There are two kinds of power sums commonly considered. The first is the sum of th powers of a set of
variables
,
(1)
|
and the second is the special case , i.e.,
(2)
|
General power sums arise commonly in statistics. For example, k-statistics are most commonly defined in terms of power sums. Power sums are related to symmetric polynomials by the Newton-Girard formulas.
The sum of
times the
th
power of
is given analytically by
(3)
|
Other analytic sums include
(4)
| |||
(5)
| |||
(6)
|
for , where
is a Pochhammer symbol.
The finite version has the elegant closed form
(7)
|
for and 2. An additional sum is given
by
(8)
|
An analytic solution for a sum of powers of integers is
(9)
| |||
(10)
| |||
(11)
|
where is the Riemann
zeta function,
is the Hurwitz zeta function, and
is a generalized harmonic
number. For the special case of
a positive integer, Faulhaber's formula gives the sum
explicitly as
(12)
|
where is the Kronecker
delta,
is a binomial coefficient, and
is a Bernoulli number.
It is also true that the coefficients of the terms
in such an expansion sum to 1, as stated by Bernoulli (Boyer 1943).
Bernoulli used the property of the figurate number triangle that
(13)
|
along with a form for
which he derived inductively to compute the sums up to
(Boyer 1968, p. 85). For
, the sum is given by
(14)
|
where the notation means the quantity in question is raised to the appropriate
power
, and all terms of the form
are replaced with the corresponding
Bernoulli numbers
.
Written explicitly in terms of a sum of powers,
(15)
| |||
(16)
| |||
(17)
|
where
(18)
|
It is also true that the coefficients of the terms sum to 1,
(19)
|
which Bernoulli stated without proof.
A double series solution for is given by
(20)
|
Computing the sums for ,
..., 10 gives
(21)
| |||
(22)
| |||
(23)
| |||
(24)
| |||
(25)
| |||
(26)
| |||
(27)
| |||
(28)
| |||
(29)
| |||
(30)
|
or in factored form,
(31)
| |||
(32)
| |||
(33)
| |||
(34)
| |||
(35)
| |||
(36)
| |||
(37)
| |||
(38)
| |||
(39)
| |||
(40)
|
A simple graphical proof of the special case of can also be given by constructing a sequence
of stacks of boxes, each 1 unit across and
units high, where
, 2, ...,
. Now add a rotated copy on top, as in the above figure. Note
that the resulting figure has width
and height
, and so has area
. The desired sum is half this, so the area
of the boxes in the sum is
.
Since the boxes are of unit width, this is also the value of the sum.
The sum
can also be computed using the first Euler-Maclaurin
integration formula
(41)
|
with . Then
(42)
| |||
(43)
| |||
(44)
|
The surprising identity
(45)
| |||
(46)
|
known as Nicomachus's theorem, can also be illustrated graphically (Wells 1991, pp. 198-199).
Schultz (1980) showed that the sum can be found by writing
(47)
|
and solving the system of
equations
(48)
|
obtained for ,
1, ...,
(Guo and Qi 1999), where
is the Kronecker delta. For example, the three
equations to be solved for
are
(49)
| |||
(50)
| |||
(51)
|
giving ,
, and
, or
(52)
|
as expected.
is related to the binomial
theorem by
(53)
|
(Guo and Qi 1999).