Sunteți pe pagina 1din 6

# A DECOMPOSITION OF RIEMANN'S ZETA FUNCTION

Andrew Granville

## It is currently very much in vogue to study sums of the form

(p1 ; p2; : : :; pg ) :=

1 1 ::: 1
p1 p2
a
apgg
a1 >a2 >>ag 1 1 a2

where all the ai s and pi s are positive integers, with p1  2. Note that it is necessary
that p1  2 else (p) diverges. These sums are related to polylogarithm functions
(see [1,2,3]) as well as to zeta functions (the Riemann zeta function is of course the
case g = 1). In this note we prove an identity that was conjectured by Moen 
and Markett :
Proposition. If g and N are positive integers with N  g + 1 then
(1)

(N) =

p1 +p2 ++pg =N
Each pj 1; and p1 2

(p1 ; p2; : : :; pg ):

This identity was proved for g = 2 by Euler, and for g = 3 by Ho man and
Moen . The above proposition has been proved independently by Zagier , who
writes of his proof, `Although this proof is not very long, it seems too complicated
compared with the elegance of the statement. It would be nice to nd a more
natural proof': Unfortunately much the same can be said of the proof that I have
presented here.
Markett  and J. Borwein and Girgensohn  were able to evaluate (p1 ; p2; p3)
in terms of values of (p) whenever p1 + p2 + p3  6, and in terms of (p) and
(a; b) whenever p1 + p2 + p3  10 | it would be interesting to know whether such
`descents' are always possible or, as most researchers seem to believe, that there is
only a small class of such sums that can be so evaluated.
Proof of (1). We may re-write the sum on the right side of (1) as

1 1 ::: 1 :
p1 p2
apgg
a1 >a2 >>ag 1 p1 +p2 ++pg =N a1 a2
Each pj 1; and p1 2

## The author is a PresidentialFaculty Fellow, supportedin part by the National ScienceFoundation.

Typeset by AMS-TEX
1

ANDREW GRANVILLE

## The second sum here is the coecient of xn in the power series

0  p 1 g 0  p 1
g
1
j
@ X x A Y @ X x A = x2=a21 Y x=aj
p1 2

a1

(1 ? x=a1) j =2 (1 ? x=aj )

j =2 pj 1 aj
g
xg+1 Y

g
g
1 = xg+1 X
1 Y
1
= a
a1 j =1 (aj ? x) i=1 (ai ? aj ) :
1 j =1 (aj ? x)
i6=j

## Therefore the sum above is

(2)
g
g
g
X
1 Y
1 =X 1 X
1X
A(m; j ?1)(?1)g?j B(m; g?j)
N ?g
N ?g
a
(a
?
a
)
m
1
i
j
a
j =1
m1
a1 >a2 >>ag 1 j =1 j i=1
i6=j

## where we take each aj = m in turn, with

jY
?1
X
1
1
1 =
A(m; j ?1) :=
a
(a
?
m)
(b
+
m)b
1 b2 : : :bj ?1
a1 >a2 >>aj?1 >m 1 i=1 i
b1 >b2>>bj?1 1 1

B(m; g?j) :=

Yg

(m ? a ) =
i

(3)

i0

B(m; i)xi =

mY
?1 
b=1

1 + xb :

## Dealing with A is somewhat more dicult. We start by noting that

 1X
m
1 X 1 ? 1
1 ;
1
=
=
(b
+
m)b
m
b
b
+
m
m
b
+
1
1
c2 =1 2 c2
b1 >b2 1
b1 >b2 1
as this is a telescoping sum. Substituting this back into the de nition for A we
next have to deal with
 X
c2
m
m 1 X
m 1 X 1
XX
X
1
1 ;
1
=
=
?
(b
+
c
)b
c
b
(b
+
c
)
c
b
+
c3
2
2
2
2
2
2
2
2
3
c2 =1 c3 =1
c2 =1 b2>b3
b2 >b3 c2 =1
X

for the same reasons. Putting this back into the de nition we have to do the same
calculation again, now with the indices moved up one. Iterating this procedure we
end up with
X
1 ;
A(m; j ? 1) := m1
c
c
: : :cj
2
3
mc2 c3 cj 1

## which has generating function

m 
m 
?1
 x 2  x 3  1 Y
Y
X
x
1
i
1 ? xc :
1+ c + c + c +::: = m
(4)
A(m; i)x = m
c=1
c=1
i0

Therefore, by (2), (3) and (4), the sum on the right side of (1) is m1 1=mN ?g+1
times the coecient of xg?1 in the power series
m 
Y
?1 
?1 X  x i
1 ? xc = 1 ? mx
=
1 ? xb
m :

mY
?1 
b=1

P
We thus get

c=1

m1 1=m

i0

## = (N), giving (1).

2. Evaluations of (r; s).

## Euler demonstrated that if N = r + s is odd wih s even then

N  
X a ? 1 a ? 1
1
(5) (r; s) = ? 2 r + 1 (N) +
s ? 1 + t ? 1 (a)(b):
a+b=N
a;b2
a odd

One can then obtain the value of (s; r), provided r > 1, from the trivial identity
(6)
(r; s) + (s; r) = (r)(s) ? (r + s):
(To prove this just write out the zeta-functions on both sides and compare terms).
In the special case r = 1 he proved, for any N  3 that
X
(a)(b):
(7)
(N ? 1; 1) = N 2? 1 (N) ? 21
a+b=N
a;b2

Pages 47-49 of , Equation (2) of  and Theorem 4.1 of  are all equivalent
to, for N  4,
? 2) (N) + 1 (2)(N ? 2) ? N ? 2 X (a)(b)
(N ? 2; 1; 1) = (N ? 1)(N
6
2
4 a+b=N

X
+ 61
(a)(b)(c):

(8)
Proof of (7).

a;b2

a+b+c=N
a;b;c2

a+b=N
a;b2

(a)(b) =

a+b=N
a;b2

=2

## ((a; b) + (b; a) + (N))

a+b=N
a;b2

(a; b) + (N ? 3)(N)

## = 2((N) ? (N ? 1; 1)) + (N ? 3)(N)

ANDREW GRANVILLE

using (1) with g = 2, and the result follows after some re-arrangement.
Proof of (8). We begin by proving

(9)

p+q=N ?1
p2; q1

## Now the sum here equals

1 ? 1
1
1 = X
p
q
N
?3 aN ?3
a>b>c1 ab(a ? c) c
a>b>c1 b p+q=N ?1 a c
p2; q1

X 1 X

## The rst term is

c?1 1
X 1 X 1 X 1 1 X 1 X 1 X
=
?
N ?2
N ?2
c1 c
c1 c
b>c b a>b a ? c a
b>c b i=0 b ? i

X 1 1 X 1 1

= (2; N ? 2) +

N ?2 i
b>c
c>i1 c

= (2; N ? 2) + (N ? 1; 1) +

b?i ? b

i?1 1
X 1 1X

N ?2 i
j =1 c ? j
c>i1 c

But the nal sum in both of the last two displays are identical (after the change of
variables (a; b; c) ! (c; i; j)), and so we have proved (9).
We next prove that
(10)

p+q=N ?1
p2; q1

## Using (7) and then (1) we have

(N ? 1)(N) ? 2(N ? 1; 1) ? (2)(N ? 2) =
=

X
N ?3a2

(a)(N ? a)

a+b+c=N
a;b2; c1

(a)(b; c)

Just as in the proof of (6) we may determine such a product in terms of zetafunctions by considering each term. We thus get (a)(b; c) = (a; b; c)+(b; a; c)+
(b; c; a)+(a+b; c)+(b; a+c). Summing up over all possibilities with a+b+c = N
and a; b  2 we get three times the sum over all (A; B; C) in the sum (1) other
than a few terms corresponding to when a = 1 or b = 1, and some multiples of

## (A; B). Precisely we get:

X
X
3
(a; b; c) ?
(2(a; 1; c) + (a; c; 1))
a+b+c=N
a;b2; c1

= 3(N) ?

a+c=N ?1
a2; c1

d+c=N
d3; c1

a+c=N ?1
a2; c1

= (N ? 2)(N) ?

(d ? 3)(d; c) +

(a; c; 1) +

a+c=N ?1
a2; c1

f +b=N
b2; f 2

(f ? 2)(b; f)

a+b=N
a2; b1

## (a; c; 1) ? (2; N ? 2) ? (N ? 1; 1)

using (1) and (9). Combining the last two displays gives (10).
We now try to evaluate the last sum in (8):
X
X X 1
(a)(b)(c) =
a b c:
a+b+c=N
a+b+c=N x;y;z1 x y z
a;b;c2

a;b;c2

By analogy with the proof of (6), with a; b; c xed we break up this sum according
to how x; y; z are ordered by size. For example, when x > y > z we get precisely
(a; b; c). We thus get the sum of (u; v; w) as u; v; w ranges over all six orderings
of a; b; c; plus the sum of (u; N ? u)+(N ? u; u) for each u 2 fa; b; cg; plus (N).
Thus, using (6), we have
X
X
X
6
(a; b; c) + 3
((c)(a + b) ? (N)) +
(N):
a+b+c=N
a;b;c2

a+b+c=N
a;b;c2

a+b+c=N
a;b;c2

## Using (1) with g = 3, we thus have

0
1
B(N) + (N ? 2; 1; 1) ? X ((a; d; 1) + (a; 1; d))CC
6B
@
A
a+d+1=N
a2; d1
N ? 4
X
+3

(d ? 3)(c)(d) ? 2

c+d=N
d4; c2

(N)

## = 6(N ? 2; 1; 1) ? (N ? 1)(N ? 8)(N) ? 3(N ? 2)(2) ? 12(N ? 1; 1)

X
(N ? 6)(c)(d)
? 23
c+d=N
d2; c2

using (9) and (10), and combining the (c; d) and (d; c) terms in the nal sum. Using
(7) to remove the (N ? 1; 1) terms, we obtain (8).
Acknowledgements: Thanks are due to Roland Girgensohn for supplying the
reference , to Don Zagier and Michael Ho man for their useful email correspondence, and to the authors of [1,2,4,6] for making available their preprints.

ANDREW GRANVILLE
1.
2.
3.
4.
5.
6.
7.
8.
9.

References
D.H. Bailey, J.M. Borwein & R. Girgensohn, Experimental evaluation of Euler sums, Experimental Math 3 (1994), 17-30.
D. Borwein, J.M. Borwein & R. Girgensohn, Explicit evaluation of Euler sums, Proc. Edin.
Math. Soc. 38 (1995), 273-294.
J.M. Borwein & R. Girgensohn, Evaluating triple Euler sums, Electronic J. Combinatorics 3
(1996), 27pp.
J.P. Buhler & R.E. Crandall, On the evaluation of Euler Sums, Experimental Math 3 (1994),
275-285.
M.E. Ho man, Multiple harmonic series, Paci c J. Math 152 (1992), 275-290..
M.E. Ho man and C. Moen, Sums of triple harmonic series, J. Number theory 60 (1996),
329-331..
C. Markett, Triple Sums and the Riemann Zeta Function, J. Number theory 48 (1994),
113-132.
N. Nielsen, Handbuch der Theorie der Gammafunktion, Chelsea, New York, 1965.
D. Zagier, Multiple zeta values (in preparation).
Department of Mathematics, The University of Georgia, Athens, Georgia 30602