On Khovanov local equivalence groups

Abstract: Inspired by work in Seiberg--Witten and Heegaard Floer homology, Dunfield, Schütz, and the author constructed an abelian "local equivalence group" built from pairs of chain complexes, and a homomorphism from the smooth concordance group to this group that uses the odd Khovanov complex and the Bar-Natan deformation of the even Khovanov complex. Lewark independently gave a related local equivalence group and concordance invariant, using only the Bar-Natan deformation. In this paper, we construct $(\mathbb{Z}/2\mathbb{Z})^\infty$ subgroups of Lewark's local equivalence group and the kernel of the map from our local equivalence group to Lewark's, as well as $\mathbb{Z}^\infty$ direct summands of Lewark's group and this kernel, invisible to the $s$-invariants. The construction gives at least six $\mathbb{Z}$-valued concordance homomorphisms linearly independent from $s$, $τ$, and Dai--Hom--Stoffregen--Truong's invariants.

• The paper establishes new algebraic subgroups in Khovanov local equivalence groups, identifying finite and infinite torsion and free summands.
• 2- Torsion elements of Lewark’s group $G$ are explicitly constructed and verified to lie in the kernels of all Rasmussen-type $s$-invariants.
• For certain coefficient rings, such as $\mathbb{Z}_{p}[h]/(ph)$, the local equivalence order is proven to be total, simplifying computations and aiding in the detection of non-trivial $2$-torsion and classes.

Scope and central problem

The paper studies the algebraic structure of Khovanov local equivalence groups introduced in work of Dunfield, Schütz, and Lipshitz and, independently, by Lewark. These groups are designed to retain information in Bar–Natan and odd/even Khovanov complexes that is invisible to Rasmussen’s $s$-invariant. The main groups are Lewark’s local equivalence group $G$ of knot-like complexes over $\mathbb{Z}[h]$, and the even–odd local equivalence group built from LEO triples. The latter records a Bar–Natan complex together with an odd Khovanov complex and their common reduction modulo $(2,h)$.

The paper establishes substantial new algebraic structure in both groups. Its principal conclusions are:

• $G$ contains, for every prime $p$, an independent subgroup isomorphic to $(\mathbb{Z}/2)^\infty$.
• The kernel of the quotient from the even–odd group to $G$ also contains a subgroup isomorphic to $(\mathbb{Z}/2)^\infty$.
• For every prime $p$, $G$0 contains a $G$1 direct summand lying in the kernel of all Rasmussen-type $G$2-invariants.
• The corresponding kernel in the even–odd group contains a $G$3 direct summand.
• At least six concordance homomorphisms derived from the even–odd construction are linearly independent of all $G$4-invariants, $G$5, and the Dai–Hom–Stoffregen–Truong homomorphisms.

The distinction between the existence of these algebraic classes and their realization by knots is maintained throughout. The paper constructs elements of the local equivalence groups, but does not show that the exhibited torsion classes are represented by actual knots.

Local equivalence groups and LEO triples

Let $G$6 be a graded $G$7-algebra. A knot-like complex is a finite free bigraded chain complex $G$8 such that, after inverting $G$9, it becomes chain homotopy equivalent to $\mathbb{Z}[h]$0 concentrated in homological grading zero. A local map $\mathbb{Z}[h]$1 is a chain map that becomes a homotopy equivalence after inverting $\mathbb{Z}[h]$2. Mutual existence of local maps defines local equivalence.

Tensor product induces an abelian group structure on local equivalence classes. The inverse of a class is represented by the dual complex. The resulting group is denoted $\mathbb{Z}[h]$3, with $\mathbb{Z}[h]$4 in the principal case. Its partial order is induced by the existence of local maps.

The LEO construction augments this data. For compatible rings $\mathbb{Z}[h]$5 and $\mathbb{Z}[h]$6, a LEO triple $\mathbb{Z}[h]$7 consists of a free complex $\mathbb{Z}[h]$8 over $\mathbb{Z}[h]$9, a knot-like complex $(2,h)$0 over $(2,h)$1, and a chain homotopy equivalence between their reductions modulo $(2,h)$2 and $(2,h)$3. The relevant local maps must be compatible on the reduced complexes. Tensor product again gives an abelian group. The forgetful map

$(2,h)$4

is a surjective homomorphism onto $(2,h)$5, and $(2,h)$6 is a direct summand of the LEO group.

This framework isolates two types of information. The Bar–Natan complex controls the image in $(2,h)$7, whereas the additional odd/even reduction data can survive in the kernel of the quotient. Consequently, invariants of the LEO group need not be detected by any invariant defined solely from the Bar–Natan complex.

Explicit $(2,h)$8-torsion in Lewark’s group

The first torsion construction begins with a five-generator complex $(2,h)$9 over $G$0. Its differential contains coefficients $G$1, $G$2, and $G$3, arranged so that after inverting $G$4 two contractible pairs can be cancelled, leaving the required Laurent polynomial tower. The complex is self-dual, while a local-equivalence invariant $G$5 shows that it is not locally trivial. Since duality gives inversion in $G$6, self-duality implies

$G$7

and hence the class has order two.

The paper generalizes this construction to complexes $G$8 indexed by a prime $G$9 and positive integers $p$0. Their differentials involve the coefficients

$p$1

and the complexes remain knot-like and self-dual. Passing to the $p$2-adic coefficient ring $p$3 makes the different parameters detectable. The invariant $p$4 is defined from the ideal generated by the images in the Laurent homology of quantum-degree-zero cycles. For a tensor product of distinct $p$5, the paper proves that the valuation $p$6 detects whether a given factor occurs.

The key structural result is

$p$7

inside $p$8. Base change to all $p$9-adic rings then separates the contributions of distinct primes. Thus, for every prime $(\mathbb{Z}/2)^\infty$0, $(\mathbb{Z}/2)^\infty$1 contains a copy of $(\mathbb{Z}/2)^\infty$2, and the copies corresponding to different primes are independent.

This result gives explicit torsion but not explicit quotient homomorphisms from $(\mathbb{Z}/2)^\infty$3 onto $(\mathbb{Z}/2)^\infty$4. The paper identifies the latter as a separate open problem.

The local algebra over $(\mathbb{Z}/2)^\infty$5 also yields constraints on torsion representatives. Every knot-like complex decomposes as a finite direct sum of indecomposable complexes, exactly one of which remains knot-like after inverting $(\mathbb{Z}/2)^\infty$6. Moreover, two indecomposable knot-like complexes that are locally equivalent are isomorphic. In particular, an indecomposable representative of a $(\mathbb{Z}/2)^\infty$7-torsion class must be isomorphic to its dual. Its total number of generators is congruent to $(\mathbb{Z}/2)^\infty$8 modulo $(\mathbb{Z}/2)^\infty$9; therefore every nontrivial indecomposable $G$0-torsion class has at least five generators.

These structural statements narrow the search space for torsion considerably. They also justify the paper’s computational focus on self-dual complexes rather than merely locally self-dual ones.

Torsion in the even–odd kernel

The second torsion construction lies in the kernel of the quotient from the LEO group to $G$1. The paper constructs a fixed knot-like complex $G$2 over $G$3 that is locally equivalent to the trivial complex, together with a family of nine-generator complexes $G$4 over $G$5 whose reductions identify with $G$6. The resulting LEO triples

$G$7

are self-dual and therefore have order two.

To distinguish them, the author reduces coefficients modulo $G$8. If $G$9, the class becomes trivial over the corresponding quotient. If $(\mathbb{Z}/2)^\infty$0, a direct analysis of a hypothetical local map leads to a parity contradiction. Consequently, the classes are pairwise distinct. Since an abelian group generated by infinitely many distinct elements of order two contains either a finite or an infinite elementary abelian subgroup, the family generates

$(\mathbb{Z}/2)^\infty$1

This construction is important because it proves that the additional even–odd information is not merely a refinement of the Bar–Natan local equivalence class. There is infinite torsion entirely invisible after projection to $(\mathbb{Z}/2)^\infty$2.

The $(\mathbb{Z}/2)^\infty$3 summand over $(\mathbb{Z}/2)^\infty$4

The main structural argument uses the coefficient ring

$(\mathbb{Z}/2)^\infty$5

which is a fiber product of the two discrete valuation rings $(\mathbb{Z}/2)^\infty$6 and $(\mathbb{Z}/2)^\infty$7. This decomposition permits a simultaneous analysis of the $(\mathbb{Z}/2)^\infty$8-adic and $(\mathbb{Z}/2)^\infty$9-adic parts of a minimal complex.

For an indecomposable minimal complex, the differential splits canonically as

$p$0

where $p$1 has coefficients divisible by $p$2 and $p$3 has coefficients divisible by $p$4. The relation $p$5 makes the mixed term vanish automatically. A complex is therefore encoded by two complexes, one over $p$6 and one over $p$7, together with an identification of their reductions modulo $p$8 and $p$9.

The paper proves that the local equivalence order on

$G$00

is total. This is a strong contrast with the original local equivalence groups, whose order is only partial. The proof constructs either a local map from the trivial complex to a given indecomposable complex or a local map in the opposite direction. The argument depends essentially on the special relation $G$01; the paper explicitly notes that the analogous order statement fails over $G$02 for $G$03.

The next step is a classification theorem. Every knot-like complex with vanishing $G$04 is locally equivalent to a standard complex

$G$05

whose generators form a linear chain and whose successive differentials alternate between powers of $G$06 and powers of $G$07. Signs record reversal of the corresponding differential. Standard complexes are locally ordered lexicographically by the reciprocal sequence

$G$08

In particular, two standard complexes are locally equivalent precisely when their parameter sequences agree.

The classification is proved in two ways. The first uses the representation theory of bunches of chains, following Bondarenko and its formulation by Burban–Drozd. The Smith normal form data of the $G$09-adic and $G$10-adic complexes form a representation of a bunch of chains; Bondarenko’s string-and-band decomposition identifies the knot-like summand with a string, which is exactly a standard complex. The second proof adapts the maximal-standard-subcomplex argument of Dai–Hom–Stoffregen–Truong. It constructs a maximal standard complex admitting a local map to the given complex and shows, using totality of the order and the dual complex, that the two classes coincide.

This theorem is the central algebraic input for the infinite-rank result. It reduces the local equivalence group to combinatorics of signed integer sequences.

Homomorphisms and independence from $G$11

For a standard complex, define integer-valued functions

$G$12

and

$G$13

The paper proves that all $G$14 and $G$15 are homomorphisms from the kernel of $G$16 to $G$17. The proof introduces an auxiliary trigraded theory over

$G$18

where the standard complexes correspond to the standard complexes used by Dai–Hom–Stoffregen–Truong in knot Floer homology. A grading construction identifies the mixed $G$19–$G$20 theory with this fixed-variable theory, and the known homomorphism properties transfer to the Khovanov setting. The $G$21 are obtained from the $G$22 by exchanging the two variables and correcting the grading asymmetry in the definition of local maps.

The resulting map from $G$23 has infinite rank. Together with the sequence of $G$24-invariants, the paper obtains a surjection of the form

$G$25

The first $G$26 records the family of $G$27-invariants, while the prime-indexed summands are detected by the $G$28 and $G$29 after base change to $G$30.

The implication is stronger than merely finding isolated invariants beyond $G$31. The group $G$32 has a direct summand of infinite rank lying in the common kernel of all Rasmussen-type invariants. Thus, the local equivalence group contains systematically organized information that the entire family of $G$33-invariants cannot detect.

The even–odd group and concordance homomorphisms

The same construction transfers to the LEO group. In characteristic two, the group of LEO triples over $G$34 is identified with a local equivalence group over $G$35. The $G$36 and $G$37 therefore define homomorphisms on this LEO group.

The classes constructed in the previous sections can be lifted to LEO triples whose projection to $G$38 is trivial. Their $G$39 and $G$40 values remain nonzero and independent. Consequently, the kernel of the quotient to $G$41 contains a $G$42 direct summand.

Applied to knots, the LEO construction yields at least six homomorphisms

$G$43

from the smooth concordance group to $G$44 that are linearly independent of one another, every characteristic $G$45-invariant, $G$46, and all Dai–Hom–Stoffregen–Truong homomorphisms. This is one of the paper’s strongest claims: the odd/even Khovanov comparison produces concordance homomorphisms not only beyond Rasmussen’s invariant, but also beyond the previously known infinite families from knot Floer homology.

The independence statement is established computationally by evaluating the homomorphisms on a collection of knots and row-reducing the resulting matrix. The argument is therefore only as strong as the correctness of the computed complexes and evaluations; the paper explicitly states that the author did not inspect the computational code directly.

Computational evidence

The computational section examines local equivalence types for knots through 16 crossings and for additional randomly generated strongly negatively amphicheiral knots. The results distinguish sharply between the torsion constructions and the infinite-rank homomorphisms.

For amphicheiral knots through 16 crossings, and for approximately 40,000 sampled strongly negatively amphicheiral knots with 19–26 crossings, the computed invariant $G$47 vanished. Since such knots are concordance self-inverse, this implies that none of the sampled knots realizes the constructed nontrivial $G$48-torsion classes. This does not suggest that the torsion classes cannot be realized by knots; it demonstrates only that they do not occur in the tested samples.

The infinite-rank invariants appear more readily. Among knots through 16 crossings, the computation found nonzero values of:

• $G$49 and $G$50 from the Bar–Natan complex over $G$51;
• $G$52 for $G$53 and $G$54 from the LEO construction;
• $G$55, $G$56, and $G$57 from knot Floer homology;
• the characteristic-zero and characteristic-two $G$58-invariants and $G$59.

The examples include knots for which the ordinary Bar–Natan invariant is trivial while the LEO invariant is nontrivial, such as $G$60 and several higher-crossing knots. Conversely, some knots have nontrivial Bar–Natan or knot Floer local equivalence types while the other constructions are trivial. This supports the algebraic distinction among the local equivalence theories.

The census also reveals relations that hold in this range but are not proved universally. For example, the data satisfy

$G$61

for the tested knots, and $G$62 whenever the Bar–Natan standard complex is locally equivalent to $G$63 or the trivial complex. The paper correctly treats these as empirical observations. In particular, it leaves open whether an auxiliary invariant $G$64 defined from the Bar–Natan complex always agrees with $G$65.

Limitations and open questions

The main limitation is the gap between algebraic local equivalence classes and topological realization. The paper constructs many torsion and infinite-rank classes in local equivalence groups, but it does not realize the explicit $G$66-torsion classes by knots. It also does not prove that every local equivalence class of knot-like complexes, or every LEO class, is represented by a knot.

Several structural problems remain unresolved:

• Whether $G$67 or the LEO group has torsion of order other than two.
• Whether either group has infinitely divisible elements.
• Whether explicit surjections onto $G$68 can be constructed.
• The full structure of $G$69, including an explicit basis if it is isomorphic to $G$70.
• Whether the computational relations among $G$71-invariants and the first $G$72 extend beyond the tested knots.

The computational conclusions also require qualification. The calculations were performed by software written or extended by Claude models, and the author reports having checked various sanity conditions but not having audited the code directly. The paper separates these computational claims from the hand-checked algebraic proofs, but the distinction is material when interpreting the knot census.

Conclusion

The paper gives the first substantial structural description of Khovanov local equivalence groups beyond their definition and their relationship with Rasmussen’s invariant. It proves abundant $G$73-torsion, constructs infinite-rank free summands invisible to all $G$74-invariants, and establishes a classification by standard complexes after passage to $G$75. The LEO theory contributes an additional infinite-rank kernel, yielding at least six concordance homomorphisms independent of $G$76, $G$77, and the Dai–Hom–Stoffregen–Truong invariants. The remaining central question is topological realization: how much of this algebraic structure is detected by actual knots and smooth concordance. (2610.37595)