Turaev Cobracket: Loop Intersection Operation

• The Turaev cobracket is a self-intersection operation on free homotopy classes of loops, defined by splitting immersed curves at transverse double points with local sign weights.
• It complements Goldman’s bracket to form the Goldman–Turaev Lie bialgebra, underpinning the algebraic and geometric structure of loop intersection theory on oriented surfaces.
• Extensions include framed and regular homotopy versions along with combinatorial models and links to the Kashiwara–Vergne framework that refine its application in topology.

The Turaev cobracket is the self-intersection operation on free homotopy classes of oriented loops on an oriented surface. In its classical form, it is defined on the vector space generated by free loop classes, usually modulo the trivial loop, by summing over transverse double points of a generic immersed representative and splitting the loop into the two branches determined at each self-intersection. Together with Goldman’s bracket, it gives the Goldman–Turaev Lie bialgebra; in framed or regular-homotopy settings, it admits natural lifts that retain rotation data and avoid quotienting by the constant loop (Kawazumi, 2019).

1. Geometric definition on immersed loops

Let $\Sigma$ be a connected oriented surface, and let $\hat{\pi}$ denote the set of free homotopy classes of loops. For a generic immersed loop $\gamma:S^1\to\Sigma$ with only transverse double points, the Turaev cobracket is defined by cutting $\gamma$ at each self-intersection point $p$ into two oriented subloops $\gamma_p^{(1)}$ and $\gamma_p^{(2)}$, and weighting the resulting tensor by the local intersection sign $\varepsilon(p)\in\{\pm1\}$ determined by the ordered tangent vectors at $p$ (Kawazumi, 2019). In the standard skew-symmetrized form,

$$\delta([\gamma])=\sum_{p}\varepsilon(p)\Big([\gamma_p^{(1)}]\otimes[\gamma_p^{(2)}]-[\gamma_p^{(2)}]\otimes[\gamma_p^{(1)}]\Big).$$

Equivalent formulations appear throughout the literature, sometimes written as a sum over ordered pairs of parameters $\hat{\pi}$0 with $\hat{\pi}$1, and sometimes as the alternating part of a based self-intersection coaction $\hat{\pi}$2 (Alekseev et al., 2017).

On negatively curved surfaces, every free homotopy class has a unique closed geodesic representative, and all self-intersections of that geodesic are transverse double points. In that setting, the cobracket can be computed directly from the geodesic representative, which makes the operation particularly concrete (Chas et al., 2010).

A standard example is the figure-eight loop with a single transverse self-intersection. If the two branches at the double point are denoted $\hat{\pi}$3 and $\hat{\pi}$4, then

$\hat{\pi}$5

so the cobracket records the decomposition of the loop into its two lobes (Alekseev et al., 2017).

A fundamental subtlety is that the unframed cobracket is not invariant under birth or death of a monogon. Inserting a positively oriented monogon changes the classical cobracket by $\hat{\pi}$6, which is why Turaev originally defined the operation on the quotient by the trivial loop (Kawazumi, 2019).

2. Lie coalgebra structure, framing, and regular homotopy

The Turaev cobracket is co-skew-symmetric and satisfies the co-Jacobi identity, so it defines a Lie coalgebra structure on the span of free loop classes (Chas et al., 2010). Combined with Goldman’s bracket, it satisfies the usual compatibility axiom: the cobracket is a $\hat{\pi}$7-cocycle for the adjoint action, giving a Lie bialgebra structure (Alekseev et al., 2017). Chas proved involutivity for the classical Goldman–Turaev Lie bialgebra (Kawazumi, 2019).

Framing removes the monogon ambiguity. If $\hat{\pi}$8 is a framing of a surface with nonempty boundary, the framed cobracket is

$\hat{\pi}$9

where $\gamma:S^1\to\Sigma$0 is the rotation number of the immersed loop relative to the framing (Kawazumi, 2019). In the punctured-disk setting, closely related formulas are written as an “enhanced” framed cobracket

$\gamma:S^1\to\Sigma$1

or equivalently

$\gamma:S^1\to\Sigma$2

with the last term supplying the framing correction (Bar-Natan et al., 25 Sep 2025). The exact sign convention varies across sources, but the structural role of the correction term is the same: it makes the cobracket well defined on free loops rather than only modulo the constant loop.

A further refinement replaces homotopy by regular homotopy of immersions. In this version, one works with regular homotopy classes of immersed loops and records monogon insertion via a free $\gamma:S^1\to\Sigma$3-action generated by adding a monogon. For surfaces with boundary and a fixed framing, the regular-homotopy refinement yields a Lie bialgebra over the group ring of the monogon action, and the regular-homotopy cobracket still vanishes on embedded loops and their powers (Kawazumi, 2014). This refinement is the setting in which the Enomoto–Satoh traces and the divergence cocycle from the Kashiwara–Vergne theory appear as part of the cobracket.

3. Combinatorial models and algebraic descriptions

For surfaces with boundary, free homotopy classes can be identified with cyclically reduced words in a free generating set of the fundamental group. In that language, the cobracket admits a purely combinatorial description via linked pairs of subwords. If $\gamma:S^1\to\Sigma$4 is a cyclically reduced word, then

$\gamma:S^1\to\Sigma$5

and this combinatorial $\gamma:S^1\to\Sigma$6 agrees with the geometric cobracket on surfaces with boundary (Chas et al., 2010). A different combinatorial reformulation uses partitions of cyclic words and an arc-diagram description; the resulting formula computes the local sign by a linking number of partitions and again reproduces the geometric Turaev cobracket exactly (Yamamoto, 2022).

The associated graded of the Goldman–Turaev structure is expressed in terms of cyclic words in homology. For genus zero, if $\gamma:S^1\to\Sigma$7 and $\gamma:S^1\to\Sigma$8 is the completed tensor Hopf algebra, then

$\gamma:S^1\to\Sigma$9

and the associated graded cobracket is the Schedler necklace cobracket defined by the Kirillov–Kostant–Souriau double bracket (Alekseev et al., 2017). In cyclic-word form, for $\gamma$0 one has

$\gamma$1

with the pairing specialized in genus zero to the KKS pairing (Alekseev et al., 2017).

For compact genus-zero surfaces, Massuyeau gave a tensorial description of the cobracket with respect to the standard group-like expansion $\gamma$2. In that description, Bernoulli numbers enter through the series

$\gamma$3

and the tensorial formula for $\gamma$4 contains explicit Bernoulli-number corrections (Kawazumi, 2015).

Taniguchi reformulated the divergence map appearing in algebraic descriptions of the cobracket using a flat noncommutative connection on $\gamma$5. For formally smooth associative algebras, the divergence is expressed as a Hattori–Stallings trace of “Lie derivative minus horizontal lift,” and in the case of a surface with boundary this recovers the Alekseev–Kawazumi–Kuno–Naef divergence description of the framed cobracket. The same framework extends to closed surfaces by replacing ordinary connections with homological connections on perfect resolutions (Taniguchi, 2024).

4. Formality, expansions, and the Kashiwara–Vergne framework

A central problem is Goldman–Turaev formality: constructing a filtered Lie bialgebra isomorphism from the geometric Goldman–Turaev Lie bialgebra to its associated graded necklace Lie bialgebra. In genus zero, the lowest-degree parts of the bracket and cobracket are canonical and independent of the chosen expansion; the cobracket part is Schedler’s Lie cobracket on cyclic words (Alekseev et al., 2017).

Alekseev, Kawazumi, Kuno, and Naef showed that in genus zero this formality problem is essentially equivalent to a Kashiwara–Vergne problem for automorphisms of a free Lie algebra. A special expansion $\gamma$6 is homomorphic precisely when its Jacobian cocycle satisfies the corresponding KV condition, and the Turaev cobracket is recovered algebraically by applying a noncommutative divergence to the derivation defined by the Goldman bracket (Alekseev et al., 2017).

An elementary proof in genus zero was later obtained from the Knizhnik–Zamolodchikov connection on $\gamma$7. In that construction, the holonomy map

$\gamma$8

is an isomorphism of Lie bialgebras, intertwining the geometric Goldman–Turaev structure with the necklace Schedler Lie bialgebra. The proof of compatibility with the bracket follows a Hitchin-type Stokes argument, and the same mechanism extends to the cobracket by using a second noncommutative variation formula and residues at self-intersections (Alekseev et al., 2017).

For arbitrary genus and boundary, the higher-genus Kashiwara–Vergne problem was introduced to solve the same formality question. Alekseev, Kawazumi, Kuno, and Naef proved that a filtration-preserving automorphism $\gamma$9 yields a Goldman–Turaev formality map if and only if it solves the higher-genus KV problem. Solutions exist for all $p$0, and for $p$1 except for specific framings where formality fails; the genus $p$2 case is constructed using elliptic associators following Enriquez (Alekseev et al., 2018).

Hain showed that after $p$3-adic completion and for a quasi-algebraic framing of a smooth affine complex curve, the Turaev cobracket is a morphism of mixed Hodge structure after a Tate twist. Combined with corresponding results for the Goldman bracket, this yields torsors of KV solutions in all genera and links the cobracket to mixed Hodge theory, mapping class group completions, and motivic structures (Hain, 2018).

A more recent genus-zero proof constructs Goldman–Turaev homomorphic expansions from the Kontsevich integral. In that approach, the cobracket is derived from a three-dimensional tangle operation and a low-degree Vassiliev quotient, and compatibility with the cobracket is established after showing that the correction terms vanish upon closure and antisymmetrization (Bar-Natan et al., 25 Sep 2025).

5. Detection theorems, rigidity, and representation-theoretic realizations

A long-standing question asked whether vanishing of the Turaev cobracket characterizes powers of simple closed curves. The basic vanishing statement is true: if a class is represented by a simple closed curve, or a power of one, then its cobracket vanishes (Chas et al., 2010). The converse, however, is subtler.

Chas and Krongold studied the family of assertions

$p$4

For orientable surfaces with boundary, they proved $p$5 for all $p$6. More precisely, if $p$7 is a nonpower cyclic reduced word and $p$8, then

$p$9

where $\gamma_p^{(1)}$0 is the Manhattan norm and $\gamma_p^{(1)}$1 is the minimal number of transversal self-intersections among representatives of $\gamma_p^{(1)}$2 (Chas et al., 2010). They also showed that $\gamma_p^{(1)}$3 fails on every surface with negative Euler characteristic except the pair of pants, while $\gamma_p^{(1)}$4 was verified computationally for hundreds of millions of the shortest classes but not proved in full generality (Chas et al., 2010). This corrects the common misconception that $\gamma_p^{(1)}$5 alone detects simplicity.

Cahn introduced a refinement $\gamma_p^{(1)}$6 modeled on chord diagrams. The Turaev cobracket factors through $\gamma_p^{(1)}$7, and $\gamma_p^{(1)}$8 if and only if $\gamma_p^{(1)}$9 is a power of a simple class. Moreover, $\gamma_p^{(2)}$0 gives an explicit formula for the minimal self-intersection number: if $\gamma_p^{(2)}$1 with $\gamma_p^{(2)}$2 primitive, then

$\gamma_p^{(2)}$3

where $\gamma_p^{(2)}$4 is the sum of absolute values of coefficients in the reduced expression (Cahn, 2010). This suggests that the classical cobracket loses information through cancellation that is retained by finer chord-diagram data.

The cobracket plus power operations also yields rigidity phenomena. If a permutation of nontrivial free homotopy classes commutes with all power maps and preserves the cobracket, then on orientable surfaces with boundary it preserves self-intersection number on nonpower classes, carries simple classes to simple classes, preserves disjointness of simple classes, and—except in the listed low-complexity cases—is induced by a unique mapping class (Chas et al., 2010).

In another direction, the Turaev cobracket admits a geometric realization on moduli spaces of flat connections for suitable Lie supergroups. For $\gamma_p^{(2)}$5, the moduli space carries a BV operator $\gamma_p^{(2)}$6, and the map

$\gamma_p^{(2)}$7

satisfies

$\gamma_p^{(2)}$8

while the induced BV bracket recovers Goldman’s bracket (Alekseev et al., 2022). In this sense, intersections of distinct loops correspond to BV brackets and self-intersections correspond to the BV Laplacian.

6. Extensions, limits, and current directions

Several extensions of the Turaev cobracket have been proposed. Taniguchi introduced a family of operations $\gamma_p^{(2)}$9 defined from noncommutative connections and parametrized by ribbon graphs. In the surface group algebra case, $\varepsilon(p)\in\{\pm1\}$0 recovers the framed Turaev cobracket, while for $\varepsilon(p)\in\{\pm1\}$1 one obtains higher algebraic operations that are not mapping-class-group equivariant (Taniguchi, 7 Feb 2025). In the free associative algebra case, these operations coincide with canonical ribbon-graph operations and define Lie algebra cohomology classes.

The relationship with the Enomoto–Satoh trace and the Kashiwara–Vergne divergence is particularly strong in regular-homotopy refinements. Kawazumi showed that part of the regular-homotopy Turaev cobracket recovers the Enomoto–Satoh traces in the one-boundary case and the divergence cocycle in genus zero (Kawazumi, 2014). Alekseev, Kawazumi, Kuno, and Naef later proved that the Johnson obstruction given by the Turaev cobracket coincides with the one given by the Enomoto–Satoh trace (Alekseev et al., 2018).

At the same time, some natural analogies fail. In the Poisson algebra of $\varepsilon(p)\in\{\pm1\}$2 Wilson loops, all Lie algebra cohomology classes of cobrackets are represented by maps of the form

$\varepsilon(p)\in\{\pm1\}$3

and there is no cohomology class corresponding to the Turaev cobracket under the Goldman-to-Wilson-loop homomorphism (Nobuta, 2017). This indicates that abelianization destroys the nontrivial two-factor structure generated by self-intersection splittings.

The higher-dimensional string-topology analogue is also limited. For surfaces, the string topology cobracket agrees with the negative of the Turaev cobracket (Hartenstein et al., 7 Oct 2025). By contrast, in higher dimensions the analogy breaks down: the $\varepsilon(p)\in\{\pm1\}$4-equivariant string cobracket is not a homotopy invariant in general, and the string bracket together with the string cobracket does not form a Lie bialgebra on the equivariant homology of the free loop space (Sumoto, 11 May 2026). A plausible implication is that the surface case is distinguished not only by low-dimensional topology but by the exceptional rigidity of loop intersection theory.

Across these developments, the Turaev cobracket remains the basic self-intersection operation on free loops: geometric in origin, combinatorial in implementation, algebraic in formal descriptions, and structurally central to the Goldman–Turaev Lie bialgebra.