Numbers, constants and computation

The Euler constant :

Xavier Gourdon and Pascal Sebah
April 14, 20041
= 0.57721566490153286060651209008240243104215933593992 . . .

Introduction

Eulers Constant was first introduced by Leonhard Euler (1707-1783) in 1734

as


1 1
1
= lim 1 + + + + log(n) .
(1)
n
2 3
n
It is also known as the Euler-Mascheroni constant. According to Glaisher [4],
the use of the symbol is probably due to the geometer Lorenzo Mascheroni
(1750-1800) who used it in 1790 while Euler used the letter C.
The constant is deeply related to the Gamma function (x) thanks to the
Weierstrass formula
 x i
Y h
1
x
= x exp(x)
1+
exp
.
(x)
n
n
n>0
This identity entails the relation
0 (1) = .

(2)

It is not known if is an irrational or a transcendental number. The question

of its irrationality has challenged mathematicians since Euler and remains a
famous unresolved problem. By computing a large number of digits of and
using continued fraction expansion, it has been shown that if is a rational
number p/q then the denominator q must have at least 242080 digits.
Even if is less famous than the constants and e, it deserves a great
attention since it plays an important role in Analysis (Gamma function, Bessel
functions, exponential-integral, ...) and occurs frequently in Number Theory
(order of magnitude of arithmetical functions for instance [11]).

Computation of the Euler constant

2.1

Basic considerations

Direct use of formula (1) to compute Euler constant is of poor interest since the
convergence is very slow. In fact, using the harmonic number notation
Hn = 1 +
1 This

1 1
1
+ + + ,
2 3
n

pages are from //numbers.computation.free.fr/Constants/constants.html

Numbers, constants and computation

we have the estimation

1
.
2n
This estimation is the first term of an asymptotic expansion which can be used
to compute effectively , as shown in next section. Nevertheless, other formulae
for (see next sections) provide a simpler and more efficient way to compute it
at a large accuracy. Better estimations are :
Hn log(n)

1
2(n + 1)

0 <
1
48n3

Hn log(n)

<

<

log(n) + log(n + 1)

2
1
1
Hn log(n + +
)
2 24n

Hn

<

<
<

1
2n

(Young [13])

1
6n(n + 1)
1
.
48(n + 1)3

(Cesaro)
(Negoi)

Application of the third relation with n = 100 gives

0.6127.109 < 0.577215664432730 < 0
and n = 1000 gives
0.6238.1013 < 0.577215664901484 < 0.
A similar estimation is given in [6].

2.2

Asymptotic expansion of the harmonic numbers

The Euler-Maclaurin summation can be used to have a complete asymptotic

expansion of the harmonic numbers. We have (see the essay on Bernoullis
numbers)
X B2k 1
1

,
(3)
Hn log(n) +
2n
2k n2k
k1

where the B2k are the Bernoulli numbers. Since B2k grows like 2(2k)!/(2)2k ,
the asymptotic expansion should be stopped at a given k. For example, the first
terms are given by
= Hn log(n)

1
1
1
1
1
1
691
1
+

+
.
2n 12n2 120n4 252n6 240n8 132n10 32760n12 12n14

This technique, directly inherited from the definition, can be employed to compute at a high precision but suffers from two major drawbacks :
It requires the computation of the B2k , which is not so easy ;
the rate of convergence is not so good compared to other formulas with .

Numbers, constants and computation

2.2.1

Eulers computation

In 1736, Euler used the asymptotic expansion 3 to compute the first 16 decimal
digits of . He went up to k = 7 and n = 10, and wrote
= H10 log(10)

1
1
1
1
1
+

+...
20 1200 1, 200, 000 252, 000, 000 24, 000, 000, 000

with
H10
log(10)

= 2.9289682539682539
= 2.302585092994045684

giving the approximation

0.5772156649015329.
2.2.2

Mascheronis mistake

During the year 1790, in Adnotationes ad calculum integrale Euleri , Mascheroni

made a similar calculation up to 32 decimal places. But, a few years later, in
1809, Johann von Soldner (1766-1833) found a value of the constant which was
in agreement only with the first 19 decimal places of Mascheronis calculation
... Embarrassing !
It was in 1812, supervised by the famous Mathematician Gauss, that a young
calculating prodigy Nicolai (1793-1846) evaluated up to 40 correct decimal
places, in agreement with Soldners value [4].
In order to avoid such miscalculations (see also William Shanks famous error
on his determination of the value of ), digits hunters are usually doing two
different calculations to check the result.
2.2.3

Stieltjes approach

In 1887, Stieltjes computed (2), (3), ..., (70) to 32 decimal places and extended a previous calculation done by Legendre up to (35) with 16 digits. He
was then able to compute Eulers constant to 32 decimal places thanks to the
fast converging series

X ((2k + 1) 1)
3
= 1 log( )
.
2
4k (2k + 1)
k=1

For large values of k we have

(2k + 1) 1 =

1
1
1
+ 2k+1 + 2k+1
22k+1
3
2

hence the series converges geometrically :

(2k + 1) 1
1

,
4k
2.16k

(4)

Numbers, constants and computation

This relation is issued from properties of the Gamma function and a proof is
given in the Gamma function essay.
The first partial sums of series (4) are
x0
x1
x5
x10
2.2.4

3
0.5(9453489189183561...) = 1 log( )
2
13
3
(3)
= 0.577(69681662853609...) =
log( )
12
2
12
= 0.57721566(733782033...)
= 0.57721566490153(417...)
=

Knuths computation

In 1962, Knuth used a computer to approximate with the Euler-Maclaurin

expansion (3), with the parameters k = 250 and n = 104 . The error is about
k,n =

B(2k+2)
1
2(2k + 2)!
101272

(2k+2)
(2k + 2) n
(2k + 2)(2n)2k+2

and Knuth gave 1271 decimal places of [8].

Some numerical results on the error function
To appreciate the rate of convergence of this algorithm we give a table of the
approximative number of digits one can find with different values for k and n.
This integer in the table is the number of digits of 1/k,n .
3

n = 10
n = 104
n = 105
n = 106

k = 10 k = 100 k = 250 k = 500

63
390
769
1235
85
592
1272
2237
107
794
1773
3239
129
996
2275
4241

This table shows that the Euler-Maclaurin summation could not be reasonably
used to compute more than a few thousands of decimal places of .

2.3

Exponential integral methods

An efficient way to compute decimal digits of the Euler constant is to start from
the identity = 0 (1) (see (2)) which entails for any positive integer n, after
integrating by part the formula
Z n
Z t
1 et
e
+ log(n) = In Rn ,
In =
dt, Rn =
dt.
t
t
0
n
Plugging the series expansion of (1 et )/t in In , we obtain
In =

X
k=1

(1)k1

nk
.
k k!

Numbers, constants and computation

The value In is an approximation to with the error bound Rn = O(en ).

By stopping the summation at the right index, we obtain the following formula
which provides an efficient way to approximate the Euler constant :
=

n
X

(1)k1

k=1

nk
log n + O(en ),
k k!

= 3.5911 . . .

(5)

The constant is such that nn /(n)! is of order en , and satisfies (log()

1) = 1. To obtain d decimal places of with (5), the formula should be used
with n ' d log(10) and computations should be done with a precision of 2d
decimal places to compensate cancellations in the sum In . This method was
used by Sweeney in 1963 to compute 3566 decimal places of [9].
A refinement is obtained by approximating Rn by its asymptotic expansion,
leading to the formula
=

n
X
k=1

(1)k1

n2
en X k!
nk
log(n)
+ O(e2n ),
k k!
n
(n)k

= 4.32 . . . (6)

k=0

The constant is such that (log() 1) = 2. This improvement, also due to

Sweeney [9], permits to take n ' d/2 log(10) and to work with a precision of
3d/2 decimal places to obtain d decimal places of .
Notice that Rn can be approximated as accurately as desired by using Eulers
continued fraction
en Rn = 1/n + 1/1 + 1/n + 2/1 + 2/n + 3/1 + 3/n +
This can be used to improve the efficiency of the technique, but leads to a much
more complicated algorithm.
More information about this technique can be found in [12].

2.4

Bessel function method

A better method (see also [12]) is based on the modified Bessel functions and
leads to the formula
An
log(n) + O(e4n ),
=
Bn
with
An =

n  k 2
X
n
k=0

k!

Hk ,

Bn =

n  k 2
X
n
k=0

k!

where = 3.5911 . . . satisfies (log() 1) = 1.

This technique is quite easy, fast and it has a great advantage compared to
Exponential integral techniques : to obtain d decimal places of , the intermediate computations can be done with d decimal places.

Numbers, constants and computation

A refinement can be obtained from an asymptotic series of the error term.

It consists in computing
2n

Cn =

1 X [(2k)!]3
.
4n
(k!)4 (16n)2k
k=0

Brent and McMillan in [12] suggest that

=

An
Cn
2 log(n) + O(e8n ).
Bn
Bn

(7)

This time, the summations in An and Bn should go up to n where =

4.970625759 . . . satisfies (log() 1) = 3. The error O(e8n ) followed an
empirical evidence but the result had not been proved by Brent and McMillan.
Formula (7) has been used by Xavier Gourdon with a binary splitting process
to obtain more than 100 millions decimal digits of in 1999.
Unlike the constant with the AGM iteration for instance, no quadratically
(or more) convergent algorithm is known for .

Collection of formulae for the Euler constant

Integral and series formulae for the Euler constant can be found in Collection
of formulae for the Euler constant.

Numbers, constants and computation

Records of computation

Number of digits
5
15
19
24
40
19
34
41
59
110
263
32
???
1271
3566
20,700
30,100
172,000
1,000,000
7,286,255
108,000,000

When
1734
1736
1790
1809
1812
1825
1857
1861
1869
1871
1878
1887
1952
1962
1962
1977
1980
1993
1997
1998 Dec.
1999 Oct.

Who
L. Euler
L. Euler
L. Mascheroni
J. von Soldner
F.B.G. Nicolai
A.M. Legendre
Lindman
Oettinger
W. Shanks
W. Shanks
J.C. Adams
T. J. Stieltjes
J.W. Wrench
D.E. Knuth
D.W. Sweeney
R.P. Brent
R.P. Brent and E.M. McMillan
J. Borwein
T. Papanikolaou
X. Gourdon
X. Gourdon and P. Demichel

Notes
He found = 0.577218.
The Euler-Maclaurin summation was u
Mascheroni computed 32 decimal place
In a work on the logarithm-integral fun
In agreement with Soldners calculation
Euler-Maclaurin summation was used w
Euler-Maclaurin summation was used w
Euler-Maclaurin summation was used w
Euler-Maclaurin summation was used w

Adams also computed the first 62 Bern

He used a series based on the zeta func
Euler-Maclaurin summation [7].
Euler-Maclaurin summation [8].
The exponential integral method was u
Brent used Sweeneys approach [10].
The Bessel function method [12] was u
A variant of Brents method was used.
This is the first gamma computation b
Sweeneys method (with N = 223 ) wit
Formula (7) was used with a binary sp

References
[1] L. Euler, Inventio summae cuiusque seriei ex dato termino generali, St
Petersbourg, (1736)
[2] A.M. Legendre, Traite des Fonctions Elliptiques, Paris, (1825-1828), vol. 2,
p. 434
[3] W. Shanks, (On Eulers constant), Proc. Roy. Soc. London, (1869), vol. 18,
p. 49
[4] J.W.L. Glaisher, History of Eulers constant, Messenger of Math., (1872),
vol. 1, p. 25-30
[5] J.C. Adams, On the value of Eulers constant, Proc. Roy. Soc. London,
(1878), vol. 27, p. 88-94
[6] G. Horton, A note on the calculation of Eulers constant, American Mathematical Monthly, (1916), vol. 23, p. 73
[7] J.W. Wrench Jr., A new calculation of Eulers constant, MTAC, (1952),
vol. 6, p. 255

Numbers, constants and computation

[8] D.E. Knuth, Eulers constant to 1271 places, Math. Comput., (1962), vol.
16, p. 275-281
[9] D.W. Sweeney, On the Computation of Eulers Constant, Mathematics of
Computation, (1963), p. 170-178
[10] R.P. Brent, Computation of the regular continued fraction for Eulers constant, Math. Comp., (1977), vol. 31, p. 771-777
[11] G.H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers,
Oxford Science Publications, (1979)
[12] R.P. Brent and E.M. McMillan, Some New Algorithms for High-Precision
Computation of Eulers constant, Math. Comput., (1980), vol. 34, p. 305312
[13] R.M. Young, Eulers constant, Math. Gazette 75, (1991), vol. 472, p. 187190