or more formulas already in the list.
Item 1 refers to a summation such as
S =

X
1
k=1

bk

1
b2

,
k
k+2
!

which, if split into two summations, has the property that the terms of the first series
cancel with offset terms of the second series, so that S reduces to a rational number (in
this example, S = b + 1/2). Item 2 refers to the fact that a formula with base b and
length n can be rewritten as a formula with base br and length rn. Item 4 refers to the
fact that the rational linear sum of two BBP series can, in many cases, be written as a
single BBP series. This is clear if the two individual series have the same base b. If one
has base br and the other has base bs , their sum can be written as a single BBP series
with base blcm(r,s) [6]. Along this line, many of the formulas listed below possess variants
that can be obtained by adding to the listed formula a rational multiple of one of the zero
relations listed in Section 11.
The formulas are listed below using a notation introduced in [6]:
P (s, b, n, A) =

X
1 X

aj
k
s
k=0 b j=1 (kn + j)

(3)

where s, b and n are integers, and A = (a1 , a2 , , an ) is a vector of integers. For

instance, using this notation we can write formulas 1 and 2 more compactly as follows:
= P (1, 16, 8, (4, 0, 0, 2, 1, 1, 0, 0))
1
log 2 =
P (1, 2, 1, (1))
2

(4)
(5)

In most cases below, the representation shown using this notation is a translation from
the original source. Also, in some cases the formula listed here is not precisely the one
mentioned in the cited reference an equivalent one is listed here instead but the
original discoverer is given due credit. In cases where the formula has been found experimentally (i.e., by using the PSLQ integer relation finding algorithm), and no formal proof
.
is available, the relation is listed here with the = notation instead of an equal sign.
The P notation formulas listed below have been checked using a computer program
that parses the LATEXsource of this document, then computes the left-hand and right-hand
sides of these formulas to 2000 decimal digit accuracy.
Additional contributions to this compendium are welcome please send a note to
the author at [email protected].
4

2. Logarithm Formulas
Clearly log n can be written with a binary BBP formula (i.e. a formula with b = 2m
for some integer m) provided n factors completely using primes whose logarithms have
binary BBP formulas one merely combines the individual series for the different primes
into a single binary BBP formula. We have seen above that log 2 possesses a binary BBP
formula, and so does the log 3, by the following reasoning:
1
log 3 = 2 log 2 + log 1
4


=
=

1
1X
2 k=0 4k

X
1

1
2k + 1

k=0

4k

= 2

X
1
k=1

k2k

X
1
k=1

2
1
1
1X
+

2k + 1 2k + 2
4 k=0 4k

k4k

2
2k + 2

= P (1, 4, 2, (1, 0))

(6)

In a similar manner, one can show, by examining the factorization of 2n + 1 and 2n 1,

where n is an integer, that numerous other primes have this property. Harley [13] further
extended this list of primes by writing
1+i
Re log 1 n
2




1
1
n log 2 + log(22n1 2n + 1),
2
2


where Re denotes the real part. He noted that the Taylor series of the left-hand side
can be written as a binary BBP-type formula and then applied Aurefeuilles factorization
formula
24n2 + 1 = (22n1 + 2n + 1)(22n1 2n + 1)
to the right-hand side. More recently, Jonathan Borwein has observed that both of these
sets of results can be derived by working with the single expression
(1 + i)k
Re log 1
2n

!!

A preliminary list of primes p such that log p has a binary BBP formula was given in
[4]. This list has now been augmented by the author to the following:
2, 3, 5, 7, 11, 13, 17, 19, 29, 31, 37, 41, 43, 61, 73, 109, 113, 127, 151,
241, 257, 337, 397, 683, 1321, 1613, 2113, 2731, 5419, 8191, 43691, 61681,
87211, 131071, 174763, 262657, 524287, 2796203, 15790321, 18837001,
22366891, 4278255361, 4562284561, 2932031007403, 4363953127297,
4432676798593
(7)
This list is certainly not complete, and it is unknown whether or not all primes have
this property, or even whether the list of such primes is finite or infinite. The actual

formulas for log p for the primes above are generally straightforward to derive and are not
shown here.
One can also obtain BBP formulas in non-binary bases for the logarithms of certain
integers and rational numbers. One example is given by the base ten formula 47 below,
which was used in [4] to compute the ten billionth decimal digit of log(9/10).
3. Arctangent Formulas
Shortly after the original BBP paper appeared in 1996, Adamchik and Wagon observed
that [1]
tan1 2 =

1
P (1, 16, 8, (8, 0, 4, 0, 2, 0, 1, 0))
8

(8)

More recently, binary BBP formulas have been found for tan1 q for a large set of rational
numbers q. These experimental results, which were obtained by the author using the
PSLQ program, coincide exactly in the cases studied so far with the set of rationals given
by q = |Im(T )/Re(T )| or |Re(T )/Im(T )|, where
T =

m
Y
k=1

(1 + i)uk
1
2vk

!wk

(9)

The arctangents of these q clearly possess binary BBP formulas, because Im(log T ) decomposes into a linear sum of terms, the Taylor series of which are binary BBP formulas.
The author is indebited to Jonathan Borwein for this observation. See also [7, pg. 344].
Alternatively, one can write 9 as
T =

m 
Y
k=1

i
1 t
2k

uk 

1+i
1 v
2k

w k

(10)

for various m-long nonnegative integer vectors t, u, v, w and choices of signs as shown.
For example, setting t = (1, 1), u = (1, 1), v = (1, 3), w = (1, 1), with signs (1, 1, 1, 1),
gives the result T = 25/325i/8, which yields q = 4/5. Indeed one can obtain the formula
tan1

4
5

 

1
P (1, 220 , 40, (0, 219 , 0, 3 217 , 15 215 , 0, 0, 5 215 , 0, 215 , 0,
217
3 213 , 0, 0, 5 210 , 5 211 , 0, 211 , 0, 210 , 0, 0, 0, 5 27 , 15 25 , 128, 0,
96, 0, 0, 0, 40, 0, 8, 5, 6, 0, 0, 0, 0))
(11)

In this manner, it can be seen that binary BBP formulas exist for the arctangents of
the following rational numbers. Only those rationals with numerators < denominators
50 are listed here.
1/2, 1/3, 2/3, 1/4, 3/4, 1/5, 2/5, 3/5, 4/5, 1/7, 3/7, 4/7, 6/7,
1/8, 7/8, 1/9, 2/9, 7/9, 8/9, 3/10, 2/11, 3/11, 7/11, 8/11, 10/11,
1/12, 5/12, 1/13, 6/13, 7/13, 9/13, 11/13, 3/14, 5/14, 1/15, 4/15,
8/15, 1/16, 11/16, 13/16, 15/16, 1/17, 6/17, 7/17, 11/17, 15/17,
6

16/17, 1/18, 13/18, 4/19, 6/19, 7/19, 8/19, 9/19, 11/19, 17/19,
1/21, 16/21, 3/22, 7/22, 9/22, 19/22, 2/23, 4/23, 6/23, 7/23,
11/23, 14/23, 15/23, 7/24, 11/24, 23/24, 13/25, 19/25, 21/25,
7/26, 23/26, 5/27, 11/27, 2/29, 3/29, 15/29, 17/29, 24/29, 28/29,
17/30, 1/31, 5/31, 8/31, 12/31, 13/31, 17/31, 18/31, 22/31, 27/31,
1/32, 9/32, 31/32, 1/33, 4/33, 10/33, 14/33, 19/33, 31/33, 32/33,
7/34, 27/34, 13/35, 25/36, 5/37, 9/37, 10/37, 16/37, 29/37, 36/37,
1/38, 5/38, 13/38, 21/38, 20/39, 23/39, 37/39, 9/40, 3/41, 23/41,
27/41, 28/41, 38/41, 11/42, 19/42, 37/42, 6/43, 19/43, 23/43,
32/43, 33/43, 7/44, 23/44, 27/44, 3/46, 9/46, 17/46, 35/46, 37/46,
1/47, 13/47, 14/47, 16/47, 19/47, 27/47, 19/48, 3/49, 8/49, 13/49,
18/49, 31/49, 37/49, 43/49, 29/50, 49/50

(12)

Note that not all small rationals appear in this list. For instance, it is not known
whether tan1 (1/6) possesses a binary BBP formula. For that matter, it has not been
proven that formulas 9 and 10 above generate all such rational numbers, although this is
a reasonable conjecture.
One can obtain BBP formulas in non-binary bases for the arctangents of certain rational numbers by employing appropriate variants of formulas 9 and 10.
4. Other Degree 1 Binary Formulas
We present here some additional degree 1 binary BBP-type formulas (in other words,
in the P notation defined in equation 3 above, s = 1, and b = 2m for some integer m > 0).
=

2 ln(1 + 2)

1
2 tan1
2

=
=
=
=

1
P (1, 16, 8, (8, 8, 4, 0, 2, 2, 1, 0))
4
P (1, 4, 4, (2, 2, 1, 0))
1
P (1, 64, 12, (32, 0, 8, 0, 8, 0, 4, 0, 1, 0, 1, 0))
8
9
P (1, 64, 6, (16, 8, 0, 2, 1, 0))
32
1
P (1, 16, 8, (8, 0, 4, 0, 2, 0, 1, 0))
8
1
P (1, 16, 8, (8, 0, 4, 0, 2, 0, 1, 0))
8

(13)
(14)
(15)
(16)
(17)
(18)

Formula 13 was first found by Ferguson [12], while 14, which is the alternating sign
equivalent of 13, was found independently by Hales and by Adamchik and Wagon [1].
Technically speaking, these formulas can be obtained from the original BBP formula for
(formula 1) by adding 1/4 times relation 58 of Section 11, but they are included here
for historical interest, since their discovery predated the discovery of relation 58. Formula

15 appeared in [4]. A formula equivalent to 16 appeared in [6, pg. 185]. Formulas 17 and
18 are due to Knuth [14, pg. 628].
5. Degree 2 Binary Formulas
Here are some degree 2 binary formulas (i.e., s = 2, and b = 2m for some integer
m > 0). The constant G here is Catalans constant, namely G = 1 1/32 + 1/52 1/72 +
= 0.9159655941 . . ..
2 = P (2, 16, 8, (16, 16, 8, 16, 4, 4, 2, 0))
9
2 =
P (2, 64, 6, (16, 24, 8, 6, 1, 0))
8
. 1
P (2, 16, 8, (16, 40, 8, 28, 4, 10, 2, 3))
log2 2 =
6
1
log2 2 =
P (2, 64, 6, (64, 160, 56, 40, 4, 1))
32
1
1
G log 2 =
P (2, 16, 8, (8, 8, 4, 0, 2, 2, 1, 0))
8
16
1
.
P (2, 212 , 24, (212 , 213 , 51 29 , 15 210 , 210 , 39 28 , 0,
log 2 =
256
45 28 , 37 26 , 29 , 0, 3 28 , 64, 0, 51 23 , 45 24 , 16, 196, 0,
60, 37, 0, 0, 0))

1
3 log 2 =
P (2, 212 , 24, (9 29 , 27 29 , 9 211 , 27 29 , 0, 81 27 ,
128
9 26 , 45 28 , 9 28 , 0, 0, 9 26 , 72, 216, 9 25 , 9 26 , 0, 162,
9, 72, 36, 0, 0, 0, ))
1
.
G = 10 P (2, 212 , 24, (210 , 210 , 29 , 3 210 , 256, 211 , 256,
2
9 27 , 5 26 , 64, 64, 0, 16, 64, 8, 72, 4, 8, 4, 12, 5,
4, 1, 0))

(19)
(20)
(21)
(22)
(23)

(24)

(25)

(26)

Formulas 19, 20, 22 and 23 were presented in [4] (although 23 appeared in a 1909
book by Nielsen [16, pg. 105]). Formulas 21 and 25 were found by the author, using the
PSLQ program. Formulas for log 2 and G were first derived by Broadhurst, although the
specific explicit formulas given here (24 and 26) were found by the author in the course
of this research.
6. Degree 3 Binary Formulas
1
P (3, 212 , 24, (3 211 , 21 211 , 3 213 , 15 211 , 3 29 , 3 210 ,
8
72
3 28 , 0, 3 210 , 21 27 , 192, 3 29 , 96, 21 25 , 3 27 , 0,
24, 48, 12, 120, 48, 42, 3, 0))
(27)
1
.
log3 2 =
P (3, 212 , 24, (0, 3 213 , 27 212 , 3 214 , 0, 93 29 , 0, 3 214 , 27 29 ,
256
(3) =

3 29 , 0, 75 26 , 0, 3 27 , 27 26 , 3 210 , 0, 93 23 , 0, 192, 216,

24, 0, 3))
(28)
. 1
2 log 2 =
P (3, 212 , 24, (0, 9 211 , 135 29 , 9 211 , 0, 99 28 , 0, 27 210 , 135 26 ,
32
9 27 , 0, 45 26 , 0, 9 25 , 135 23 , 27 26 , 0, 396, 0, 72, 135, 18, 0, 0)) (29)
1
.
log2 2 = 56 P (3, 260 , 120, (7 259 , 37 260 , 63 258 , 85 259 , 3861 256 ,
2
3357 255 , 0, 655 258 , 347 254 , 79 253 , 0, 4703 252 , 7 253 , 0,
1687 252 , 655 254 , 7 251 , 4067 249 , 0, 6695 248 , 347 248 ,
0, 0, 7375 246 , 3861 246 , 37 248 , 63 246 , 85 247 , 7 245 ,
933 245 , 0, 655 246 , 347 242 , 37 244 , 875 243 , 4703 240 ,
7 241 , 0, 63 240 , 3105 238 , 7 239 , 4067 237 , 0, 85 239 , 441 239 ,
0, 0, 7375 234 , 7 235 , 79 233 , 63 234 , 85 235 , 7 233 ,
3357 231 , 875 233 , 655 234 , 347 230 , 37 232 , 0, 167 232 ,
7 229 , 0, 63 228 , 655 230 , 3861 226 , 4067 225 , 0, 85 227 ,
347 224 , 375 223 , 0, 7375 222 , 7 223 , 37 224 , 1687 222 ,
85 223 , 7 221 , 3357 219 , 0, 3105 218 , 347 218 , 37 220 , 0,
4703 216 , 3861 216 , 0, 63 216 , 655 218 , 7 215 , 923 215 , 0,
85 215 , 347 212 , 0, 875 213 , 7375 210 , 7 211 , 37 212 ,
63 210 , 6695 28 , 7 29 , 3357 27 , 0, 655 210 , 441 29 ,
37 28 , 0, 4703 24 , 224, 375 23 , 63 24 , 655 26 , 56, 8134,
875 23 , 85 23 , 347, 0, 0, 0))
(30)
1
.
3 = 54 P (3, 260 , 120, (5 259 , 15 260 , 225 258 , 95 259 , 4115 256 ,
2
3735 255 , 0, 685 258 , 505 254 , 5 253 , 0, 5485 252 , 5 253 , 0,
1775 252 , 685 254 , 5 251 , 3945 249 , 0, 7365 248 , 505 248 ,
0, 0, 8125 246 , 4115 246 , 15 248 , 225 246 , 95 247 , 5 245 ,
965 245 , 0, 685 246 , 505 242 , 15 244 , 125 246 , 5485 240 ,
5 241 , 0, 225 240 , 2835 238 , 5 239 , 3945 237 , 0, 95 239 ,
905 238 , 0, 0, 8125 234 , 5 235 , 5 233 , 225 234 , 95 235 ,
5 233 , 3735 231 , 125 236 , 685 234 , 505 230 , 15 232 , 0,
165 232 , 5 229 , 0, 225 228 , 685 230 , 4115 226 , 3945 225 ,
0, 95 227 , 505 224 , 125 223 , 0, 8125 222 , 5 223 , 15 224 ,
1775 222 , 95 223 , 5 221 , 3735 219 , 0, 2835 218 , 505 218 ,
15 220 , 0, 5485 216 , 4115 216 , 0, 225 216 , 685 218 , 5 215 ,
955 215 , 0, 95 215 , 505 212 , 0, 125 216 , 8125 210 , 5 211 ,
15 212 , 225 210 , 7365 28 , 5 29 , 3735 27 , 0, 685 210 ,
9

905 28 , 15 28 , 0, 5485 24 , 160, 125 23 , 225 24 , 685 26 , 40,

7890, 125 26 , 95 23 , 505, 0, 0, 0))
(31)
The existence of BBP formulas for these constants was originally established by Broadhurst [9]. However, except for 27, which appeared in [6], the specific explicit formulas
listed here were produced by the authors PSLQ program. The results for log2 and
3 were produced by a special parallel version of this program, running on the IBM SP
parallel computer system in the NERSC supercomputer facility at the Lawrence Berkeley
National Laboratory.
7. Degree 4 Binary Formulas
1
P (4, 212 , 24, (27 211 , 513 211 , 135 214 , 27 211 , 27 29 ,
164
621 210 , 27 28 , 729 210 , 135 211 , 513 27 , 27 26 ,
189 29 , 27 25 , 513 25 , 135 28 , 729 26 , 216, 621 24 ,
108, 216, 135 25 , 1026, 27, 0))
(32)
1
.
P (4, 212 , 24, (73 212 , 2617 212 , 8455 212 , 2533 212 ,
log4 2 =
205 25
73 210 , 25781 29 , 73 29 , 6891 211 , 8455 29 , 2617 28 ,
73 27 , 23551 26 , 73 26 , 2617 26 , 8455 26 , 6891 27 ,
73 24 , 25781 23 , 73 23 , 2533 24 , 8455 23 , 10468,
146, 615))
(33)
1
.
2 log2 2 =
P (4, 212 , 24, (121 211 , 3775 211 , 10375 211 , 1597 211 ,
41 25
121 29 , 3421 211 , 121 28 , 7695 210 , 10375 28 , 3775 27 ,
121 26 , 3539 28 , 121 25 , 3775 25 , 10375 25 , 7695 26 ,
121 23 , 3421 25 , 484, 1597 23 , 41500, 7550, 121, 0))
(34)
.
4 =

The existence of BBP-type formulas for these constants was originally established by
Broadhurst [9], although the explicit formulas given here were found by the authors PSLQ
program.
8. Degree 5 Binary Formulas
.
(5) =

1
P (5, 260 , 120, (279 259 , 7263 260 , 293715 257 ,
62651 249
13977 260 , 1153683 256 , 28377 260 , 279 256 , 83871 259 ,
293715 254 , 7263 256 , 279 254 , 889173 253 , 279 253 ,
7263 254 , 429705 252 , 83871 255 , 279 251 , 28377 254 ,
279 250 , 1041309 249 , 293715 248 , 7263 250 , 279 248 ,
1153125 247 , 1153683 246 , 7263 248 , 293715 245 , 13977 248 ,
279 245 , 28377 248 , 279 244 , 83871 247 , 293715 242 ,
10

7263 244 , 1153683 241 , 889173 241 , 279 241 , 7263 242 ,
293715 239 , 188811 239 , 279 239 , 28377 242 , 279 238 ,
13977 240 , 429705 237 , 7263 238 , 279 236 , 1153125 235 ,
279 235 , 7263 236 , 293715 233 , 13977 236 , 279 233 ,
28377 236 , 1153683 231 , 83871 235 , 293715 230 , 7263 232 ,
279 230 , 16497 233 , 279 229 , 7263 230 , 293715 227 ,
83871 231 , 1153683 226 , 28377 230 , 279 226 , 13977 228 ,
293715 224 , 7263 226 , 279 224 , 1153125 223 , 279 223 ,
7263 224 , 429705 222 , 13977 224 , 279 221 , 28377 224 ,
279 220 , 188811 219 , 293715 218 , 7263 220 , 279 218 ,
889173 217 , 1153683 216 , 7263 218 , 293715 215 , 83871 219 ,
279 215 , 28377 218 , 279 214 , 13977 216 , 293715 212 ,
7263 214 , 1153683 211 , 1153125 211 , 279 211 , 7263 212 ,
293715 29 , 1041309 29 , 279 29 , 28377 212 , 279 28 ,
83871 211 , 429705 27 , 7263 28 , 279 26 , 889173 25 ,
279 25 , 7263 26 , 293715 23 , 83871 27 , 279 23 ,
28377 26 , 2307366, 13977 24 , 293715, 29052, 279, 0))
(35)
1
.
log5 2 =
P (5, 260 , 120, (2783 259 , 32699 262 , 7171925 257 ,
2021 252
187547 261 , 41252441 256 , 9391097 257 , 2783 256 ,
52183 265 , 7171925 254 , 32699 258 , 2783 254 ,
29483621 253 , 2783 253 , 32699 256 , 17037475 252 ,
52183 261 , 2783 251 , 9391097 251 , 2783 250 ,
38246123 249 , 7171925 248 , 32699 252 , 2783 248 ,
41307505 247 , 41252441 246 , 32699 250 , 7171925 245 ,
187547 249 , 2783 245 , 9391097 245 , 2783 244 ,
52183 253 , 7171925 242 , 32699 246 , 41252441 241 ,
29483621 241 , 2783 241 , 32699 244 , 7171925 239 ,
12188517 239 , 2783 239 , 9391097 239 , 2783 238 ,
187547 241 , 17037475 237 , 32699 240 , 2783 236 ,
41307505 235 , 2783 235 , 32699 238 , 7171925 233 ,
187547 237 , 2783 233 , 9391097 233 , 41252441 231 ,
52183 241 , 7171925 230 , 32699 234 , 2783 230 ,
5881627 230 , 2783 229 , 32699 232 , 7171925 227 ,
52183 237 , 41252441 226 , 9391097 227 , 2783 226 ,
187547 229 , 7171925 224 , 32699 228 , 2783 224 ,
11

41307505 223 , 2783 223 , 32699 226 , 17037475 222 ,

187547 225 , 2783 221 , 9391097 221 , 2783 220 ,
12188517 219 , 7171925 218 , 32699 222 , 2783 218 ,
29483621 217 , 41252441 216 , 32699 220 , 7171925 215 ,
52183 225 , 2783 215 , 9391097 215 , 2783 214 , 187547 217 ,
7171925 212 , 32699 216 , 41252441 211 , 41307505 211 , 2783 211 ,
32699 214 , 7171925 29 , 38246123 29 , 2783 29 , 9391097 29 ,
2783 28 , 52183 217 , 17037475 27 , 32699 210 , 2783 26 ,
29483621 25 , 2783 25 , 32699 28 , 7171925 23 , 52183 213 ,
2783 23 , 9391097 23 , 82504882, 187547 25 , 7171925,
32699 24 , 2783, 30315))
(36)
1
.
2 log3 2 =
P (5, 260 , 120, (21345 259 , 464511 261 , 47870835 257 ,
2021 253
1312971 261 , 236170815 256 , 1579179 262 , 21345 256 ,
286131 265 , 47870835 254 , 464511 257 , 21345 254 ,
173704605 253 , 21345 253 , 464511 255 , 94128645 252 ,
286131 261 , 21345 251 , 1579179 256 , 21345 250 ,
215120589 249 , 47870835 248 , 464511 251 , 21345 248 ,
236128125 247 , 236170815 246 , 464511 249 , 47870835 245 ,
1312971 249 , 21345 245 , 1579179 250 , 21345 244 ,
286131 253 , 47870835 242 , 464511 245 , 236170815 241 ,
173704605 241 , 21345 241 , 464511 243 , 47870835 239 ,
56870019 239 , 21345 239 , 1579179 244 , 21345 238 ,
1312971 241 , 94128645 237 , 464511 239 , 21345 236 ,
236128125 235 , 21345 235 , 464511 237 , 47870835 233 ,
1312971 237 , 21345 233 , 1579179 238 , 236170815 231 ,
286131 241 , 47870835 230 , 464511 233 , 21345 230 ,
1950735 234 , 21345 229 , 464511 231 , 47870835 227 ,
286131 237 , 236170815 226 , 1579179 232 , 21345 226 ,
1312971 229 , 47870835 224 , 464511 227 , 21345 224 ,
236128125 223 , 21345 223 , 464511 225 , 94128645 222 ,
1312971 225 , 21345 221 , 1579179 226 , 21345 220 ,
56870019 219 , 47870835 218 , 464511 221 , 21345 218 ,
173704605 217 , 236170815 216 , 464511 219 , 47870835 215 ,
286131 225 , 21345 215 , 1579179 220 , 21345 214 , 1312971 217 ,
47870835 212 , 464511 215 , 236170815 211 , 236128125 211 ,
12

21345 211 , 464511 213 , 47870835 29 , 215120589 29 , 21345 29 ,

1579179 214 , 21345 28 , 286131 217 , 94128645 27 , 464511 29 ,
21345 26 , 173704605 25 , 21345 25 , 464511 27 ,
47870835 23 , 286131 213 , 21345 23 , 1579179 28 , 472341630,
1312971 25 , 47870835, 464511 23 , 21345, 0))
(37)
1
.
4 log 2 =
P (5, 260 , 120, (5157 259 , 89127 261 , 7805295 257 ,
2021 250
195183 261 , 32325939 256 , 1621107 259 , 5157 256 ,
37287 265 , 7805295 254 , 89127 257 , 5157 254 ,
24620409 253 , 5157 253 , 89127 255 , 12255165 252 ,
37287 261 , 5157 251 , 1621107 253 , 5157 250 ,
29192697 249 , 7805295 248 , 89127 251 , 5157 248 ,
32315625 247 , 32325939 246 , 89127 249 , 7805295 245 ,
195183 249 , 5157 245 , 1621107 247 , 5157 244 ,
37287 253 , 7805295 242 , 89127 245 , 32325939 241 ,
24620409 241 , 5157 241 , 89127 243 , 7805295 239 ,
5866263 239 , 5157 239 , 1621107 241 , 5157 238 ,
195183 241 , 12255165 237 , 89127 239 , 5157 236 ,
32315625 235 , 5157 235 , 89127 237 , 7805295 233 ,
195183 237 , 5157 233 , 1621107 235 , 32325939 231 ,
37287 241 , 7805295 230 , 89127 233 , 5157 230 ,
480951 233 , 5157 229 , 89127 231 , 7805295 227 ,
37287 237 , 32325939 226 , 1621107 229 , 5157 226 ,
195183 229 , 7805295 224 , 89127 227 , 5157 224 ,
32315625 223 , 5157 223 , 89127 225 , 12255165 222 ,
195183 225 , 5157 221 , 1621107 223 , 5157 220 ,
5866263 219 , 7805295 218 , 89127 221 , 5157 218 ,
24620409 217 , 32325939 216 , 89127 219 , 7805295 215 ,
37287 225 , 5157 215 , 1621107 217 , 5157 214 , 195183 217 ,
7805295 212 , 89127 215 , 32325939 211 , 32315625 211 , 5157 211 ,
89127 213 , 7805295 29 , 29192697 29 , 5157 29 , 1621107 211 ,
5157 28 , 37287 217 , 12255165 27 , 89127 29 , 5157 26 ,
24620409 25 , 5157 25 , 89127 27 , 7805295 23 , 37287 213 ,
5157 23 , 1621107 25 , 64651878, 195183 25 , 7805295,
89127 23 , 5157, 0))
(38)
As before, the existence of BBP-type formulas for these constants was originally es13

tablished by Broadhurst [9], although the explicit formulas given here were found by the
authors PSLQ program.
9. Ternary Formulas
No ternary BBP formulas (i.e. formulas with b = 3m for some integer m > 0) were
presented in [4], but several have subsequently been discovered. Here are some that are
now known:
2
P (1, 9, 2, (1, 0))
(39)
log 2 =
3
!

3
1
3 tan1
=
P (1, 9, 3, (3, 1, 0))
(40)
7
6

1
3 =
P (1, 36 , 12, (81, 54, 0, 9, 0, 12, 3, 2, 0, 1, 0, 0)) (41)
9
1
log 3 =
P (1, 36 , 6, (729, 81, 81, 9, 9, 1))
(42)
729
2
2 =
P (2, 36 , 12, (243, 405, 0, 81, 27, 72, 9, 9, 0,
27
5, 1, 0))
(43)
1
log2 3 =
P (2, 36 , 12, (4374, 13122, 0, 2106, 486, 243 23 ,
729
162, 234, 0, 162, 18, 8))
(44)

2
3 log 3 =
P (2, 36 , 12, (243, 405, 486, 135, 27, 0, 9, 15, 18,
27
5, 1, 0))
(45)
2

X
n=1

1
n

2n1
X

 
2n
n

k=n

1
1
=
P (2, 27, 6, (18, 18, 24, 6, 2, 0))
k
9

(46)

Formulas 39 and 40 appeared in [6]. Formulas 41 through 45 are due to Broadhurst

[8]. Formula 46 was numerically discovered by Broadhurst and proven in [3, pg. 88-92].
10. Other BBP-Type Formulas
Here are several interesting results in other bases, including results
for certain irrational

bases, and two formulas for an arbitrary base b. Here = (1 + 5)/2 is the golden mean.
9
1
log
=
P (1, 10, 1, (1))
10
10

! 5
25
781 57 5 5

log
= P (1, 5 , 5, (0, 5, 1, 0, 0))
2
256 57 + 5 5
!
!
q
1/4
1/4
1
233 329 5
939 + 281 5
1 5
1 5
tan

+
tan
5938
5938

1
=
P (1, 55 , 5, (125, 25, 5, 1, 0))
2 513/4


14

(47)
(48)

(49)

1111111111
log
387420489


1
P (1, 1010 , 10, (108 , 107 , 106 , 105 , 104 , 103 ,
8
10
102 , 101 , 1, 0))
(50)

b2 + b + 1
= 3P (1, b3 , 3, (b, 1, 0))
b log 2
b 2b + 1
!
bb 1
b2
b log
= P (1, bb , b, (bb2 , bb3 , , b2 , b, 1))
(b 1)b
2

= P (1, 1/(2)5 , 5, (1, 1/(22 ), 1/(43 ),

5 2+
1/(83 ), 0))
22

= P (1, 1/10 , 10, (1, 1, 1/, 1/3 , 0, 1/5 ,

5 2+
1/5 , 1/6 , 1/8 , 0))
2

= P (1, 1/5 , 5, (1, 1/2 , 1/3 , 1/3 , 0))

5 2+
6

= P (1, (/2)10 , 10, (1, 2 /2, 3 /4, 3 /8, 0,

5 2+
5 /32, 7 /64, 8 /128, 8 /256, 0))
 



1
1 X
Dk
9
1
cos1
=
10
10 k=0 10k k + 1
19
!

(51)
(52)

(53)

(54)
(55)

(56)
(57)

where in formula 57 the D coefficients satisfy the recurrence D0 = D1 = 1, and Dk+1 =

Dk 5Dk1 for k 2. It is possible that a variant of the original BBP algorithm can
be fashioned for this case, on the idea that the Dk comprise a Lucas sequence, and as
is known, evaluations of sequence elements mod n can be effected via exponential-ladder
methods. Formula 47 appeared in [4] (although it is an elementary observation). Formulas
48 through 52 appeared in [6]. Formulas 53 through 56 appeared in [11], although the
equivalent of formula 56 appeared earlier in [6, pg. 185]. Formula 57 appeared in [6].
11. Zero Relations
Below are some of the known BBP zero relations, or in other words BBP-type formulas
that evaluate to zero. These have been discovered using the authors PSLQ program, and
most are new with this compilation. For brevity, not all of the zero relations that have
been found are listed here some of the larger ones are omitted although the author
has a complete set. Further, zero relations that are merely a rewriting of another on the
list, such as by expanding a relation with base b and length n to one with base br and
length rn, are not included in these listings. For convenience, however, the total number
of linearly independent zero relations for various choices of s, b and n, including rewritings
and unlisted relations, are tabulated in Table 1.
Knowledge of these zero relations is essential for finding formulas such as those above
using integer relation programs (such as PSLQ). This is because unless these zero relations
are excluded from the search for a conjectured BBP-type formula, the search may only
15

recover a zero relation. A zero relation may be excluded from a integer relation search by
setting the input vector element whose position corresponds to the zero relations smallest
nonzero element to some value that is not linearly related to the other entries of the input
vector.
For example, note in Table 1 below that there are five zero relations with s = 1, b = 212
and n = 24. These relations are given below as formulas 60 through 64. If one is searching
for a conjectured formula with these parameters using PSLQ, then these five zero relations
must be excluded. This can be done by setting entries 19 through 23 of the PSLQ input
vector to e, e2 , e3 , e4 and e5 , respectively, where e is the base of natural logarithms.
Positions 19 through 23 are specified here because in relations 60 through 64 below, the
smallest nonzero entries appear in positions 23, 22, 21, 20 and 19, respectively. Powers of e
are specified here because, as far as anyone can tell (although this has not been rigorously
proven), e is not a polylogarithmic constant in the sense of this paper, and thus it and its
powers are not expected to satisfy BBP-type linear relations (this assumption is confirmed
by extensive experience using the authors PSLQ programs). In any event, it is clear that
many other sets of transcendental constants could be used here.
Note that by simply adding a rational multiple of one of these zero relations to one the
formulas above (with matching arguments s, b and n), one can produce a valid variant of
that formula. Clearly infinitely many variants can be produced in this manner.
Aside from the discussion in [10], these zero relations are somewhat mysterious
it is not understood why zero relations occur for certain s, b and n, but not others. It
should also be noted that in most but not all cases where a zero relation has been found,
nontrivial BBP-type formulas have been found with the same parameters. This suggests
that significant BBP-type results may remain to be discovered. In any event, it is hoped
that this compilation will spur some additional insight into these questions.
Note that all of these formulas except for the last two are binary formulas (i.e. b = 2m
for some integer m > 0).

0 = P (1, 16, 8, (8, 8, 4, 8, 2, 2, 1, 0))

(58)
0 = P (1, 64, 6, (16, 24, 8, 6, 1, 0))
(59)
.
0 = P (1, 212 , 24, (0, 0, 211 , 211 , 0, 29 , 256, 3 28 , 0, 0, 64, 128, 0, 32, 32,
48, 0, 24, 4, 8, 0, 2, 1, 0))
(60)
.
12
9
10 10
8
8
7
0 = P (1, 2 , 24, (2 , 2 , 2 , 7 2 , 256, 3 2 , 64, 3 2 , 0, 0, 0, 0, 8, 32, 16,
12, 4, 4, 1, 8, 0, 1, 0, 0))
(61)
.
12
9
10
9
7
0 = P (1, 2 , 24, (2 , 2 , 2 , 256, 0, 256, 64, 3 2 , 64, 0, 0, 0, 8, 16, 8, 12, 0,
4, 1, 2, 1, 0, 0, 0))
(62)
.
12
9
10
7
0 = P (1, 2 , 24, (3 2 , 3 2 , 0, 256, 0, 0, 192, 3 2 , 0, 0, 0, 64, 24, 48,
0, 12, 0, 0, 3, 2, 0, 0, 0, 0))
(63)
.
12
10
9 9
0 = P (1, 2 , 24, (2 , 3 2 , 2 , 256, 128, 128, 64, 192, 0, 32, 0, 32, 16, 16,
8, 0, 2, 2, 1, 0, 0, 0, 0, 0))
(64)
16

s
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

b
16
64
28
212
212
216
218
220
224
224
228
230
230
232
236
236
240
242
242
244
36

n
8
6
16
12
24
32
18
40
24
48
56
30
60
64
36
72
80
42
84
88
12

No. zero
relations
1
1
1
1
5
1
1
3
1
5
1
1
1
1
1
5
3
1
1
1
2

s
1
1
1
1
1
1
1
2
2
2
2
2
2
2
3
3
3
3
3
4

b
48

2
248
252
254
256
260
260
212
220
224
236
240
248
260
212
224
236
248
260
260

n
48
96
104
54
112
60
120
24
40
48
72
80
96
120
24
48
72
96
120
120

No. zero
relations
1
5
1
1
1
1
7
2
1
2
2
1
2
4
1
1
1
1
2
1

Table 1: Zero relation counts for various parameters

17

.
0 = P (1, 220 , 40, (0, 218 , 218 , 217 , 0, 5 216 , 216 , 5 215 , 0, 216 , 214 , 213 ,
0, 5 212 , 214 , 5 211 , 0, 210 , 210 , 211 , 0, 5 28 , 256, 5 27 , 0, 64, 64,
32, 0, 0, 16, 40, 0, 4, 16, 2, 0, 5, 1, 0))
(65)
.
0 = P (1, 220 , 40, (218 , 219 , 0, 217 , 3 215 , 216 , 0, 0, 214 , 213 , 0, 213 , 212 , 212 ,
5 210 , 0, 210 , 211 , 0, 29 , 256, 256, 0, 0, 96, 128, 0, 32, 16, 24,
0, 0, 4, 8, 5, 2, 1, 1, 0, 0))
(66)
.
0 = P (1, 220 , 40, (218 , 3 218 , 0, 218 , 13 215 , 0, 0, 5 215 , 214 , 213 , 0, 214 ,
212 , 0, 5 210 , 5 211 , 210 , 3 210 , 0, 3 29 , 256, 0, 0, 5 27 , 13 25 ,
192, 0, 64, 16, 40, 0, 40, 4, 12, 5, 4, 1, 0, 0, 0))
(67)
.
12
10
10 9
10
7
7
0 = P (2, 2 , 24, (0, 2 , 3 2 , 2 , 0, 2 , 0, 9 2 , 3 2 , 64, 0, 128, 0, 16, 48, 72,
0, 16, 0, 2, 6, 1, 0, 0))
(68)
.
12
11
11
10 9
10
8
0 = P (2, 2 , 24, (2 , 0, 17 2 , 17 2 , 2 , 15 2 , 256, 63 2 ,
17 28 , 0, 64, 5 28 , 32, 0, 17 25 , 63 24 , 8, 240, 4, 68, 68,
0, 1, 0))
(69)
.
20
19
20
18
18
20
18 16
0 = P (2, 2 , 40, (2 , 3 2 , 2 , 13 2 , 3 2 , 3 2 , 2 ,
25 216 , 215 , 3 216 , 214 , 13 214 , 213 , 3 214 , 3 215 , 25 212 ,
211 , 3 212 , 210 , 3 212 , 29 , 3 210 , 256, 25 28 , 3 210 ,
3 28 , 64, 13 26 , 32, 192, 16, 25 24 , 8, 48, 96, 52, 2, 12, 1, 0)) (70)
.
0 = P (3, 212 , 24, (211 , 19 211 , 5 214 , 211 , 29 , 23 210 , 256, 27 210 ,
5 211 , 19 27 , 64, 7 29 , 32, 19 25 , 5 28 , 27 26 , 8,
23 24 , 4, 8, 160, 38, 1, 0))
(71)
.
0 = P (1, 729, 12, (0, 81, 162, 0, 27, 36, 0, 9, 6, 4, 1, 0))
(72)
.
0 = P (1, 729, 12, (243, 324, 162, 81, 0, 36, 9, 0, 6, 1, 0, 0))
(73)
Relation 58 appeared in [4]. Relation 59 and 60 were given in [6]. Relations 61 through
73 were found by the author using his PSLQ program, and are new with this compilation.
13. Acknowledgements
The author wishes to acknowledge some very helpful comments and suggestions from
J. Borwein, P. Borwein, R. Crandall and S. Wagon.

18

References
[1] Victor Adamchik and Stan Wagon, A Simple Formula for Pi, American Mathematical Monthly, Nov. 1997, pg. 852-855.
[2] Victor Adamchik and Stan Wagon, Pi: A 2000-Year-Old Search Changes Direction,
Mathematica in Science and Education, vol. 5 (1996), no. 1, pg. 11-19.
[3] David H. Bailey and Jonathan M. Borwein, Mathematics by Experiment: Plausible
Reasoning in the 21st Century, AK Peters, 2004.
[4] David H. Bailey, Peter B. Borwein and Simon Plouffe, On The Rapid Computation
of Various Polylogarithmic Constants, Mathematics of Computation, vol. 66, no.
218, 1997, pg. 903913.
[5] David H. Bailey and David J. Broadhurst, Parallel Integer Relation Detection: Techniques and Applications, Mathematics of Computation, vol. 70, no. 236 (2000), pg.
1719-1736.
[6] David H. Bailey and Richard E. Crandall, On the Random Character of Fundamental Constant Expansions, Experimental Mathematics, vol. 10, no. 2 (June 2001), pg.
175-190.
[7] Jonathan M. Borwein and Peter B. Borwein, Pi and the AGM, John Wiley and Sons,
New York, 1987.
[8] David J. Broadhurst, Massive 3-loop Feynman Diagrams Reducible to SC Primitives of Algebras of the Sixth Root of Unity, preprint, Mar. 1998, available from
http://xxx.lanl.gov/format/hep-ph/9803091.
[9] David J. Broadhurst, Polylogarithmic Ladders, Hypergeometric Series and the
Ten Millionth Digits of (3) and (5), preprint, March 1998, available from
http://xxx.lanl.gov/format/math/9803067.
[10] David J. Broadhurst, Vanishing Polylogarithmic Ladders, manuscript, March 2000,
available from author.
[11] Hei-Chi Chan, Pi in Terms of Phi, manuscript, 2004.
[12] Helaman R. P. Ferguson, David H. Bailey and Stephen Arno, Analysis of PSLQ,
An Integer Relation Finding Algorithm, Mathematics of Computation, vol. 68, no.
225 (Jan. 1999), pg. 351-369.
[13] Robert Harley, personal communication to Peter Borwein, 1995.
[14] Donald E. Knuth, The Art of Computer Programming, vol. 2, third edition, AddisonWesley, 1998.
19

[15] Jeffrey C. Lagarias, On the Normality of Arithmetical Constants, manuscript, Sept.

2000.
[16] N. Nielsen, Der Eulersche Dilogarithmus, Halle, Leipzig, 1909.

20