How Euler Did It

by Ed Sandifer
Bernoulli numbers
September 2005

As we learned in last month’s column, in the 1760’s Euler wrote only two articles on series.
There was E-326, a paper on the so-called central trinomial coefficients and the subject of that column.
In researching that column, I also looked at the other paper, E-393, De summis serierum numeros
Bernoullianos involventium, or “On sums of series involving Bernoulli numbers.” While the paper itself
didn’t turn out to be all that interesting, the path that leads to the paper is fascinating. It is the story of
the Bernoulli numbers.

Bernoulli numbers are a sequence of rational numbers that arise in a dazzling variety of
applications in analysis, numerical analysis and number theory. When Charles Babbage designed the
Analytical Engine in the 19th century, one of the most
important tasks he hoped the Engine would perform was
the calculation of Bernoulli numbers.

The first few Bernoulli numbers [K] are

B0 = 1
−1
B1 =
2
1
B2 =
6
B3 =0
−1
B4 =
30
B5 =0
1
B6 =
42

After B1 all Bernoulli numbers with odd index are

zero, and the non-zero ones alternate in sign. They first
appeared in 1713 in Jakob Bernoulli’s pioneering work on
probability, Ars Conjectandi. Jakob Bernoulli (1654-1705) was the older brother of Johann Bernoulli
(1667-1748), who was, in turn, Euler’s teacher and mentor at the University of Basel.
1
 Sometimes people simply omit the Bernoulli numbers with odd index from the list, and write Bk*
where we write B2k . They, of course, must then make certain modifications to their formulas, and, in
general, their formulas are a bit simpler.

Bernoulli was studying sums of powers of consecutive integers, like sums of squares,

1 + 4 + 9 + 16 + 25 = 55
or sums of cubes
1 + 8 + 27 + 64 + 125 + 216 + 343 = 784

In modern notation (Bernoulli did not use subscripts, nor did he use Σ for summations or ! for
factorials) Bernoulli found that

n −1 p
Bk
∑kp =∑
p!
n p+1− k
k =1 k=0 k ! ( p + 1 − k ) !

If n is large and p is small, that means that the left hand side is a sum of a relative large number
of relatively small powers, and if we know the necessary Bernoulli numbers then the sum on the right is
simpler to evaluate than the sum on the left. Bernoulli himself is said [G+S] to have used this formula to
find the sum of the tenth powers of numbers 1 to 1000 in less than eight minutes. The answer is a 32-
digit number.
Bernoulli numbers arise in Taylor series in the expansion
∞
x xk
= ∑ k k! .
e x − 1 k =0
B

Bernoulli numbers are also involved in the expansions of several other functions, including
x  sin x 
tan x, , log   and others.
sin x  x 

Euler encountered Bernoulli numbers in his great solution to the Basel problem when he showed
that
∞
1 1 1 1 1 π2
∑
k =1 k
2
= 1 + + + +
4 9 16 25
+etc. =
6
,

though he did not recognize them at the time. In the same paper, Euler also evaluated

∞
1
∑k n
k =1

for the first several even values of n. Only later would he realize that these other sums involved the
Bernoulli numbers. In fact, if n is even, then

∞
1 2n | Bn | π n
∑ k n = 2 ( n !) .
k =1

2
 Euler also failed to recognize the Bernoulli numbers in 1732 when he first did his work on the
Euler-Maclaurin formula. Maclaurin also missed them when he discovered the formula independently in
1742. Again in modern form, the result says that for sufficiently smooth functions f, a series based on f
and an integral of f are related by

f (1) + f ( n )
( f 2 k −1 ( n ) − f 2 k −1 (1) ) + Rn ( f , p )
n p

∑ f ( k ) = ∫ f (x ) dx + +∑
n B2 k
2 k =1 ( 2k ) !
1
k =1

where Rn (f, p) is a remainder term that usually disappears rapidly as p increases. [G+S] The formula
can be used either to estimate the series on the left knowing the integral on the right, or conversely, to
estimate the integral by evaluating the series. Euler used the series on the left hand side of this formula
∞
1
in 1732 to estimate the values of infinite series and to find ∑ 2 to six decimal places. This is also how
k =1 k
he found the first properties of γ, the so-called Euler constant. Maclaurin used the other side of the
formula to estimate the values of integrals from series.

The Bernoulli numbers are related to Euler’s constant γ by

1 ∞ B2k
γ = +∑ .
2 k =1 2k

There is also an astonishing result due to Kummer [G+S] relating Bernoulli numbers to Fermat’s
Last Theorem. Kummer noticed that a prime number p is special if it does not divide the numerators of
any of the Bernoulli numbers B2 , B4 , B6 , … Bp-3 . Such primes are now called regular. The first prime
that is not regular is 37. Kummer showed that if p is a regular prime, then Fermat’s Last Theorem, that
x p + y p = z p has no non-trivial integer solutions, is true for p.

In 1921, Eric Temple Bell, author of the well-known popular mathematics history book Men of
Mathematics, proved [B]:

Theorem: If p is an odd prime which does not divide 4 r − 1 , then the

numerator of B2 pr is divisible by p.

There must be thousands of such results, and Bernoulli numbers continue to be studied today.
JSTOR reports over 150 “hits” on the key words “Bernoulli numbers” since 1990.

Let us turn now to 1755, when Euler published his Institutiones calculi differentialis [E212]. At
that time, only a few of the results above were known, and their links to Bernoulli numbers were
apparently not yet recognized. The results that were known seem to be:

1. Bernoulli’s own results on summing powers of integers. Bernoulli showed how

this involved Bernoulli numbers, hence the name.
2. The Euler-Maclaurin summation formula.
3. Taylor series for various functions.
4. Euler’s evaluation of ζ(2n).

3
 Then, through all the trees, Euler sees the forest. It must have been a wonderful feeling to see
how so many different aspects of mathematics are linked through these mysterious Bernoulli numbers.

Euler devotes almost all of chapters 5 and 6 of Part 2 of his Calculus differentialis to results
related to Bernoulli numbers, and on page 420 (page 321 of the Opera Omnia edition) he attributes them
to Jakob Bernoulli and calls them Bernoulli numbers. Unfortunately, only Part 1 of the Calculus
differentialis has been translated into English, so readers who want to enjoy it in Euler’s words must
either brave the Latin or find a copy of the rare 1790 German translation.

Euler begins his chapter 5, “Investigation of the sums of series from their general term” with a
quick treatment of Bernoulli’s results on summing sequences of powers. Then he repeats his own
results from the 1730’s [E25] on the Euler-Maclaurin formula and gives the recursive relation on the
coefficients in that formula. Euler doesn’t mention Maclaurin, so he is probably unaware of his work on
the subject.
x
Then he shows how those coefficients arise from the Taylor series expansions of −x
and
1− e
1 1 
cot  x  .
2 2 

Eventually, after quite a bit of work, he lists the Bernoulli numbers, naming them after Bernoulli
in the process, and shows how they are related to the coefficients in the Euler-Maclaurin formula.

1 1 
This done, he extends occurrence of Bernoulli numbers in the expansion of cot  x  to the
2 2 
π  mπ 
more general form cot   and uses that to relate Bernoulli numbers to the values of ζ(2n). To end
n  n 
the theoretical parts of his exposition, he gives some of the properties of the Bernoulli polynomials and
notes that Bernoulli numbers grow faster than any geometric series.

Euler spends the rest of these two chapters doing applications of Bernoulli numbers, including
calculating the Euler-Mascheroni constant, γ, to 15 decimal places.

All this is rather unexpected in a textbook on differential calculus.

With this, Euler did not write again on Bernoulli numbers until 1768. In fact, in the intervening
13 years, he wrote only four papers on series. Besides the one on central trinomial coefficients that was
the subject of last month’s column, he wrote one paper on approximating pi, one on trig functions, and
one on continued fractions.

We have already said that the 1768 paper, E-393, “didn’t turn out to be all that interesting,” but it
might be worth summarizing its results. Euler opens E-393 with a list of Bernoulli numbers and a list of
the coefficients that arise in ζ(2n), and shows how two lists are related. Then he gives his recursive
relation on the zeta coefficients.

Then he leaps to the Euler-Maclaurin formula. Up to this point, most of the essay is just a new
version of what he had presented in the Calculus differentialis. From here, though, he gives a different
way to show the relation between the Bernoulli numbers and the expansion of − cot  x  . Then he
1 1 1
2 2 2 
4
uses this same technique to give new relations between the Bernoulli numbers and a variety of other
x e y + e− y 1 1
functions and numbers, including −y
− and 2 . Finally, he gives the values of integrals
2e −e e −1
y
2
1 ( ln x )
n

like ∫0 dx , for n odd, in terms of the (n + 1)st Bernoulli number. None of this would have been
(1 − x )
appropriate to include in the Calculus differentialis.

Bernoulli numbers are still a bit mysterious. They appear frequently in Julian Havel’s recent
book Gamma, about Euler’s constant, and people continue to discover new properties and to publish
articles about them.

Simon Singh [S] quotes Andrew Wiles as describing the process of mathematical discovery with
the colorful words “You enter the first room of the mansion and it’s completely dark. You stumble
around bumping into the furniture but gradually you learn where each piece of furniture is. Finally, after
six months or so, you find the light switch, you turn it on, and suddenly it’s all illuminated.” It must
have been something like this for Euler, when he saw how the “furniture” was arranged around the
Bernoulli numbers.

In Euler’s time, though, light switches hadn’t yet been invented.

References:

[B] Bell, E. T., Note on the Prime Divisors of the Numerators of Bernoulli's Numbers, The American Mathematical
Monthly, Vol. 28, No. 6/7. (Jun. - Jul., 1921), pp. 258-259.
[E25] Euler, Leonhard, Methodus generalis summandi progressiones, Commentarii academiae scientiarum Petropolitanae
6 (1732/3) 1738, p. 68-97. Reprinted in Opera Omnia Series I vol 14, p. 42-72. Available on line at
www.EulerArchive.org.
[E212] Euler, Leonhard, Institutiones calculi differentialis, St. Petersburg, 1755, reprinted in Opera Omnia Series I vol 10.
English translation of Part I by John Blanton, Springer, NY, 2000.
[E393] Euler, Leonhard, De summis serierum numeros Bernoullianos involventium, Novi commentarii academiae
scientiarum Petropolitanae 14 (1769): I, 1770, p. 129-167, reprinted in Opera Omnia Series I vol 15 pp. 91-130.
[G+S] Gourdon, Xavier and Pascal Sebah, Numbers, constants and computation, online at
numbers.computation.free.fr/Constants/constants.html, link to Constants, Miscellaneous, Bernoulli numbers,
consulted July 25, 2005
[H] Havil, Julian, Gamma: Exploring Euler’s Constant, Princeton University Press, 2003.
[K] Kellner, Bernd C., The Bernoulli Number Page online at Bernoulli.org. Consulted July 25, 2005.
[S] Singh, Simon, Who is Andrew Wiles? SimonSingh.net, on line at http://www.simonsingh.com/Andrew_Wiles.html.
Consulted August 1, 2005.

Ed Sandifer ([email protected]) is Professor of Mathematics at Western Connecticut State

University in Danbury, CT. He is an avid marathon runner, with 33 Boston Marathons on his shoes, and
he is Secretary of The Euler Society (www.EulerSociety.org)

How Euler Did It is updated each month.

Copyright ©2005 Ed Sandifer