Power-sum problem, Bernoulli Numbers and Bernoulli Polynomials.

Arkady M. Alt

Definition 1 (Power Sum Problem) Find the sum S p (n) := 1 p + 2 p + ... + n p where p, n ∈ N (or, using sum notation,
n
S p (n) = k p ) in closed form.
P
k=1

Recurrence for S p (n)

Exercise 2 Using representations 1 = (k + 1) − k, 2k = k (k + 1) − (k − 1) k, 3k = k (k + 1) (k + 2) − (k − 1) k (k + 1)

find S p (n) for p = 1, 2, 3 and n ∈ N.

Exercise 3 By summing differences k2 − (k − 1)2 = 2k − 1, k3 − (k − 1)3 = 3k2 − 3k + 1,

k4 − (k − 1)4 = 4k3 − 6k2 + 4k − 1 for k running from 1 to n find S p (n) for p = 2, 3, 4.

General case

p + 1 p+1−i
p+1
!
Exercise 4 For any p ∈ N by summing differences (k + 1) p+1
k p+1 =
P
− k for k running from 1 to
i=1 i
n prove that

p+1
p−1
!
(n + 1) p+1 P
−n−1− S i (n)
i=1 i
S p (n) = (1)
p+1

p + 1 p+1−i
p+1
!
Exercise 5 For any p ∈ N by summing differences k p+1 − k − 1 p+1 =
P
k for k running from 1 to n prove
i=1 i
that

p+1
p
! !
1
S p (n) = n p+1 + (−1)i+1
P
S p−i (n) (2)
p+1 i=1 i+1
n
Recurrences (1) and (2) give opportunity, starting from S 0 (n) = k0 = n, constructively find representation of
P
k=1
S p (n) as polynomial of n.
Since any polynomial degree of m uniquely defined by their values in m +1 distinct points ((1) or (2) holds for any
natural n) then, by such,
 polynomials
 S p (x) are defined for any x ∈ R and p ∈ N, more precisely, defined sequence of polynomials
S p (x) by recurrence
p∈N
P p+1
p−1
!
(x + 1) p+1 − 1 − S i (x)
i=0 i
S p (x) = (1’)
p+1
(or by recurrence
i+1 p + 1
p
! !
1
S p (x) = p+1 + (−1)
P
x S p−i (x) ) (2’)
p+1 i=1 i+1
with initial condition S 0 (x) = x.

Mathematical Reflections 3 (2018) 1

1 Bernoulli Numbers and Bernoulli Polynomials
Our goal is to solve this recurrence in closed form, that is to find a regular polynomial representation of S p (x).
Since S p (0) = 0 for any p = 0, 1, 2, ... then we should find real numbers s1 , ..., s p+1 such that S p (x) = s1 x +
...s p+1 x p+1 .
Note that the problem would simply be solved if we had known for some polynomial H (x) of degree p + 1 such
that H (x + 1) − H (x) = cx p where c is some constant.
n 1 P n H (n + 1) − H (1)
Then S p (n) = kn = (H (k + 1) − H (k)) = .
P
k=1 c k=1 c
In a sense, we already have one such polynomial (up to an arbitrary constant c), H (x) = S p (x − 1) + c since
H (x + 1) − H (x) = S p (x) − S p (x − 1) = x p
But our problem is that S p (x) is not yet represented in terms of powers of x.
By differentiation of S p+1 (x)−S p+1 (x − 1) = x p+1 we obtain S 0p+1 (x)−S 0p+1 (x − 1) = (p + 1) x p ; then S 0p+1 (x − 1) can be co
sidered as another candidate for the role of H (x) , which does not look better than S p (x − 1) for the same reason.
x (x + 1) x (x + 1) (2x + 1) x2 (x + 1)2
We know that S 0 (x) = x, S 1 (x) = , S 2 (x) = , S 3 (x) =
2 6 4
Applying the recurrences (1)  or (2) we obtain   
x (x + 1) (2x + 1) 3x2 + 3x − 1 x2 (x + 1)2 2x2 + 2x − 1
S4 (x) = and S 5 (x) = .
30 12
1 1 3x2 x
Accordingly, we also have S 00 (x) = 1, S 10 (x) = x + , S 20 (x) = x2 + x + , S 30 (x) = x3 + + ,
2 6 2 2
1 5 4 5 3 x
S 4 (x) = x + 2x + x − , S 5 (x) = x + x + x −
0 4 3 2 0 5
30 2 !3 6
1
S 0 (x) = 0, S 1 (x) = 1, S 2 (x) = 2x + 1 = 2 x +
00 00 00 = 2S 10 (x) ,
2
1  
S 300 (x) = 3x2 + 3x + = 3 x2 + x + 16 = 3S 20 (x) ,
2
3x2 x
!
S 4 (x) = 4x + 6x + 2x = 4 x +
00 3 2 3 + = 4S 30 (x) ,
2 2 !
1 1
S 5 (x) = 5x + 10x + 5x − = 5 x + 2x + x −
00 4 3 2 4 3 2 = 5S 40 (x) .
6 30
The above equations lead to the conclusion that the correlation S 00p (x) = pS 0p−1 (x) holds for any p ∈ N.
In fact, assuming S i00 (x) = pS i−1 0 (x) , i = 1, 2, ..., p − 1,and by differentiating (1’) twice, we obtain

P p+1 0 P p+1 0
p−1
! p−1
!
(p + 1) (x + 1) − p
S i (x) (p + 1) (x + 1) − 1 − p
S i (x)
i=0 i i=1 i
S p (x) =
0 = and
p+1 p+1
P p + 1 00
p−1
!
(p + 1) p (x + 1) p−1 − S i (x)
i=1 i
S p (x) =
00 =
p + 1!
P p+1 0
p−1
(p + 1) p (x + 1) p−1 − iS i−1 (x)
i=1 i
=
p+1 !
p−1
P p
(p + 1) p (x + 1) p−1 − (p + 1) 0 (x)
S i−1
i=1 i − 1
=
p+1 !
p−2
P p 0
p−2
! (x + 1) p−1
− S i−1 (x)
P p 0 i=0 i
p (x + 1) p−1
− S i−1 (x) = p · = pS 0p−1 (x) .
i=0 i p

Exercise 6 Prove that S 00p (x) = pS 0p−1 (x), for any p ∈ N using (2’).

Thus, by induction, S 00p (x) = pS 0p−1 (x) for any p ∈ N.

Coming back to the polynomial S 0p (x − 1), we denote it by B p (x), and then by replacing x with x − 1 in the
recurrence

Mathematical Reflections 3 (2018) 2

 p+1 0
p−1
!
(p + 1) (x + 1) −
p P
S i (x)
i=0 i
S 0p (x) =
p+1
we obtain the following recurrence for polynomials B p (x) , p ∈ N :
+
p
!
p 1
(−1)i+1
P
B p−i (x)
i=1 i+1
B p (x) = (x − 1) +
p
. (B2)
p+1

2 Properties.
P0. deg B p (x) = deg S 0p (x − 1) = p;
P1. B0 (x) = S 10 (x − 1) = 1;
 0
P2. B0p (x) = S 0p (x − 1) = S 00p (x − 1) = pS 0p−1 (x) = pB p−1 (x) ;
P3. B p (x + 1) − B p (x) = S 0p (x) − S 0p (x − 1) = px p−1 , p ∈ N.
We call such polynomials Bernoulli Polynomials.
We already have the first few polynomials B p (x) , namely,
1 1 1 1
B1 (x) = S 10 (x − 1) = x − 1 + = x − , B2 (x) = S 20 (x − 1) = (x − 1)2 + (x − 1) + = x2 − x + ,
2 2 6 6
3 (x − 1)2 (x − 1) 3 2 1
B3 (x) = S 3 (x − 1) = (x − 1) +
0 3
+ = x − x + x;
3
2 2 2 2
1 5x4 5x3 x
B4 (x) = S 40 (x − 1) = x4 − 2x3 + x2 − , B5 (x) = S 50 (x − 1) = x5 − + − .
30 2 3 6
1 1 1
We can see that B1 (0) = − , B1 (1) = , but B2 (0) = B2 (1) = , B3 (0) = B3 (1) = 0, B4 (0) = B4 (1) =
2 2 6
1
− , B5 (0) = B5 (1) = 0
30
and in general B p (0) = B p (1) for any p ≥ 2. Furthermore, B2p+1 (0) = B2p+1 (1) = 0.
Since B p (x + 1) − B p (x) = px p−1 , then for x = 0 we obtain B p (1) − B p (0) = p · 0 p−1 ⇐⇒ B p (1) = B p (0), for
all p ≥ 2.
(Hypothesis B2p+1 (0) = B2p+1 (1) = 0, p ∈ N is equivalent to dividing B2p+1 (x) by x which we will prove later).
Note that the recursion B0p (x) = pB p−1 (x) , p ∈ N with initial condition B0 (x) = 1 allows us to obtain polynomials
B1 (x) , B2 (x) , B3 (x) , ....and thus easier than by recurrence (B1) or (B2). Rx Rx
Indeed, assume that we already know polynomial B p−1 (x) ,then B p (x)−B p (1) = 1 B0p (t) dt = 1 pB p−1 (t) dt ⇐⇒
Rx
B p (x) = B p (0) + p 1 B p−1 (t) dt.
Let B p := B p (0) , p ∈ N ∪ {0} . We call such numbers Bernoulli Numbers.
By replacing x in (B1) or in !(B2) with 0 we obtain
P p+1
p−1
− Bi
i=0 i
Bp = (B3)
p+1
or
i+1 p + 1
p
!
P
(−1) B p−i
i=1 i+1
B p = (−1) p + . (B4)
p+1
Any of these recurrences allows to get consistently numbers B1 , B2 , B3 , ...

Exercise 7 Find the first 5 terms of sequence (B p ) p≥0 .

We show that by Bk , k = 1, 2, .... , we can obtain polynomial B p (x).

Let B p (x) = b p x p + b p−1 x p−1 + ... + b1 x + b0 ,where bk should be determined.
Since B p (0) = B p then b0 = B p . Also since B0p (x) = pB p−1 (x) then B(k) p (x) = p (p − 1) ... (p − k + 1) B p−k (x) and
 (k)  (k)
B(k)
p (x) = b p x + b p−1 x
p p−1 + ... + b x + b
1 0 = b p x p + b p−1 x p−1 + ... + bk+1 xk + bk k! yields
p − 1) ... (p − k + 1)
(p
B(k)
p (0) = bk k! ⇐⇒ p (p − 1) ... (p − k + 1) B p−k (0) = bk k! ⇐⇒ bk = B p−k ⇐⇒
k!

Mathematical Reflections 3 (2018) 3

 !
p
bk = B p−k , k = 1, 2, ..., p.
k ! ! !
p p p p
Hence, B p (x) = B p + B p−1 x + ... +
1 p−1 + B0 x =
p B p−k xk .
P
B1 x
1 p−1 k=0 k
1 1
In particular B0 (x) = x, B1 (x) = x − , B2 (x) = x2 − x + ,
2 ! 6
1 2 1 3 1
B3 (x) = B0 x + 3B1 x + 3B2 x + B3 = x + 3 − x + 3 · x = x3 − x2 + x.
3 2 3
2 6 2 2
More properties of Bernoulli polynomials and numbers.
R1
P4. 0
B p (x) dx = 0 for any p ∈ N.
Proof. Because of P2. we have B0p+1 (x) = (p + 1) B p (x) then
R1 R1 0  1  
(p + 1) 0 B p (x) dx = (p + 1) 0 B p+1 (x) dx = (p + 1) B p+1 (x) = (p + 1) B p+1 (1) − B p+1 (0) = (p + 1) · 0 =
R1 0
0 =⇒ 0 B p (x) dx = 0.
We will
 prove  that properties P1.,P2.,P3. determine polynomials B p (x) uniquely.
Let Q p (x) be a sequence of polynomials such that Q0 (x) = 1, Q0n (x) = nQn−1 (x) , n ∈ N and Q p (x + 1) −
p≥0
Q p (x + 1) = px p−1 , p ∈ N.
First note that Q0 (x) = 1 = B0 (x) .Also, Qn (1) = Qn (0) for n ≥ 2 since Q p (1) − Q p (0) = p · 0 p−1 = 0, p ≥ 2.
R1
This yields 0 Q p (x) dx = 0, p ∈ N.
R1 R1
Indeed, p 0 Q p (x) dx = 0 Q0p+1 (x) dx = Qn+1 (1) − Qn+1 (0) = 0. Since Q01 (x) = 1 · Q0 (x) = 1 then Q1 (x) =
x + c and, therefore, Q02 (x) = 2Q1 (x)
yields Q2 (x) = x2 + 2cx + d. Then Q2 (x + 1) − Q2 (x) = 2x ⇐⇒ (x + 1)2 + 2c (x + 1) − x2 − 2cx = 2x ⇐⇒
1 1
2c + 1 = 0 ⇐⇒ c = − .Hence, Q1 (x) = x − = B1 (x)
2 2
Assume that Q p (x) = B p (x) then Q0p+1 (x) = (p + 1) Q p (x) = (p + 1) B p (x) = B0p+1 (x) ⇐⇒
R1 R1  R1
Q p+1 (x) = B p+1 (x) + [Link] 0 = 0 Q p+1 (x) dx = 0 B p+1 (x) + c dx = 0 B p+1 (x) dx + c = c.
So, by induction Q p (x) = B p (x) for any p ∈ N.
P5. B p (x) = (−1) p B p (1 − x) , p ≥ 0.(Complement property)
Proof. Let Q p (x) := (−1) p B p (1 − x) , p ∈ N ∪ {0} then:
1. By P1 Q0 (x) = B0 (1 − x) = 1; 0  0
2. By P2. Q0p (x) = (−1) p B p (1 − x) = (−1) p B p (1 − x) = − (−1) p B0p (1 − x) = p (−1) p−1 B p−1 (1 − x) =
pQ p−1 (x) ;
3. By P3. Q p (x + 1) − Q p (x) = p
 (−1) B p (1  − (x + 1)) − (−1) B p (1
p
 − x) =
(−1) B p (−x) − B p (1 + (−x)) = (−1)
p p+1
B p ((−x) + 1) − B p (−x) = p (−1) p+1 (−x) p−1 = px p−1 .
Therefore, by property of uniqueness we get (−1) p B p (1 − x) = B p (x) .

Corollary 8 For p = 2m + 1, m ∈ N holds B p (0) = 0.

Indeed, if p = 2m + 1 then B p (x) = −B p (1 − x) and, therefore, for x = 0 we have B p (0) = −B p (1) = −B p (0) =⇒
2B p (0) = 0 ⇐⇒ B p (0) = 0.

Corollary 9 By replacing x in B p (x) = (−1) p B p (1 − x) with x + 1 we obtain

p p
!
B p (x + 1) = (−1) p B p (1 − (x + 1)) = (−1) p B p (−x) = (−1) p B p−k (−x)k =
P
! ! k=0 k
p p
n−k p k p
B p−k x =
k B p−k x .
k
P P
(−1) (−1)
k=0 k k=0 k
Now, we write S p (n) in polynomial form by powers of n.
n
Since B p+1 (x + 1) − B p+1 (x) = (p + 1) x p then (p + 1) S p (n) = (p + 1) kp =
P
k=1
n  
B p+1 (k + 1) − B p+1 (k) = B p+1 (n + 1) − B p+1 (1) = B p+1 (n + 1) − B p+1 (0)
P
k=1
and, therefore, (p + 1) S p (n) = B p+1 (n + 1) − B p+1 ⇐⇒

Mathematical Reflections 3 (2018) 4

 B p+1 (n + 1) − B p+1 k p+1
p+1
! !
1
S p (n) = = B p+1−k n − B p+1 =
k
P
(−1)
p + 1! p + 1 k=0 k
1 p+1 p+1
(−1)k B p+1−k nk .
P
p + 1 k=1 k
k p+1
!
1 p+1
S p (n) = B p+1−k nk (Faulhaber’s Formula).
P
(F) (−1)
p + 1 k=1 k
Problem 1
Prove that B2m+1 (x) is divisible by S 2 (x − 1) for any m ∈ N.
Problem 2
Prove that S ign (B2m ) = (−1)m+1 and max B4m−2 (x) = B4m−2 , min B4m (x) = B4m , m ∈ N.
[0,1] [0,1]
Hint (use induction).

1. A. M. Alt-Variations on a theme-The sum of equal powers of natural numbers,

part 1 Crux vol.40,n.8;
2. A. M. Alt-Variations on a theme-The sum of equal powers of natural numbers,
part 2 Crux vol.40,n.10.
.

Mathematical Reflections 3 (2018) 5