11.

Bernoullis Power Sum Problem

Determine the sum

S  1p  2p  3p  C  np
of the p th powers of the first n natural numbers for positive integer exponents p.

The problem, posed in this general form, was first solved in Ars Conjectandi (The Art
of Conjecture an early work on probability) in 1713 by the Swiss mathematician Jacob
Bernoulli (1654-1705).

The following elegant solution is based on the binomial theorem. Let P  p  1. Then

7  SP  7 P  P7 p S  P
2
7 p"1 S 2 . . . and
7  S " 1 P  7 P  P7 p S " 1  P
2
7 p"1 S " 1 2 . . .

and subtraction gives us

(I) 7  S P " 7  S " 1 P  P 7 p  P

2
7 p"1 S 2 " S " 1 2  P
3
7 p"2 S 3 " S " 1 3 . . .

Making the substitution S 7 for S 7 in the equations

(1) S " 12  S2
S " 1 
3
(2) S3
S " 1 
4
(3) S4
etc.

gives us "2S 1  1  0 and S 1  12 , "3S 2  3S 1 " 1  0 and S 2  16 , "4S 3  6S 2 " 4S 1  1  0

and S 3  0. Similarly S 4  "301 . These are the Bernoulli numbers.

Then from (I), we conclude that

P7 p  7  S P " 7  S " 1 P | S 7 S 7
or just

(Ia) P 7 p  7  S P " 7  S " 1 P ,

the substitutions being understood. Now let 7  1, 2, 3, . . . , n to get n equations:

1
 P  1p   SP " SP
1

P  2p  2  S P " 1  S P
B

P  np  n  S P " n " 1  S P ,

addition of which gives us

(II) PS  n  S P " S P

or

nS
P "S P
(II) 1p  2p  3p  C  np  P
| S 7 S 7 .

Using the S 7 values found above, we obtain the following results:

nS
2 "S 2
1  2  3 . . . n  2
| S 7 S 7
n 2 2nS
 2
1

n 2 n
 2
nn1
 2
,

nS
3 "S 3
1 2  2 2  3 2 . . . n 2  3
| S 7 S 7
n 3 3n 2 S 1 3nS 2
 3
2n 3 3n 2 n
 6
nn1 2n1
 6
,

nS
4 "S 4
1 3  2 3  3 3 . . . n 3  4
| S 7 S 7
n 4 4n 3 S 1 6n
2S
2 4nS 3
 4
n 4 2n 3 n 2
 4
nn1 2
 2
,

2
 nS
5 "S 5
1 4  2 4  3 4 . . . n 4  5
| S 7 S 7
n 5 5n 4 S 1 10n 3 S 2 10n 2 S 3 5nS 4
 5
6n 5 15n 4 10n 3 "n
 30
nn1 2n1 3n 2 3n"1
 30
.

PS  n  S P " S P  n P  P
1
S 1 n P"1  P
2
S 2 n P"2 . . . can be rewritten as

P P
S  1  S1 S2
1
 2
. . .
nP P Pn Pn 2

which as n v . appears to approach P1  p11 . [One should know something about the
long-term behavior of S 7 to be sure of this.] This suggests that the average value of
1 p , 2 p , . . . , n p is p11 . One application of the definite integral in an elementary calculus course
is to find the average value of a function. More specifically, the average value of ft on the
interval a, b is b"1a ; ftdt. It follows that the average value of t p from 0 to x is
b

x ; t dt  x p1  p1 , agreeing with the result above, and extending it to any p  0.
a
1 x p 1 x p1 xp
0