Enriquez Kernels and Higher-Genus Polylogarithms

Updated 10 July 2026

Enriquez kernels are meromorphic integration kernels on compact Riemann surfaces that serve as the coefficient functions in flat Lie algebra-valued connections and iterated integrals.
They are constructed using normalized holomorphic Abelian differentials, prime forms, and convolution integrals, with recursive Fay-type identities ensuring analytic closure.
Degeneration techniques show that these kernels localize to genus-one or rational data, linking higher-genus polylogarithms to arithmetic phenomena like multiple zeta values.

Enriquez kernels are meromorphic integration kernels on compact Riemann surfaces of genus $h \geq 1$ , usually denoted $g^{I_1\cdots I_r}{}_J(x,y)$ , $g_{I_1\ldots I_r|J}(x,y)$ , or, in generating-function formalisms, $\omega_{i_1\cdots i_r j}(z,x)$ . They were introduced by Enriquez through functional properties and serve as the coefficient functions of a meromorphic flat connection with simple poles valued in a free Lie algebra or, in multivariable versions, in the Lie algebras $t_{h,n}$ or $\hat t_{h,n}$ . Their iterated integrals generate higher-genus analogues of ordinary and elliptic polylogarithms, and recent work has made their construction, identities, degeneration theory, and arithmetic content increasingly explicit (D'Hoker et al., 20 Feb 2025, Baune et al., 2024).

1. Definition and analytic characterization

The basic object is a family of kernels indexed by a word $I_1,\ldots,I_r$ and a terminal index $J$ , with $I_k,J \in \{1,\ldots,h\}$ . In one common convention, the kernels appear in the expansion

${\bf K}_J(x,y;B)=\sum_{r=0}^{\infty} g^{I_1\cdots I_r}{}_J(x,y)\,B_{I_1}\cdots B_{I_r},$

while in the generating-function formalism one writes

$g^{I_1\cdots I_r}{}_J(x,y)$ 0

The generating function is characterized by quasi-periodicity under deck transformations and a prescribed residue at the diagonal, namely

$g^{I_1\cdots I_r}{}_J(x,y)$ 1

(Baune et al., 2024).

The kernels are meromorphic in the surface variables and locally holomorphic in the moduli. For $g^{I_1\cdots I_r}{}_J(x,y)$ 2, they reduce to the normalized holomorphic Abelian differentials,

$g^{I_1\cdots I_r}{}_J(x,y)$ 3

For $g^{I_1\cdots I_r}{}_J(x,y)$ 4, they have a simple pole at the diagonal,

$g^{I_1\cdots I_r}{}_J(x,y)$ 5

while for $g^{I_1\cdots I_r}{}_J(x,y)$ 6 they are holomorphic in both variables $g^{I_1\cdots I_r}{}_J(x,y)$ 7 and $g^{I_1\cdots I_r}{}_J(x,y)$ 8 (D'Hoker et al., 20 Feb 2025).

Their monodromy data are part of the definition. The $g^{I_1\cdots I_r}{}_J(x,y)$ 9-cycle monodromies are trivial, whereas the $g_{I_1\ldots I_r|J}(x,y)$ 0-cycle monodromies close recursively on lower-rank kernels. In the conventions used for cyclic products of Szegő kernels, their $g_{I_1\ldots I_r|J}(x,y)$ 1-cycle periods are independent of $g_{I_1\ldots I_r|J}(x,y)$ 2 and are expressed through Bernoulli numbers: $g_{I_1\ldots I_r|J}(x,y)$ 3 (D'Hoker et al., 12 May 2025).

Different papers use different index placements and normalizations. In particular, one rescaled convention is

$g_{I_1\ldots I_r|J}(x,y)$ 4

chosen so that the genus-one limit aligns smoothly with the Kronecker–Eisenstein kernels $g_{I_1\ldots I_r|J}(x,y)$ 5 (D'Hoker et al., 2024). This multiplicity of conventions is a notational issue rather than a substantive one.

2. Construction from Abelian differentials, prime forms, and convolutions

A direct construction of Enriquez kernels uses only holomorphic Abelian differentials and the prime form. For the lowest nontrivial rank, homotopy-invariant convolution integrals over canonical $g_{I_1\ldots I_r|J}(x,y)$ 6-cycles yield

$g_{I_1\ldots I_r|J}(x,y)$ 7

where $g_{I_1\ldots I_r|J}(x,y)$ 8 is the prime form and $g_{I_1\ldots I_r|J}(x,y)$ 9 is the normalized holomorphic Abelian differential (D'Hoker et al., 20 Feb 2025). A closely related formula in a different normalization was obtained in terms of Abelian differentials and the prime form following recent advances by D’Hoker and Schlotterer (Ichikawa, 1 Oct 2025).

Higher-rank kernels are generated recursively. In schematic form, the recursion is obtained from the Fay identity and expresses a higher-rank kernel as a homology-cycle convolution of lower-rank kernels together with lower-rank correction terms: $\omega_{i_1\cdots i_r j}(z,x)$ 0 The omitted terms include explicit lower-rank contributions and Bernoulli-number corrections (D'Hoker et al., 20 Feb 2025, Ichikawa, 1 Oct 2025).

A structural result is that the space of Enriquez kernels closes under convolution over homology cycles. More precisely, the convolution over an $\omega_{i_1\cdots i_r j}(z,x)$ 1-cycle of products of arbitrary Enriquez kernels again produces a linear combination of Enriquez kernels. The same closure phenomenon holds under variation of the moduli: explicit deformation equations show that moduli derivatives of Enriquez kernels are again linear combinations of Enriquez kernels (D'Hoker et al., 20 Feb 2025). This is the technical basis for treating them as a functionally closed class of integration kernels.

These constructions make the kernels explicit, algorithmic, and intrinsically geometric. The prime form controls singularities and transformation properties; the Abelian differentials encode the complex structure and periods; and the convolution representation organizes the recursive buildup from rank $\omega_{i_1\cdots i_r j}(z,x)$ 2 and rank $\omega_{i_1\cdots i_r j}(z,x)$ 3 data (D'Hoker et al., 20 Feb 2025).

3. Fay-type identities, interchange relations, and flatness

The modern theory of Enriquez kernels is organized around the relation between kernel identities and flatness of the associated connection. In genus one, Kronecker–Eisenstein kernels satisfy the classical Fay trisecant identity. For arbitrary genus, higher-genus Fay identities were first established for single-valued non-holomorphic kernels, and the meromorphic Enriquez kernels were conjectured to satisfy identities of exactly the same form (D'Hoker et al., 2024).

That conjectural status was sharpened substantially by the study of meromorphic generating functions. A quadratic Fay-like identity was derived for the Enriquez generating function, and its coefficientwise expansion yields explicit quadratic identities for the kernels. The key result is that all possible quadratic three-point identities among Enriquez kernels are generated by this Fay-like identity together with trivial linear identities. Equivalently, the resulting relations are exhaustive: no additional nontrivial quadratic three-point kernel relations exist (Baune et al., 2024). In genus one, the identity reduces exactly to the classical Fay identity.

The multivariable picture makes the link to flatness explicit. The Enriquez connection on the configuration space of $\omega_{i_1\cdots i_r j}(z,x)$ 4 points has the form

$\omega_{i_1\cdots i_r j}(z,x)$ 5

with each $\omega_{i_1\cdots i_r j}(z,x)$ 6 assembled from kernel generating series and Lie algebra generators. Flatness requires

$\omega_{i_1\cdots i_r j}(z,x)$ 7

or, equivalently, commuting derivatives together with

$\omega_{i_1\cdots i_r j}(z,x)$ 8

For $\omega_{i_1\cdots i_r j}(z,x)$ 9, the Lie algebra is not freely generated, so the commutator condition imposes nontrivial functional relations on the kernels (D'Hoker et al., 9 Feb 2026).

Recent work shows that the flatness conditions of the multivariable Enriquez connection imply the full union of interchange identities $t_{h,n}$ 0 and Fay identities $t_{h,n}$ 1 for Enriquez kernels. The converse is more delicate in the meromorphic case: it requires control of coincident limits and modular or dihedral symmetries of those limits, and this remains subtle for Enriquez kernels (D'Hoker et al., 9 Feb 2026). Accordingly, the current status is asymmetric: flatness implies the kernel identities, while the reverse implication is not stated without further hypotheses.

4. Degeneration and localization on nodal limits

A central recent development concerns the behavior of Enriquez kernels under degeneration of the underlying Riemann surface. For degenerating families of pointed Riemann surfaces, the variation of the Enriquez connection can be tracked explicitly through Abelian differentials, prime forms, and the kernels themselves. In one class of degenerations, a smooth surface $t_{h,n}$ 2 degenerates to a singular curve assembled from genus-one curves and projective lines with marked points. In that limit, the Enriquez connection becomes the connection previously constructed by the same author for degenerating families, obtained by gluing the Knizhnik–Zamolodchikov and elliptic KZB connections along the components (Ichikawa, 1 Oct 2025).

A characteristic phenomenon is localization. In the degenerate limit, only kernels attached to the component containing the relevant cycles survive, while the others vanish. In the notation of that analysis,

$t_{h,n}$ 3

as $t_{h,n}$ 4, for $t_{h,n}$ 5 and $t_{h,n}$ 6 fixed (Ichikawa, 1 Oct 2025). This localization supplies a precise mechanism by which global higher-genus kernels reduce to componentwise genus-one or rational data.

A complementary non-separating degeneration takes a genus- $t_{h,n}$ 7 surface to one of genus $t_{h,n}$ 8 with two additional punctures $t_{h,n}$ 9. In this setting, the kernels close under degeneration: the surviving genus- $\hat t_{h,n}$ 0 kernels become kernels or trace-type objects on the lower-genus surface with punctures. The explicit degeneration formulas distinguish whether the index $\hat t_{h,n}$ 1 occurs among the upper indices or in the terminal position. In particular,

$\hat t_{h,n}$ 2

and kernels with $\hat t_{h,n}$ 3 in the last position produce puncture terms involving $\hat t_{h,n}$ 4-functions and Bernoulli numbers (Biancotto et al., 6 Jul 2026). The Bernoulli generating function

$\hat t_{h,n}$ 5

controls the relation between the Lie algebra generators before and after degeneration (Biancotto et al., 6 Jul 2026).

These results show that degeneration is not merely a limiting procedure on coefficients. It reorganizes the kernel algebra, the Lie algebra presentation of the connection, and the relevant polylogarithmic periods in a way compatible with lower-genus geometry.

5. Higher-genus polylogarithms and multiple zeta values

The principal role of Enriquez kernels is to generate polylogarithms by iterated integration. In both the meromorphic and the generating-function formalisms, polylogarithms on a higher-genus surface are the coefficients of holonomies or iterated integrals built from the Enriquez connection (D'Hoker et al., 2024). In this sense, Enriquez kernels are the higher-genus analogue of the rational kernels of genus zero and the Kronecker–Eisenstein kernels of genus one.

Degeneration theory yields an arithmetic application. The monodromies of the Enriquez connection, described as higher-genus analogues of ordinary and elliptic polylogarithms, can be expanded explicitly as power series in deformation parameters $\hat t_{h,n}$ 6 and their logarithms: $\hat t_{h,n}$ 7 The coefficients $\hat t_{h,n}$ 8 are linear combinations of multiple zeta values and can be described via noncommutative polynomials in the Lie algebra generators (Ichikawa, 1 Oct 2025). Explicit formulas for these monodromies had already been established for the glued degeneration model, and the degeneration analysis shows that the same formulas extend to all families through the Enriquez connection (Ichikawa, 1 Oct 2025).

The appearance of multiple zeta values generalizes earlier genus-zero and genus-one results. This suggests a deep arithmetic structure for higher-genus polylogarithms. The kernel identities are the mechanism behind that structure: they govern the functional relations among the polylogarithms, ensure closure under integration, and organize specializations at degenerate configurations. In the meromorphic theory, the resulting functional equations can yield higher-genus analogues of multiple zeta values, while the boundary terms appearing in the kernel identities involve genus-zero multiple zeta values (Baune et al., 2024).

The recent literature places Enriquez kernels within a broader ecosystem of higher-genus integration kernels. One important comparison is with the single-valued DHS kernels and modular tensors; another is with Schottky–Kronecker forms and their Poincaré-series expansions.

Formalism	Main properties	Relation to Enriquez kernels
Enriquez kernels	Meromorphic, multiple-valued, trivial $\hat t_{h,n}$ 9-monodromy, recursive $I_1,\ldots,I_r$ 0-monodromy	Coefficients of the Enriquez connection
DHS kernels	Single-valued, modular invariant, non-meromorphic, real analytic	Parallel recursion and descent structure
Schottky–Kronecker kernels	Poincaré-series construction on the Schottky cover	Coincide with Enriquez differentials under convergence assumptions

In the analysis of cyclic products of Szegő kernels for even spin structure, one descent procedure expresses the point dependence through meromorphic multiple-valued Enriquez kernels, while the spin-structure dependence is isolated in constants given by multiple convolution integrals over homology cycles. A complementary descent uses DHS kernels and modular tensors instead. Although the analytic properties of the two sets of building blocks differ sharply, the combinatorial structure of the two decompositions is described as virtually identical (D'Hoker et al., 12 May 2025). This is the precise sense in which Enriquez kernels parallel the non-holomorphic modular tensors developed in the single-valued approach.

The modular distinction is essential. Enriquez kernels are meromorphic and adapted to flat meromorphic connections, but they are not modular tensors and do not enjoy simple $I_1,\ldots,I_r$ 1-covariance. DHS kernels, by contrast, are single-valued and modular invariant, but not meromorphic (D'Hoker et al., 12 May 2025). The choice between the two formalisms is therefore a choice between meromorphicity and modular covariance rather than between two unrelated theories.

A further development uses Schottky uniformization. Schottky–Kronecker forms are defined as Poincaré series on the Schottky cover, and their expansion produces higher-genus integration kernels $I_1,\ldots,I_r$ 2. Under convergence assumptions, these kernels coincide with the differentials defined by Enriquez (Baune et al., 2024). This is significant because Enriquez’s original definition is not well-suited for numerical evaluation, whereas the Poincaré-series realization can be evaluated numerically for real hyperelliptic curves, and genus-two examples have been computed explicitly (Baune et al., 2024).

Taken together, these developments present Enriquez kernels as a unifying structure across explicit construction, moduli variation, kernel identities, degeneration theory, worldsheet correlators, and numerical realization. Their defining features remain constant throughout: meromorphicity, recursive control by monodromy and convolution, and their role as the local-to-global mechanism connecting the geometry of Riemann surfaces to the algebra of higher-genus polylogarithms (Ichikawa, 1 Oct 2025).