arxiv: v1 [math.nt] 16 Jan 2018

Transcription

1 ROOTED TREE MAPS AND THE KAWASHIMA RELATIONS FOR MULTIPLE ZETA VALUES HENRIK BACHMANN AND TATSUSHI TANAKA arxiv: v1 [math.nt] 16 Jan 2018 Abstract. Recently, inspired by the Connes-Kreimer Hopf algebra of rooted trees, the second named author introduced rooted tree maps as a family of linear maps on the noncommutative polynomial algebra in two letters. These give a class of relations among multiple zeta values, which are known to be a subclass of the socalled linear part of the Kawashima relations. In this paper we show the opposite implication, that is the linear part of the Kawashima relations is implied by the relations coming from rooted tree maps. 1. Introduction Rooted tree maps were introduced in [T] and they assign to a rooted tree a linear map on the space H = Q x,y of noncommutative polynomials in x and y. One application of these maps is that any admissible word evaluated at a rooted tree map gives a Q-linear relation between multiple zeta values. To prove this result, one shows that these type of relations follow from a special case, the linear part, of the so called Kawashima relations. The purpose of this note is to show, that the rooted tree maps relations are actually equivalent to the linear part of the Kawashima relations. For k 1 2,k 2,...,k r 1 the multiple zeta values are defined by ζk 1,...,k r = m 1 > >m r>0 m k m kr r To state the Kawashima relations we need to introduce some notations. Denote by H 1 = Q+Hy the subspace of words in H, which end in y. The space H 1 is spanned by the monomials z k1...z kr with k 1,...,k r 1, where z k = x k 1 y. On H 1 one can define recursively the harmonic product by 1 w = w 1 = w and. z k1 v z k2 w = z k1 v z k2 w+z k2 z k1 v w+z k1 +k 2 v w 2010 Mathematics Subject Classification. 05C05, 05C25, 11M32, 16T05. Key words and phrases. Hopf algebra of rooted trees, noncommutative polynomial algebra, multiple zeta values, Kawashima relations. 1

2 2 HENRIK BACHMANN AND TATSUSHI TANAKA for k 1,k 2 1 and v,w H 1. Let H 0 = Q + xhy H be the subspace of admissible words and define the Q-linear map Z : H 0 R on a monomial w = z k1...z kr by Zw = ζk 1,...,k r. Equipped with the harmonic product both H 1 and H 0 are commutative Q-algebras and it is a well-known fact, that Z is an algebra homomorphism from H 0 to the algebra of multiple zeta values. Define the automorphism ϕ AutH with respect to the concatenation on the generators by ϕx = x+y and ϕy = y and define for words v,w H 1 the operator z p v z q w = z p+q v w. With this the Kawashima relations can be stated as follows: Theorem 1.1. [K, Corollary 5.4] For all v,w Hy and m 1 we have 1.2 Zϕv y i Zϕw y j = Zϕv w y m. i+j=m i,j 1 It is expected that Theorem 1.1 gives all Q-linear relations between multiple zeta values after evaluating the product on the left-hand side by the shuffle product formula, whose definition we will omit here. Moreover numerical experiment suggest, that the two cases m = 1,2 are enough to obtain all linear relations. Rooted tree maps assign to a rooted tree a linear map on the space H. Evaluated at any admissible word these give also relations between multiple zeta values. Denoting by H d 1 EndH the space of all rooted tree maps of non-zero degree see Section 2 for precise definitions, we have the following result proven in [T]. Theorem 1.2. [T, Theorem 1.3] For any rooted tree map f H d 1 we have fh 0 kerz. The main result of this work is the following. Theorem 1.3. The rooted tree maps relations are equivalent to the linear part of the Kawashima relations, i.e. Theorem 1.2 implies the m = 1 case of Theorem 1.1. Let R k be the number of linearly independent rooted tree maps relations Theorem 1.2 among multiple zeta values in weight k the number k 1 + +k r in 1.1 and denote by C k the conjectured number of all linearly independent relations of multiple zeta values in weight k. For small k these are given by k R k C k

3 ROOTED TREE MAPS AND THE KAWASHIMA RELATIONS FOR MULTIPLE ZETA VALUES 3 Acknowledgment. The authors would like to thank the Max-Planck-Institut für Mathematik and the Hausdorff Research Institute for Mathematics in Bonn for hospitality and support. The second author was also supported by Kyoto Sangyo University Research Grants. 2. Rooted tree maps A rooted tree is as a finite graph which is connected, has no cycles, and has a distinguished vertex called the root. We draw rooted trees with the root on top and we just consider rooted trees with no plane structure, which means that we for example do not distinguish between and. A product given by the disjoint union of rooted trees will be called a rooted forest and by H we denote the Q-algebra of forests generated by all trees. The unit of H, given by the empty forest, will be denoted by I. Since we just consider trees without plane structure the algebra H is commutative. Due to the work of Connes and Kreimer [CK] the space H has the structure of a Hopf algebra. To define the coproduct on H we first define the linear map B + on H, which connects all roots of the trees in a forest to a new root. For example we have B + =. Clearly for every tree t H there exists a unique forest f t H with t = B + f t, which is just given by removing the root of t. The coproduct on H can then be defined recursively for a tree t H by t = t I+id B + f t and for a forest f = gh with g,h H multiplicatively by f = g h and I = I I. For example we have = I+ +2 +I. In [T] the second named author uses the coproduct to assign to a forest f H a Q-linear map on the space H = Q x,y, called a rooted tree map, by the following: Definition 2.1. For any non-empty forest f H, we define a Q-linear map on H, also denoted by f, recursively: For a word w H and a letter u {x,y} we set 2.1 fwu := M fw u, where Mw 1 w 2 = w 1 w 2 denotes the multiplication on H. This reduces the calculation to fu for a letter u {x,y}, which is defined by the following: i If f =, then fx := xy and fy := xy. ii For a tree t = B + f we set tu := R y R x+2y R 1 y fu, where R v is the linear map given by R v w = wv v,w H.

4 4 HENRIK BACHMANN AND TATSUSHI TANAKA iii If f = gh is a forest with g,h I, then fu := ghu. The rooted tree map of the empty forest I is given by the identity. By H EndH we denote the space spanned by all rooted tree maps and by H d the space spanned by rooted tree maps of degree d number of vertices. Note that for any f H d with d 1 we have f1 = 0, which follows by induction on the degree of f from 2.1 with w = 1. In [BT] it was shown that the derivation n on H, defined for n 1 by n x = xx+y n 1 y and n y = xx+y n 1 y can be written in terms of rooted tree maps. In particular the rooted tree map = 1 is a derivation, which will be used for various calculations in the remaining parts of this work. 3. The space F Denote by H d H the subspace spanned by all rooted forests of degree d. In this section we will consider a subspace of H d defined recursively by F 1 = Q and for d 2 by F d = B + F + F. For example for d = 2,3,4 the F d are given by F 2 = Q +Q, F 3 = Q +Q +Q +Q, F 4 = Q +Q +Q +Q +Q +Q +Q +Q. Notice that the space F = d 1 F d H is a subspace of H, but not a subalgebra since for example F 2 but / F 4. By definition we have dim Q F d = 2. Denote by F d EndH the space spanned by all rooted tree maps corresponding to the rooted trees in F d and set F = F d 1 d H. The main goal of this section is to prove the following. Theorem 3.1. We have an isomorphism of Q-vector spaces Θ : F xhy f fx. Before we can prove Theorem 3.1 at the end of this section, we need to introduce some notation and prove some Lemma. Define the 2 n 1 2 n -matrix A n by A 1 = 1 1 and for n 2 recursively by A n := A n 1 E 2 n E 2 n 2 A n 1 =: A 1 n A2 n,

5 ROOTED TREE MAPS AND THE KAWASHIMA RELATIONS FOR MULTIPLE ZETA VALUES 5 where E n denotes the n n identity matrix. The matrices A j n are both 2 n 1 2 n 1 square matrices. Denote for d 2 by w d H 0 d the vector of all 2 d 2 monomials 2d 2 in H 0 of degree d, ordered in lexicographical order x < y from the top to the bottom. For example we have for d = 2,3,4 w 2 = x 2 y xy, w 3 = xy 2 x 3 y x 2 y 2, w 4 = xyxy. xy 3 For a rooted forest f H we denote by fw d H 0 2d 2 the component-wise evaluation of the rooted tree map f on the entries of w d. In the following we denote for v,w H by R v and L v the linear maps given by R v w = wv and L v w = vw. Lemma 3.2. For all d 2 we have w d A w d+1 mod 2. Proof. We prove this statement by induction on d. For d = 2 this follows immediately, since w 2 = xy = xyy xxy and A 1 w 3 = xxy+xyy. By definition of w d one can check that L x 2xw d w d = where we write w d 2 y = xy yields 3.2 and 3.3 L x w L xy L 1 x w L x 2yw d 2 = L xy xw d 2, L xy yw d 2 := L 1 x w d 2. Using that is a derivation with x = xy and L x w = x 2 w = xxw +x xw +x 2 w = L xy w +L x 2L y + w L xy L 1 x w = xyw = xyw +x yw +xy w Now by assumption we have L xy L y + w +L x 2yw mod 2. A d 2 w d w = xw = L xy w +L x w mod 2. This together with 3.2 and 3.3 yields L x w A d 2 L x w d +L xy w mod 2,

6 6 HENRIK BACHMANN AND TATSUSHI TANAKA L xy L 1 x w A d 2 L xy L 1 x w d+l x 2L y w mod 2. This concludes the statement in the Lemma, since by definition of A and 3.1 the right-hand side gives the entries of A w d+1. Define for d 1 the 2 d 2 -matrices E 1 d = Lemma 3.3. We have for all d 2, E2 d = w d+1 = E 2 R yw d +E 1 R xyr 1 y w d. Proof. This follows from the fact that if one orders the monomials m d in xhy of degree d lexicographically by m d 1 < m d 2 < < m d, then m d+1 2 d 2 2j 1 = R xyr y 1m d j and m d+1 2j = R y m d j for j = 1,...,2 d 2. Define on H the anti-automorphism τ by τx = y and τy = x. Clearly we have τxhy = xhy which we can use to define for d 1 the unique permutation matrix T d M 2 d 2Z satisfying T d w d = τw d. Lemma 3.4. We have for all d det A E 2 1 mod 2. Proof. By Lemma 3.2 and 3.3 we have 3.6 w d A w d+1 mod 2 = A E 2 R yw d +A E 1 R xyr 1 y w d. Since for all w H we have τwy = xyτw+x τw and τ wy = xτ w xyτw the operators and τ commute modulo 2. Together with the definition of the matrix T d, Lemma 3.2, τl x = R y τ and τl xy = R xy τ we get T d w d = τw d τ w d τa w d+1 = τ A 1 L xw d +A 2 L xyl 1 x w d. = A 1 T dr y w d +A 2 T dr xy R 1 y w d mod 2. Since the entries inthe first summand of this expression end all in y 2 andthe entries of thesecondexpressionendallinxy, wegetby3.6theidentitiesa E 2 T da 1 T d

7 ROOTED TREE MAPS AND THE KAWASHIMA RELATIONS FOR MULTIPLE ZETA VALUES 7 mod 2 and A E 1 T da 2 T d mod 2. Since A 1 is an upper triangle matrix with 1 on the diagonal we have deta 1 = 1 which together with dett d 2 = 1 gives 3.5. Now define for d 1 the matrices B d M 2 Z by B 1 = 1 and for d 2 by B E 1 B d = t. B A Notice that B E 1 t is just B with zero columns added at the even places, i.e. if B = b 1,...,b 2 d 2 then B E 2 t = b 1,0,b 2,0,...,b 2 d 2,0. Lemma 3.5. For all d 1 we have detb d 1 mod 2. Proof. With the matrix E d := E 1 d E 2 d M 2Z one checks that B 0 B d E d = B A E 1 B A E 2. The result now follows inductively together with Lemma 3.4. Before we can finally prove Theorem 3.1 we will now give the connection of the map Θ : F xhy given by f fx and the matrix B d. For this we define the vector f d Fd 2 by f 1 = and B + f f d =. f So in particular f d contains all generators of F d. By f d x xhy 2 we will denote the vectors obtained by evaluating all entries at x. Lemma 3.6. For all d 1 we have f d x B d w d+1 mod 2. Proof. We will again use induction on d. For d = 1 the statement can be checked easily. Now for d 2 we get by assumption, the definition of rooted tree maps and Lemma 3.2 f d x = B + f x f x B R y R x R 1 y w d B w d R y R x Ry 1f x f x B R y R x R 1 y w d B A w d mod 2. Together with B R y R x Ry 1w d = B E 1 t w d+1, which follows from the lexicographical ordering of the w d, we obtain the desired result.

8 8 HENRIK BACHMANN AND TATSUSHI TANAKA Proof of Theorem 3.1. That Θ is an isomorphism follows now directly from Lemma 3.2 and 3.6, which imply that every monomial in xhy of degree d+1 can be written as some fx with f F d. Therefore the map Θ is surjective on the degree graded parts F d and xhy d+1, which both have dimension Kawashima relations In this section we want to give the proof of Theorem 1.3, i.e. that the rooted tree map relations imply the linear part of the Kawashima relation. Recall that with the automorphism ϕ AutH defined by ϕx = x+y and ϕy = y the linear part of the Kawashima relations, i.e. the m = 1 case in Theorem 1.1, states that 4.1 ϕv w y = L x ϕv w kerz for any w,v Hy. In [T] it was shown, that the rooted tree maps relations follow from the linear part of the Kawashima relations see proof of [T, Corollary 4.7], i.e. we have in particular {fw f F, w H 0 } L x ϕhy Hy. We will now show that also the opposite inclusion holds. Theorem 4.1. We have L x ϕhy Hy {fw f F, w H 0 }. Proof. Due to Lemma 4.2 in [T] we know that 1 τxhy L x ϕhy Hy. So we have 1 τxhy = τ1 τxhy τl x ϕhy Hy and hence L x ϕhy Hy = 1 τ+τl x ϕhy Hy 1 τxhy+τl x ϕhy Hy τl x ϕhy Hy. The opposite inclusion τl x ϕhy Hy L x ϕhy Hy was shown in the proof of Corollary 4.7 in [T] and therefore we have 4.2 L x ϕhy Hy = τl x ϕhy Hy. Following [T] we define χ x := τl x ϕ. Due to Corollary 4.5. in [T] there exist for any f F a w Hy such that 4.3 fχ x = χ x H w,

9 ROOTED TREE MAPS AND THE KAWASHIMA RELATIONS FOR MULTIPLE ZETA VALUES 9 where H w v = w v for w,v Hy. The element w is uniquely determined by w = χ 1 x fy. Since we have for any rooted tree map f F that fy = fx, we get χ x w = fx. In Theorem 3.1 we proved that the map Θ : F xhy with f fx is an isomorphism and hence for an arbitrary w Hy the rooted tree map f = Θ 1 χ x w F satisfies 4.3. Now let w,v be arbitrary elements in Hy. To proof the statement of the theorem we want to show that we can find a rooted tree map f F and a u H 0, such that L x ϕw v = fu. Because of 4.2 it suffices to show that we can find such an f and u with τl x ϕw v = χ x H w v = fu = fχ x χ 1 x u. By the above discussion we can choose u = χ x v and f = Θ 1 χ x w. Remark 4.2. The space F is a proper subset of H, since H but / F. Due to Theorem 4.1 all rooted tree maps relations are already obtained by considering just the rooted trees in F, that is the space F. Due to numerical experiments we actually expect that the space F is the whole space H, i.e. the recursively defined elements at the beginning of Section 3 might give a basis of the space of rooted tree maps and in particular dim Q Hd = 2. Since the number of rooted tree maps in degree d is larger than 2 for d 4, this would give linear relations between rooted tree maps. For example we expect that we have for all w H w = 2 w+ w w w. References [BT] H. Bachmann, T. Tanaka: Rooted tree maps and the derivation relation for multiple zeta values, arxiv: [math.nt]. [CK] A. Connes, D. Kreimer: Hopf algebras, renormalization and noncommutative geometry, Commun. Math. Phys , [K] G. Kawashima: A class of relations among multiple zeta values, J. Number Theory , no. 4, [T] T. Tanaka: Rooted Tree Maps, arxiv: [math.nt]. Graduate School of Mathematics, Nagoya University, Nagoya, Japan. address: henrik.bachmann@math.nagoya-u.ac.jp Department of Mathematics, Kyoto Sangyo University, Kyoto, Japan. address: t.tanaka@cc.kyoto-su.ac.jp