You are on page 1of 10

Journal of Number Theory 132 (2012) 314–323

Contents lists available at ScienceDirect

Journal of Number Theory


www.elsevier.com/locate/jnt

Some identities for multiple zeta values


Zhongyan Shen a,b,1 , Tianxin Cai a,∗,2
a
Department of Mathematics, Zhejiang University, Hangzhou 310027, PR China
b
Department of Mathematics, Zhejiang International Study University, Hangzhou 310012, PR China

a r t i c l e i n f o a b s t r a c t

Article history: In this note, we obtain the following identities,


Received 6 December 2010
Revised 13 June 2011  5 1
Accepted 22 June 2011 ζ (2a, 2b, 2c ) = ζ (2n) − ζ (2)ζ (2n − 2), for n > 2,
Available online 4 October 2011 8 4
a+b+c =n
Communicated by Wenzhi Luo
 35 5
Keywords: ζ (2a, 2b, 2c , 2d) = ζ (2n) − ζ (2)ζ (2n − 2),
64 16
Multiple zeta values a+b+c +d=n
Harmonic shuffle relation
for n > 3.
Bernoulli numbers

Meanwhile, some weighted version of sum formulas are also


obtained.
© 2011 Elsevier Inc. All rights reserved.

1. Introduction

The multiple zeta sums

 1
ζ (s1 , s2 , . . . , sk ) = s s s
n 1 n 2 · · · nkk
n1 >n2 >···>nk >0 1 2

are also called Euler–Zagier sums, which have attracted considerable interest in recent times. The
study of such sums is not only important to general zeta function theory, but also touches upon such

*
Corresponding author.
E-mail addresses: huanchenszyan@yahoo.com.cn (Z. Shen), caitianxin@hotmail.com (T. Cai).
1
Research of the author supported by Educational Commission of Zhejiang Province of China, Project Y200803340 and
Foundation of Zhejiang International Study University, Project 09Z08.
2
Research of the author supported by the National Natural Science Foundation of China, Project 10871169.

0022-314X/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jnt.2011.06.011
Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323 315

domains as arithmetic geometry, Galois representations, invariants for knots, quantum groups, etc.
In [3], H. Gangl, M. Kaneko and D. Zagier proved the following results,

n −1
 3
ζ (2m, 2n − 2m) = ζ (2n), for n > 1, (1)
4
m =1


n −1
1
ζ (2m + 1, 2n − 2m − 1) = ζ (2n), for n > 1. (2)
4
m =1

Recently, Takashi Nakamura [6] gave a new proof of (1). In this paper, we consider the sums in three
and four variables, by using harmonic shuffle relations, we prove the following theorems.

Theorem 1. For n > 2, we have

 5 1
ζ (2a, 2b, 2c ) = ζ (2n) − ζ (2)ζ (2n − 2).
8 4
a+b+c =n
a,b,c 1

Theorem 2. For n > 3, we have

 35 5
ζ (2a, 2b, 2c , 2d) = ζ (2n) − ζ (2)ζ (2n − 2).
64 16
a+b+c +d=n
a,b,c ,d1

In 2008, Ohno and Zudilin [7] proved a weighted form of Euler’s sums formula (the weight Euler
sum formula)

n −1

2i ζ (i , n − i ) = (n + 1)ζ (n), n  2,
i =2

and applied it to study multiple zeta star values. One year later, Guo and Xie [5] generalized the
weight Euler sum formula of Ohno and Zudilin to higher dimensions. For positive integers k  2 and
n  k + 1, they proved

  k−1  
   
s 1 −1 s 1 −1 S i −s1 −(i −1) S k−1 −s1 −(k−2)
2 + 2 −1 2 +2 ζ (s1 , . . . , sk ) = nζ (n),
s1 2,si 1 i =2
s1 +···+sk =n

where S i = s1 + · · · + si for i = 1, . . . , k − 1.

In this paper, we prove the following weighted version of sum formula.

Theorem 3. For n > 1, we have


2n −1
1
(−1)a ζ (a, 2n − a) = ζ (2n). (3)
2
a =2
316 Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323

For n > 2, we have


3
(−1)a + (−1)b + (−1)c ζ (a, b, c ) = ζ (2n) − ζ (2)ζ (2n − 2). (4)
2
a+b+c =2n
a2,b,c 1

2. Preliminaries

Throughout this paper, let a, b, c, d be positive integers, B k is the kth Bernoulli number.

Lemma 1. (See Nakamura [6].) For n > 1, we have

n −1
 2n + 1
ζ (2m)ζ (2n − 2m) = ζ (2n).
2
m =1

Lemma 2. (See Granville [4].) For any integers n > 2 and 1  k < n,


ζ (a1 , a2 , . . . , ak ) = ζ (n).
a1 +a2 +···+ak =n
a1 >1,a2 ,...,ak 1

Lemma 3. (See Eie [2].) For n > 2, we have


2n n(2n − 1)
B 2a B 2b B 2c = (2n + 1)(n + 1) B 2n + B 2n−2 . (5)
2a, 2b, 2c 2
a+b+c =n
a,b,c 1

For n > 3, we have


2n
B 2a B 2b B 2c B 2d
2a, 2b, 2c , 2d
a+b+c +d=n
a,b,c ,d1

(2n + 1)(2n + 2)(2n + 3) 4n2 (2n − 1)
=− B 2n + B 2n−2 . (6)
6 3

Lemma 4. (See Chen and Sun [1].) Let n  0 be integer, then

n
 n n−1 n+3
B 1+i B 1+n−i = B n − B n +1 − B n +2 .
i 6 6
i =0

(Take k = m = 1 in [1, Theorem 4.4].)

Lemma 5. For n > 1, we have


n −1
n 2n − 3
m(n − m)ζ (2m, 2n − 2m) = ζ (2n) + ζ (2)ζ (2n − 2).
16 4
m =1
Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323 317

Proof. By Lemma 4, when 2n − 2 > 0, we have


2n −2
2n − 2 2n − 3 2n + 1
B 1+i B 2n−1−i = B 2n−2 − B 2n . (7)
i 6 6
i =0

Let 1 + i = j, we obtain


2n −2 
2n −1
2n − 2 2n − 2
B 1+i B 2n−1−i = B j B 2n− j . (8)
i j−1
i =0 j =1

When n  2,


2n −1 n −1

2n − 2 2n − 2
B j B 2n− j = B 2m B 2n−2m . (9)
j−1 2m − 1
j =1 m =1

Using the well-known formula of Euler,

(2π )2n B 2n (−1)n+1


ζ (2n) = , for n > 0, (10)
2(2n)!

we have

n −1

2n − 2
B 2m B 2n−2m
2m − 1
m =1

n −1
 (2n − 2)! 2(2m)!ζ (2m)(−1)m+1 2(2n − 2m)!ζ (2n − 2m)(−1)n−m+1
=
(2m − 1)!(2n − 2m − 1)! (2π )2m (2π )2n−2m
m =1

n −1
 (2n − 2)!(2m)(2n − 2m)(−1)n
= ζ (2m)ζ (2n − 2m), (11)
(2π )2n
m =1

and

2n − 3 2n + 1 (2n − 2)!(2n − 3)(−1)n


B 2n−2 − B 2n = ζ (2n − 2)
6 6 3(2π )2n−2
(2n)!(2n + 1)(−1)n+1
− ζ (2n). (12)
3(2π )2n

From (7) to (9) and (11), (12), we obtain


n −1
π 2 (2n − 3) n(2n − 1)(2n + 1)
(2m)(2n − 2m)ζ (2m)ζ (2n − 2m) = ζ (2n − 2) + ζ (2n). (13)
3 6
m =1

By the harmonic shuffle relation, we find that

ζ (2m)ζ (2n − 2m) = ζ (2m, 2n − 2m) + ζ (2n − 2m, 2m) + ζ (2n). (14)
318 Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323

Multiply (14) on both sides by (2m)(2n − 2m), and summing over m from 1 to n − 1, we have

π 2 (2n − 3) n(2n − 1)(2n + 1)


ζ (2n − 2) + ζ (2n)
3 6
n −1
 2n(n2 − 1)
=2 (2m)(2n − 2m)ζ (2m, 2n − 2m) + ζ (2n). (15)
3
m =1

2
Combining (13) and (15) and noting that ζ (2) = π6 , we prove Lemma 5. 2

3. Proofs of the theorems

Proof of Theorem 1. By (10), we have


2n
B 2a B 2b B 2c
2a, 2b, 2c
a+b+c =n
a,b,c 1

 (2n)! 2(2a)!ζ (2a)(−1)a+1 2(2b)!ζ (2b)(−1)b+1 2(2c )!ζ (2c )(−1)c +1


=
(2a)!(2b)!(2c )! (2π )2a (2π )2b (2π )2c
a+b+c =n
a,b,c 1

 8(2n)!(−1)n+3
= ζ (2a)ζ (2b)ζ (2c ), (16)
(2π )2n
a+b+c =n
a,b,c 1

and

n(2n − 1) 2(2n − 2)!n(2n − 1)(−1)n


B 2n−2 + (n + 1)(2n + 1) B 2n = ζ (2n − 2)
6 (2π )2n−2
2(2n)!(n + 1)(2n + 1)(−1)n+1
− ζ (2n). (17)
(2π )2n

Combining (5), (16) and (17), we have

 (n + 1)(2n + 1) π2
ζ (2a)ζ (2b)ζ (2c ) = ζ (2n) − ζ (2n − 2). (18)
4 4
a+b+c =n
a,b,c 1

Again by the harmonic shuffle relation, we find that

 
ζ (2a)ζ (2b)ζ (2c ) = 6 ζ (2a, 2b, 2c )
a+b+c =n a+b+c =n
a,b,c 1 a,b,c 1
  
+3 ζ (2a, 2b + 2c ) + ζ (2b + 2c , 2a)
a+b+c =n
a,b,c 1

+ ζ (2n), (19)
a+b+c =n
a,b,c 1
Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323 319

and

  
ζ (2a, 2b + 2c ) + ζ (2b + 2c , 2a)
a+b+c =n
a,b,c 1

n −1 n −
 a −1
 
= ζ (2a, 2n − 2a) + ζ (2n − 2a, 2a)
a =1 b =1

n −1
  
= (n − a − 1) ζ (2a, 2n − 2a) + ζ (2n − 2a, 2a) . (20)
a =1

Using the shift from a to n − a, (20) is equal to

n −1
  
(a − 1) ζ (2n − 2a, 2a) + ζ (2a, 2n − 2a) , (21)
a =1

adding (20) and (21), divided by 2 and from (1), (20) equals to

n −1
 3(n − 2)
(n − 2) ζ (2a, 2n − 2a) = ζ (2n). (22)
4
a =1

It’s easy to see that

 (n − 1)(n − 2)
ζ (2n) = ζ (2n). (23)
2
a+b+c =n
a,b,c 1

Combining from (18) to (20), (22) and (23), we have

 5 π2
ζ (2a, 2b, 2c ) = ζ (2n) − ζ (2n − 2).
8 24
a+b+c =n
a,b,c 1

2
Noting that ζ (2) = π6 , this completes the proof of Theorem 1. 2

Proof of Theorem 2. By (10), we have


2n
B 2a B 2b B 2c B 2d
2a, 2b, 2c , 2d
a+b+c +d=n
a,b,c ,d1

 (2n)! 2(2a)!ζ (2a)(−1)a+1 2(2b)!ζ (2b)(−1)b+1


=
(2a)!(2b)!(2c )!(2d)! (2π )2a (2π )2b
a+b+c +d=n
a,b,c ,d1

2(2c )!ζ (2c )(−1)c +1 2(2d)!ζ (2d)(−1)d+1


×
(2π )2c (2π )2d
320 Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323

 16(2n)!(−1)n
= ζ (2a)ζ (2b)ζ (2c )ζ (2d), (24)
(2π )2n
a+b+c +d=n
a,b,c ,d1

and

(2n + 1)(2n + 2)(2n + 3) 4n2 (2n − 1)


B 2n + B 2n−2
6 3
2(2n)!(2n + 1)(2n + 2)(2n + 3)(−1)n+1
= ζ (2n)
6(2π )2n
2(2n − 2)!4n2 (2n − 1)(−1)n
+ ζ (2n − 2). (25)
3(2π )2n−2

Combining (6), (24) and (25), we have

 (2n + 1)(n + 1)(2n + 3) nπ 2


ζ (2a)ζ (2b)ζ (2c )ζ (2d) = ζ (2n) − ζ (2n − 2). (26)
48 3
a+b+c +d=n
a,b,c ,d1

Using harmonic shuffle relation, we find that


ζ (2a)ζ (2b)ζ (2c )ζ (2d)
a+b+c +d=n
a,b,c ,d1

= 24 ζ (2a, 2b, 2c , 2d)
a+b+c +d=n
a,b,c ,d1
 
+6 ζ (2a + 2b, 2c , 2d) + ζ (2a + 2b, 2d, 2c ) + ζ (2c , 2a + 2b, 2d)
a+b+c +d=n
a,b,c ,d1

+ ζ (2d, 2a + 2b, 2c ) + ζ (2c , 2d, 2a + 2b) + ζ (2d, 2c , 2a + 2b)
  
+6 ζ (2a + 2b, 2c + 2d) + 4 ζ (2a + 2b + 2c , 2d)
a+b+c +d=n a+b+c +d=n
a,b,c ,d1 a,b,c ,d1
 
+ ζ (2d, 2a + 2b + 2c ) + ζ (2n). (27)
a+b+c +d=n
a,b,c ,d1

Obviously,

 (n − 1)(n − 2)(n − 3)
ζ (2n) = ζ (2n). (28)
6
a+b+c +d=n
a,b,c ,d1

Now we consider the three sums in the middle of (27), respectively. Firstly, from (1) and Lemma 5,
we have
Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323 321

  
ζ (2a + 2b + 2c , 2d) + ζ (2d, 2a + 2b + 2c )
a+b+c +d=n
a,b,c ,d1


n −1 n −
d−1 n−
a−d−1
 
= ζ (2n − 2d, 2d) + ζ (2d, 2n − 2d)
d =1 a =1 b =1


n −1
(n − d − 1)(n − d − 2)  
= ζ (2n − 2d, 2d) + ζ (2d, 2n − 2d)
2
d =1

n −1 n −1
2(n − 1)(n − 2)  n−3  
= ζ (2n − 2d, 2d) − d ζ (2n − 2d, 2d)
2 2
d =1 d =1

n −1
 
+ ζ (2d, 2n − 2d) − d(n − d)ζ (2n − 2d, 2d)
d =1

n −1
3 n−3   
= (n − 1)(n − 2) ζ (2n) − d ζ (2n − 2d, 2d) + ζ (2d, 2n − 2d)
4 2
d =1

n 2n − 3
− ζ (2n) − ζ (2)ζ (2n − 2). (29)
16 4
n−1 3n
Similarly as (20), we can treat d=1 d(ζ (2n − 2d, 2d)+ζ (2d, 2n − 2d)), it is equal to 4
ζ (2n), therefore,
(29) is equal to

6n2 − 19n + 24 2n − 3
ζ (2n) − ζ (2)ζ (2n − 2). (30)
16 4

Secondly, from (1) and Lemma 5, we obtain


ζ (2a + 2b, 2c + 2d)
a+b+c +d=n
a,b,c ,d1


n −1 m
 −1 n −
m −1
= ζ (2m, 2n − 2m)
m =1 a =1 c =1

n −1

= (m − 1)(n − m − 1)ζ (2m, 2n − 2m)
m =1


n −1 
n −1
= m(n − m)ζ (2m, 2n − 2m) − (n − 1) ζ (2m, 2n − 2m)
m =1 m =1

n (2n − 3) (3(n − 1))


= ζ (2n) + ζ (2)ζ (2n − 2) − ζ (2n)
16 4 4
12 − 11n 2n − 3
= ζ (2n) + ζ (2)ζ (2n − 2). (31)
16 4

For the third sum, using harmonic shuffle relation, we have


322 Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323

ζ (2a + 2b)ζ (2c , 2d) + ζ (2a + 2b)ζ (2d, 2c )



= ζ (2a + 2b, 2c , 2d) + ζ (2a + 2b, 2d, 2c ) + ζ (2c , 2a + 2b, 2d) + ζ (2d, 2a + 2b, 2c )
 
+ ζ (2c , 2d, 2a + 2b) + ζ (2d, 2c , 2a + 2b) + ζ (2a + 2b + 2c , 2d)

+ ζ (2d, 2a + 2b + 2c ) + ζ (2a + 2b + 2d, 2c ) + ζ (2c , 2a + 2b + 2d) . (32)

The second part on the right of (32) could be treated as the first sum of (29). By harmonic shuffle
relation and Lemma 1, the left of (32) is equal to


ζ (2a + 2b)ζ (2c , 2d) + ζ (2a + 2b)ζ (2d, 2c )
a+b+c +d=n
a,b,c ,d1

=2 ζ (2a + 2b)ζ (2c , 2d)
a+b+c +d=n
a,b,c ,d1

n −2 
 
=2 ζ (2m) ζ (2c , 2d)
m=2 a+b=m c +d=n−m

n −2
3
= (m − 1)ζ (2m)ζ (2n − 2m)
2
m =2

n −1
3 3
= (m − 1)ζ (2m)ζ (2n − 2m) − (n − 2)ζ (2)ζ (2n − 2)
2 2
m =1

n −1
3 3
= (n − m − 1)ζ (2n − 2m)ζ (2m) − (n − 2)ζ (2)ζ (2n − 2)
2 2
m =1

3(n − 2)(2n + 1) 3
= ζ (2n) − (n − 2)ζ (2)ζ (2n − 2). (33)
8 2

Where again we use the shift as in the proof of Theorem 1, together with (29), (30) and (32), we have

 
ζ (2a + 2b, 2c , 2d) + ζ (2a + 2b, 2d, 2c ) + ζ (2c , 2a + 2b, 2d)
a+b+c +d=n
a,b,c ,d1

+ ζ (2d, 2a + 2b, 2c ) + ζ (2c , 2d, 2a + 2b) + ζ (2d, 2c , 2a + 2b)
5n − 15 n−3
= ζ (2n) − ζ (2)ζ (2n − 2). (34)
4 2
2
Combining from (26) to (31), (34) and noting that ζ (2) = π6 , we complete the proof of Theorem 2. 2

Proof of Theorem 3. Note that


2n −1 
n −1 
n −1
(−1)a ζ (a, 2n − a) = ζ (2i , 2n − 2i ) − ζ (2i + 1, 2n − 2i − 1).
a =2 i =1 i =1

Then (3) follows from (1) and (2). Meanwhile,


Z. Shen, T. Cai / Journal of Number Theory 132 (2012) 314–323 323

(−1)a + (−1)b + (−1)c ζ (a, b, c )


a+b+c =2n
a2,b,c 1
   
=3 ζ (2i , 2 j , 2k) − ζ (2i + 1, 2 j − 1, 2k) + ζ (2i + 1, 2 j , 2k − 1)
i + j +k=n i + j +k=n
i , j ,k1 i , j ,k1

+ ζ (2i , 2 j − 1, 2k − 1). (35)
i + j +k=n+1
i , j ,k1

By Theorem 1 and Lemma 2, for n > 2, we obtain

   
ζ (2i + 1, 2 j − 1, 2k) + ζ (2i + 1, 2 j , 2k − 1) + ζ (2i , 2 j − 1, 2k − 1)
i + j +k=n i + j +k=n+1
i , j ,k1 i , j ,k1

3 1
= ζ (2n) + ζ (2)ζ (2n − 2). (36)
8 4

Combining (35), (36) and Theorem 1, we complete the proof of Theorem 3. 2

Remark. Similarly, we can consider the sum

(−1)a + (−1)b + (−1)c + (−1)d ζ (a, b, c , d).


a+b+c +d=2n
a2,b,c ,d1

But it will be much more complicated.

Acknowledgment

The authors thank the referee for his constructive suggestion and careful reading of the manuscript.

References

[1] W.Y.C. Chen, L.H. Sun, Extended Zeilberger’s algorithm for identities on Bernoulli and Euler polynomials, J. Number The-
ory 129 (2009) 2111–2132.
[2] M. Eie, A note on Bernoulli numbers and Shintani generalized Bernoulli polynomials, Trans. Amer. Math. Soc. 248 (1996)
1117–1136.
[3] H. Gangl, M. Kaneko, D. Zagier, Double zeta values and modular forms, in: Siegfried (Ed.), Automorphic Forms and Zeta
Functions. In Memory of Tsuneo Arakawa, Proc. of the Conf., Rikkyo University, Tokyo, Japan, 4–7 September 2004, World
Scientific Böherer, Hackensack, NJ, 2006, pp. 71–106.
[4] A. Granville, A decomposition of Riemann’s zeta function, in: Analytic Number Theory, in: London Math. Soc. Lecture Note
Ser., vol. 247, Cambridge University Press, 1997, pp. 95–101.
[5] L. Guo, B. Xie, Weighted sum formula for multiple zeta values, J. Number Theory 129 (2009) 2747–2765.
[6] T. Nakamura, Restricted and weighted sum formulas for double zeta values of even weight, Šiauliai Math. Semin. 12 (2009)
151–155.
[7] Y. Ohno, W. Zudilin, Zeta stars, Commun. Number Theory Phys. 2 (2008) 327–349.

You might also like