4-Ended Tangle Khovanov Theory

• 4-ended tangle Khovanov theory is a collection of bordered, categorical, and geometric formalisms that assign invariants to tangles with four boundary points.
• The approach employs type D and type A structures, arc algebra modules, and immersed multicurve models to compute link homology through algebraic tensoring and Floer-theoretic pairings.
• Extensions of the theory include spectral refinements, odd/even variants, and operator constructions that enhance its applications in mutation and symplectic topology.

4-ended tangle Khovanov theory is the collection of bordered, categorical, and geometric formalisms that assign invariants to tangles with four boundary points and recover ordinary Khovanov-type link homology after gluing complementary tangles. In its original bordered form, a tangle properly embedded in a half-space is assigned a type D structure over a differential bigraded boundary algebra, and the complementary tangle carries a type A structure whose box tensor reconstructs the bigraded Khovanov chain complex of the closed link obtained by gluing (Roberts, 2013, Roberts, 2013). Subsequent developments re-expressed the same four-ended theory in terms of modules over arc algebras, immersed multicurves with local systems in the 4-punctured sphere, and twisted complexes in wrapped Fukaya categories, yielding equivalent tangle invariants and new applications to mutation, Floer-theoretic pairings, and categorical operations (Manion, 2015, Kotelskiy et al., 2019, Kotelskiy et al., 2020).

1. Foundational setting

A 4-ended tangle is a properly embedded $1$-manifold with four boundary points. In the Fukaya-categorical formulation, a pointed 4-ended tangle is such a tangle together with a distinguished end, conventionally the top-left one, marked by $\ast$, and removing the four endpoints from $\partial D^3$ gives the 4-punctured sphere $(S^2,4)$, which serves as the boundary condition for the wrapped Fukaya category (Kotelskiy et al., 2020). In the bordered combinatorial formulation, one fixes four boundary points on the $y$-axis and encodes boundary data by decorated cleaved links transverse to that axis (Roberts, 2013).

The basic algebraic input is a boundary algebra. In Roberts’ construction this algebra is $\mathcal B\Gamma_2$, generated by idempotents corresponding to decorated cleaved links, bridge generators associated to surgery arcs, and decoration generators that flip a $+$-decoration to $-$ (Roberts, 2013, Roberts, 2013). The algebra carries a bigrading and a differential of bidegree $(1,0)$. The geometric input is the cube of resolutions of a tangle diagram together with decorations on the resulting circles, exactly as in Khovanov’s original construction, but now retaining the boundary pattern rather than closing the diagram immediately (Roberts, 2013).

A recurrent point of terminology is that several equivalent-looking theories use different choices of boundary presentation. Roberts works with half-space tangles and cleaved-link algebras; Kotelskiy–Watson–Zibrowius formulate pointed 4-ended tangles in the 3-ball and the 4-punctured sphere; Manion relates the bordered algebra to Khovanov’s arc algebra $H^2$ (Manion, 2015). These are not independent theories of unrelated objects. Rather, they are distinct models for the same four-ended cut-and-paste phenomenon, with different emphases on combinatorics, algebra, or symplectic geometry.

2. Bordered algebraic model

Roberts’ type D construction begins with a graded left module $\ast$0 over the idempotent ring $\ast$1, where $\ast$2, together with a structure map

$\ast$3

of bidegree $\ast$4 satisfying the curvature-zero relation

$\ast$5

Here $\ast$6 is multiplication in $\ast$7, $\ast$8 is the differential on $\ast$9, and $\partial D^3$0 is the sign twist coming from the internal $\partial D^3$1-grading (Roberts, 2013).

For a generic 4-ended tangle $\partial D^3$2, the module $\partial D^3$3 is freely generated over $\partial D^3$4 by states $\partial D^3$5, where $\partial D^3$6 is a $\partial D^3$7-resolution of the crossings, $\partial D^3$8 is a left matching closing the four loose ends, and $\partial D^3$9 decorates each resulting circle by $(S^2,4)$0 or $(S^2,4)$1 (Roberts, 2013). The structure map $(S^2,4)$2 is assembled from three types of terms: interior contributions coming from the usual Khovanov merge/split rules, boundary terms recording surgeries along right-bridges in the closure matching, and decoration terms associated to cleaved circles carrying $(S^2,4)$3-labels. The construction is explicitly modeled on bordered Heegaard Floer homology, but uses only combinatorial and diagrammatic methods (Roberts, 2013).

The complementary type A theory assigns to an inside tangle a strictly unital right $(S^2,4)$4-module $(S^2,4)$5 over the same algebra. The differential $(S^2,4)$6 is the Asaeda–Przytycki–Sikora differential restricted to the inside region, $(S^2,4)$7 gives the right action of idempotents, decoration generators, and bridge generators, and $(S^2,4)$8 appears when successive boundary changes interact through special relations in the algebra; one has $(S^2,4)$9 for $y$0 (Roberts, 2013). This asymmetry between type D and type A is intrinsic to the bordered formalism rather than an artifact of notation.

Manion gave an alternate construction in the language of Khovanov’s modules over $y$1, the four-point arc algebra generated by the two crossingless matchings $y$2 and $y$3 (Manion, 2015). In that reformulation, the dg module $y$4 of a 4-ended tangle is equivalent to a type D structure over $y$5, and Roberts’ bordered modules arise from the same data viewed over the smaller algebra $y$6 (Manion, 2015).

3. Gluing theorem and isotopy invariance

The central structural statement is the pairing theorem. If $y$7 is the type A structure for an inside tangle and $y$8 is the type D structure for the complementary outside tangle, then one forms the box tensor

$y$9

whose differential combines the $\mathcal B\Gamma_2$0-operations on the type A side, the type D operation $\mathcal B\Gamma_2$1, and multiplication in the boundary algebra. Roberts proves that this glued complex is canonically isomorphic, up to the overall shift $\mathcal B\Gamma_2$2, to the usual bigraded Khovanov chain complex of the closed link $\mathcal B\Gamma_2$3 obtained by gluing the two tangles (Roberts, 2013). In the type A paper the same statement is written as

$\mathcal B\Gamma_2$4

with chain homotopy equivalence of $\mathcal B\Gamma_2$5-graded differentials (Roberts, 2013).

This pairing formalizes a modular computation principle. The tangle pieces can be simplified independently, then recombined without recomputing the full cube of resolutions for the closed link. Roberts emphasizes this in the connect-sum example: the type A structure of the inside tangle for the right-handed trefoil and the type D structure of the outside tangle for the left-handed trefoil pair to recover the bigraded Khovanov homology of $\mathcal B\Gamma_2$6, including four torsion summands (Roberts, 2013).

Invariance is established at the level of bordered modules. For the type D side, planar isotopy of the endpoints and Reidemeister I–III moves in the interior change the resulting structure only by basis cancellations and homotopy equivalences, so its homotopy class is an invariant of the isotopy class of the tangle (Roberts, 2013). For the type A side, changing the crossing order yields a strictly isomorphic $\mathcal B\Gamma_2$7-module, and Reidemeister moves yield homotopy equivalent $\mathcal B\Gamma_2$8-modules (Roberts, 2013). Manion reproved these invariance and pairing properties in the $\mathcal B\Gamma_2$9-module framework, using explicit generators-and-relations descriptions of $+$0 and standard cancellation arguments (Manion, 2015).

A common misconception is that the bordered package is merely a repackaging of the link complex after closure. The gluing theorem shows the opposite: the bordered objects are genuine tangle invariants, and the closed-link Khovanov complex is recovered only after pairing compatible boundary data.

4. Immersed-curve and Fukaya-category reformulations

A major reinterpretation identifies four-ended tangle invariants with immersed multicurves in the 4-punctured sphere. Kotelskiy, Watson, and Zibrowius associate to a 4-ended tangle $+$1 a Bar–Natan multicurve $+$2, defined as the multicurve corresponding to the bounded bigraded type D structure associated to $+$3 under a classification theorem for such modules (Kotelskiy et al., 2019). The multicurve consists of immersed curves with local systems, taken up to homotopy of the underlying curves and equivalence of the local systems.

In this language, gluing becomes Floer theory. If a link $+$4 is obtained by gluing two pointed 4-ended tangles $+$5 and $+$6, then the reduced Bar–Natan homology of $+$7 is computed by the Floer homology of the corresponding multicurves. The same paper defines immersed-curve invariants $+$8 and $+$9 via mapping cones of the self-action $-$0 on the Bar–Natan object, and their pairings recover reduced and unreduced Khovanov homology of glued links (Kotelskiy et al., 2019). The theory also yields mutation statements: Conway mutation preserves reduced Bar–Natan homology over $-$1, and the reduced Khovanov curve is mutation-invariant; consequently $-$2 and Rasmussen’s $-$3-invariant are mutation-invariants in the sense stated there (Kotelskiy et al., 2019).

The Fukaya-categorical version places these tangle invariants inside the wrapped Fukaya category $-$4, whose objects are exact, graded, possibly immersed Lagrangians equipped with local systems, and whose morphisms are Floer complexes generated by intersection points, with $-$5 counting immersed bigons and higher $-$6 counting $-$7-gons (Kotelskiy et al., 2020). In that setting there are two Khovanov-theoretic tangle invariants for a pointed 4-ended tangle, $-$8 and $-$9, both realized as twisted complexes over $(1,0)$0. The main theorem of (Kotelskiy et al., 2020) proves that they agree after identifying the Bar–Natan algebraic model with a suitable subalgebra of the pillowcase algebra via an explicit $(1,0)$1-quasi-isomorphism.

This equivalence has two consequences emphasized in the literature. First, it shows that the immersed-curve description is not only heuristic but fully compatible with algebraic Khovanov tangle invariants. Second, it provides a concrete immersed-curve model whose wrapped Floer homology computes reduced Khovanov homology after gluing, and whose explicit nature admits algorithmic implementation (Kotelskiy et al., 2020).

5. Refined theories and alternative algebraic models

Several extensions preserve the four-ended cut-and-paste architecture while changing coefficients or target categories. Naisse and Putyra developed a covering arc algebra $(1,0)$2 for four boundary points over

$(1,0)$3

together with the theory of quasi-associative algebras and bimodules graded over a category with a $(1,0)$4-cocycle (Naisse et al., 2020). For a $(1,0)$5-tangle, the resulting chain complex is a $(1,0)$6-graded dg $(1,0)$7-bimodule, and specializations recover the even and odd theories: $(1,0)$8 gives ordinary Khovanov tangle homology over $(1,0)$9, while $H^2$0, $H^2$1 gives odd Khovanov tangle homology (Naisse et al., 2020). The gluing property is retained at the level stated იქ, and the construction is related to a level $H^2$2 cyclotomic half $H^2$3-Kac–Moody action.

Lawson, Lipshitz, and Sarkar defined stable homotopy refinements of Khovanov’s arc algebras and tangle invariants. In the four-ended case, the arc algebra $H^2$4 is promoted to a ring spectrum $H^2$5, and a 4-ended tangle $H^2$6 is assigned an $H^2$7-bimodule spectrum $H^2$8 whose chain homology recovers the ordinary tangle complex (Lawson et al., 2017). Applying a generalized homology theory produces a convergent cube spectral sequence whose $H^2$9-page is the classical bigraded Khovanov homology of the tangle (Lawson et al., 2017).

This spectral package extends to Chen–Khovanov platform algebras. The four-point arc algebra $\ast$00 admits a platform quotient $\ast$01, and the corresponding spectral platform algebra $\ast$02 and spectral tangle bimodules lift Chen–Khovanov’s invariants to spectra (Lawson et al., 2019). Their topological Hochschild homology recovers the classical Hochschild picture for annular Khovanov homology, and the spectrum-level theory carries the additional structure of THH and Steenrod operations in the manner described there (Lawson et al., 2019).

Another algebraic reformulation uses Ozsváth–Szabó strand algebras and DA bimodules. For four boundary points, Alishahi and Dowlin construct elementary DA bimodules for the identity, crossings, cups, and caps over the once-truncated strands algebra $\ast$03; iterated box tensor products assemble an oriented cube of resolutions, and forgetting the filtration produces objects homotopy equivalent to suitable Ozsváth–Szabó bimodules (Alishahi et al., 2019). This places Khovanov-theoretic 4-ended tangle invariants and knot Floer-type tangle invariants in a common bordered framework.

6. Extensions and current directions

The Roberts formalism has been extended beyond links in $\ast$04. For links in connected sums of orientable interval bundles over surfaces, cutting along a separating sphere yields type D and type A structures for tangles in the two halves, and gluing them along the common boundary recovers the Khovanov homology of the link (Du, 7 Mar 2026). In the $\ast$05 specialization, this reproduces the same four-ended cleaved-link algebra, type D map, type A module, and box-tensor differential familiar from the $\ast$06 setting (Du, 7 Mar 2026).

Recent work also studies operators on the four-ended category itself. Marian defines a 2-cabling operator on pointed framed 4-ended tangles, inducing an endofunctor on the homotopy category of type D structures over the Bar–Natan algebra $\ast$07 and, via the curve correspondence, an operator on the Fukaya category of the 4-punctured sphere (Marian, 1 Aug 2025). For cap-trivial tangles, the induced operator factors through the quotient setting $\ast$08, and the resulting geography theorem constrains the unique noncompact component of the associated immersed curve to be either the basic trefoil-axis curve or the straight line, up to mirroring and overall framing shift (Marian, 1 Aug 2025).

Across these developments, one structural theme remains constant. Four-ended tangle Khovanov theory is a gluing theory: local tangle data are encoded as modules, bimodules, twisted complexes, or immersed curves living on the boundary, and closed-link homology is recovered by algebraic tensor product or Floer-theoretic pairing. The coexistence of bordered, quiver-theoretic, immersed-curve, and spectral models does not indicate incompatibility. The established equivalences instead show that the four-ended setting is a particularly rigid and productive interface between Khovanov homology, bordered methods, and symplectic-categorical topology (Roberts, 2013, Kotelskiy et al., 2020).