INTEGERS 14 (2014) #A51 MULTI-POLY-BERNOULLI-STAR NUMBERS AND FINITE MULTIPLE ZETA-STAR VALUES Kohtaro Imatomi Graduate School of Mathematics, Kyushu University, Nishi-ku, Fukuoka, Japan k-imatomi@math.kyushu-u.ac.jp Received: 11/27/13, Revised: 5/3/14, Accepted: 7/22/14, Published: 10/8/14 Abstract We define the multi-poly-Bernoulli-star numbers which generalize classical Bernoulli numbers. We study the basic properties for these numbers and establish sum formulas and a duality theorem, and discuss a connection to the finite multiple zeta-star values. As an application, we present alternative proofs of some relations on the finite multiple zeta-star values. 1. Introduction For any multi-index (k1 , . . . , kr ) with ki 2 Z, we define two kinds of multi-poly(k ,...,kr ) (k ,...,kr ) Bernoulli-star numbers Bn,?1 , Cn,?1 by the following generating series: Li?k1 ,...,kr (1 1 e e t) t Li?k1 ,...,kr (1 e t ) et 1 = = 1 X n=0 1 X n=0 n (k ,...,kr ) t Bn,?1 n! n (k ,...,kr ) t Cn,?1 n! , , where Li?k1 ,...,kr (z) is the non-strict multiple polylogarithm given by Li?k1 ,...,kr (z) = X m1 ··· mr z m1 . mk11 · · · mkr r 1 When r = 1, these numbers are poly-Bernoulli numbers studied in [1], [9]. Further, when r = 1 and k1 = 1, both numbers are classical Bernoulli numbers since Li?1 (1 (1) (1) (1) (1) e t ) = t. We note that Bn,? = Cn,? (n 6= 1) with B1,? = 1/2 and C1,? = 1/2. We call k = k1 + · · · + kr the weight of multi-index (k1 , . . . , kr ). We set Q p Z/pZ A := L = {(ap )p ; ap 2 Z/pZ}/ ⇠, p Z/pZ 2 INTEGERS: 14 (2014) where (ap )p ⇠ (bp )p is equivalent to the equalities ap = bp for all but finitely many primes p. The finite multiple zeta-star values are defined by ? ⇣A (k1 , . . . , kr ) := Hp? (k1 , . . . , kr ) mod p p 2 A, where Hn? (k1 , . . . , kr ) is the non-strict multiple harmonic sum defined by Hn? (k1 , . . . , kr ) = X n 1 m1 ··· mr 1 . k1 m · · · mkr r 1 1 For more details on the finite multiple zeta(-star) values, we refer the reader to [7], [10] and [12]. We use “star” to indicate that the inequalities in the sum are nonstrict in contrast to the strict ones usually adopted in the references above. Each of these is expressed as a linear combination of the other. This article is organized as follows. In §2, we give fundamental properties for the multi-poly-Bernoulli-star numbers. In §3, we describe the sum formula and the duality relation for the multi-poly-Bernoulli-star numbers. In §4, we study connections between the finite multiple zeta-star values and the multi-poly-Bernoulli-star numbers. As a result, we obtain alternative proofs of some relations for the finite multiple zeta-star values. 2. Basic Properties for the Multi-Poly-Bernoulli-Star Numbers In this section, we introduce basic results for the multi-poly-Bernoulli-star numbers. We first give the recurrence relations for the multi-poly-Bernoulli-star num(k ,...,kr ) (k ,...,kr ) bers Bn,?1 , Cn,?1 . Before stating them, we provide the following identity for the non-strict multiple polylogarithm, whose proof is straightforward and is omitted. Lemma 2.1. For any multi-index (k1 , . . . , kr ) with ki 2 Z, we have 8 1 > Li? (t) (k1 6= 1), > > > t k1 1,k2 ,...,kr < 1 d ? Li? (t) (k1 = 1, r 6= 1), Li (t) = t(1 t) k2 ,...,kr > dt k1 ,...,kr > > > : 1 (k1 = r = 1). 1 t Proposition 2.1. For any multi-index (k1 , . . . , kr ), we have the following recursions: 3 INTEGERS: 14 (2014) (i) When k1 6= 1, (k ,...,kr ) = (k ,...,kr ) = Bn,?1 Cn,?1 (ii) When k1 = 1, (1,k ,...,kr ) = (1,k ,...,kr ) = Bn,? 2 Cn,? 2 0 1 @ (k1 Bn,? n+1 0 1 @ (k1 Cn,? n+1 1,k2 ,...,kr ) 1,k2 ,...,kr ) 1 ◆ n (k ,...,kr ) A Bj,?1 , j 1 j=1 1 n X1 ✓n + 1◆ (k ,...,k ) r A Cj,?1 . j j=0 n X1 ✓ 0 1 ◆ n ✓ ◆ n 1✓ X X 1 @ n n (k ,...,kr ) (1,k ,...,kr ) A Bj,?2 Bj,? 2 , n + 1 j=0 j j 1 j=1 0 1 ✓ ◆ n 1 X 1 @ (k2 ,...,kr ) n (1,k ,...,kr ) A Cn,? ( 1)n j Cj,? 2 , n+1 j 1 j=1 where an empty sum is understood to be 0. (k ,...,kr ) Proof. We prove the relations for Bn,?1 Li?k1 ,...,kr (1 e t ) = (1 (k ,...,kr ) : those for Cn,?1 e t) 1 X n (k ,...,kr ) t Bn,?1 n! n=0 are similar. . (1) We di↵erentiate both sides of (1): When k1 6= 1, we obtain (LHS) = e 1 (RHS) = e t t e t 1 X Li?k1 1,k2 ,...,kr (1 n (k ,...,kr ) t Bn,?1 n=0 So we have 1 ⇣ X (k Bn,?1 n=0 1,k2 ,...,kr ) (k ,...,kr ) Bn,?1 n! ⌘ tn n! + (1 e t ), e t) 1 X n (k ,...,kr ) t 1 Bn+1,? n=0 = (et 1) 1 X n! . n (k ,...,kr ) t 1 Bn+1,? n! n=0 ✓ ◆ 1 n n X X1 n (k1 ,...,kr ) t = Bj+1,? . j n! n=1 j=0 Comparing the coefficients of tn /n! on both sides, we obtain (i). When k1 = 1, 1 Li? (1 e t ), 1 e t k2 ,...,kr 1 1 ⇣ n X X (1,k2 ,...,kr ) t (1,k ,...,kr ) (RHS) = Bn+1,? +e t Bn,? 2 n! n=0 n=0 (LHS) = (1,k ,...,kr ) 2 Bn+1,? ⌘ tn n! . 4 INTEGERS: 14 (2014) So we have 1 ⇣ X (1,k ,...,kr ) Bn,? 2 (1,k2 ,...,kr ) Bn+1,? n=0 ⌘ tn n! 1 ⇣ X (k ,...,kr ) = e Bn,?2 (1,k ,...,kr ) t = 2 Bn+1,? n=0 1 X n ✓ X n=0 j=0 ◆ n ⇣ (k2 ,...,kr ) Bj,? j ⌘ tn n! (1,k ,...,kr ) 2 Bj+1,? ⌘ tn n! and by this we obtain (ii). We proceed to describe explicit formulas for the multi-poly-Bernoulli-star numbers in terms of the Stirling numbers of the second kind. We recall that the Stirling ⇢ n numbers of the second kind are the integers for all integers m, n satisfying m the following recursions and the initial values: ⇢ ⇢ ⇢ n+1 n n = +m (8n, m 2 Z), m m 1 m ⇢ ⇢ ⇢ 0 n 0 = 1, = = 0 (n, m 6= 0). 0 0 m Proposition 2.2. For any multi-index (k1 , . . . , kr ), ki 2 Z, we have ⇢ X ( 1)m1 +n 1 (m1 1)! n (k1 ,...,kr ) Bn,? = k1 kr m 1 m1 · · · mr 1 n+1 m1 ··· mr 1 and (k ,...,kr ) Cn,?1 X = n+1 m1 ··· mr ⇢ 1)! n + 1 . m1 ( 1)m1 +n 1 (m1 mk11 · · · mkr r 1 Proof. Using the following identity (cf. [3, eqn. 7.49]) 1 ⇢ X j tj t m (e 1) = m! (m m j! j=m 0), (2) we have 1 X = n=0 Li?k1 ,...,kr (1 = m1 = n (k ,...,kr ) t Bn,?1 n! e t) 1 e t X (1 e t )m1 1 mk11 · · · mkr r ··· mr 1 X 1 X m1 ··· mr 1 j=m1 ( 1)m1 +j 1 (m1 mk11 · · · mkr r 1 1)! ⇢ j m1 tj 1 j! 5 INTEGERS: 14 (2014) = 1 X X j=0 j+1 m1 ··· mr ( 1)m1 +j 1 (m1 mk11 · · · mkr r 1 1)! ⇢ j m1 tj . 1 j! Thus comparing the coefficients of tn /n! on both sides, we obtain the explicit for(k ,...,kr ) mula for Bn,?1 . (k ,...,kr ) The explicit formula for Cn,?1 is obtained similarly by using the identity ⇢ 1 X e t (1 e t )m 1 j + 1 tj = ( 1)j+m+1 , (m 1)! m j! j=m 1 which follows from (2) by di↵erentiation. We finish this section by giving some simple relations among the multi-polyBernoulli-star numbers. Proposition 2.3. For any multi-index (k1 , . . . , kr ) with ki 2 Z, we have n ✓ ◆ X n (k1 ,...,kr ) (k ,...,kr ) Bn,? = Cj,?1 j j=0 and (k ,...,kr ) Cn,?1 n X = ( 1)n j j=0 (k ,...,kr ) Proof. The generating functions of Bn,?1 the above identities follow immediately. ✓ ◆ n (k ,...,kr ) Bj,?1 . j (k ,...,kr ) , Cn,?1 di↵er by the factor et , and Proposition 2.4. For any multi-index (k1 , . . . , kr ), we have (k ,...,kr ) Cn,?1 (k ,...,kr ) = Bn,?1 (k ,...,kr ) Proof. By the explicit formula for Cn,?1 (k ,...,kr ) Cn,?1 ( 1)m1 +n 1 (m1 mk11 · · · mkr r 1 n+1 m1 ··· mr X = n+1 m1 ··· mr + X , we obtain ⇢ 1)! n + 1 m1 ⇢ m1 +n 1 ( 1) (m1 1)! n k1 kr m1 1 m1 · · · mr 1 ⇢ ( 1)m1 +n 1 (m1 1)! n m1 mk11 1 mk22 · · · mkr r X = (k1 1,k2 ,...,kr ) . 1,? Cn n+1 m1 ··· mr 1 (k ,...,kr ) = Bn,?1 (k1 1,k2 ,...,kr ) . 1,? Cn The second equality above is by the recursion for the Stirling numbers of the second kind. 6 INTEGERS: 14 (2014) 3. Main Results We give the following sum formulas for multi-poly-Bernoulli-star numbers. Theorem 3.1. We have (k ,...,kr ) = ✓ ◆ ( 1)k n (1) B k k 1 n k+1,? (3) (k ,...,kr ) = ✓ ◆ ( 1)k n (1) C . k k 1 n k+1,? (4) X ( 1)r Bn,?1 X ( 1)r Cn,?1 k1 +···+kr =k 1rk,ki 1 and k1 +···+kr =k 1rk,ki 1 Proof. We multiply both sides of (3) by tn /n! and sums on n. Hence we have 1 X (LHS) = X n (k ,...,kr ) t ( 1)r Bn,?1 n! n=0 k1 +···+kr =k 1rk,ki 1 X = ( 1)r k1 +···+kr =k 1rk,ki 1 (RHS) = = = = Li?k1 ,...,kr (1 1 e e t) t , ◆ 1 ✓ ( 1)k X n tn (1) Bn k+1,? k n=0 k 1 n! 1 ( 1)k X (1) B k! n=0 n k+1,? (n 1 ( 1)k X (1) tn+k Bn,? k! n=0 n! ( 1)k tk k! 1 e t tn k + 1)! 1 . Since both sides have the same denominator, it suffices to prove the following identity: X k1 +···+kr =k 1rk,ki 1 ( 1)r Li?k1 ,...,kr (1 e t) = ( 1)k k t . k! (5) This equality is proved by induction on the weight. When k = 1, the left-hand side is Li?1 (1 e t ) = t and is equal to the right-hand side. Next we assume the 7 INTEGERS: 14 (2014) identity holds when the weight is k. Then by di↵erentiating the left-hand side of the identity of weight k + 1, we obtain 0 X d B B dt @ 1 C e t )C A ( 1)r Li?k1 ,...,kr (1 k1 +···+kr =k+1 1rk+1,ki 1 = X ( 1)r k1 +···+kr =k 1rk,ki 1 = X ✓ e 1 t e t 1 1 e ( 1)r+1 Li?k1 ,...,kr (1 ◆ t Li?k1 ,...,kr (1 e t) e t) k1 +···+kr =k 1rk,ki 1 = ( 1)k+1 k t . k! We used the induction hypothesis in the last equality. Therefore we have X ( 1)r Li?k1 ,...,kr (1 e t) = k1 +···+kr =k+1 1rk+1,ki 1 ( 1)k+1 k+1 t +C (k + 1)! with some constant C, which we find is 0 by putting t = 0. The equation (4) follows from (5) since the generating function of Cn,? di↵ers from that of Bn,? only by a factor e t . Next we describe the duality relation for the multi-poly-Bernoulli-star numbers. We recall the duality operation of Ho↵man [6, p. 65]. We define a function S from the set of multi-indices (k1 , . . . , kr ) with ki 1 and weight k to the power set of {1, 2, . . . , k 1} by S((k1 , . . . , kr )) = {k1 , k1 + k2 , . . . , k1 + · · · + kr 1 }. Obviously, the map S is a one-to-one correspondence. Then (k10 , . . . , kl0 ) is said to be the dual index for (k1 , . . . , kr ) in Ho↵man’s sense when (k10 , . . . , kl0 ) = S 1 ({1, 2, . . . , k 1} S((k1 , . . . , kr ))). It is easy to see that Ho↵man’s duality operation is an involution. Note that k1 > 1 if and only if k10 = 1. Theorem 3.2. For any multi-index (k1 , . . . , kr ) with ki (k ,...,kr ) Cn,?1 (k10 ,...,kl0 ) = ( 1)n Bn,? 1(1 i r), we have , where (k10 , . . . , kl0 ) is the dual index of (k1 , . . . , kr ) in Ho↵man’s sense. 8 INTEGERS: 14 (2014) Proof. As in the previous proof, consider the generating functions of both sides: (LHS) = (RHS) = 1 X n (k ,...,kr ) t Cn,?1 n=0 1 X n! n ( 1) = Li?k1 ,...,kr (1 e t ) , et 1 n (k0 ,...,kl0 ) t Bn,?1 n! n=0 = Li?k0 ,...,k0 (1 1 l 1 et et ) . Hence we have to show the following identity: Li?k1 ,...,kr (1 e t ) + Li?k10 ,...,k0 (1 et ) = 0. l This identity also follows from induction on the weight. First, it is trivial in the case k = 1. Thus, we assume the above identity holds when the weight is k. Since 0 k1 = 1 is equivalent to k1 6= 1, we may assume k1 = 1 by the symmetry of the identity. Then when the weight is k + 1, the derivative of the left-hand side yields ⌘ d ⇣ ? Li1,k2 ,...,kr (1 e t ) + Li?k0 ,...,k0 (1 et ) 1 l dt t 1 e = Li? (1 e t ) + Li? 0 et ) 0 0 (1 1 e t k2 ,...,kr 1 et k1 1,k2 ,...,kl ⇣ ⌘ 1 ? t ? t = Li (1 e ) + Li (1 e ) 0 0 0 k2 ,...,kr k1 1,k2 ,...,kl 1 e t = 0. Therefore we obtain Li?1,k2 ,...,kr (1 e t ) + Li?k0 ,...,k0 (1 1 et ) = C l with some constant C, and by putting t = 0, we conclude C = 0. 4. Connection to the Finite Multiple Zeta-Star Values In this section, we give alternative proofs for some relations of the finite multiple zeta-star values using the multi-poly-Bernoulli-star numbers. The following congruence is the “star-version” of the congruence given in [8, Theorem.8], and is proved in exactly the same manner: Hp? (k1 , . . . , kr ) ⌘ Thus we find ? ⇣A (k1 , . . . , kr ) = ⇣ (k1 1,k2 ,...,kr ) 2,? Cp (k1 1,k2 ,··· ,kr ) 2,? Cp mod p. ⌘ mod p . p (6) 9 INTEGERS: 14 (2014) Corollary 4.1 (M. Ho↵man [7]). For any multi-index (k1 , . . . , kr ) with ki 0 0 1(1 i r), let (k1 , . . . , kl ) be the dual index for (k1 , . . . , kr ) in Ho↵man’s sense. Then we have 0 0 ? ? ⇣A (k1 , . . . , kr ) = ⇣A (k1 , . . . , kl ). Proof. It is sufficient to prove the case k1 = 1. By (6) and the duality relation for the multi-poly-Bernoulli-star numbers, we obtain ⇣ ⌘ (0,k2 ,··· ,kr ) (LHS) = Cp 2,? mod p , p ✓ ◆ 0 0 0 (k1 1,k2 ,··· ,kl ) (RHS) = Cp 2,? mod p p ⇣ ⌘ (k ,··· ,k ) = ( 1)p Bp 22,? r mod p . p (0,k ,··· ,kr ) Hence we complete the proof if we prove Cn,? 2 consider the generating functions of these numbers: 1 X n (0,k ,··· ,kr ) t Cn,? 2 n! n=0 Li?0,k2 ,...,kr (1 et 1 = 1 = et 1 X m2 ··· mr (0,k2 ,··· ,kr ) 2,? Cp for all n. We e t) 1 k2 m2 · · · mkr r 1 1 = Li? (1 e t (et 1) k2 ,...,kr 1 n X (k ,··· ,kr ) t = Bn,?2 . n! n=0 From this we have (k ,··· ,kr ) = Bn,?2 1 X (1 e t )m1 m1 =m2 e t) (k2 ,··· ,kr ) 2,? = ( 1)p Bp for any odd prime p. The following corollary is a weaker version of the sum formula for the finite multiple zeta-star values proved in [11]. Corollary 4.2. We have X ? ( 1)r ⇣A (k1 , . . . , kr ) = (Bp k1 +···+kr =k r 1,k1 2,ki 1 where Bn is the classical Bernoulli numbers. k mod p)p , 10 INTEGERS: 14 (2014) Proof. Equations (4) and (6) yield X ( k1 +···+kr =k+1 r 1,k1 2,ki 1 ? 1)r+1 ⇣A (k1 , . . . , kr ) = = So replacing k by k ✓ ⇣ ✓ ( 1)k p k k (1) k 1,? Cp ◆ 2 (1) C 1 p k 1,? ◆ mod p p ⌘ mod p . p 1, we obtain the desired identity. Acknowledgments. The author expresses his sincere gratitude to Professor M. Kaneko for useful discussions and invaluable advice. References [1] T. Arakawa and M. Kaneko, Multiple zeta values, poly-Bernoulli numbers, and related zeta functions, Nagoya Math. J. 153 (1999), 189–209. [2] A. Bayad and Y. Hamahata, Multiple Polylogarithms and Multi-Poly-Bernoulli Polynomials, Funct. Approx. Comment. Math. 46 (2012), 45–61. [3] R. Graham, D. Knuth, and O. Patashnik, Concrete Mathematics, Addison-Wesley, 1989. [4] Y. Hamahata and H. Masubuchi, Special Multi-Poly-Bernoulli Numbers, J. Integer Seq. 10, (2007). [5] Y. Hamahata and H. Masubuchi, Recurrence Formulae for Multi-Poly-Bernoulli Numbers, Integers 7 (2007), #A46, 15pp. [6] M. Ho↵man, Algebraic aspects of multiple zeta values, Zeta functions, Topology and Quantum Physics, Developments in Math. 14, Springer, New York, 2005, pp. 51–73. [7] M. Ho↵man, Quasi-symmetric functions and mod p multiple harmonic sums, preprint, arXiv:math/0401319v2 [math.NT] 17 Aug. 2007. [8] K. Imatomi, M. Kaneko and E. Takeda, Multi-Poly-Bernoulli Numbers and Finite Multiple Zeta Values, J. Integer Seq. 17 (2014). [9] M. Kaneko, Poly-Bernoulli numbers, J. Théor. Nombres Bordeaux 9 (1997), 199–206. [10] M. Kaneko and D. Zagier, Finite Multiple Zeta Values, preprint. [11] S. Saito and N. Wakabayashi, Sum formula for finite multiple zeta values, J. Math. Soc. Japan, to appear. [12] J. Zhao, Wolstenholme type theorem for multiple harmonic sums, Int. J. Number Theory 4 (2008), 73–106.