MPP–2013–278
Closed Superstring Amplitudes, Single–Valued Multiple Zeta Values and Deligne Associator
S. Stieberger
Max–Planck–Institut f¨ ur Physik
Werner–Heisenberg–Institut, 80805 M¨ unchen, Germany
Abstract
We revisit the tree–level closed superstring amplitude and identify its α
′–expansion as
series with single–valued multiple zeta values as coefficients. The latter represent a subclass
of multiple zeta values originating from harmonic polylogarithms at unity. Moreover, in
terms of a non–commutative Hopf algebra the closed superstring amplitude can be cast
into the same algebraic form as the open superstring amplitude: the closed superstring
amplitude essentially is the single–valued version of the open superstring amplitude. This
fact points into a deeper connection between gauge and gravity amplitudes than what is
implied by Kawai–Lewellen–Tye relations. Finally, we give an explicit representation of
the Deligne associator in terms of beta functions modulo squares of commutators of the
underlying Lie algebra. This form of the associator can be interpreted as the four–point
closed superstring amplitude.
1. Introduction
During the last years a great deal of work has been addressed to the problem of re- vealing and understanding the hidden mathematical structures of scattering amplitudes in both field– and string theory, for a recent review cf. [1]. Particular emphasis on the underlying algebraic structure of amplitudes seems to be especially fruitful and might eventually yield an alternative way of constructing perturbative amplitudes by methods residing in arithmetic algebraic geometry. In particular, studying motivic aspects of am- plitudes has dramatically changed our view of how to write amplitudes in terms of simple objects, cf. [2] for an early and [3] for a recent reference. Although motivic amplitudes seem to be mathematically more complicated, they are much more structured, organized and canonical objects.
In perturbative string theory, it is the dependence on the inverse string tension α
′, i.e.
the nature of the underlying string world–sheet describing the string interactions, which provides an extensive and rich structure in the analytic expressions of the amplitudes.
Some of the motivic concepts have recently matured in describing tree–level superstring amplitudes [4]. By passing from the multiple zeta values (MZVs) entering as coefficients in the α
′–expansion of the amplitude to their motivic versions [2,5] and then mapping the latter to elements of a Hopf algebra reveals the motivic structure of the superstring amplitude. In this way the motivic superstring amplitude becomes a rather simple and well organized object. At the same time it is completely insensitive to a change of the basis of the underlying MZVs.
Perturbative gauge and gravity amplitudes in string theory seem to be rather different due to the unequal world–sheet topology of open and closed strings. Although in prac- tice some properties of scattering amplitudes in both gauge and gravity theories suggest a deeper relation originating from string theory, it is not clear how and whether more symmetries or analogies between open and closed string amplitudes can be found. Finding more similarities between the latter is one aim of this article.
In Ref. [4] the closed superstring tree–level amplitude has been given in terms of a Hopf algebra structure and it has been observed, that in contrast to the open string case only MZVs of a special class show up in its α
′–expansion. In Section 2 we revisit the tree–level closed superstring amplitude and identify the coefficients of its power series in α
′as single–valued multiple zeta values (SVMZVs). The latter represent a subclass of MZVs originating from single–valued harmonic polylogarithms (SVMPs) at unity [6,7]. In his recent work [8] Brown has introduced the map sv, which maps the algebra of non–
commutative words describing the open superstring amplitude to a smaller subalgebra,
which describes the space of SVMZVs. In Section 3 we find, that the closed superstring
amplitude essentially follows from the open superstring amplitude by applying this map sv.
The Drinfeld associator [9,10], which is an infinite series in two non–commutative variables with coefficients being MZVs, has been argued to be the generating function of the open superstring amplitudes [11,12]. In Section 4 we identify the Deligne associator [13], which has SVMZVs as coefficients in its series, to be the relevant object describing closed su- perstring amplitudes. More precisely, we give an explicit representation of the Deligne associator in terms of beta functions modulo squares of commutators of the underlying Lie algebra and this form can be interpreted as the four–point closed superstring amplitude.
Finally, in Section 5 we give some concluding remarks.
2. Closed superstring amplitudes and single–valued multiple zeta values
In this Section we want to illuminate the observations on the α
′–expansion of the graviton amplitude [4] in view of Browns recent work on SVMZVs [8].
The string world–sheet describing the tree–level string S–matrix of N gravitons has the topology of a complex sphere with N insertions of graviton vertex operators. Of the latter N − 3 are integrated on the whole sphere leading to the following type of complex
integrals
N Y
−2j =2
Z
z
j∈Cd 2 z j
Y
1≤i<j≤N
−1|z i − z j | s
ij(z j − z i ) n
ij, (2.1)
with z 1 = 0, z N
−1= 1, z N = ∞, the set of integers n ij ∈ Z and the real numbers s ij = α
′(k i + k j ) 2 = 2α
′k i k j . The latter describe the 1 2 N (N − 3) independent kinematic invariants of the scattering process involving N external momenta k i , i = 1, . . . , N and α
′is the inverse string tension. The integrals (2.1) can be considered as iterated integrals on P 1 \{0, 1, ∞} integrated independently on all choices of paths.
One of the key properties of graviton amplitudes in string theory is that at tree–level they can be expressed as sum over squares of (color ordered) gauge amplitudes in the left–
and right–moving sectors. This map, known as Kawai–Lewellen–Tye (KLT) relations [14], gives a relation between a closed string tree–level amplitude M involving N closed strings and a sum of squares of (partial ordered) open string tree–level amplitudes. We may write these relations in matrix notation as follows
M(1, . . . , N ) = A t S A , (2.2)
with the vector A encoding a basis of (N − 3)! open string subamplitudes and some
(N −3)! ×(N −3)! intersection matrix S. The KLT relations are insensitive to the compact-
ification details or the amount of supersymmetries of the superstring background. Hence,
the following discussions and results on N –graviton tree–level scattering are completely general. In the following let us first review some aspects of open superstring amplitudes A.
Tree–level scattering of N open strings involves (N − 3)! independent color ordered subamplitudes [15,16]. The latter can be collected in an (N − 3)!–dimensional vector A, which can be expressed as [4]
A = F A , (2.3)
with the (N − 3)!–dimensional vector A encoding the Yang–Mills basis and the period matrix F , given by 1
F = P Q : exp
X
n
≥1ζ 2 n +1 M 2 n +1
: , (2.4)
with the (N − 3)! × (N − 3)! matrices P = 1 + X
n
≥1ζ 2 n P 2 n , P 2 n = F | ζ
n2
, M 2 n +1 = F | ζ
2n+1,
(2.5)
and:
Q := 1 + X
n
≥8Q n = 1 + 1
5 ζ 3,5 [M 5 , M 3 ] + 3
14 ζ 5 2 + 1 14 ζ 3,7
[M 7 , M 3 ] +
9 ζ 2 ζ 9 + 6
25 ζ 2 2 ζ 7 − 4
35 ζ 2 3 ζ 5 + 1 5 ζ 3 , 3 , 5
[M 3 , [M 5 , M 3 ]] + . . . . (2.6) Above we have have adapted to the following definition of MZVs
ζ n
1,...,n
r:= ζ(n 1 , . . . , n r ) = X
0<k
1<...<k
rY r l=1
k l
−n
l, n l ∈ N + , n r ≥ 2 , (2.7)
with r specifying the depth and w = P r
l =1 n l denoting the weight of the MZV ζ n
1,...,n
r. Furthermore, we have used the MZV basis constructed in [17]. Note, that for any N the tree–level open superstring amplitude assumes the form (2.3) with (2.4). The only ingre- dients are the (N − 3)! × (N − 3)! matrices P 2n and M 2n+1 , whose entries are polynomials in degree 2n and 2n + 1 in the kinematic invariants s ij , respectively. The matrices P 2n and
1 The ordering colons : . . . : are defined such that matrices with larger subscript multiply ma- trices with smaller subscript from the left, i.e. : M
iM
j:=
M
iM
j, i ≥ j ,
M
jM
i, i < j . The generalization
to iterated matrix products : M
i1M
i2. . . M
ip: is straightforward.
M 2n+1 have been thoroughly investigated in [18]. Moreover, the form of the expressions (2.4), (2.5) and (2.6) is bolstered by the algebraic structure of motivic MZVs and their decomposition [5]. In fact, the operator F is isomorphic to the decomposition operator of motivic MZVs [4].
Applying the open string results (2.4) to the graviton amplitude (2.2) gives rise to [4]:
M = A t G A , (2.8)
with the matrix 2 G = F t S F = S 0
exp
X
r
∈2N++1
ζ r M r t
t
Q Q e exp
X
s
∈2N++1
ζ s M s
, (2.9) and the intersection form S 0 defined by:
S 0 = P t S P . (2.10)
An other interpretation of S 0 is, that it makes sure, that the field–theory limit of the graviton amplitude (2.8) is correctly reproduced: M(1, . . . , N )| α
′→0= A t S 0 A, i.e.
G| α
′→0= S 0 .
It has already been observed in [4] (extending the results [19]), that one implication of the specific form of (2.9) is, that only a certain subclass of MZVs appears in the α
′– expansion of the graviton amplitude (2.8). In fact, in Eq. (2.9) the product QQ e is given by [4]
Q Q e = 1 + 2 Q 11 + 2 Q 13 + 2 Q 15 + . . . , (2.11) with Q e = Q |
Q(r)→(−1)r+1Q(r)
and Q (r) any nested commutator of depth r appearing in (2.6). As a consequence the product (2.11) is free of odd powers in even depth commutators Q (2n) . Furthermore, the specific form of (2.9) involves, that MZVs of even weight or depth
≥ 2 only enter through the product (2.11) starting at weight w = 11 [4].
The subclass of MZVs appearing in (2.9) can be identified as single–valued multiple zeta values
ζ sv (n 1 , . . . , n r ) ∈ R (2.12) originating from single–valued multiple polylogarithms at unity and studied recently in [8]
from a mathematical point of view. Let us now illuminate (2.9) in view of this subclass of
2 Note, that the transpositions involved in the expression : exp P
r
ζ
rM
rt:
tlead to a reversal of the matrix multiplication order compared to the ordered product : exp P
s
ζ
sM
s: without transposition, i.e. : : exp P
r∈2N++1
ζ
rM
rt:
t= 1 + ζ
3M
3+ ζ
5M
5+
12ζ
32M
32+ ζ
7M
7+
ζ
3ζ
5M
3M
5+
16ζ
33M
33+ ζ
9M
9+
12ζ
52M
52+ ζ
3ζ
7M
3M
7+
12ζ
32ζ
5M
32M
5+ ζ
11M
11+ . . . .
MZVs. The numbers (2.12) satisfy the same double shuffle and associator relations than the usual MZVs and many more relations [8]:
ζ sv (2) = 0 ,
ζ sv (2n + 1) = 2 ζ 2n+1 , n ≥ 1 . (2.13) Furthermore, for instance we have:
ζ sv (3, 5) = −10 ζ 3 ζ 5 ,
ζ sv (3, 7) = −28 ζ 3 ζ 7 − 12 ζ 5 2 ,
ζ sv (3, 3, 5) = 2 ζ 3 , 3 , 5 − 5 ζ 3 2 ζ 5 + 90 ζ 2 ζ 9 + 12
5 ζ 2 2 ζ 7 − 8
7 ζ 2 3 ζ 5 2 , (2.14) ζ sv (3, 5, 5) = 2 ζ 3 , 5 , 5 + 10 ζ 5 ζ 3 , 5 + 50 ζ 3 ζ 5 2 + 275 ζ 2 ζ 11 + 20 ζ 2 2 ζ 9 ,
ζ sv (3, 3, 7) = 2 ζ 3 , 3 , 7 + 12 ζ 5 ζ 3 , 5 + 14 ζ 3 2 ζ 7 + 60 ζ 3 ζ 5 2 + 407 ζ 2 ζ 11 + 112
5 ζ 2 2 ζ 9 − 64 35 ζ 2 3 ζ 7 . The matrix (2.9) can be written purely in terms of SVMZVs (2.12) as follows:
G = S 0 1 + ζ sv (3) M 3 + ζ sv (5) M 5 + 1
2 ζ sv (3) 2 M 3 2 + ζ sv (7) M 7 + 1
2 ζ sv (3) ζ sv (5) {M 3 , M 5 } + ζ sv (9) M 9 + 1
3! ζ sv (3) 3 M 3 3 + 1
2 ζ sv (5) 2 M 5 2 + 1
2 ζ sv (3) ζ sv (7) {M 3 , M 7 } + Q sv (11) + ζ sv (11) M 11 + 1
8 ζ sv (3) 2 ζ sv (5) {M 3 , {M 3 , M 5 }}
+ 1
4! ζ sv (3) 4 M 3 4 + 1
2 ζ sv (3) ζ sv (9) {M 3 , M 9 } + 1
2 ζ sv (5) ζ sv (7) {M 5 , M 7 } + Q sv (13) + ζ sv (13) M 13 (2.15) + 1
8 ζ sv (3) 2 ζ sv (7) {M 3 , {M 3 , M 7 }} + 1
4 ζ sv (3) ζ sv (5) 2 {M 3 , M 5 2 } + 1
2 ζ sv (3) {M 3 , Q sv (11)} + 1
2 ζ sv (7) 2 M 7 2 + 1
2 ζ sv (3) ζ sv (11) {M 3 , M 11 } + 1
2 ζ sv (5) ζ sv (9) {M 5 , M 9 } + 1
48 ζ sv (3) 3 ζ sv (5) {M 3 , {M 3 , {M 3 , M 5 }}} + . . . , with:
Q sv (11) = 1
5 ζ sv (3, 3, 5) + 1
8 ζ sv (3) 2 ζ sv (5)
[M 3 , [M 5 , M 3 ]] , Q sv (13) =
1
25 ζ sv (3, 5, 5) − 1
4 ζ sv (3) ζ sv (5) 2
[M 5 , [M 5 , M 3 ]] (2.16) +
1
14 ζ sv (3, 3, 7) − 3
35 ζ sv (3, 5, 5) − 1
16 ζ sv (3) 2 ζ sv (7)
[M 3 , [M 7 , M 3 ]] .
Note, that in (2.15) the terms (2.11) containing MZVs of even weight or depth ≥ 2 comprise into the expressions Q sv (n), which can be written purely in terms of SVMZVs.
A different representation in terms of the SVMZV basis chosen in [8] gives:
Q sv (11) =
− 1
10 ζ sv (3, 5, 3) − 1
8 ζ sv (3) 2 ζ sv (5) + 299
20 ζ sv (11)
[M 3 , [M 5 , M 3 ]] , Q sv (13) =
− 1
50 ζ sv (5, 3, 5) − 3
10 ζ sv (3) ζ sv (5) 2 + 1003
100 ζ sv (13)
[M 5 , [M 5 , M 3 ]]
+ 3
70 ζ sv (5, 3, 5) − 1
28 ζ sv (3, 7, 3) − 1
8 ζ sv (3) 2 ζ sv (7) + 571
140 ζ sv (13)
[M 3 , [M 7 , M 3 ]].
(2.17)
3. Motivic open and closed superstring amplitudes
In this Section we find a striking similarity between the open superstring amplitude A and the closed superstring amplitude M thus giving a new relation between gauge and gravity amplitudes at the level of the underlying Hopf algebra.
Motivic MZVs ζ m are defined as elements of a certain algebra H = L
w≥0 H w over Q, which is graded for the weight and equipped with the period homomorphism per : H → R, which maps ζ n m
1,...,n
rto ζ n
1,...,n
r, i.e. per(ζ n m
1,...,n
r) = ζ n
1,...,n
r. The motivic versions of the SVMZVs (2.12) have been defined in [8] and are denoted by ζ sv m (n 1 , . . . , n r ). The latter satisfy all motivic relations of MZVs and ζ sv m (2) = 0 [8]. The motivic SVMZVs span a subalgebra H sv ⊂ H. There exist a homomorphism H → H sv , which maps each ζ n m
1,...,n
rto ζ sv m (n 1 , . . . , n r ).
A list of generators of H sv w up to weight w = 14 is collected in Tables 1–2 below.
w 2 3 4 5 6 7 8 9 10
H
wsv− ζ
svm(3) − ζ
svm(5) ζ
svm(3)
2ζ
svm(7) ζ
svm(3) ζ
svm(5) ζ
svm(9) ζ
svm(5)
2ζ
svm(3)
3ζ
svm(3) ζ
svm(7)
dim(H
svw) 0 1 0 1 1 1 1 2 2
Table 1: Generators of H
svwfor 2 ≤ w ≤ 10 .
w 11 12 13 14
H
svwζ
svm(11) ζ
svm(3) ζ
svm(9) ζ
svm(13) ζ
svm(3) ζ
svm(3, 3, 5) ζ
svm(3, 3, 5) ζ
svm(5) ζ
svm(7) ζ
svm(3, 5, 5) ζ
svm(3)
3ζ
svm(5) ζ
svm(3)
2ζ
svm(5) ζ
svm(3)
4ζ
svm(3, 3, 7) ζ
svm(7)
2ζ
svm(3)
2ζ
svm(7) ζ
svm(3) ζ
svm(11) ζ
svm(3) ζ
svm(5)
2ζ
svm(5) ζ
svm(9)
dim(H
svw) 3 3 5 5
Table 2: Generators of H
svwfor 11 ≤ w ≤ 14 .
Note, that the generators of Tables 1–2 are those, which appear in (2.15), subject to the period map per.
To explicitly describe the structure of the algebra H Brown has introduced an auxiliary algebra U , the (trivial) algebra–comodule [5]:
U = Qhf 3 , f 5 , . . .i ⊗
QQ[f 2 ] . (3.1) The first factor U
′= U
f 2 U is a cofree Hopf–algebra on the cogenerators f 2r+1 in degree 2r + 1 ≥ 3, whose basis consists of all non–commutative words in the f 2i+1 . The multipli- cation on U
′is given by the shuffle product III . The Hopf–algebra U
′is the algebra of all words constructed from the alphabet {f 3 , f 5 , f 7 , . . .} and is isomorphic to the space of non–
commutative polynomials in f 2i+1 . The element f 2 commutes with all f 2r+1 . Again, there is a grading U k on U and we have the non–canonical isomorphism: H ≃ U . Furthermore, there exists a morphism φ of graded algebra–comodules
φ : H −→ U , (3.2)
normalized by:
φ ζ n m
= f n , n ≥ 2 . (3.3)
The map (3.2), which respects the shuffle multiplication rule
φ(x 1 x 2 ) = φ(x 1 ) III φ(x 2 ) , x 1 , x 2 ∈ H , (3.4) sends every motivic MZV to a non–commutative polynomial in the f i .
The motivic period matrix F m , where all MZVs ζ n
1,...,n
rare replaced by their motivic objects ζ n m
1,...,n
r, has been introduced and studied in [4]. Furthermore, in this reference the map of F m under φ has been computed, with the result:
φ(F m ) = X
∞k=0
f 2 k P 2k
!
X
∞p =0
X
i1,...,ip
∈2N++1
f i
1f i
2. . . f i
pM i
p. . . M i
2M i
1
. (3.5)
Furthermore, in [4] the motivic version of G m has been introduced and an expression for φ(G m ) has been given. We want to rewrite the latter in view of the recent work of Brown [8] and eventually find a striking similarity between φ(F m ) and φ(G m ).
Similar to the construction of U in Ref. [8] Brown has introduced a model U sv for H sv via the the homomorphism
sv : U
′−→ U
′, (3.6)
with
w 7−→ X
uv=w
u III e v , (3.7)
and e v being the reversal of the word v. For the image U sv under (3.6) we have the isomorphism: H sv ≃ U sv . For instance we have [8]
sv(f a ) = 2f a , sv(f a f b ) = 2 (f a f b + f b f a ) , sv(f a f b f c ) = 2 (f a f b f c + f a f c f b + f c f a f b + f c f b f a ) , sv(f a f b f c f d ) = 2 (f a f b f c f d + f a f b f d f c + f a f d f b f c + f a f d f c f b
+ f d f a f b f c + f d f a f c f b + f d f c f a f b + f d f c f b f a ) ,
(3.8)
for the odd integers a, b, c, d. Evidently, we can extend the map (3.6) to U with:
sv(f 2 ) = 0 . (3.9)
Let us now return to the matrix G, given in (2.9) and (2.15) and compute the image φ(G m ). The latter has been given in [4] as:
φ(G m ) = S 0
X
∞p =0
X
i1,...,ip
∈2N+ +1
M i
1M i
2. . . M i
pX p k=0
f i
1f i
2. . . f i
kIII f i
pf i
p−1. . . f i
k+1
.
(3.10) By making profit out of the map (3.6) and using relations like (3.8) we can cast (3.10) into the compact form:
φ(G m ) = S 0
X
∞p=0
X
i1,...,ip
∈2N+ +1
M i
1M i
2. . . M i
psv(f i
1f i
2. . . f i
p)
. (3.11)
After comparing (3.11) with (3.5) and using (3.9) we find:
φ(G m ) = S 0 sv(φ(F m )) . (3.12)
Finally, due to (3.12) the motivic open superstring amplitude A m
φ(A m ) = X
∞k=0
f 2 k P 2k
!
X
∞p=0
X
i1,...,ip
∈2N+ +1
f i
1. . . f i
pM i
p. . . M i
1
A , (3.13)
and the motivic closed superstring amplitude M m
φ(M m ) = A t S 0
X
∞p=0
X
i1,...,ip
∈2N+ +1