Bernoulli Numbers

Junior Number Theory Seminar – University of Texas at Austin

September 6th, 2005
Matilde N. Lalı́n

I will mostly follow [2].

Definition and some identities

Definition 1 Bernoulli numbers are defined as B0 = 1 and recursively as

m−1
X µm + 1¶
(m + 1)Bm = − Bk ,
k
k=0

so we find B1 = − 21 , B2 = 61 , B3 = 0, B4 = − 30
1
, B5 = 0, B6 = 1
42 , . . . , B12
691
= − 2730 ,...
Lemma 2
∞
t X tm
= Bm .
et − 1 m!
m=0

t P∞ tm
PROOF. Write et −1 = m=0 am m! and multiply by et − 1,
∞
∞ n X
X t tm
t= am ,
n! m!
n=1 m=0

equate coefficients for tm+1 gives a0 = 1 and

m µ ¶
X m+1
ak = 0.
k
k=0

Theorem 3 (J. Bernoulli) Let m be a positive integer and define

Sm (n) = 1m + . . . + (n − 1)m ,
then
m µ ¶
X m+1
(m + 1)Sm (n) = Bk nm+1−k .
k
k=0
P∞ tm
PROOF. In ekt = m=0 k
m
m! substitute k = 0, 1, . . . , n − 1 and add,
∞
X tm ent − 1 t
Sm (n) = 1 + et + . . . + e(n−1)t =
m! t et − 1
m=0
∞ ∞
X tk−1 X tj
= nk Bj
k! j!
k=1 j=0

now equate the coefficients of tm and multiply by (m + 1)!. ¤

1
Definition 4
m µ ¶
X m
Bm (x) = Bk xm−k
k
k=0

are called Bernoulli polynomials.

So B0 (x) = 1, B1 (x) = x − 12 , etc.

Then Theorem 3 may be stated as
1
Sm (n) = (Bm+1 (n) − Bm+1 ).
m+1
Lemma 5
∞
text X tm
= Bm (x) .
et − 1 m!
m=0

PROOF. First note that

m−1
Xµ ¶
0 m
Bm (x) = (m − k)Bk xm−1−k = mBm−1 (x).
k
k=0

Also
1 m µ ¶
1 X m+1
Z
Bm (x)dx = Bk = 0, m ≥ 1.
0 m+1 k
k=0
tm
Now let F (x, t) = ∞
P
m=0 Bm (x) m! , differentiating,

∞
∂ X tm
F (x, t) = Bm−1 (x) = tF (x, t).
∂x (m − 1)!
m=1

Now we solve using separation of variables, F (x, t) = T (t)ext , then

Z 1 Z 1
et − 1
F (x, t)dx = T (t)ext dx = T (t)
0 0 t
but
1 ∞
tm 1
Z X Z
F (x, t)dx = Bm (x)dx = 1
0 m! 0
m=0

and this proves the statement (Castellanos, [1]). ¤

Pm ¡m¢
Proposition 6 1. Bm (x + 1) − Bm (x) = k=0 k Bm−k (x) = mxm−1 (Roman [5]).

2. Bm (1 − x) = (−1)m Bm (x).

3. Bm = m 1 Pk r k m (Rademacher [4]).
P ¡ ¢
k=0 k+1 r=0 (−1) r r
Pm k+m m B (k) = m! (Ruiz [6]).
¡ ¢
4. k=0 (−1) k m

Pk−1 ³ ´
5. Bm (kx) = k q−1 j=0 Bm x + kj .

2
 Euler MacLaurin sum formula (Rademacher, [4]).

Let f (x) smooth. Since B10 (x) = 1,

Z 1 Z 1
f (x)dx = B1 (x)f (x)|10 − B1 (x)f 0 (x)dx
0 0

q
Bm (x) (m−1) ¯¯1
¯ Z 1
X
m−1 q Bq (x) (q)
= ... = (−1) f (x)¯ + (−1) f (x)dx
m! 0 0 q!
m=1

Evaluating in x = 1,
1 q 1
Bq (x) (q)
Z Z
m Bm
X
(m−1) (m−1) q−1
f (1) = f (x)dx + (−1) (f (1) − f (0)) + (−1) f (x)dx.
0 m! 0 q!
m=1

Changing f (x) by f (n − 1 + x) and adding, we obtain the formula

b Z b q
X X Bm (m−1)
f (n) = f (x)dx + (−1)m (f (b) − f (m−1) (a)) + Rq
a m!
n=a+1 m=1

where
b
(−1)q−1
Z
Rq = Bq (x − [x])f (q) (x)dx
q! a

An integral and some identities

Proposition 7 We have:
³ ´ ³ ´
2 log a 2 log b
Z ∞
x log xdx k ³ π ´k+1 Pk π − Pk π
= .
0 (x2 + a2 )(x2 + b2 ) 2 a2 − b2

where
2ik+1 ³ ³x´ ³ x ´´ (2k+1 − 2)ik+1
Pk (x) = Bk+1 − 2k Bk+1 + Bk+1
k+1 i 2i k+1

PROOF. (Idea) We first prove that

∞
xα dx π(aα−1 − bα−1 )
Z
= for 0 < α < 1, a, b 6= 0
0 (x2 + a2 )(x2 + b2 ) 2 cos πα 2 2
2 (b − a )

by first computing
∞
xα dx πaα−1
Z
= .
0 x2 + a2 2 cos πα
2
We note that the polynomials Pk may be defined recursively as
k+1
xk+1
µ ¶
1 X j+1 k+1
Pk (x) = + (−1) 2 Pk+1−j (x).
k+1 k+1 j
j>1 (odd)

3
 The idea, suggested by Rodriguez-Villegas, is to obtain the value of the integral in the
statement by differentiating k times the integral of with α and then evaluating at α = 1.
Let
π(aα−1 − bα−1 )
f (α) =
2 cos πα 2 2
2 (b − a )
which is the value of the integral with α. In other words, we have
Z ∞
(k) x logk xdx
f (1) = .
0 (x2 + a2 )(x2 + b2 )
By developing in power series around α = 1, we obtain
∞
X logn a − logn b
πα π
f (α) cos = (α − 1)n .
2 2(b2 − a2 ) n!
n=0

By differentiating k times,
k µ ¶ ∞
X k ³ πα ´(j) π X logn+k a − logn+k b
f (k−j) (α) cos = (α − 1)n .
j 2 2(b2 − a2 ) n!
j=0 n=0

We evaluate in α = 1,
k
π(logk a − logk b)
µ ¶
X j+1 k (k−j) ³ π ´j
(−1) 2 f (1) = .
j 2 2(b2 − a2 )
j=0 (odd)

As a consequence, we obtain
k+1
k + 1 (k+1−j) ³ π ´j−1 logk+1 a − logk+1 b
µ ¶
(k) 1 X j+1
f (1) = (−1) 2 f (1) + .
k+1 j 2 (k + 1)(a2 − b2 )
j>1 (odd)

When k = 0,
³ ´ ³ ´
2 log a 2 log b
log a − log b π P0 π − P0 π
f (0) (1) = f (1) = = .
a2 − b2 2 a2 − b2
The general result follows by induction on k and the definition of Pk . ¤

Theorem 8 We have the following identities:

• For 1 ≤ l ≤ n:
sn−l (12 , . . . , (2n − 1)2 )
n−l µ ¶
X
2 2 1 2(l + s)
=n sn−l−s (2 , . . . , (2n − 2) ) B2s (22s − 2)(−1)s+1 .
l+s 2s
s=0

• For 1 ≤ n:
µ ¶2 n
(2n)! X 1
= 2n sn−s (22 , . . . , (2n − 2)2 ) B2s (22s − 1)(−1)s+1 .
2n n! s
s=1

4
 • For 0 ≤ l ≤ n:
(2l + 1)sn−l (22 , . . . , (2n)2 )
n−l µ ¶
X
2 2 2(l + s)
= (2n + 1) sn−l−s (1 , . . . , (2n − 1) )B2s (22s − 2)(−1)s+1 .
2s
s=0

• For 1 ≤ n:
n
X 22s (22s − 1)
sn−s (22 , . . . , (2n − 2)2 )(−1)s+1 B2s = 2(2n − 1)!
s
s=1

where 
1
 P if l=0
sl (a1 , . . . , ak ) = i1 <...<il ai1 . . . ail if 0<l≤k
0 if k<l


are the elementary symmetric polynomials, i.e.,

k
Y k
X
(x + ai ) = sl (a1 , . . . , ak )xk−l
i=1 l=0

Some big classic results

Theorem 9 (Euler)
(2π)2m
2ζ(2m) = (−1)m+1 B2m .
(2m)!

PROOF. We will need

∞
1 X x
cot x = −2 .
x n2 π 2 − x2
n=1

This identity may be deduced by applying the logarithmic derivative to

∞ µ
x2
Y ¶
sin x = x 1− 2 2 .
n π
n=1

Then
∞ ∞ ∞
X x2 X ³ x ´2k X x2m
x cot x = 1 − 2 = 1 − 2 ζ(2m)
n2 π 2 nπ π 2m
n=1 k=0 m=1

On the other hand,

∞
cos x i(eix + e−ix ) 2ix X (2ix)n
x cot x = x = x ix = ix + = 1 + Bn
sin x e − e−ix e2ix − 1 n!
n=2

and compare coefficients of x2m .

2 π4
For instance, ζ(2) = π6 , ζ(4) = 90 , etc.

Corollary 10 1. (−1)m+1 B2m > 0.

5
 2(2m)!
2. ¯ B2m
¯ ¯
2m ¯
→ ∞ or B2m ∼ (−1)m+1 (2π) 2m as m → ∞.

PROOF. The first assertion is consequence of the fact that ζ(2m) is positive. The
second is consequence of the fact that ζ(2m) > 1 implies

2(2m)!
|B2m | > .
(2π)2m
¤

Theorem 11 (Claussen, von Staudt) For m ≥ 1

X 1
B2m ≡ − (mod 1)
p
(p−1)|2m, p prime

We will need the following

Definition 12 For every rational number r and p prime write r = pk ab where a, b are
integers such that p 6 |ab. Then ordp (r) = k. We say that r is p-integral if ordp (r) ≥ 0.

Lemma 13 Let p be a prime number and k a positive integer, then

pk
1. k+1 is p-integral.
pk
2. k+1 ≡ 0 (mod p) if k ≥ 2.
pk−2
3. k+1 is p-integral if k ≥ 3 and p ≥ 5.

p k k−a
PROOF. By induction, k + 1 ≤ pk . Let k + 1 = pa q. Then k+1 = p q ≥ 1 implies
k ≥ a. For the second case use that k + 1 < pk for k ≥ 2. The third case is consequence of
k + 1 < pk−2 for k ≥ 3 and p ≥ 5. ¤

Proposition 14 Let p be a prime and m a positive integer. Then pBm is p-integral. Also,
if m is even pBm ≡ Sm (p) (mod p)

PROOF. For the first statement we will use induction. It is clear for m = 1. Now note
that for m ≥ k we have µ ¶ µ ¶
m+1 m+1 m
=
k m−k+1 k
Then Theorem 3 becomes
m µ ¶ m µ ¶
X m nm+1−k X m nk+1
Sm (n) = Bk = Bm−k (1)
k m+1−k k k+1
k=0 k=0

Now set n = p and since Sm (p) is integer, it suffices to prove that

pk
µ ¶
m
pBm−k
k k+1
is p-integral for k = 1, . . . , m, but that is true by induction and Lemma 13.
6
 For the congruence it suffices to see that

pk
µµ ¶ µ ¶¶
m
ordp pBm−k ≥1
k k+1

for k ≥ 1. By Lemma 13 this is true for k ≥ 2. The case with k = 1 corresponds to

m
2 (pBm−1 )p and it is true because m is even and the only nontrivial case is with m = 2.

Lemma 15 Let p be a prime. If p − 1 6 |m, then Sm (p) ≡ 0 (mod p). If p − 1|m then
Sm (p) ≡ −1 (mod p)

PROOF. First suppose that p − 1 6 |m. Let g be a primitive root modulo p. Then

Sm (p) = 1m + . . . + (p − 1)m ≡ 1m + g m + . . . + g (p−2)m (mod p)

and
(g m − 1)Sm (p) ≡ g m(p−1) − 1 ≡ 0 (mod p)
the result follows. Now suppose that p − 1|m, then

Sm (p) ≡ 1 + 1 + . . . + 1 ≡ p − 1 (mod p)

¤
PROOF. (Theorem 11) Assume m is even. By Proposition 14, pBm is p-integral and
≡ Sm (p) (mod p). By Lemma 15, Bm is a p-integer if p − 1 6 |m and pBm ≡ −1 (mod p) if
p − 1|m. Then
X 1
Bm +
p
p−1|m

is a p-integer for all primes p, then it must be integral. ¤

More Congruences

Corollary 16 If p − 1 6 |2m, then B2m is p-integral. If p − 1|2m then pB2m + 1 is p-integral

and µ µ ¶¶
1
ordp (pB2m + 1) = ordp p B2m + ≥1
p
so pB2m ≡ −1 (mod p). Also 6 always divides the denominator of B2m .
Um
From now on write Bm = Vm as a fraction in lowest terms with Vm > 0.

Proposition 17 For m even and > 1,

Vm Sm (n) ≡ Um n (mod n2 )

PROOF. We will use equation (1), for k ≥ 1 write

nk−1
µ ¶µ ¶
m
Bm−k n2 = Am 2
k n .
k k+1

7
If we show that for p|n, p 6= 2, 3, then ordp (Am m
k ) ≥ 0 and if p|n, p = 2 or 3, ordp (Ak ) ≥ −1,
m
then (Ak , n) must divide 6 and the same is true for the greater common divisor between
the sum of Amk and n. Then we may write

An2
Sm (n) = Bm n +
lB
with (B, n) = 1 and l|6. Multiplying by BVm and using the fact that 6|Vm (by Corollary
16) the result is proved.
Use Corollary 16 to see that ordp (Bm−k ) ≥ −1. Assume p|n and p 6= 2, 3. The cases
k = 1, 2 are simple. If k ≥ 3,

nk−1
µ ¶
ordp Bm−k ≥ −1 + (k − 1)ordp n − ordp (k + 1) ≥ k − 2 − ordp (k + 1) ≥ 0
k+1

by Lemma 15.
Now let p = 2. If k = 1, then Bm−1 = 0 for m > 2 and A21 = 2B1 12 = − 12 . For k > 1
note that Bm−k = 0 unless k is even or k = m − 1. k even implies ord2 (k + 1) = 0 and
1 m−2
k = m − 1, Am
m−1 = − 2 n which has order greater or equal to −1.
When p = 3, ord (Am ) ≥ −1 and ord (Am ) ≥ 1. For k ≥ 4, one shows that
³ ´ 3 2 3 3
3k−2
ord3 k+1 ≥ 0. ¤

Corollary 18 Let m be even and p prime with p − 1 6 |m. Then

Sm (p) ≡ Bm p (mod p2 ).

PROOF. By Theorem 11, p 6 |Vm . Now put n = p in the above Proposition and divide
by Vm which is permissible since p 6 |Vm . ¤

Proposition 19 (Voronoi’s congruence) Let m even and > 1. Suppose that a and n are
positive coprime integers. Then
n−1 · ¸
X ja
(am − 1)Um ≡ mam−1 Vm j m−1 (mod n).
n
j=1

PROOF. Write ja = qj n + rj with 0 ≤ rj < n. Then

j m am ≡ rjm + mqj nrjm−1 (mod n2 ).

But rj ≡ ja (mod n), then

j m am ≡ rjm + mam−1 qj nj m−1 (mod n2 ).

Summing for j = 1, . . . , n − 1,
n−1 · ¸
X ja
Sm (n)a m
≡ Sm (n) + ma m−1
n j m−1
(mod n2 ).
n
j=1

Now multiply by Vm and use Proposition 17. ¤

8
 Bm
Proposition 20 If p − 1 6 |m, then m is p-integral.

PROOF. By Theorem 11, Bm is a p-integer. Let m = pt m0 with p 6 |m0 . In Voronoi

congruence put n = pt . Then (am − 1)Um ≡ 0 (mod pt ). Now let a be a primitive root
Um
modulo p. Since p − 1 6 |m, then p 6 |am − 1. Then Um ≡ 0 (mod pt ). Then Bmm = mVm
is
p-integer. ¤

Theorem 21 (Kummer congruences) Suppose m ≥ 2 is even, p prime, and p − 1 6 |m. Let

m−1
Cm = (1−p m )Bm . If m0 ≡ m (mod φ(pe )), then Cm0 ≡ Cm (mod pe ).

PROOF. We will see the case e = 1. Let t = ordp (m). By Proposition 20, pt |Um . In
Voronoi’s congruence, set n = pe+t . Since pt divides both Um and m, and mVpt is prime to
m

p, we obtain,
pe+t−1
(am − 1)Bm
· ¸
m−1
X
m−1 ja
≡a j (mod pe ).
m pe+t
j=1

The right-hand side is unchanged if we replace m by m0 ≡ m (mod p − 1). Then

0
(am − 1)Bm0 (am − 1)Bm
≡ (mod p).
m0 m
0
Choose a to be a primitive root modulo p. Since p − 1 6 |m we have am − 1 ≡ am − 1 6≡
0 (mod p). Then
Bm0 Bm
0
≡ (mod p).
m m
¤

Definition 22 An odd prime number p is said to be regular if p does not divide the nu-
merator of any of the numbers B2 , B4 , . . . , Bp−3 . The prime 3 is regular. Equivalently, p
is regular if it does not divide the class number of Q(ξp )

The first irregular primes are 37 and 59.

Theorem 23 (Kummer) Let p be a regular prime. Then xp + y p = z p has no solution in

positive integers.

Theorem 24 (Jensen) The set of irregular primes is infinite.

PROOF. Let {p1 , . . . , ps } be the set of irregular primes. Let k ≥ 2 be even¡ and¢ n =
k(p1 −1) . . . (ps −1). Choose k large such that n > 1 and p prime such that ordp Bnn > 0.
¯B ¯
¯ n¯

Then p − 1 6 |n and so p 6= pi . We will prove that p is also irregular.

Let n ≡ m (mod p − 1) where 0 < m < p − 1. Then m is even and 2 ≤ m ≤ p − 3. By
the Kummer congruence,
Bn Bm
≡ (mod p).
n m
Since ordp Bnn > 0 and ordp Bnn − Bmm > 0, then
¡ ¢ ¡ ¢

µ ¶
Bm
ordp = ordp Bm > 0
m
and p is irregular. ¤
9
References
[1] D. Castellanos, The Ubiquitous Pi. Part I.Math. Mag. 61, 67–98, 1988.

[2] K. Ireland, M. Rosen, A classical introduction to modern number theory. Second edition.
Graduate Texts in Mathematics, 84. Springer-Verlag, New York, 1990. xiv+389 pp.

[3] M. N. Lalı́n, Mahler measure of some n-variable polynomial families, (preprint, Septem-
ber 2004, to appear in J. Number Theory)

[4] H. Rademacher, Topics in Analytic Number Theory, Die Grundlehren der Mathematis-
chen, Wissenschaften. New York: Springer-Verlag, 1973.

[5] S. Roman, The Bernoulli Polynomials. 4.2.2 The Umbral Calculus. New York: Academic
Press, pp. 93–100, 1984.

[6] Eric Weisstein’s World of Mathematics, http://mathworld.wolfram.com/BernoulliNumber.html

10