Abstract
We relate the low-energy expansions of world-sheet integrals in genus-one amplitudes of open- and closed-string states. The respective expansion coefficients are elliptic multiple zeta values (eMZVs) in the open-string case and non-holomorphic modular forms dubbed 'modular graph forms (MGFs)' for closed strings. By inspecting the differential equations and degeneration limits of suitable generating series of genus-one integrals, we identify formal substitution rules mapping the eMZVs of open strings to the MGFs of closed strings. Based on the properties of these rules, we refer to them as an elliptic single-valued map which generalizes the genus-zero notion of a single-valued map acting on MZVs seen in tree-level relations between the open and closed string.
Export citation and abstract BibTeX RIS
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
1. Introduction
One-loop amplitudes in string theories are computed from integrals over moduli spaces of punctured genus-one world-sheets. For open and closed strings, the punctures are integrated over a cylinder boundary and the entire torus, respectively, which is often done in a low-energy expansion, i.e. order by order in the inverse string tension α'. The coefficients of such α'-expansions involve special numbers and functions which have triggered fruitful interactions between number theorists, particle phenomenologists and string theorists. For instance, elliptic polylogarithms [1, 2] and elliptic multiple zeta values (eMZVs) [3] were identified to form the number-theoretic backbone of genus-one open-string integrals [4–6].
For the closed string, the analogous genus-one integrals involve non-holomorphic modular forms [7–9] dubbed modular graph forms (MGFs) [10, 11] which inspired mathematical research lines [12–16]. As a unifying building block shared by open and closed strings, both eMZVs [3, 17] and MGFs [10, 11, 18] can be reduced to iterated integrals over holomorphic Eisenstein series, or iterated Eisenstein integrals. Similar iterated integrals over holomorphic modular forms play a key role in recent progress on the evaluation of Feynman integrals [19–39]. As a main result of this work, we identify infinite families of closed-string integrals, where the appearance of iterated Eisenstein integrals is in precise correspondence with those in open-string α'-expansions.
More specifically, we give an explicit proposal for a single-valued map at genus one, mapping individual eMZVs to combinations of iterated Eisenstein integrals and their complex conjugates which should be contained in Brown's single-valued iterated Eisenstein integrals [13, 14]. This generalizes the genus-zero result that the sphere integrals in closed-string tree amplitudes are single-valued versions of the disk integrals in open-string tree amplitudes [40–45]. The notion of single-valued periods [46, 47] and single-valued integration [48, 49] is very general, and in the case of MZVs amounts to evaluating single-valued polylogarithms [50] at unit argument. While the single-valued map for the MZVs in tree-level α'-expansions has been pinpointed in [46, 48], the genus-one studies of single-valued maps from mathematical [12–14] and physical [51, 52] viewpoints 6 have not yet led to a consensus for the single-valued version of individual eMZVs.
Our proposal for single-valued eMZVs can be seen as a correspondence between integration cycles and antimeromorphic forms that is akin to Betti–deRham duality [49, 57, 58]. In a tree-level context, Betti–deRham duality relates the ordering of open-string punctures on a disk boundary to Parke–Taylor factors [40–45]—cyclic products of propagators on the sphere. As a genus-one generalization, we spell out certain antielliptic (i.e. antimeromorphic and doubly-periodic) functions on the torus which will be referred to as the Betti–deRham duals 7 of integration cycles on a cylinder boundary.
It will be important to collect the various eMZVs and MGFs in generating series similar to those in [18, 59–61] as the genus-one single-valued map SV is most conveniently described at the level of these generating series. The α'-expansion of genus-one closed-string integrals—using the techniques of [18]—yields an explicit form of the proposed single-valued map of the eMZVs in open-string integrals. The open-string punctures on a cylinder boundary are ordered according to the cycle which is Betti–deRham dual to the additional antielliptic functions in the closed-string integrand. For the purpose of this work, it will be sufficient to place all the open-string punctures on the same cylinder boundary which corresponds to planar genus-one amplitudes: as will be discussed in future work, single-valued non-planar open-string integrals yield the same collection of MGFs as the planar ones. Apart from a characterization via iterated Eisenstein integrals, we will arrive at a closed formula for the single-valued versions of any convergent eMZV that straightforwardly yields the familiar lattice-sum representations of MGFs.
The main evidence for our proposal for an elliptic single-valued map stems from its consistency with holomorphic derivatives in the modular parameters τ of the surfaces and the degeneration τ → i∞ of the torus to a nodal sphere. Compatibility with the holomorphic derivative is a simple consequence of recent results on the differential equations of genus-one open-string integrals [59, 60] and closed-string integrals [61] in τ. Our antielliptic integrands on the torus ensure that the closed-string differential equations match those of the open string apart from the disappearance of ζ2 as expected from the single-valued map of MZVs. Moreover, the antielliptic integrands are engineered such as to reproduce Parke–Taylor factors in the degeneration τ → i∞. Hence, compatibility of the single-valued maps at genus zero and one is supported by the identification of sphere integrals as single-valued disk integrals [40–45]. The logic of our construction is illustrated in figure 1.
1.1. Summary of main results
The main result of this work is the proposal
for a single-valued map SV at genus one which relates generating series and of open- and closed-string integrals, respectively, see (3.35). As summarized in figure 1, this induces an SV action on the eMZVs in the α'-expansion of the cylinder integrals to be defined in (3.1). By comparing coefficients of dimensionless Mandelstam invariants α'ki ⋅ kj and formal expansion variables ηj , SV maps each eMZV generated by to combinations of MGFs at the same order in the analogous expansion of the torus integrals in (3.13). The integrands of and are assembled from combinations of doubly-periodic Kronecker–Eisenstein series in (2.13) known from [18, 59–61] and antielliptic functions that we introduce in (3.8) as tentative Betti–deRham duals of integration cycles on a cylinder boundary.
A key motivation and evidence for this construction stems from the degeneration limit τ → i∞ of the series and . Within this limit, genus-zero integrals similar to those in open- and closed-string tree-level amplitudes are recovered, the latter being related by the single-valued map of MZVs [40–45]. The leading terms of the eMZVs in the τ → i∞ limit of are certain Laurent polynomials in the modular parameter τ of the cylinder with MZVs in its coefficients. As visualized in the lower part of figure 1, the known single-valued map sv of MZVs [46, 48] is conjectured to yield the analogous Laurent polynomials in the degeneration limit τ → i∞ of the torus integrals . This is made precise in the conjecture (3.23)—a central prerequisite for (1.1)—which generalizes earlier observations in [51, 52] and has been proven at the leading orders in the formal expansion variable ηj at two points [16].
The earlier proposal for an elliptic single-valued map 'esv' in [51] concerns the full τ-dependence of certain generating series of eMZVs or the composing iterated Eisenstein integrals. This reference associates open-string prototypes (i.e. esv preimages) to the simplest closed-string integrals at genus one whose integrands are solely built from Green functions involving any number of punctures. On the one hand, the proposal for the single-valued map in [51] is contained in (1.1) upon symmetrizing over the integration cycles on its left-hand side and extracting the lowest order in ηj . On the other hand, the implementation of the single-valued map at the level of iterated Eisenstein integrals in the reference is very different from the proposal in the present work. In comparison to the proposal of [51], our SV action on iterated Eisenstein integrals in (4.30) does not necessarily generate real combinations and is therefore applicable to imaginary cusp forms and MGFs of different holomorphic and antiholomorphic modular weights. Moreover, in contrast to esv in [51], the SV map in (1.1) is observed to be compatible with shuffle multiplication in all known examples. Our SV map additionally introduces combinations of (conjecturally single-valued) MZVs and antiholomorphic terms, which are absent in [51].
At the time of writing, the antiholomorphic admixtures introduced by our SV map on iterated Eisenstein integrals at depth ⩾ 2 can only be fixed by indirect methods beyond the reach of open-string data. Instead, the explicit form of the SV action on iterated Eisenstein integrals has so far been extracted from the reality properties of closed-string generating series in [18] that extend the series as described below. However, this limitation does not affect the formulation of our SV map at the level of the lattice-sum representation of MGFs: by virtue of the antielliptic functions in (3.8), the SV image of an arbitrary convergent eMZV given in (5.39) can be straightforwardly expressed in terms of lattice sums using the integration techniques of [8–10, 52] and, for certain weights, further simplified using the Mathematica package [62].
1.2. Outline
This work is organized as follows. We start by reviewing open- and closed-string integrals at genus zero and genus one as well as the basic definitions of single-valued MZVs, eMZVs and MGFs in section 2. Then, section 3 is dedicated to the modified generating series of open- and closed-string integrals as well as their relation through our proposed single-valued map at genus one. In particular, the central antielliptic integrands and the resulting proposal for an elliptic single-valued map can be found in sections 3.2 and 3.5, respectively. In section 4, we set the stage for generating explicit examples of single-valued eMZVs by introducing a new expansion method for open-string integrals over B-cycles and relating it to similar closed-string α'-expansions. This leads to the identifications of MGFs as single-valued eMZVs in section 5, where examples of the antielliptic integrands are related to earlier approaches to an elliptic single-valued map in the literature. The resulting lattice-sum representations of all single-valued convergent eMZVs are discussed in section 5.7. In the concluding section 6, we comment on the relation of string amplitudes to the generating series of this work and further directions.
2. Review of genus-zero and genus-one integrals
In this section, we collect background material on world-sheet integrals at genus zero and one, including the genus-zero single-valued map, and review various definitions relevant to the single-valued map at genus one.
2.1. Genus-zero integrals
We briefly review the basic disk (open-string) and sphere (closed-string) integrals for genus-zero world-sheets and how they are related by the genus-zero single-valued map.
2.1.1. Definitions of disk and sphere integrals
Massless tree-level n-point amplitudes of the open superstring [63] and the open bosonic string [64] can be expanded in a basis of iterated integrals [65]
over the boundary of a disk which we parametrize through the real line
The disk integrands involve dimensionless Mandelstam invariants
and Parke–Taylor factors
The inverse in (2.1) instructs us to set any triplet of punctures to 0, 1, ∞, where the invariance of genus-zero integrands hinges on momentum conservation . Both the domains and the Parke–Taylor integrands are indexed via permutations γ, ρ ∈ Sn of the external legs 1, 2, ..., n. One can arrive at smaller bases of (n − 3)! cycles γ and Parke–Taylor orderings ρ via monodromy relations [66, 67] and integration by parts [63, 65], respectively.
Closed-string tree amplitudes in turn can be reduced to sphere integrals
involving and permutations γ, ρ ∈ Sn of meromorphic and antimeromorphic Parke–Taylor factors subject to the same integration-by-parts relations as in the open-string case.
2.1.2. Single-valued map between disk and sphere integrals
The disk and sphere integrals (2.1) and (2.5) converge for a suitable range of the Re(sij ) and they admit a Laurent expansion in α', i.e. around the value sij = 0 of the dimensionless Mandelstam invariants (2.3). The coefficients in the α'-expansions of disk integrals Ztree are MZVs [68, 69],
whose weight n1 + n2 +⋯+ nr matches the order in α' beyond the low-energy limit (i.e. beyond the leading order in α'). The polynomial structure of the Ztree in sij can for instance be generated from the Drinfeld associator [70] or Berends–Giele recursions [71], with explicit results available for download from [72, 73].
When applying the single-valued map [46, 48] of motivic [74] MZVs 8
order by order in α', the disk and sphere integrals (2.1) and (2.5) are related by [40–45]
The first permutation γ in Ztree and Jtree refers to a disk ordering (2.2) and an antimeromorphic Parke–Taylor factor (2.4), respectively, which are connected by a Betti–deRham duality [49, 57, 58]. The key result of this work is to identify similar pairs of cycles and antimeromorphic functions at genus one.
2.2. Genus-one integrals
As a preparation for our proposal of a genus-one single-valued map, we now introduce the basic genus-one world-sheet integrals and the objects appearing in their α'-expansion.
2.2.1. Genus-one open-string A-cycle integrals
In the same way as disk integrals can be cast into a Parke–Taylor-type basis (2.1), the basis integrals for massless genus-one open-string amplitudes are claimed to be generated by [59, 60]
where we have set z1 = 0 by translation invariance. In this work we restrict to planar amplitudes with all state insertions on a single cylinder boundary (as opposed to non-planar amplitudes with punctures on both boundaries of the cylinder). We do not impose momentum conservation in a genus-one context and treat all the sij with 1 ⩽ i < j ⩽ n as independent. The ordering of the open-string punctures on a cylinder boundary is encoded in an integration domain on the A-cycle of a torus (see figure 2 for the standard parametrization) with [75]
with similar integration domains [76] for the non-planar open-string integrals.
Download figure:
Standard image High-resolution imageThe integrand of (2.9) features the open-string Green function on an A-cycle (which is chosen to enforce and [51, 77])
and the following combination of the doubly-periodic Kronecker–Eisenstein series [78]
with ηij...k = ηi + ηj +⋯+ ηk . 9 The permutation ρ ∈ Sn−1 in is taken to act on both the zj and the formal expansion variables in (2.13). The conjectural basis (2.9) is a generating function of the world-sheet integrals over the Kronecker–Eisenstein coefficients f(w)
that occur in the integrands of genus-one open- and closed-string amplitudes [4, 52, 79], e.g.
While the massless four-point genus-one amplitude of the open superstring [80] is proportional to the most singular -order of , the analogous amplitude of the open bosonic string additionally involves contributions of (and its permutations in 2, 3, 4) at the orders of [76]. 10 The short-distance behavior introduces kinematic poles into the α'-expansion of (2.9), and the remaining f(w≠1)(z, τ) are regular for any .
2.2.2. Genus-one closed-string integrals
In the same way as (2.9) is claimed to be a universal basis of genus-one open-string integrals, the integrals over the torus punctures for massless genus-one amplitudes in type II, heterotic and bosonic string theories should be generated by [61]
with z1 = 0. The remaining zj are integrated over the torus with modular parameter . The closed-string Green function
is chosen to be modular invariant and to obey , and its holomorphic derivatives parallel those of the open-string Green function in (2.11),
where parametrize the covering space of the torus and the f(w)(z, τ) with z = uτ + v are defined by (2.15). The second arguments and of the Kronecker–Eisenstein series and their complex conjugates in (2.16) have been chosen such that each order in the ηj - and α'-expansion gives rise to modular forms of purely antiholomorphic modular weight 11 .
When assembling genus-one amplitudes of open and closed strings from the series and , it remains to dress the component integrals in their ηj -expansions with kinematic factors that carry the dependence on the external polarizations. The latter are determined from the conformal-field-theory correlators of the vertex operators, see e.g. [81, 82], and are unaffected by our proposal for the single-valued map at genus one.
2.2.3. Differential equations in τ
Based on the differential equations (2.18) of the Green functions and integration by parts in the zj , the open- and closed-string integrals (2.9) and (2.16) were shown in [60, 61] to obey the differential equations
respectively. The right-hand sides involve holomorphic Eisenstein series G0 = −1 and
as well as (n − 1)! × (n − 1)! matrices independent of τ that vanish for k = 2 and . This means in particular that G2(τ) does not appear in (2.19).
The two-point instances are
The notation k reflects the expectation that the are matrix representations of Tsunogai's derivation algebra [83] and obey relations such as (see [17, 84, 85] for similar relations at higher weight and depth)
The all-multiplicity formulae for these (n − 1)! × (n − 1)! representations in [60, 61] manifest that the are linear in the sij , i.e. proportional to α', and their closed-string analogues additionally involve terms independent of α' (with ):
2.2.4. Basic definitions of eMZVs and MGFs
We shall now review the definitions of the eMZVs and MGFs that occur as the expansion coefficients of the above genus-one integrals. The ηj - and α'-expansion of the open-string integrals in (2.9) gives rise to A-cycle eMZVs [4–6]
introduced by Enriquez [3] which are said to carry weight n1 + n2 +⋯+ nr and length r. Endpoint divergences in case of n1 = 1 or nr = 1 are shuffle-regularized as in section 2.2.1 of [4]. The specific eMZVs at a given order of in sij and ηj can be obtained from the differential equations (2.19) along with the initial values in [60] or from matrix representations of the elliptic KZB associator [86, 87].
The closed-string integrals in (2.16) in turn introduce multiple sums over the momentum lattice of a torus [52, 61]
that are known as MGFs [10, 11]. With the removal of p = 0 from Λ, they can be thought of as infrared-regulated and discretized versions of Feynman integrals on a torus. The MGFs associated with Feynman graphs of dihedral topology are defined by 12
and more general topologies are for instance discussed in [11, 62]. The simplest examples of dihedral MGFs (2.26) have two columns and are associated with one-loop graphs on the world-sheet
whereas are referred to as (r − 1)-loop MGFs. As long as the entries obey a + b > 2, the lattice sums (2.27) are absolutely convergent and the one-loop MGFs are expressible in terms of non-holomorphic Eisenstein series Ek (τ) and their Cauchy–Riemann derivatives
where ∇ = 2i(Im τ)2∂τ and . As will be detailed below, both eMZVs (2.24) and MGFs such as (2.26) can be represented via iterated integrals of holomorphic Eisenstein series defined by (2.20). Both eMZVs [17] and MGFs [11, 62, 88–90] exhibit a multitude of relations over rational combinations of MZVs, all of which are automatically exposed in their iterated-Eisenstein-integral representation 13 . A computer implementation for the decomposition of a large number of eMZVs and MGFs into basis elements is available in [62, 92], respectively.
3. New types of genus-one integrals
The goal of this paper is to relate the α'-expansions of suitable generating functions of genus-one open- and closed-string integrals. The and in (2.9) and (2.16) can be anticipated to not yet furnish the optimal building blocks for this purpose since
- (a)The τ-dependence ∼Gk (τ) and of the open- and closed-string differential equations (2.19) does not match, even in absence of .
- (b)
Both of these shortcomings will be fixed by the improved open- and closed-string generating functions and to be introduced in this section.
3.1. Genus-one open-string B-cycle integrals
Instead of parametrizing the cylinder boundary through the A-cycle of a torus as in (2.9), one can perform a modular S transformation
to attain a parametrization through the B-cycle (recalling that z1 = 0 and )
where , and the B-cycle Green function is constructed in two steps: first, we define for z on the line (0, τ) by [77]
Then, we extend this to z ∈ (−τ, 0) by imposing for compatibility with (2.11) under modular S transformations, leading to the combined expression
Instead of integrating over zi = τui with ui ∈ (0, 1), we have chosen the representative of the B-cycle in order to facilitate the comparison with genus-zero integration cycles as τ → i∞. Figure 3 illustrates the integration cycle (3.2) in both the zj and variables (the latter becoming the coordinates on the sphere as τ → i∞), where and for purely imaginary choices of τ. Note that non-planar versions of the B-cycle integrals involve additional punctures at or negative σj ∈ (−q−1/2, −q1/2).
Download figure:
Standard image High-resolution imageThe modular transformation of the doubly-periodic Kronecker–Eisenstein series (2.12) leads to the rescaling ηj → τηj in the subscript of the ρ-dependent integrand of (3.1).
3.2. Dual closed-string integrals
The doubly-periodic integrands in (2.13) are non-holomorphic, so their complex conjugates in (2.16) obey
which leads to the terms in the closed-string derivations in (2.23). This introduces a tension between the open- and closed-string differential equation (2.19) such that the do not qualify as Betti–deRham duals of open-string integration cycles. In order to generalize the interplay of Parke–Taylor factors (2.4) with single-valued integration [48, 49] to genus one, the factor of in the integrals (2.16) needs to be replaced by an antimeromorphic function that is still well-defined on the torus, i.e. the complex conjugate of an elliptic function in all of z1, z2, ..., zn .
Such elliptic functions of n punctures can be generated by cycles of Kronecker–Eisenstein series [79]
where Vw has holomorphic modular weight w. Even though the individual Kronecker–Eisenstein series Ω are not meromorphic in the zj , the Vw are elliptic functions since the non-holomorphic phase factors in (2.12) cancel from the cyclic product in (3.6). The simplest examples are
with zn+1 = z1 and Kronecker–Eisenstein coefficients f(w) defined by (2.14), also see (5.2) for the analogous expressions at general w.
As will be detailed below, these elliptic functions degenerate to suitable combinations of Parke–Taylor factors when forming the linear combinations
such as (see sections 5.4 and 5.5 for detailed discussions of the three- and four-point examples)
To lend credence to this definition of the V-function, let us see how their properties parallel those of the genus-zero case: the Betti–deRham duality at genus zero relies on the simple-pole residues
of the Parke–Taylor factors (2.4). These residues correspond to the situation when two neighboring points of the disk ordering (2.2) at zj = zj±1 come together, which is crucial for sphere integrals being single-valued disk integrals [43, 45].
Similarly, at genus one, the generating function (3.6) of the elliptic Vw functions exposes the recursive structure of their simple-pole residues
and the absence of higher poles in zj − zj±1. Consequently, the pole structure of the elliptic combinations (3.8)
mirrors the boundaries of the open-string integration cycles as zj = zj±1, i.e. one recovers mutually consistent V-functions and cycles at multiplicity n − 1 in both cases 14 .
The absence of Vw with w ⩾ n − 1 in (3.8) can be understood from
- The vanishing of Vn−1(1, 2, ..., n|τ) since (3.6) would otherwise be an elliptic function of ξ with a simple pole at the origin [79]
- The breakdown of uniform transcendentality when expanding Koba–Nielsen integrals involving Vn (1, 2, ..., n|τ) [52] (which is in tension with the transcendentality properties of open-string integrals [60])
- The fact that Vw⩾n+1(1, 2, ..., n|τ) is expressible in terms of Gw−k Vk (1, 2, ..., n|τ) with k ⩽ n − 2 [79]
Similar to the closed-string integrals , we define an (n − 1)! × (n − 1)! matrix of torus integrals
indexed by permutations γ, ρ ∈ Sn−1 of (3.8) and (2.13). Note that the cyclic symmetry
exposed by the generating function (3.6) has been used to bring the integrand of (3.13) into the form of .
3.3. Asymptotics at the cusp
The modular S transformation in (3.1) maps the A-cycle eMZVs (2.24) in the ηj - and α'-expansion of to B-cycle eMZVs [3] in the analogous expansion of . As detailed in [51, 77, 93], the asymptotic expansion of B-cycle eMZVs as τ → i∞ is governed by Laurent polynomials in whose coefficients are -linear combinations of MZVs, for instance
The suppressed terms are series in q = e2πiτ = e2iT with Laurent polynomials in T as their coefficients.
The MGFs (2.26) in the ηj - and α'-expansion of (2.16) admit similar expansions around the cusp, where the leading term is a Laurent polynomial in y = π Im τ instead of T. The coefficients in the Laurent polynomials of MGFs were shown to be -linear combinations of MZVs 15 [15] and are conjectured to be single-valued MZVs [10, 12]. Simple examples of the asymptotics of MGFs include
see (2.28) for the lattice-sum representations of the non-holomorphic Eisenstein series.
In a variety of examples, the Laurent polynomials of MGFs and B-cycle eMZVs have been related by an extension of the single-valued map (2.7) to [16, 51, 52]
By (3.15) and (3.16), for instance, the Laurent polynomials of as well as and are related by (3.17).
The A-cycle eMZVs in , by contrast, enjoy a Fourier expansion in q = e2πiτ whose coefficients are combinations of MZVs [3, 17] and do not feature any analogues of the Laurent polynomials in the expansion of . This is yet another indication besides their differential equations that the B-cycle integrals (3.1) are a more suitable starting point for comparison with closed-string integrals than their A-cycle counterparts (2.9).
3.4. Single-valued correspondence of the Laurent polynomials
As a particular convenience of the elliptic combinations (3.8) in the integrands of , their degeneration at the cusp gives rise to Parke–Taylor factors in n + 2 punctures (σ1 = 1 by z1 = 0)
Since the non-holomorphic exponentials of cancel from the cyclic products in (3.6), one can determine the asymptotics of V(...|τ) as τ → i∞ by using the degeneration of the holomorphic Kronecker–Eisenstein series
The relative factors of the Vw in (3.8) have been engineered to obtain the following cyclic combinations of Parke–Taylor factors at the cusp,
which have featured in the context of one-loop gauge-theory amplitudes in ambitwistor string theories [94]. The denominators on the left-hand side of (3.20) arise from , and the factor of |σ−|2 on the right-hand side identifies functions on a degenerate torus with SL2-fixed expressions at genus zero [65]. Given that Parke–Taylor factors are Betti–deRham dual to disk orderings in (2.2), the τ → i∞ asymptotics of should yield the single-valued map of suitably chosen disk integrals. In fact, upon rewriting the B-cycle ordering in terms of the σj variables (3.18), each of the contributions in (3.2) and figure 3 degenerates to a single disk ordering
such that the overall B-cycle ordering at the cusp becomes the Betti–deRham dual to the cyclic combination of Parke–Taylor factors in (3.20),
Hence, the tree-level result (2.8) provides evidence for our central conjecture
where the notation |LP instructs to only keep the Laurent polynomials in τ and Im τ while discarding any contribution . The conjectural part of (3.23) concerns the non-constant terms in the Laurent polynomials, i.e. corrections ∼(log q)±1 to the expansion around the cusp q = 0, so it is not implied by the Betti–deRham duality of (3.2) and (3.20) which only holds at the cusp. That is why we support (3.23) by extensive tests at low orders in ηj , α' as detailed below, and by the fact that the asymptotic expansions of the Green functions are related by the single-valued map with sv log σij = log|σij |2,
Note that the absolute value in (3.24) is due to the argument |u|τ of θ1 in (3.4). For the two-point instances and of the open- and closed-string integrals (3.1) and (3.13), the Laurent polynomials in the asymptotics at the cusp can be determined [95] by mild generalizations of the techniques in [16, 55] (also see [56] for an alternative approach to the closed-string case):
These two-point expressions are easily seen to line up with the all-multiplicity claim (3.23) since
and the last line of (3.26) therefore vanishes under sv. Moreover, we have checked the three-point Laurent polynomials to obey (3.23) to the orders in the sij - and ηj -expansions where MGFs such as (2.26) of total modular weight 10 occur 16 . Finally, we have checked (3.23) to hold at four points to the orders where MGFs of total modular weight 8 occur, at least for contributions from in (2.13) without any singular factors of f(1)(zij , τ). 17 These checks are based on Enriquez' methods [3] (also see appendix B of [51]) to determine the Laurent polynomials of B-cycle eMZVs. The Laurent polynomials for all B-cycle eMZVs with (length + weight) ⩽16 obtained from an FORM implementation [96] of these methods are available for download [97].
While the two-point Laurent-polynomials generated by (3.26) and (3.27) only involve Riemann zeta values, higher-point examples also introduce irreducible MZVs of depth ⩾ 2. The appearance of ζ3,5 in B-cycle Laurent polynomials is later on exemplified in (4.27) and (5.37). Moreover, the appearance of ζ3,5,3 in open- and closed-string calculations at three points in agreement with (3.23) was observed in section 3.3.5 of [51], based on earlier closed-string computations [12].
3.5. Single-valued correspondence of the differential equations
The holomorphic derivatives of the and -integrals (3.1) and (3.13) can be easily deduced from (2.19): in the open-string case, the modular S transformation relating and the modular weight (k, 0) of the holomorphic Eisenstein series Gk give rise to
see [60] for the n-point derivations and (2.21) for their two-point examples. Since the single-valued map at genus zero acts on transcendental constants, we have passed to the differential operator in comparison to (2.19) and in preparation for the extended single-valued map to be introduced below around (3.32).
In the closed-string case, the in (3.13) are not affected by holomorphic derivatives, and one can import a simplified version of the differential equations in [61] where contributions are absent,
By the differential equation (2.18) of the Green functions, also the term ∼ζ2 in is absent which we have indicated through the sv notation,
where . The building blocks of the closed-string differential operator in (3.30) are related to those in the open-string analogue (3.29) through an extension SV of the single-valued map (recall that T = πτ and y = π Im τ)
which is taken to preserve the properties of sv,
and to furthermore preserve (2πi)−k Gk (τ) (the inverse powers of π ensuring rational coefficients in the q-expansion) and the ηj -variables, cf (3.32). In other words, the differential operator appearing in (3.29) and its closed-string analogue in (3.30) are related by
From the above discussion, both the τ → i∞ asymptotics and the differential operators of the open- and closed-string integrals and are related by the SV map (3.33). Hence, we propose that the solutions of (3.34) yield an appropriate extension of the SV map
This proposal is key the result of this work, relating the open-string integrals in (3.1) with integration ordering γ to the closed-string integrals , where the ordering γ governs the singularity structure of the antielliptic integrand in (3.13). By construction, this SV map commutes with the holomorphic τ-derivative and, under the assumption (3.23), it is consistent at the level of the Laurent polynomials at the cusp. Compatibility of a single-valued map at genus one with τ → i∞ generalizes the fact that the single-valued map of multiple polylogarithms commutes with evaluation [46]. Moreover, by the evidence to be discussed in section 4.3, the SV map is expected to be compatible with the shuffle product. As we will see in the next section, the α'-expansion of (3.35) induces an elliptic single-valued map for the eMZVs generated by which yields the MGFs generated by .
Let us consider the scope of our definition (3.35). Firstly, not all the holomorphic iterated Eisenstein integrals appear in the α'-expansion of . As was discussed in [17, 18] and will become clearer when we discuss the α'-expansion of the solution of (3.34), relations among the such as (2.22) lead to dropouts of certain iterated Eisenstein integrals from eMZVs and and thereby from and . Hence, (3.35) does not comprise the SV map for the combinations of iterated Eisenstein integrals affected by these dropouts, starting with double integrals involving G4 and G10.
By contrast, the SV map of arbitrary convergent eMZVs can be extracted from (3.35) at sufficiently high multiplicity: as will be detailed in section 5.7, for ω(n1, ..., nr |τ) in (2.24) with given entries nj (where n1, nr ≠ 1), one can engineer a combination of genus-one open-string integrals, where the desired eMZV occurs at the zeroth order in sij . 18
Finally, one could wonder whether holomorphic cusp forms lead to ambiguities in the definition of the V(1, ..., n|τ) in (3.8): 19 Starting from n = 14 points, their defining properties including simple-pole residues, the modularity of their constituents and their behavior (3.20) at the cusp are unchanged when adding combinations of holomorphic cusp forms and lower-weight Vw (1, ..., n|τ). However, adding a cusp form without any zj -dependent coefficient to V(1, ..., n|τ) leads to a contradiction with the requirement that the τ-independent η1−n order of is mapped to the same term ∼η1−n in the corresponding integral. Products of Vw (1, ..., n|τ) with cusp forms in turn would violate the pole structure (3.12) that reflects the boundary structure of the dual cycles. Hence, the above requirements do not leave any room to modify V(1, ..., n|τ) by holomorphic cusp forms.
3.6. Dual modular weights for cycles
Given that the antielliptic -functions (3.6) carry modular weight (0, w), their combinations (3.8) mix different modular weights. Hence, the α'-expansion of the generating function (3.13) with in the integrand mixes modular forms of different weight, even at fixed order in ηj . One may wish to isolate the contributions at fixed modular weights and study
with 0 ⩽ w ⩽ n − 2, where the terms at homogeneity degree m in the ηj are modular forms of weight (0, 1 − n − m + w). One can still identify combinations of integration cycles (3.2) to write (3.36) at fixed modular weight w and ordering γ as the single-valued version of known open-string integrals: each Vw (1, 2, ..., n|τ) with w ⩽ n − 2 is expressible via permutation sums
with coefficients , e.g.
These relations and coefficients cw,γ can be traced back to the symmetries of the Vw -functions including the cyclicity (3.14), the reflection property
and corollaries of the Fay identity [98] which have been discussed in [79, 99]. An independent method based on the degeneration (3.20) to determine the cw,γ
is described in appendix
Table 1. Examples of the unsigned Stirling numbers Sn−1,n−w−1 which count the number of independent permutations γ ∈ Sn−1 of Vw (1, γ(2, ..., n)|τ).
w | ||||||
---|---|---|---|---|---|---|
n | 0 | 1 | 2 | 3 | 4 | 5 |
2 | 1 | 0 | 0 | 0 | 0 | 0 |
3 | 1 | 1 | 0 | 0 | 0 | 0 |
4 | 1 | 3 | 2 | 0 | 0 | 0 |
5 | 1 | 6 | 11 | 6 | 0 | 0 |
6 | 1 | 10 | 35 | 50 | 24 | 0 |
7 | 1 | 15 | 85 | 225 | 274 | 120 |
In particular, permutations of Vw=n−2(1, ..., n|τ) are related by Kleiss–Kuijf relations [99, 100]
such as
consistent with the counting Sn−1,1 = (n − 2)! of independent permutations.
Given the decomposition (3.37) of a given Vw function with rational coefficients cw,γ , one can by (3.35) write each integral (3.36) as a combination of single-valued B-cycle integrals
For instance, the equivalent
of (3.35) together with (3.38) and (3.39) suggests to assign a formal 'dual modular weight' 0 and 1 to the symmetric and antisymmetric three-point cycles, respectively,
Similarly, combining (3.35) with (3.38), (3.40) and (3.41) leads to the following dual modular weights (d.m.w.) for four-point cycles
see section 5.5 for a more detailed discussion of the weight-two case. Finally, the all-multiplicity formula (3.37) translates into
see appendix
4. Single-valued iterated Eisenstein integrals from α'-expansions
The goal of this section is to provide the explicit form of the single-valued map SV for the iterated-Eisenstein-integral representation of eMZVs [17] by reading (3.35) at the level of the α'- and ηj -expansions of and . We will employ the formulation of iterated Eisenstein integrals with integration kernels τj Gk , k ⩾ 4 [101],
The entries are taken to obey kr ⩾ 4 and 0 ⩽ jr ⩽ kr − 2, and we use tangential-base-point regularization for the divergences as qr → 0 [101], which implies that the iterated Eisenstein integrals vanish in the regularized limit τ → i∞.
4.1. Improving the differential equations
We shall now derive the structure of the α'-expansion of and by repeating the key steps of [18] in solving the differential equation (2.19) of . The first step is to introduce redefined generating series and by
After this redefinition, the k = 0 terms involving and with G0 = −1 are absent from the analogues of the differential equations (3.29) and (3.30), see (4.4) and (4.5) below. Throughout this section, we suppress the permutations γ, ρ labeling and , and all matrix representations are understood to act matrix-multiplicatively on the ρ-entry.
Since the are expected (as tested for a wide range of k and n) to inherit the ad-nilpotency relations of the derivation algebra,
the combinations in the differential equations of the redefined integrals (4.2) truncate to a finite number of terms and we obtain
We have used that (since the term ∼ζ2 in (3.31) suppressed by sv is commutative) and employ the shorthands
In equations (4.4), (4.5) and below, we use the derivative with respect to τ instead of log q as compared to section 3.5 above.
4.2. The α'-expansion of
By the differential equation
of the iterated Eisenstein integrals (4.1), one can solve the differential equation (4.4) of the generating series through the path-ordered exponential
for some initial value to be discussed below. By inverting the redefinition (4.2) and moving the exponential to act directly on the initial value, we obtain the open-string analogue
of the key result for the α'-expansion of in (3.11) of [18]. In commuting past the , the iterated Eisenstein integrals are rearranged into the combinations
and more generally
Note that (4.9) is an alternative 20 organization of open-string α'-expansions at genus one as compared to [59, 60]. Non-planar B-cycle integrals obey the same differential equation (3.29) as the planar ones and therefore have an α'-expansion of the same form (4.9), only their initial values need to be adapted to the non-planar integration cycle.
The modified iterated Eisenstein integrals β[...] in (4.11) satisfy the differential equations
which allows us to directly check that (4.9) obeys (3.29). The integrals β[...] inherit the property that they vanish for τ → i∞ from the . Note that the definition (4.11) is equivalent to integral representations such as
The definition (4.11) of the β[...] preserves the shuffle relations of the iterated Eisenstein integrals (4.1), for instance
4.3. The α'-expansion of
One can extend the above strategy to expand via (4.4) to the integrals. The idea is to solve their differential equation (4.5) order by order in α' via
using the combinations of holomorphic iterated Eisenstein integrals (4.1) and their complex conjugates introduced in [18]. Their depth ℓ = 1 instances are completely known from the reference
and their generalizations to depth ℓ ⩾ 2 involve antiholomorphic integration constants ,
The integration constants are invariant under τ → τ + 1 since the and the contributions from the to (4.17) are. They are known on a case-by-case basis, for instance
and the complete list of at k1 + k2 ⩽ 12 can be found in an ancillary file on the journal website of this work. The integration constants at arbitrary depth can be determined from the reality properties of the integrals [18]. The method in the reference to fix the hinges on the fact that the coefficients in the ηj - and -expansion of are closed under complex conjugation. For the n-point -series in turn the antiholomorphic modular weights of the integrands in (3.13) are bounded by , so the complex conjugates of higher orders in the ηj -expansion are not part of the series. Hence, in the present formulation, the expansion of the in [18] is a necessary input to obtain well-defined . This expansion depends on the knowledge of the initial values of which is currently available from sphere integrals to arbitrary weight only for two points and is under investigation for higher multiplicity [95]. Still, the torus-integral- and lattice-sum representations of single-valued eMZVs in section 5.7 do not require any knowledge of and .
By repeating the steps toward (4.8) and (4.9), we arrive at the structure of the α'-expansion
with an initial value to be discussed below and the combinations analogous to (4.11) [18]
The expansion (4.19) solves (3.30) since the βsv inherit their differential equation from (4.15),
see (4.12) for the holomorphic counterpart for ∂τ β[...]. Both the and the βsv[...] are expected to preserve the shuffle multiplication of their holomorphic counterparts (4.1) and (4.11): the differential equations (4.15) and (4.20) recursively imply that shuffle relations among and the βsv[...] can at most be violated by antiholomorphic functions such as the integration constants in (4.17). 21 All examples of up to including k1 + k2 = 12 were checked to preserve the shuffle relations, and their explicit form can also be found in an ancillary file to the arXiv submission of this work. Note that these checks cover the more intricate cases with (k1, k2) = (4, 6) and (k1, k2) = (4, 8) where imaginary cusp forms occur among the MGFs [18, 62].
The and βsv are expected to occur in Brown's generating series of single-valued iterated Eisenstein integrals [13, 14, 101]. The construction of non-holomorphic modular forms in the references—so-called equivariant iterated Eisenstein integrals—are obtained by augmenting their single-valued counterparts by combinations of MZVs and objects of lower depth. At depth one, the equivariant iterated Eisenstein integrals are non-holomorphic Eisenstein series along with their Cauchy–Riemann derivatives [13, 14, 101]. From their representation [18]
the βsv are seen to take the role of the single-valued rather than equivariant iterated Eisenstein integrals at depth one. At higher depth, the precise relation of the βsv to Brown's construction is an open question at the time of writing.
4.4. Initial values
It remains to specify the initial values and in the α'-expansions (4.9) and (4.19). The Laurent-polynomial contributions from the asymptotics (3.24) and (3.25) of the Green functions are still functions of τ and need to be translated into series that solely depend on ηj and sij . Following the construction of a similar initial value for in section 3.4 of [18], we import the constant parts ∼τ0 and ∼(Im τ)0 of the respective Laurent polynomials
In both cases, the exponentials ensure that the negative powers of T in and y in disappear order by order in α'. Hence, (4.23) and (4.24) pick up the lowest powers of T, y present in and . The leading α'- and η2-orders of the two-point initial values following from the expressions in (3.26) and (3.27) are
as well as
Higher orders in s12 and η2 are readily available through the straightforward expansion of the exponentials and Γ-functions in (3.26) and (3.27). In particular, these two-point expressions imply that all the coefficients in the s12- and η2-expansions are combinations of Riemann zeta values for and odd Riemann zeta values for .
Starting from n = 3 points, the initial values will also feature irreducible MZVs of higher depth. Based on Enriquez' method to generate the Laurent polynomial of B-cycle eMZVs [3] (also see appendix B of [51]) we have determined the three-point initial values to certain orders, and the results are included in an ancillary file to the arXiv submission of this article. To the orders under consideration, we find the following coefficients of ζ3,5
which by the single-valued maps sv ζ3,5 = −10ζ3 ζ5 and sv ζ3 ζ5 = 4ζ3 ζ5 enter the closed-string initial values via
The coefficients of ζ3,5 and ζ3
ζ5 in (4.27) and (4.28) are extracted after reducing all MZVs at weight 8 to -linear combinations of [102]. Similarly, the MZV ζ3,5,3 seen in Laurent polynomials of both B-cycle integrals [51] and modular graph functions [12] will occur in both and . The contributions to involving MZVs of weight up to and including four can be found in appendix
Since the initial values are obtained from the Laurent polynomials and the exponents in (4.23) and (4.24) are related by the single-valued map, the conjecture (3.23) supported by tree-level results and extensive genus-one tests is equivalent to
in agreement with (4.25) and (4.26).
The τ → i∞ asymptotics of n-point A-cycle integrals (2.9) has been expressed in terms of (n + 2)-point disk integrals (2.1) in suitable kinematic limits [60]. Similarly, the Laurent polynomials of n-point genus-one integrals are determined by genus-zero integrals at multiplicity n + 2 and below, see (3.26) and (3.27) for the explicit two-point result. As will be further investigated in [95], the main challenge is to determine the admixture of lower-point genus-zero integrals that generalize the subtraction of and from the Γ-functions in (3.26) and (3.27).
4.5. The single-valued map on iterated Eisenstein integrals
The proposed single-valued map (3.35) can now also be studied at the level of the α'-expansions. Using (4.29) and (3.31), we find that one obtains as the single-valued version of if the coefficients obey
This follows from the relation (4.29) among the initial values and the form of the operators in the respective α'-expansions, recalling that .
On the one hand, (4.30) fixes the single-valued map of the eMZVs in the expansion of that enter through the iterated Eisenstein integrals β[...]. On the other hand, (4.30) only applies to the combinations β[...] and βsv[...] that occur in the path-ordered exponentials (4.9) and (4.19). The SV map of individual remains undetermined whenever the relations in the derivation algebra such as (2.22) lead to dropouts of certain β[...] and βsv[...] from and (starting with cases at (k1, k2) = (10, 4) at depth ℓ = 2 and any instance where ji > ki − 2).
Since the factors of and in (4.11) and (4.20) are furthermore related by SV, (4.30) is equivalent to
again up to cases where the relations in the derivation algebra cause dropouts. For instance, (4.16) implies that the single-valued version of holomorphic Eisenstein integrals (4.1) at depth one is given by
where the contributions on the right-hand side can be recognized as
respectively. Note that (4.30) fixes the SV map of all the at depth one with k ⩾ 4 and 0 ⩽ j ⩽ k − 2 since the caveats related to relations in the derivation algebra beyond (4.3) only affect iterated Eisenstein integrals of depth ℓ ⩾ 2.
The iterated Eisenstein integrals defined in (4.1) may be reorganized in terms of [17]
with and . By subtracting the zero mode of the holomorphic Eisenstein series
the integrals (4.34) are made to converge if k1 > 0, and all other cases are shuffle-regularized based on .
At depth one, they are related to the holomorphic iterated Eisenstein integrals via [51]
with Bernoulli numbers Bk . From this, we see that the implicit action of SV on these functions at depth one is given by
By (4.32), (4.36) and the shuffle relation
two equivalent formulations of (4.37) are
which match the expectation from [103].
5. Examples
We shall now spell out a variety of examples that illustrate both the pairing of cycles with dual antielliptic integrands and the action of the single-valued map on eMZVs. Both eMZVs and MGFs occur in the simultaneous expansion of the generating series in sij and ηj which results in a lattice-sum representation of all convergent in section 5.7. The coefficients in the ηj -expansions will be referred to as component integrals, and we will use the shorthand
for the Kronecker–Eisenstein coefficients defined by (2.14) that occur in the integrands. More precisely, the building blocks (2.13) and (3.6) of the integrands of involve the following combinations of (5.1) with ηj,j+1...n = ηj + ηj+1 +⋯+ ηn ,
5.1. Two-point α'-expansions
At two points, the general definitions (3.1) and (3.13) only admit a single permutation of the integrands and cycle in and ,
and we introduce the following notation for component integrals
Then, combining the initial values (4.25) and (4.26) with the α'-expansions (4.9) and (4.19) yields expressions like
as well as
upon extracting suitable powers of η2. The action of SV on the ζn , T, β[...] as in (3.33) and (4.30) relates as expected from (3.35). Examples of β[...] beyond depth one occur at the next orders in sij , e.g.
as well as
5.2. Extracting single-valued eMZVs
The above α'-expansions at two points have been generated in earlier work in terms of eMZVs [51, 59] and MGFs [10, 11, 61], respectively. The results in the references include
where for instance (by comparison with (5.5) and (5.6))
and the modular transformation may be evaluated to yield [51, 93]
For the -order in (5.9), the component version of implies that
and a similar analysis for higher orders in s12 and at a ≠ 0 yields for instance
At depth two, relating (see (5.7) and (5.8)) or yields
where the combinations
of MGFs (2.26) are engineered to avoid G8 in the differential equations [51]. The systematics of depth-one relations between eMZVs and non-holomorphic Eisenstein series including higher-weight generalizations of (5.13) is detailed in appendix
for combinations of eMZVs of different length that are weighted by Bernoulli numbers Bj
[104]. Similarly, the analogue of (5.14) for the MGF and its holomorphic derivatives is spelled out in appendix
Note that the single-valued map of A-cycle eMZVs at argument τ rather than generically leads to combinations of MGFs of different modular weights. For instance, changing the argument to τ in (5.12) gives rise to
instead of a single modular invariant E2(τ). This can be seen by expressing all of ω(0, 0, 2|τ), E2(τ) and in terms of convergent iterated Eisenstein integrals (4.34) and applying their single-valued map (4.39). 22 Alternatively, (5.17) can be deduced by setting in (5.11) and exploiting the result that will be extracted from a four-point example in section 5.5. The much cleaner result (5.12) for as compared to SV ω(0, 0, 2|τ) is another manifestation of the fact that the differential equation (3.29) of B-cycle integrals are more closely related to the closed-string counterparts (3.30) than the A-cycle differential equations in (2.19).
5.3. Symmetrized cycles and graph functions
At n ⩾ 3 points, most of the antielliptic functions in (3.9) introduce non-constant into the closed-string integrands, except for the simplest case V0(1, 2, ..., n|τ) = 1 dual to a permutation sum over B-cycles, see (3.38),
As indicated by the *-notation, the integral (3.36) on the left-hand side is independent of the ordering * since its integrand V0 is. The symmetrized open-string integrals on the right-hand side were studied in [16, 51, 77] as the generating series of holomorphic graph functions,
where each puncture is integrated independently over the entire B-cycle. More specifically, the references considered the components at the most singular order in the ηj ,
where the dependence on the permutation ρ drops out, and the integrands at fixed order in sij are polynomials in B-cycle Green functions. In passing to the second line, each monomial in is mapped to a graph Γ that labels the B-cycle graph functions B[Γ], where a factor of is represented by an edge connecting vertices zi and zj . One-particle reducible graphs Γ1PR lead to vanishing B[Γ1PR] since , i.e. higher orders of (5.20) stem from all combinations of one-particle irreducible graphs with four and more edges in total. Any B[Γ] is expressible in terms of B-cycle eMZVs [51] since the α'-expansion of each component integral of the series is.
Similarly, modular graph functions D[Γ] (as opposed to MGFs) were defined [10, 11] by n-point torus integrals over monomials in , where each torus Green function is again visualized through an edge between vertices zi and zj . The D[Γ] associated with dihedral graphs Γ are proportional to the lattice sums (2.26) with aj = bj , and also more complicated graph topologies can be straightforwardly translated into lattice sums.
The generating series of n-point modular graph functions resides at the most singular order of (5.18) w.r.t. ηj where the insertions of are absent,
and we have D[Γ1PR] = 0 by . As a consequence of (5.18) at the most singular order in the ηj , modular graph functions are single-valued B-cycle graph functions,
which ultimately follows from the 'Betti–deRham duality' between V0 = 1 and the symmetrized cycles .
The relations in (5.22) have firstly appeared in [51] with a proposal 'esv' for an elliptic single-valued map in the place of SV. The esv map of [51] has the same action (3.33) on MZVs and Laurent polynomials in τ as the SV map in this work. In particular, all pairs of B-cycle eMZVs and modular graph functions related via in the reference are also related via as a consequence of (5.22). For suitable representations of the q-series of eMZVs via defined by (4.34), the Fourier expansions of all modular graph functions up to weight six could be reproduced from the replacement prescribed by esv [51]. However, it was an open problem in the reference to reconcile esv with the shuffle property of iterated Eisenstein integrals. The SV action (4.30) in turn is expected to be compatible with the shuffle multiplication of the β[...] and βsv[...] by the discussion in section 4.3 as detailed below (4.21).
Note that subleading orders in the ηj -expansion of (5.18) generate infinite families of additional relation between MGFs and single-valued eMZVs beyond (5.22). The comparison of open- and closed-string integrals with additional insertions of identifies MGFs of various modular weights as single-valued B-cycle eMZVs.
5.4. Three-point cycles and
The simplest instance of with non-constant antielliptic integrands occurs at three points. The single-valued map relates an antisymmetric integration cycle on the open-string side in (3.46) to the closed-string integral
Since contributions with even a + b integrate to zero, the simplest component integrals involve permutations of or at the orders of or ,
which furnish the simplest examples of kinematic poles in a -series. The corresponding antisymmetrized B-cycle integral features the same types of kinematic poles in component integrals involving , e.g. 23
We have used that, by the antisymmetry of the integrand, the contribution from the ordering is minus that of the ordering . Comparison of (5.25) with (5.24) confirms the relation (3.46) under the SV map at the respective orders in sij and ηj . Up to the restriction of the Koba–Nielsen factor to three instead of five punctures, (5.24) and (5.25) are the type of integrals over seen in genus-one five-point amplitudes of type II superstrings [105, 106].
5.5. Four-point cycles and
The esv map [51] has also been applied to the four-gluon amplitude of the heterotic string [52], where the torus integral 24
was related to the open-string integration cycle dual to (3.39). More specifically, the MGFs in [52]
were proposed to be the single-valued versions of the eMZVs in the α'-expansion of
namely [4]
As indicated by , the α'-expansions (5.27) and (5.29) have been obtained in the limit of four-point on-shell kinematics with two independent Mandelstam invariants instead of six. However, the relation (3.35) between n-point closed-string and single-valued open-string integrals is conjectured to be valid for the independent Mandelstam variables {sij , 1 ⩽ i < j ⩽ n} with sij = sji . At four points, the corollary
of the relation (3.41) between V2(1, 2, 3, 4|τ) and permutations of the V(1, 2, 3, 4|τ) functions is claimed to hold for all of {s12, s13, s23, s14, s24, s34} independent. The coefficient of in (5.30) with ρ = 2, 3, 4 then implies
and explains the relations between the α'-expansions (5.27) and (5.29) observed in [52] in the on-shell limit . In particular, the prescription (4.30) for the single-valued map of the iterated-Eisenstein-integral representation of produces the complete -expansion of the MGFs in (5.27), whereas certain antiholomorphic contributions could not be reproduced by esv in [52].
By applying (5.31) at the level of the α'-expansions (5.27) and (5.29), one can infer
Moreover, higher orders in the ηj -expansion of (5.30) yield infinite families of relations between the α'-expansions of open- and closed-string integrals over additional factors .
5.6. Imaginary cusp forms and double zetas
We shall finally exemplify the appearance of cuspidal MGFs from single-valued open-string integrals whose Laurent polynomial at the order of vanishes. A systematic study of imaginary cusp forms among the two-loop MGFs can be found in [54], also see [62] for examples of real cusp forms. The simplest imaginary cusp forms occur among the lattice sums (2.26) at modular weights (5, 5) whose basis can be chosen 25 to include [18]
The βsv-representations involve double-integrals over G4 G6 [18],
and the associated integration constants can be found in the reference and in an ancillary file within the arXiv submission of this article. Both B2,3, B2,3' and their Cauchy–Riemann derivatives drop out from and at two points. At three points, one can identify their derivatives as single-valued eMZVs,
by inspecting the contributions of or to and to . The appearance of the undifferentiated B2,3 and B2,3' is relegated to the -series at four points (or the -series at three points [18]), and comparison with the B-cycle integrals yields
The open-string counterparts of B2,3, B2,3' and their Cauchy–Riemann derivatives involve the simplest combinations of B-cycle eMZVs with an irreducible ζ3,5 in their Laurent polynomials: the methods of [3] (also see appendix B of [51]) yield the following examples of τ → i∞ degenerations in (5.35) and (5.36),
One can see from the order of T−3 or y−3 that the cuspidal nature of B2,3 hinges on the depth-two result sv ζ3,5 = −10ζ3 ζ5. The non-vanishing Laurent polynomial of (π∇)3 B2,3' is due to the real MGFs in (5.33).
Note that the simplest instances of ζ3,7 and ζ3,5,3 arise in the Laurent polynomials of B-cycle eMZVs with MGFs (2.26) of weights and 14 in their SV mage. The appearance of ζ3,5,3 in modular graph functions and eMZVs can be found in [12, 51], respectively. While ζ3,7 drops out from MGFs under the single-valued map, it enters for instance the T0-order of the Laurent polynomial of whose SV image contributes to the quantity (π∇)4 B'2,4 in section 9.2 of [62]
One can eventually find all -independent MZVs 26 in the Laurent polynomials of B-cycle eMZVs. This follows from both the degeneration limits of the elliptic KZB associator [3] and from the fact that any MZV is expressible via -linear combinations of multiple modular values [107].
Note that the Laurent polynomials of all B-cycle eMZVs with length + weight ⩽ 16 obtained from an FORM implementation [96] of the methods of [3, 51] are available for download from [97].
5.7. Single-valued map of individual eMZVs
While the above combinations of single-valued eMZVs were tailored to obtaining a single MGF in the bases of [18, 51], we shall now give a closed formula for the single-valued map of individual eMZVs. The integrands of convergent A-cycle eMZVs (2.24) with length r and n1, nr ≠ 1 arise at the -order of the series at r + 1 points. After modular S transformation, one can obtain any convergent by isolating suitable ηj -orders in the sij → 0 limit of , where the permutation sum over the orderings of the integrands (2.13) yields the integrands in the definition (2.24) of eMZVs. Hence, our proposal (3.35) implies that occurs at the corresponding orders of sij and ηj in the series , so their definition (3.13) leads to (n1, nr ≠ 1 and z1 = 0)
By the techniques in [52], the integral on the right-hand side can be straightforwardly performed in terms of lattice sums over p = mτ + n ∈ Λ' in (2.25): after expressing the -functions in terms of via (5.2) and inserting the double Fourier expansions
the integrals lead to momentum-conserving delta functions as seen in the dihedral MGFs (2.26). When visualizing each factor of and in the integrand of (5.39) through an edge between vertices j and k, contributions from one-particle reducible graphs integrate to zero. There are at most r − 1 factors of from the in (5.39), and the admissible pairs (j, k) are visualized via dashed lines in figure 4.
Download figure:
Standard image High-resolution imageOne can anticipate from the example (reproducing (5.12))
that only small subset of the terms in contribute to generic single-valued eMZVs—the right-hand side of (5.41) entirely stems from . Apart from the restriction to one-particle irreducible graphs, only those with parity contribute since lattice sums with odd overall modular weight vanish.
Note that the torus integrals in the expression (5.39) for single-valued eMZVs converge whenever the eMZVs themselves do: the convergence criterion n1, nr ≠ 1 rules out any double pole |zjk |−2 in the integrand (and kinematic poles in the Koba–Nielsen integral) since the only overlap between the solid and dashed lines in figure 4 occurs via and .
Based on the conjectural relation (3.23) between the Laurent polynomials, one can use (5.39) to infer the asymptotics of the MGFs on the right-hand side by importing the Laurent polynomials of the B-cycle eMZVs from [97] and applying the single-valued map. Moreover, any relation among eMZVs induces a relation among the MGFs through the lattice-sum representation of their single-valued images. Hence, the database of MGF relations [62] can be complemented by applying (5.39) to the eMZV relations on the website [92]. The lattice sums contributing to have (anti-)holomorphic modular weights subject to w = n1 +⋯+ nr and . Accordingly, the Laurent polynomials of B-cycle eMZVs of length + weight ⩽ 16 [97] give access to those of various combinations of MGFs with .
In the remainder of this section, we will study the single-valued eMZVs in (5.39) at fixed length r and comment on consistency checks with the properties of their SV preimages [17]. The condition n1, nr ≠ 1 for convergence is taken to hold throughout. At length r = 1, for instance, and with k ⩾ 1 are annihilated by SV, in lines with the vanishing of the torus integral over a single at w ≠ 0.
5.7.1. Length-two examples ω(n1, n2)
Single-valued eMZVs of length r = 2 take the form (5.39)
where the distinction between even and odd weight n1 + n2 stems from the vanishing of lattice sums with odd modular weight. At even n1 + n2 > 0, the integrand of (5.42) is proportional to which corresponds to a one-particle reducible graph with
in agreement with and [17]. For odd weight n1 + n2, we keep n2 ≠ 0 and distinguish the two cases n1 = 0 and n1 ≠ 0, where the only contributions of to the integral (5.42) stem from and , respectively:
The resulting relation is consistent with (2.33) of [17] after discarding any SV ζ2k with k ⩾ 1 from the equation of the reference. In combination with the vanishing of , we conclude that all single-valued eMZVs of length two do not exceed one-loop MGFs, in lines with ω(0, 2k + 1) being iterated Eisenstein integrals of depth one.
5.7.2. Length-three examples ω(n1, n2, n3)
Single-valued eMZVs of length three can be written as
after discarding lattice sums of odd modular weight. For , this results in the one-loop MGFs in the second line of (C.2). Starting from weight 8, the bases of ω(n1, n2, n3) require representatives with two non-zero entries such as ω(0, 3, 5), ω(0, 3, 7) [17], and their single-valued versions correspond to MGFs of depth two as exemplified in section 5.6. For any combination of three non-vanishing entries, the torus integral in (5.45) can be expressed in terms of two-loop MGFs in (2.26) and products of one-loop MGFs,
leading to iterated Eisenstein integrals of depth two.
5.7.3. Length four and beyond
Starting from single-valued eMZVs of length four
MGFs of different modular weights may mix through the contributions of and to
The first term drops out when adding in (5.44) and explains the combinations of eMZVs of different length in the third line of (C.2). Similarly, the more general combinations (5.16) of single-valued eMZVs of different length that isolate one-loop MGFs can be understood from the combinations of that contribute to higher-point -functions (3.8).
With multiple non-zero entries in of odd weight, the lattice sums from integrating correspond to MGFs of trihedral topology. Similarly, single-valued eMZVs at length five introduce four-point MGFs of kite topology introduced in section 4.3 of [62]. We hope that their identification with single-valued eMZVs will facilitate the study of MGFs beyond the dihedral topology and result in efficient methods to determine their Laurent polynomials and relations at arbitrary weight.
6. Conclusions and outlook
In this work, we have studied generating series of configuration-space integrals that arise in open- and closed-string amplitudes at genus one. The differential equations and τ → i∞ degenerations of these generating series served as a framework to propose the explicit form of an elliptic single-valued map. Our construction is based on a tentative genus-one uplift of the Betti–deRham duality between integration cycles on a disk boundary and antiholomorphic Parke–Taylor integrands which drives the relation between closed-string and single-valued open-string tree amplitudes [40–45]. These considerations lead us to construct closed-string genus-one integrals over specific antielliptic functions which are thought of as Betti–deRham dual to open-string integration cycles in view of their singularities at zi → zj and their degeneration at τ → i∞.
Most importantly, the differential equations of the open- and closed-string integrals under investigation only differ by τj Gk (τ) vs in the respective differential operators with holomorphic Eisenstein series Gk . Accordingly, we generate the eMZVs and MGFs in their α'-expansions via path ordered exponentials with the same polynomial structures in kinematic invariants and formal expansion variables. The τ-dependent building blocks are iterated Eisenstein integrals in both cases—holomorphic ones with kernels τj Gk (τ) for the open-string integrals and their single-valued versions involving kernels for closed strings.
Our proposal for an elliptic single-valued map is defined through the relation between the generating series of open- and closed-string integrals. By their respective α'-expansion, we obtain the single-valued map for all iterated Eisenstein integrals occurring in the open-string series. This in turn determines the single-valued map of any convergent eMZV in terms of MGFs.
This construction hinges on the compatibility of the initial values at τ → i∞ under the single-valued map of MZVs [46, 48]. We have given evidence for their compatibility by identifying the key building blocks of genus-zero integrals at the cusp—appropriate pairs of disk orderings and Parke–Taylor integrands. However, the detailed expressions for the asymptotic expansions beyond two points in terms of genus-zero integrals is left for future work. At present, the procedure also relies on the reality properties of a generating series of a more general class of closed-string integrals. Our method does not yet provide a direct construction of single-valued iterated Eisenstein integrals solely from open-string data.
6.1. Genus-one integrals versus string amplitudes
The results of this work concern infinite families of configuration-space integrals at genus one, and their application to genus-one string amplitudes requires the following leftover steps:
For both open and closed strings, it remains to integrate over the modular parameter τ of the respective surface. In the closed-string case, τ-integrals over MGFs are typically performed on the basis of their Laplace equations [9, 108–112] and Poincaré-series representations [54, 113–118]. The τ-integration of open-string integrals has for instance been discussed in [119–122], and a general method applicable to arbitrary depth may be based on the representation of eMZVs in terms of iterated Eisenstein integrals (including their 'over-integrated' instances with kernels τj Gk at j > k − 2 [118]) and properties of multiple modular values [101]. It would be particularly interesting to relate closed-string and single-valued open-string integrals at genus one after integration over τ.
For open strings, the - or -series are claimed to exhaust all the configuration-space integrands built from f(k)(zi − zj , τ) that are inequivalent under Fay identities and integration by parts. Similarly, the -series built from double copies of the open-string integrands is expected to contain all torus integrals of this type. Hence, by the arguments of [4, 52, 79], and should 27 capture the conformal-field-theory correlators in the integrands of n-point genus-one amplitudes of massless states (and possibly also of massive states) in bosonic, heterotic and type-II string theories. In all cases, the component integrals in the ηj -expansions of the -, or series need to be dressed with kinematic factors that are determined by the correlators and carry the polarization dependence of the respective string amplitude.
The integrands of only involve antielliptic combinations of and omit infinite classes of component integrals of . For a given genus-one closed-string amplitude, it is therefore not a priori clear if its correlator is generated by the integrand of . Still, the correlators for the four- and five-point type-II amplitudes can be recovered from the subsectors of the -series at fixed modular weights: the four-point correlator of [80] resides at the η−3 order of , and the five-point correlators of [105, 106] can be assembled from the most singular η-orders of at w = 0, 1. Similarly, the four- and five-point amplitudes of gluons and gravitons in heterotic string theories can in principle be extracted from the same which also appear in type II, where higher orders in ηj are needed to capture the bosonic sectors. It would be interesting to see if this pattern persists at higher points in supersymmetric amplitudes, and whether the are sufficient to generate bosonic-string amplitudes at low multiplicity.
6.2. Further directions
This work spawns a variety of further directions and open questions of relevance to both physicists and mathematicians:
The single-valued image of eMZVs is proposed to contain combinations of holomorphic iterated Eisenstein integrals and their complex conjugates denoted by βsv and constructed from the α'-expansion of closed-string integrals in [18]. It would be important to work out their detailed relation to Brown's earlier construction of single-valued iterated Eisenstein integrals [13, 14]. In particular, it remains to relate the MZVs in the antiholomorphic contributions to βsv (fixed from reality properties of Koba–Nielsen integrals in [18]) to the combinations of multiple modular values entering Brown's construction. This will hopefully bypass the need to use these reality properties as independent input for the construction of βsv as done so far.
Several aspects of our construction are based on conjectures with strong support from a variety of non-trivial examples. As pointed out in the relevant passages in earlier parts of this work, it would be desirable to find mathematically rigorous proofs that
- Any Koba–Nielsen integral at genus one involving products and derivatives of Kronecker–Eisenstein coefficients f(k)(z, τ) can be expanded in the coefficients of the series and
- The matrices and in open- and closed-string differential equation (2.19) preserve the commutation relations of Tsunogai's derivations k
- The single-valued images βsv of iterated Eisenstein integrals satisfy shuffle relations, i.e. that the antiholomorphic integration constants α do not introduce any obstructions
- The coefficients of the sij - and ηj -expansion of the initial values and are single-valued multiple zeta values
The proposal of the present work concerns single-valued integration [48, 49] in the modular parameter τ. An alternative approach is to recover MGFs from single-valued functions of torus punctures [10, 15]. In this context, it would be rewarding to find an explicit realization of single-valued integration in z for elliptic polylogarithms and their complex conjugates, for instance by building upon the ideas of [15] and the depth-one results in [123].
At genus zero, the identification of sphere integrals as single-valued disk integrals is equivalent to the Kawai–Lewellen–Tye (KLT) relations between closed-string and squares of open-string tree-level amplitudes [124]. Accordingly, one could wonder if the combinations of holomorphic and antiholomorphic iterated Eisenstein integrals in the βsv or MGFs can arise from products of open-string type generating functions and their complex conjugates. If such a genus-one echo of KLT relations exists, then one can expect a close connection to the monodromy relations among open-string integrals [122, 125–127] and in particular their study in the light of twisted deRham theory [128]. And it could open up a new perspective on the quest for loop-level KLT relations to revisit the generating functions of closed-string integrals in the framework of chiral splitting [129, 130], by performing the α'-expansion at the level of the loop integrand.
A particularly burning question concerns a higher-genus realization of single-valued integration and the associated relations between open- and closed-string amplitudes. A promising first step could be to identify suitable holomorphic open-string analogues of the MGFs [131, 132] and modular graph tensors [133] at higher genus. More generally, the simplified correlators of maximally supersymmetric genus-two amplitudes at four points [134, 135] and five points [136, 137] provide valuable showcases of Koba–Nielsen integrals relevant to open- and closed-string scattering. Furthermore, the construction of the generating series in this work was inspired by extended families of genus-one Koba–Nielsen integrals that arise from heterotic or bosonic strings [52]. Hence, the genus-two correlators of the heterotic string and the combinations of theta functions studied in [138, 139] could give important clues on higher-genus versions of the elliptic functions and generating series in this work.
Acknowledgments
We would like to thank Johannes Broedel, Eric D'Hoker, Daniele Dorigoni, Clément Dupont, André Kaderli, Erik Panzer, Oliver Schnetz, Federico Zerbini and in particular Nils Matthes for combinations of valuable discussions and collaboration on related topics. Moreover, we are grateful to Johannes Broedel, Eric D'Hoker, André Kaderli and an anonymous referee for valuable comments on a draft. OS thanks AEI Potsdam and AK & JG thank Uppsala University for hospitality during various stages of this work. JG, CRM and OS are grateful to the organizers of the program 'Modular forms, periods and scattering amplitudes' at the ETH Institute for Theoretical Studies for providing a stimulating atmosphere and financial support during early stages of this project. JG was supported by the International Max Planck Research School for Mathematical and Physical Aspects of Gravitation, Cosmology and Quantum Field Theory during most stages of this work. CRM is supported by a University Research Fellowship from the Royal Society. OS and BV are supported by the European Research Council under ERC-STG-804286 UNISCAMP.
Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).
Appendix A.: Relations among the elliptic V and Vw functions
In this appendix, we spell out a method to determine the rational coefficients cw,γ in the expansion (3.37) of elliptic functions Vw (...) of fixed modular weights in terms of the V(...) functions in (3.8). This will be done by exploiting the τ → i∞ degeneration (3.20) of the V(...) which fixes the cw,γ in the ansatz (3.37) via
The combinations PT(1) are known as one-loop Parke–Taylor factors from an ambitwistor-string context [94], and we have used σ+ = 0 in passing to the last line. In order to determine the degeneration of the left-hand side of (A.1), we expand the elliptic functions
in terms of the meromorphic Kronecker–Eisenstein coefficients instead of the in (5.2), starting with and . Their τ → i∞ limits [3, 4]
generated by (3.19) ensure that the combination (2πi)−w Vw (1, 2, ..., n|τ) in (A.1) degenerates to a rational function of the σj , where all factors of iπ cancel. Hence, the only σj dependence of Vw (...|τ → i∞) occurs via .
By applying the degeneration (A.4) to the elliptic function Vw in (A.3), the leftover challenge in determining the cw,γ in (A.1) is to expand the terms of the form on the left-hand side in terms of Parke–Taylor factors. For the choices of that arise from the degeneration of Vw⩽n−2, these Parke–Taylor decompositions can be performed by the methods of [140]: as explained in section 3 of the reference, the net effect of the rational factor is to modify the signs of the Parke–Taylor factors on the right-hand side of
More specifically, with the notation
the modification of (A.5) by degenerations of can be written as [140]
The contributions of in turn degenerate to rational constants by (A.4) which multiply the overall sum over permutations ρ. Hence, (A.7) allows to straightforwardly expand the left-hand side of (A.1) in terms of Parke–Taylor factors in an n!-element basis of PT(+, ..., −). Matching the Parke–Taylor coefficients with those on the right-hand side determines the cw,γ in (3.37). It is a special property of the elliptic functions Vw that their degeneration conspires to the cyclic combinations PT(1) in (A.2), i.e. that the (n − 1)! independent cw,γ are sufficient to accommodate the n! permutations of PT(+, 1, 2, ..., n, −) in 1, 2, ..., n.
For instance, the decompositions in (3.39) to (3.41) follow from the special cases of (A.7)
once the right-hand sides are matched with the combinations of one-loop Parke–Taylor factors PT(1) in (A.1) and (A.2).
Appendix B.: The initial value at three points
This appendix gathers the three-point initial values for the α'-expansion (4.9) of B-cycle integrals up to and including weight four. The corresponding orders of relevant to the α'-expansion (4.19) of J-integrals are obtained from the single-valued map (ζ2, ζ3, ζ4) → (0, 2ζ3, 0). Since even (odd) orders in the ηj -expansion integrate to zero on the odd (even) integration cycles , we will separate the two types of contributions in order to infer from a relabeling of .
The expressions in this appendix along with various higher-order terms in the sij - and ηj -expansions can be found in the supplementary file at https://stacks.iop.org/A/55/025401/ mmedia, also see (4.27) for the appearance of ζ3,5.
B.1. Even orders in ηj
The terms of even orders in ηj in the three-point initial values are given by
and
with MZVs of weight ⩾ 5 in the ellipsis.
B.2. Odd orders in ηj
The terms of odd orders in ηj in the three-point initial values are given by
and
again with MZVs of weight ⩾ 5 in the ellipsis.
Appendix C.: Examples of single-valued eMZVs
C.1. Systematics at depth one
The simplest examples (5.12) and (5.13) of single-valued eMZVs extracted from the two-point integrals (5.4) are special cases of the SV map (4.30) on holomorphic iterated Eisenstein integrals. For their depth-one combinations in (4.10), the SV image yields the following Cauchy–Riemann derivatives ∇ = 2i(Im τ)2∂τ of non-holomorphic Eisenstein series [18]
also see (4.22). While the objects on the right-hand side are expressible in terms of the lattice sums in (2.27) via (2.28), the are simple combinations of B-cycle eMZVs , where 0p stands for a sequence 0, 0, ..., 0 of p successive zeros. On these grounds, translates into simple relations such as
as well as
The relative factors of and among the eMZVs of different lengths are engineered to streamline the iterated-Eisenstein-integral representation [17] and generalize as follows [104]
In obtaining (C.2) and (C.3) from (C.4), we have used SV ω(m) = 0 ∀ m ⩾ 1 and the following simplifications of the only eMZV ω(0ℓ−1, 2k + ℓℓ) whose length and weight adds up to an even number [17],
Based on the dictionary (2.28) between lattice sums and non-holomorphic Eisenstein series, one can reformulate (C.4) as
where k = 0 needs to be excluded if ℓ = 1, for instance
Moreover, by extending (C.4) to k → −k and applying the complex conjugate of (2.28), we also obtain antiholomorphic Cauchy–Riemann derivatives as single-valued eMZVs (with ℓ − 2k > 0),
The simplest examples include
and the first two lines are equivalent to those in (5.32).
C.2. Examples with real MGFs at depth two
By inspecting the order of the two-point integrals and the order of , we have obtained the representations (5.14) of E2,2 and π∇E2,2 as single-valued eMZVs. One can extract similar representations for E2,3, π∇E2,3 and (π∇)2 E2,3 from the order of , the order of and the order of , respectively:
The corresponding lattice-sum representations [51, 62] and βsv representations [18] are given by
Footnotes
- 6
- 7
We shall use this terminology at genus one even though we are not aware of any explicitly worked out notion of Betti–deRham duality beyond genus zero.
- 8
Strictly speaking, MZVs need to be replaced by their motivic versions to have a well-defined single-valued map.
- 9
Our conventions for the standard odd Jacobi theta function are
and η(τ) is the Dedekind eta function. In order to avoid confusion with the expansion parameters ηj , we always spell out the argument τ of the Dedekind eta function. Representations of the open-string Green function in terms of elliptic polylogarithms are discussed in [4, 5, 51], and we follow the conventions of [4] for regularizing endpoint divergences.
- 10
By using Fay identities and integration by parts, the massless four-point genus-one amplitude of open bosonic strings in section 8.1.1 of [76] can be rewritten in terms of the coefficients in the ηj -expansion of .
- 11
Functions F(τ) on the upper half plane with transformations under are said to carry holomorphic and antiholomorphic modular weight w and , respectively.
- 12
- 13
This relies on the linear-independence result of [91] on holomorphic iterated Eisenstein integrals.
- 14
On the closed-string side of the 'genus-one Betti–deRham duality' we note that, by double-periodicity of the V-functions, additional poles with identical residues occur as zj → zj±1 + mτ + n (). On the open-string side in turn, the delimiters of the integration cycles in the B-cycle parametrization of figure 3 are separated by τ.
- 15
See [12] for an earlier proof of the weaker statement that the Laurent polynomials of modular graph functions are -linear combinations of cyclotomic MZVs.
- 16
This amounts to performing the α'- and η-expansion to order 10 in the terminology of section 3.4.2 of [18].
- 17
We have excluded the singular functions in the integrand from our checks to avoid the tedious treatment of the resulting kinematic poles in the α'-expansion. For the contributions of V0 and V1 to the integrand in (3.9), we have checked the Laurent polynomials from up to one factor of f(1)(zij , τ) in the integrand to obey (3.23), see section 3.6 for the disentanglement of different Vw entering .
- 18
By a similar argument, each MGF can be realized in the -order of Yτ -integrals at sufficiently high multiplicity, see section 2.5 of [61].
- 19
We are grateful to Nils Matthes for valuable discussions on this point.
- 20
On top of the modular S transformation relating and , the in (4.8) involve integration kernels τj Gk with 0 ⩽ j ⩽ k − 2 instead of the in [59, 60]. In other words, the relations (4.3) in the derivation algebra are built into (4.8), whereas the results in the references may require the use of shuffle relations to manifest the absence of .
- 21
Moreover, any such violation of shuffle relations would need to be a combination of antiholomorphic iterated Eisenstein integrals (by the differential equation [18, (2.37)] for ) but at the same time line up with the modular weights in the ηj -expansion of , and the reality properties of the latter. It would be interesting to find a rigorous argument to rule out the existence of such antiholomorphic functions.
- 22
The representations in terms of convergent iterated Eisenstein integrals needed to verify (5.17) are
- 23
By slight abuse of notation, we denote the ordering of punctures zi , zj on the imaginary axis by .
- 24
- 25
- 26
See [102] for a computer implementation of -relations among MZVs.
- 27
It has been shown in [52] that the integrands of massless genus-one amplitudes in bosonic, heterotic and type-II theories are expressible in terms of products of f(k)(zi − zj , τ) and their zi -derivatives. The conjectural part is that arbitrary products of f(k)(zi − zj , τ) (possibly including derivatives) are expressible in terms of the in (2.13) with their specific chain structure via repeated use of Fay identities and integration by parts [59, 60].