Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

✓ 13 results · all verified · 7 also independently AI-judged
Every result on this page is machine-checked by a proof checker and read in full and owner-audited; the judge is an additional, independent cross-model AI review of the proofs. The 6 not AI-judged were verified by owner audit (typically over a confirmed judge false positive), not failures.

Hochschild Homology and Triply-Graded Link Homology

1 · Prerequisites

2 · Summary

This page develops Khovanov's realization of triply-graded link homology as Hochschild homology of Soergel bimodules. It fixes the reduced type-A polynomial ring R=Q[x1−x2,…,xm−1−xm] with its Sm-action, the invariant coordinates ti=xi+xi+1 and the unshifted bimodules Bi=R⊗RsiR, together with the unreduced cousins Bi′=R′⊗(R′)siR′ and the trivial polynomial factor Bi′≅Bi⊗QQ[ti] that records the source's R′=R⊗QQ[x1] convention in the si-equivariant form. Khovanov's generator complexes [R{2}→Bi] and [Bi{−2}→R{−2}] are defined and compared with the library's shifted Rouquier generators, and the termwise Hochschild complex of the resulting braid complex is defined, with the groups HHHc,h,p.

The main theorem identifies HHH with the reduced Khovanov-Rozansky homology of the closure. The proof runs through the a=0 specialization of the matrix-factorization construction: setting a=0 turns the local factorizations into folded Koszul complexes, the first rm relations of a closed marked resolution form a regular sequence with quotient the unreduced Soergel bimodule B′(D), and the remaining closure relations reproduce the diagonal Hochschild complex of B′(D), so that Hochschild homology is computed by the Koszul complex of the resolution and splits off two copies of the reduced theory, one of which matches the reduced Khovanov-Rozansky complex. The comparison also tracks the differentials through the wide-edge morphisms χ0,χ1 (with the corrected χ1-cone for negative crossings) and fixes the trigrading dictionary a=−h, q=p−h, t=c after the global correction (1,−1,0), anchored by the unknot and (2,n) normalizations. Two consequences close the page: HHH is an oriented-link invariant up to an overall trigrading shift, and its explicitly normalized graded Euler series recovers the v2 HOMFLYPT invariant of the closure. The Axiom of Choice enters only through the comparison of the bar and diagonal Koszul resolutions, the Markov invariance of the Khovanov-Rozansky theory, and the rationality and oriented-link descent of the Euler series, together with homogeneous basis choices when transferring unreduced invariance to the reduced graded vector spaces.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicableprecheck passjudge pass (gpt-6.1-sol)Open item page →

The reduced type-A polynomial ring and Soergel bimodules for the HHH construction

Definition

Fix m≥1 and let R′=Q[x1,…,xm] be the polynomial ring of The polynomial ring over a commutative ring as finitely supported coefficient sequences with convolution in m commuting indeterminates, with the place-permutation action of the symmetric group Sm, the grading deg⁡xi=2, the adjacent transpositions si=(i i+1) and the invariant rings (R′)si of The standard type-A reflection realization and its polynomial ring. Write xi for the i-th coordinate function and let Sm act by w⋅xi=xw(i).

The reduced ring. Put yi:=xi−xi+1 for 1≤i≤m−1 and R:=Q[y1,…,ym−1]=Q[x1−x2, x2−x3,…,xm−1−xm]⊂R′, the reduced polynomial ring of the HHH construction, with the restricted Sm-action and the induced grading deg⁡yi=2. On the displayed generators the restricted action is si(yi)=−yi,si(yi+1)=yi+yi+1,si(yi−1)=yi−1+yi,si(yj)=yj (∣i−j∣≥2), and it extends uniquely to a Q-algebra automorphism of R, because the displayed polynomials lie in R. Write Rsi⊂R for the invariant subring of the involution si; put zj=yj+12yi when ∣j−i∣=1 and zj=yj when ∣j−i∣≥2. Then si fixes each zj and sends yi to −yi; the linear change of variables is invertible, and explicitly Rsi=Q[zj (j≠i), yi2].

The invariant coordinate. For 1≤i≤m−1 put ti:=xi+xi+1∈R′. Then si fixes ti, and the substitution xi=12(ti+yi),xi+1=12(ti−yi),xi+2=xi+1−yi+1,xi−1=xi+yi−1, … expresses every coordinate as a polynomial in ti and the yj with coefficients in Q (2 is invertible). The substitution φ ⁣:Q[y1,…,ym−1,T]→R′ with yj↦yj and T↦ti is the linear change of variables (x1,…,xm)↔(y1,…,ym−1,ti), which is invertible over Q by the preceding display, hence an isomorphism of graded rings onto R′; consequently R′=R[ti]=R⊗QQ[ti] as graded rings. This polynomial extension is a free graded R-module on the infinite basis 1,ti,ti2,…. The extension of the invariant subring instead has rank two: R′ is free over (R′)si on 1,yi (or on 1,xi), since R=Rsi⊕yiRsi. Since si fixes ti and acts on the coefficients through R, (R′)si=Rsi[ti]=Rsi⊗QQ[ti]. Concretely (R′)si=Q[x1,…,xi−1, ti, xixi+1, xi+2,…,xm] and xixi+1=14(ti2−yi2); both displays generate the same subring because yi2=ti2−4xixi+1.

The reduced simple-reflection bimodule. Since 2 is invertible in Q, the averaging idempotent 12(1+si) splits R=Rsi⊕yiRsi as graded Rsi-modules: for f∈R the element 12(f−si(f)) equals yig with g=12(f−si(f))/yi∈Rsi. Hence the balanced tensor product of the graded (R,Rsi)-bimodule R with the graded (Rsi,R)-bimodule R Bi:=R⊗RsiR is free of rank two as a left R-module and as a right R-module, with left basis 1⊗1,1⊗yi and right basis 1⊗1,yi⊗1, of degrees 0,2 on either side; we use the library's internal shift convention (M{r})d=Md−r for graded modules, in which Bi carries no internal shift.

Comparison with the published type-A bimodule. The published type-A Soergel bimodule of a simple reflection works with the ambient ring R′=Q[x1,…,xm] in place of the reduced ring and with the shifted bimodule Bilib:=R′⊗(R′)siR′(1)=Bi′{−1}, where Bi′:=R′⊗(R′)siR′ is the corresponding unshifted balanced tensor; the dictionary is extended on the next item of this page. In particular the element 1⊗1 of the published bimodule has degree −1, while the element 1⊗1 of Bi has degree 0.

Source convention. Khovanov writes R′=R⊗QQ[xj] for the coordinate xj and chooses j=1. Read with x1 this is exact as a statement about graded rings, but the identification of invariant subrings (R′)si=Rsi[x1] is a literal equality of subrings of R′ only when si fixes x1, that is for i≥2; for i=1 the s1-invariant coordinate is t1=x1+x2, and all statements of this page are therefore stated with the invariant coordinate ti, the two forms being related by the substitution above. This is a convention correction, not a change of the source's construction.

Small cases. For m=1 there are no differences and R=Q; there are no simple reflections, and the empty-word bimodule is R itself. For m=2 one has R=Q[y1], Rs1=Q[y12] and B1=Q[y1]⊗Q[y12]Q[y1].

No choice principle is used in this definition.

DefinitionDefinition: Literature-sourcedProof: Not applicableprecheck passjudge pass (gpt-6.1-sol)Open item page →

Unreduced type-A Soergel bimodules and the trivial polynomial factor

Definition

Keep the ambient ring R′=Q[x1,…,xm], the reduced ring R⊂R′ generated by the differences yj=xj−xj+1, the invariant rings Rsi, the invariant coordinates ti=xi+xi+1 and the reduced bimodules Bi=R⊗RsiR of The reduced type-A polynomial ring and Soergel bimodules for the HHH construction, with the internal shift convention (M{r})d=Md−r of Associative graded algebras, bimodules, and internal shifts.

The unreduced simple-reflection bimodule. For a simple reflection si put Ri′:=Q[x1,…,xi−1, xi+xi+1, xixi+1, xi+2,…,xm]=(R′)si, the si-invariant subring of R′ (symmetric polynomials in xi,xi+1), and Bi′:=R′⊗Ri′R′ with no internal shift. Since R′=R[ti] and (R′)si=Rsi[ti] by The reduced type-A polynomial ring and Soergel bimodules for the HHH construction, and R′=Rsi[ti]⊕yiRsi[ti], the bimodule Bi′ is free of rank two as a left and as a right R′-module, with left basis 1⊗1,1⊗yi and right basis 1⊗1,yi⊗1, of degrees 0,2 on either side.

The trivial polynomial factor. Write t=ti and let Rt⊆R′ denote the polynomial ring Q[t] regarded as a subring of R′; the coordinate t is si-invariant, hence central. The assignment on Ri′-balanced tensors Φ ⁣:Bi′⟶Bi⊗QQ[t],Φ(r ta⊗r′ tb)=(r⊗r′)⊗ta+b(r,r′∈R, a,b≥0) is an isomorphism of graded (R′,R′)-bimodules, where R′=R[t] acts on the target by R on the Bi-factor and by multiplication on Q[t]. It is well defined: if h=∑ahata∈Rsi[t]=Ri′ with ha∈Rsi, then Φ(htc⊗1)=∑a(ha⊗1)⊗ta+c and Φ(1⊗htc)=∑a(1⊗ha)⊗ta+c agree because ha⊗1=1⊗ha in Bi. It is bijective because both sides are free of rank two over R′ on the respective generators 1⊗1,1⊗yi and (1⊗1)⊗1,(1⊗yi)⊗1, which Φ matches. The factor Q[t] is the trivial polynomial factor: the coordinate t acts by the same multiplication on both sides of the target, which is exactly the statement that it lies in the invariant ring (R′)si and hence balances in Bi′.

The reduced maps and their trivial extensions. Define degree-zero maps of graded (R,R)-bimodules bri ⁣:Bi→R,bri(a⊗b)=ab, rbi ⁣:R{2}→Bi,rbi(1)=yi⊗1+1⊗yi. Both are well defined: bri kills the balancing relation h⊗1−1⊗h for h∈Rsi, and rbi is left R-linear by construction while the identity r(yi⊗1+1⊗yi)=(yi⊗1+1⊗yi)r holds for every invariant generator zj because it balances, and for r=yi because yi2∈Rsi. These elements generate R, as proved in The reduced type-A polynomial ring and Soergel bimodules for the HHH construction. Both maps have degree zero in the internal grading, since deg⁡yi=2 and deg⁡1∈R{2}=2. Likewise bri′ ⁣:Bi′→R′,bri′(a⊗b)=ab, rbi′ ⁣:R′{2}→Bi′,rbi′(1)=(xi−xi+1)⊗1+1⊗(xi−xi+1), are well-defined degree-zero (R′,R′)-bimodule maps by the same argument with R′ in place of R and (R′)si in place of Rsi. Under Φ the elements 1⊗1↦(1⊗1)⊗1 and yi⊗1+1⊗yi↦rbi(1)⊗1 correspond, so bri′ and rbi′ restrict to the trivial-factor extensions of bri and rbi.

Products over the layers. Let D be a marked MOY resolution with r wide edges, at positions s1,…,sr in the layer order, and put B′(D):=⨂j=1rBsj′,B(D):=⨂j=1rBsj, the balanced tensor products over R′ respectively over R in the layer order. Choose the common invariant coordinate t=(x1+⋯+xm)/m; then R′=R[t] and t is fixed by every simple reflection. Applying the preceding single-factor construction with this t gives an isomorphism of graded (R′,R′)-bimodules B′(D)⟶B(D)⊗QQ[t]. Indeed, each factor is Bsj⊗Q[t], and balancing over R[t] combines the polynomial factors by multiplication. The inverse sends (b1⊗R⋯⊗Rbr)⊗ta to the tensor with ta in its first factor. For r=0 this is the ring identity R′=R[t]. The two maps preserve the left and right actions and grading. An arbitrary coordinate such as x1 need not act equally on both sides; its left and right actions are recovered from x1=t+(x1−t), with x1−t∈R.

Normalization dictionary with the published bimodule. The published The Soergel bimodule Bi of a simple reflection has Bilib=R′⊗(R′)siR′(1)=Bi′{−1} because the Elias-Williamson shift (1) is the internal shift {−1} in the library convention, that is M(r)=M{−r}; equivalently Bi′=Bilib(−1)=Bilib{1}. The dictionary M(r)=M{−r} is used throughout this page, and the generator 1⊗1 of Bilib has degree −1 while the generator 1⊗1 of Bi′ has degree 0.

Source convention. Khovanov writes the polynomial factor as Q[x1]. The ring splitting R′=R[x1] is valid, but a literal trivial factor for the bimodule actions uses a common invariant coordinate t as above. The change of coordinates x1=t+(x1−t) translates the source's ring notation into this bimodule description.

No choice principle is used in this definition.

DefinitionDefinition: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

Khovanov's generator complexes for the HHH construction

Definition

In the reduced setting of The reduced type-A polynomial ring and Soergel bimodules for the HHH construction define, for 1≤i≤m−1, degree-zero maps of graded (R,R)-bimodules bri ⁣:Bi→R,bri(a⊗b)=ab(bri(1⊗1)=1), rbi ⁣:R{2}→Bi,rbi(1)=yi⊗1+1⊗yi,yi=xi−xi+1. Khovanov's generator complexes are the bounded complexes of graded (R,R)-bimodules with degree-zero differentials F(σi):=[ R{2}→  rbi  Bi ],F(σi−1):=[ Bi{−2}→  bri  R{−2} ], where in F(σi) the term R{2} sits in cohomological degree −1 and the term Bi in degree 0, and in F(σi−1) the term Bi{−2} sits in cohomological degree 0 and the term R{−2} in degree 1. For a signed word σ=σi1ϵ1⋯σirϵr put F(σ):=F(σi1ϵ1)⊗R⋯⊗RF(σirϵr), the signed tensor totalization of Bounded graded bimodule complexes and signed tensor totalization, with the empty word giving the unit complex R; it is a bounded complex of graded (R,R)-bimodules with degree-zero differentials, whose terms carry a cohomological and an internal grading.

Normalization comparison with the library's Rouquier complexes. The library's The positive and negative Rouquier generator complexes is stated for the ambient polynomial ring Q[x1,…,xm]; its reduced instance is obtained by replacing that ring by the reduced ring R of The reduced type-A polynomial ring and Soergel bimodules for the HHH construction and Rsi by the invariant subring. In that instance the shifted bimodule is Bilib=R⊗RsiR(1)=Bi{−1},soFi=[ Bi{−1}→εiR{−1} ],Fi−1=[ R{1}→ηiBi{−1} ], with the B-term in cohomological degree 0 in both complexes and ηi(1)=αi⊗1+1⊗αi, where αi is the balanced root. Letter by letter, F(σi)≅Fi−1{1},F(σi−1)≅Fi{−1} in the library's {r} notation: the shift {1} turns R{1} into R{2} and Bi{−1} into Bi, and the balanced root satisfies αi=κi yi with κi=±1 a unit, so rbi and ηi differ by the unit κi; the same computation, with multiplication in both differentials, matches F(σi−1) with Fi{−1}. Hence for a word σ the complex F(σ) is, up to the overall internal shift {ϵ1+⋯+ϵr} (the writhe), the Rouquier complex The Rouquier complex of a braid word of the generator-inverted word σi1−ϵ1⋯σir−ϵr; by The Rouquier complex is well defined up to canonical homotopy equivalence it is determined by the underlying braid up to canonical homotopy equivalence and that shift.

Caveats. The pairing of generator and complex (the R-term below the B-term for σi) is the one matched by the positive-crossing cone of The positive and negative Khovanov-Rozansky crossing complexes; the comparison with the library's complexes is an isomorphism in the homotopy category, not an equality of complexes; the direction of the word comparison (generator inversion) is forced by the opposite pairing used in The positive and negative Rouquier generator complexes; and the reduced instance inherits the ambient homotopy comparisons by the polynomial-extension and specialization argument in step 3.1.

Facts & Assumptions

Given: the reduced ring R, its invariant subrings Rsi, the bimodules Bi=R⊗RsiR and the elements yi=xi−xi+1 of The reduced type-A polynomial ring and Soergel bimodules for the HHH construction, and the maps bri,rbi of the definition.

[L1]

The maps bri ⁣:Bi→R and rbi ⁣:R{2}→Bi are well-defined degree-zero maps of graded (R,R)-bimodules, with rbi(1)=yi⊗1+1⊗yi (Unreduced type-A Soergel bimodules and the trivial polynomial factor).

[L2]

The reduced instance of the library's generator complexes has Bilib=Bi{−1} and differentials the multiplication map εi(a⊗b)=ab and ηi, with ηi(1)=αi⊗1+1⊗αi and αi=κi yi, κi=±1; the B-term sits in cohomological degree 0 in both complexes (The positive and negative Rouquier generator complexes, Unreduced type-A Soergel bimodules and the trivial polynomial factor).

[L3]

The signed tensor totalization of bounded complexes of graded bimodules has the Koszul total differential, terms Fp⊗RGq in degree p+q, and internal degrees adding; shifts satisfy (M{r1})d=Md−r1 and tensor products of shifts add (Bounded graded bimodule complexes and signed tensor totalization).

[L4]

The Rouquier complex F(β) of a braid element β is well defined up to canonical homotopy equivalence: two signed words for the same braid give complexes isomorphic in the homotopy category by the canonical normalized maps (The Rouquier complex of a braid word, The Rouquier complex is well defined up to canonical homotopy equivalence).

Proof

technique · direct
1.1L1L3givenalgebra

Both displayed two-term complexes are complexes of graded bimodules: the differentials are degree-zero bimodule maps by [L1], and a composite of two differentials has no source or no target, so it is zero. The tensor totalization of finitely many such complexes is a bounded complex by [L3], and the empty word gives R.

2.1L1L2step 1.1algebra

Comparison at the positive generator. F(σi) has terms R{2} in cohomological degree −1, Bi in degree 0, differential rbi. The complex Fi−1{1} has terms R{1}{1}=R{2} in degree −1, Bi{−1}{1}=Bi in degree 0, differential ηi. By [L2], ηi(1)=κi rbi(1) with κi a unit; multiplying the generator of the degree −1 term by the unit κi−1 therefore intertwines ηi with rbi and gives an isomorphism of complexes, hence a homotopy equivalence.

2.2L1L2step 1.1algebra

Comparison at the negative generator. F(σi−1) has terms Bi{−2} in cohomological degree 0, R{−2} in degree 1, differential bri (multiplication). The complex Fi{−1} has terms Bi{−1}{−1}=Bi{−2} in degree 0, R{−1}{−1}=R{−2} in degree 1, and differential εi=bri, the ordinary multiplication map. Thus these two shifted negative-letter complexes are equal, in particular isomorphic and homotopy equivalent.

3.1L2L3L4step 2.1step 2.2algebra∎

Transfer the ambient comparisons and keep track of the word. Put R′=R[t], with t=(x1+⋯+xm)/m. By Unreduced type-A Soergel bimodules and the trivial polynomial factor, every ambient generator and differential is the reduced one extended by Q[t]; balanced products have the same property. Ambient bimodule maps and homotopies are t-linear. Quotienting their identities by t therefore gives homotopy-inverse maps between the reduced word complexes, with the transitivity and tensor compatibility of [L4]. This uses specialization of actual homotopy identities, so no flatness of R′/(t) is required. These are the comparisons inherited from the fixed ambient normalized system. Tensoring steps 2.1 and 2.2 then identifies F(σ) with the reduced Rouquier complex of the generator-inverted word, shifted internally by the writhe {ϵ1+⋯+ϵr}; no cohomological shift occurs. Generator inversion preserves the Artin relations, and writhe is invariant under those relations and inverse cancellation, so words for the same braid have the stated canonical homotopy comparison. For m=1 only the unit complex occurs.

DefinitionDefinition: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)Open item page →

The termwise Hochschild complex of a Rouquier complex and the groups HHH

Definition

Let F(σ) be the generator complex of Khovanov's generator complexes for the HHH construction, a bounded complex of graded (R,R)-bimodules with degree-zero differentials, and let HHh(R,−) be Hochschild homology, computed by the chain complexes C∙(R,−) of Hochschild chains and Hochschild homology with coefficients. Since every differential dj ⁣:F(σ)j→F(σ)j+1 is a map of R-bimodules, it commutes with the Hochschild faces and induces a chain map C∙(R,F(σ)j)→C∙(R,F(σ)j+1); since dj+1dj=0, the induced maps give a cochain complex of graded Q-vector spaces for every h≥0, (HHh(R,F(σ)∙), HHh(R,d∙)), the termwise Hochschild complex in Hochschild degree h (Termwise Hochschild homology and iterated homology applied to the bounded complex of R-bimodules F(σ); its internal grading is inherited). The HHH groups of σ are HHHj,h,d(σ):=Hj(HHh(R,F(σ)∙))d,j∈Z, h≥0, d∈Z, the cohomology in Rouquier degree j and internal degree d of the termwise complex, a trigraded Q-vector space.

This is the componentwise construction of Beliakova-Putyra-Wehrli, section 3.8.6, equation (3.44) (Khovanov's construction is the case of the Rouquier complex of a braid word), and it is not the total Hochschild hyperhomology of the complex F(σ) of Hochschild hyperhomology of a bounded bimodule complex: the latter is the abutment of a spectral sequence whose second page consists of the iterated homology groups of the termwise complex (Termwise Hochschild spectral sequence of a bounded bimodule complex), so the termwise groups carry the finer (j,h,d) trigrading while the hyperhomology retains only j−h up to filtration.

Caveats. The definition uses the chain-level Hochschild complexes and is choice-free: the identification of HHh with Tor⁡hRe(R,−) assumes the Axiom of Choice and is not needed here. The trigrading (j,h,d) (Rouquier degree, Hochschild degree, internal degree) is not the trigrading (a,q,t) of the comparison, which is fixed in The Koszul-Hochschild comparison respects crossing differentials and trigradings; the source's HHH is the cohomology of the termwise complex, that is the second page of the spectral sequence, not the hyperhomology abutment.

Remarks

The difference between the termwise theory and the total Hochschild hyperhomology is worked out on the later examples page in Termwise and total Hochschild theories have different grading outputs ↗; nothing in the definition depends on that item.

Facts & Assumptions

Given: the reduced ring R, the generator complex F(σ) of Khovanov's generator complexes for the HHH construction, and the Hochschild chain complexes C∙(R,−) of Hochschild chains and Hochschild homology with coefficients.

[L1]

F(σ) is a bounded complex of graded (R,R)-bimodules, every differential is a degree-zero map of R-bimodules, and each term is a finitely generated free graded R-module (Khovanov's generator complexes for the HHH construction, Bounded graded bimodule complexes and signed tensor totalization).

[L2]

Ch(R,M)=M⊗QR⊗Qh for h≥1 and C0(R,M)=M, with faces δi and boundary bh=∑i(−1)iδi; HHh(R,M)=Hh(C∙(R,M)), and a map of R-bimodules induces a chain map of the Hochschild complexes (Hochschild chains and Hochschild homology with coefficients).

[L3]

If every differential dFi ⁣:Fi→Fi+1 of a bounded complex F of k-central A-bimodules is a map of A-bimodules, then the induced maps HHj(A,dFi) make (HHj(A,F∙),HHj(A,dF∙)) a cochain complex, with Hi(HHj(A,F∙)) its cohomology; the construction is not defined to coincide with the hyperhomology, and its relationship to the hyperhomology is through the spectral sequence (Termwise Hochschild homology and iterated homology).

[L4]

The Hochschild hyperhomology HHhyper,n(A,F) of a bounded complex of k-central A-bimodules is the cohomology of the total complex T∙(A,F) with D=dF+(−1)ib on the summand Cj(A,Fi), and the iterated homology of the termwise complex is the second page of the associated spectral sequence, which abuts to the hyperhomology with the image filtration (Hochschild hyperhomology of a bounded bimodule complex, Termwise Hochschild spectral sequence of a bounded bimodule complex).

Proof

technique · direct
1.1L1L2givenalgebra

Each differential of F(σ) commutes with the Hochschild faces, hence induces a chain map of the Hochschild complexes. By [L1] dj is a degree-zero map of R-bimodules; by [L2] the Hochschild faces are built from the bimodule actions and multiplication in R, so dj∘δi=δi∘dj for every face, and therefore dj commutes with the boundary b and defines a chain map C∙(R,F(σ)j)→C∙(R,F(σ)j+1).

2.1L1L2L3step 1.1algebra

The induced maps make a cochain complex in each Hochschild degree. Since dj+1dj=0 as bimodule maps, the composite chain map is induced by the zero map, hence is zero and induces the zero map on homology; identities induce identities and composition is respected because the construction is functorial in the coefficient bimodule. By [L3] the family (HHh(R,F(σ)∙),HHh(R,d∙)) is therefore a cochain complex of graded Q-vector spaces for every h, with cohomology Hj as in the definition.

3.1L1L2L3step 2.1algebra

The trigrading is well defined and finite in each degree. The internal grading is preserved by [L1], so all induced Hochschild maps have degree zero. For fixed h and internal degree d, the chain group F(σ)j⊗QR⊗h has finite-dimensional degree-d part: F(σ)j is a finite direct sum of shifts of the positive-degree polynomial ring R, and the other h factors are the same polynomial ring, so only finitely many monomials of the required total degree occur. Its subquotient HHh and the subsequent bounded cochain homology therefore have finite-dimensional graded pieces. Thus HHHj,h,d is defined for every j∈Z, h≥0, d∈Z; the possible j are bounded by the finite word complex.

4.1L4step 2.1algebra∎

Distinguish termwise from total. By [L4] the total hyperhomology complex combines b and dF into one differential, its homology is the abutment, and the iterated homology of the termwise complex of step 2.1 is the second page of the spectral sequence; consequently the termwise groups carry the separate (j,h,d) trigrading whereas the hyperhomology retains the total degree and the internal degree together with a finite filtration. The definition therefore records the second-page groups, as in the source, and asserts nothing about degeneration of the spectral sequence.

DefinitionDefinition: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

The reduced Khovanov-Rozansky homology

Definition

Let D be a marked braid diagram with nonempty closure, so its label count satisfies m≥1 and let H(D)=⨁j,k,lHk,lj(D) be the triply graded groups of The Khovanov-Rozansky complex and trigraded braid homology, built over the polynomial ring Q[a,x1,…,xm] generated by the coefficient variable a and the m strand labels. The reduced Khovanov-Rozansky homology H‾(D) is defined by repeating that construction verbatim with the label ring replaced by the ring of differences from one chosen coordinate, Q[a,x1,…,xm] ⇝ Q[a,x2−x1,…,xm−x1], the coefficient a being retained: the same marked MOY resolutions, arc and wide-edge factorizations, positive and negative crossing cones, removal of contractible summands and induced differentials, but with the m strand labels replaced by the m−1 differences x2−x1,…,xm−x1; the trivial variable of the one-mark circle is thereby dropped. The result H‾(D)=⨁j,k,lH‾k,lj(D) is again a triply graded Q-vector space, with the trigrading of that item.

By Khovanov-Rozansky II, end of section 1, one has a trigraded isomorphism H(D)≅H‾(D)⊗QQ[x] in which x is the trivial variable, and in the reduced theory the unknot has one-dimensional homology. The splitting is a statement about the two constructions, not a definition of H‾(D).

Grading normalization and the unknot. In the trigrading of The Khovanov-Rozansky complex and trigraded braid homology the one-mark circle has cohomology Q[x]{−1,1} at the resolution level (the companion computation of the matrix-factorization page, indexed there by the first and second bigrading); the reduction removes the trivial factor, leaving the one-dimensional reduced group in bidegree (−1,1) at that level, and in the full trigraded theory the unknot class of the reduced homology sits in the raw tridegree (−1,1,0) in (k,l,j) order, whose image under the global correction (1,−1,0) of The Koszul-Hochschild comparison respects crossing differentials and trigradings is the class (0,0,0).

Caveats. H‾(D) is a construct of the reduced label ring and is not defined here by quotienting an arbitrary presentation of H(D); the splitting H(D)≅H‾(D)⊗QQ[x] is part of the source's assertion, recorded here with its grading conventions. The trivial variable x is a polynomial variable carried by the one-mark circle and is not the coefficient variable a. Under AC (The Axiom of Choice), invariance of H‾(D) up to an overall trigrading shift is inherited from Khovanov-Rozansky braid homology is an oriented link invariant up to shift and is used in HHH is an oriented-link invariant up to an overall trigrading shift; the trigrading normalization is the one of The Khovanov-Rozansky complex and trigraded braid homology, and no Wu regrading is asserted here.

The coordinate reduction and polynomial splitting are choice-free. The oriented-link-invariance claim uses AC through the indicated Markov supplier and for homogeneous basis choices in step 3.1. The zero-strand tensor unit Q[a] has no chosen label x1 and is excluded from this reduced construction.

Facts & Assumptions

Given: the nonempty marked braid diagram D, its label ring Q[a,x1,…,xm], and the chosen coordinate x1.

[F1]

The raw braid complex is assembled from arc and wide-edge rows and crossing maps over the shared polynomial coefficient ring; its outer cohomology is H(D) (The Khovanov-Rozansky complex and trigraded braid homology, The factorization of a marked MOY graph).

[F2]

Compatible homogeneous Koszul row operations are factorization isomorphisms. In particular the pair (0,q),(a,β) may be replaced by (0,q−tβ),(a,β) for homogeneous t of bidegree (0,2) (Koszul row operations and variable exclusion preserve homotopy type).

[F3]

After the indicated local row changes, the two crossing maps have forms 1⊗ψ(x4−x2) and 1⊗ψ′(x4−x2), with common first row (a,x1+x2−x3−x4) and second rows (0,x2−x3) or (0,(x2−x3)(x4−x2)) (The wide-edge morphisms chi-zero and chi-one).

[F4]

Under AC, a finite Markov sequence compares two ambient-isotopic closed braids by the explicit local equivalences and their stated shifts (Khovanov-Rozansky braid homology is an oriented link invariant up to shift, The Axiom of Choice).

Proof

technique · homogeneous row gauges remove the common translation coordinate; free polynomial extension preserves both cohomology operations
1.1F1F2algebra

Put t=x1 and uj=xj−x1, j>1, with u1=0. This gives the graded polynomial-ring isomorphism Q[a,x1,…,xm]=Q[a,u2,…,um][t], with deg⁡t=(0,2). Every arc entry is a difference and is independent of t. A wide-edge linear row has entry β independent of t, whereas its quadratic entry is q=q0+tβ, where q0 is the same quadratic expression in the u labels. Apply [F2] to replace that second row by (0,q0). Thus all resolution factorizations are polynomial extensions of the corresponding reduced factorizations.

2.1F1F2F3step 1.1algebra

The same reduction respects the crossing cube. In the local forms of [F3], the entries x1+x2−x3−x4, x2−x3 and x4−x2 are all unchanged by the common translation. The wide-row operation subtracting x2β=(t+u2)β is the gauge of step 1.1 followed by subtraction of u2β. Both flip matrices are therefore independent of t in these bases. Consequently the entire outer complex of factorizations is isomorphic to the reduced one extended by Q[t], including both crossing signs and all gradings. Taking inner and then outer cohomology commutes with this free extension: its monomial basis makes it a direct sum of shifted copies of each complex, with componentwise differentials, kernels and images. This proves H(D)≅H‾(D)⊗QQ[t].

2.2F1step 1.1algebra

For the one-mark circle the row is (a,0); setting its only label to zero leaves Q[a]→aQ[a]{−1,1}→0Q[a]. Its odd inner cohomology is Q{−1,1} and its even inner cohomology is zero because multiplication by a is injective. The sole outer term is in degree0, giving raw (k,l,j)=(−1,1,0). This verifies the stated reduced unknot value and grading.

3.1F1F4step 1.1step 2.1algebra∎

Inherit invariance at the level of graded vector spaces. Let D1,D2 have isotopic nonempty closures. Under [F4] their unreduced dimensions agree with a shift (j0,k0,l0). Write bD(j,k,l)=dim⁡Hk,lj(D) and rD(j,k,l)=dim⁡H‾k,lj(D); these dimensions are finite because both constructions use finite crossing cubes and finitely many polynomial variables of positive second degree. The splitting of step 2.1 gives bD(j,k,l)=∑d≥0rD(j,k,l−2d), a finite sum in each degree since the second gradings are bounded below. Consequently rD(j,k,l)=bD(j,k,l)−bD(j,k,l−2), so the same shift relates the reduced dimensions. Under AC choose bases in all homogeneous pieces; equal finite dimensions then give a trigraded vector-space isomorphism with that shift. This proves the asserted noncanonical invariance without assuming that arbitrary unreduced equivalences preserve a chosen label coordinate. Changing the chosen label only makes an invertible linear change among the differences, so the constructions of steps 1.1 and 2.1 give the same polynomial splitting. AC is used through [F4] and for the homogeneous basis choices.

Remarks

The one-strand normalization recorded above is computed on the later examples page in The trivial one-braid and the grading normalization ↗; nothing in the definition depends on that item.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)Open item page →

Setting a to zero in a closed KR factorization gives the layer-by-layer Koszul complex

Statement

Let D be a closed marked MOY resolution with r wide edges and m strands, carrying marks (i,j) with 0≤i≤r, 1≤j≤m, as in Khovanov's figure 4: for each layer 1≤i≤r the i-th wide edge occupies two adjacent positions s,s+1, and the variables xi,j label the marks. Let C(D) be the factorization of The factorization of a marked MOY graph built from the arc and wide-edge factorizations of Arc and wide-edge Khovanov-Rozansky factorizations over the shared polynomial ring R~:=Q[a, xi,j:0≤i≤r, 1≤j≤m] (the internal marks are retained as coefficients), with potential wD=a∑pϵpxp over the boundary points; for a closed graph wD=0 and C(D) is a genuine complex of graded free R~-modules.

Write S:=R~/(a)=Q[xi,j] and form, in S, the (r+1)m-element sequence βi=xi,s+xi,s+1−xi−1,s−xi−1,s+1,γi=xi,sxi,s+1−xi−1,sxi−1,s+1(1≤i≤r), the differences xi,j−xi−1,j for the positions j∉{s,s+1}, and the closure differences x0,j−xr,j (1≤j≤m); let K(D) be the Koszul complex of this sequence over S (Koszul Complex Of A Sequence With Coefficients). Then:

(a) the specialization C(D)∣a=0:=C(D)⊗Q[a]Q is a complex (d2=0) of graded free S-modules;

(b) C(D)∣a=0 is isomorphic, as a Z/2-graded complex of graded S-modules (a factorization with d2=0), to the folding by parity of K(D): the even part is ⨁p evenK(D)p and the odd part is ⨁p oddK(D)p, with the Koszul differential, up to a global sign on the odd part that is absorbed by a change of basis;

(c) the isomorphism carries the bidegree shifts of the arc and wide-edge factorizations: a linear relation xi,j−xi−1,j or βi occupies a term of bidegree (−1,1) and a quadratic relation γi a term of bidegree (−1,3), so that every differential has bidegree (1,1) for deg⁡a=(2,0), deg⁡xi,j=(0,2) and the library shift convention (M{r1,r2})(k,l)=M(k−r1,l−r2).

Facts & Assumptions

Given: a closed marked MOY resolution D with r wide edges and m strands, marks (i,j), positions s of the wide edges, the shared ring R~, the factorization C(D) and the sequence and Koszul complex of the statement.

[L1]

A bigraded matrix factorization with potential w over R~ consists of free bigraded modules and differentials of bidegree (1,1) with d2=w⋅id; on the closed graph the potential is wD=a∑pϵpxp, and setting a=0 makes d2=0 and kills exactly the a-linear matrix entries (Bigraded matrix factorizations with a potential).

[L2]

The arc factorization of an arc with endpoint labels x1,x2 is Cc=[S→aS{−1,1}→x1−x2S], the two-term row (a, x1−x2): its even part is S in bidegree (0,0), its odd part is S{−1,1}, the map even-to-odd is multiplication by a and the map odd-to-even is multiplication by x1−x2. The wide-edge factorization of a wide edge with four labels x1,x2,x3,x4 is the tensor product of the rows (a, x1+x2−x3−x4) and (0, x1x2−x3x4), with middle terms S{−1,1}⊕S{−1,3} (Arc and wide-edge Khovanov-Rozansky factorizations).

[L3]

C(D)=⨂cCc⊗⨂tCt is the tensor product over the shared variables of the local arc and wide-edge factorizations, its potential is the sum of the local potentials, and for a closed graph wD=0, so C(D) is a genuine complex of graded R~-modules (The factorization of a marked MOY graph).

[L4]

For a sequence f1,…,fN in a commutative ring S the Koszul complex K(f1,…,fN;S) has degree-p term ⋀pSN and differential d(ei1∧⋯∧eip)=∑t(−1)t−1ei1∧⋯eit^⋯∧eip⊗fit; its degree-zero term is S with differential zero (Koszul Complex Of A Sequence With Coefficients).

[L5]

For finite sequences x,y there is a signed chain isomorphism K(x,y;S)≅K(x;S)⊗SK(y;S) (Koszul Complex Concatenation Tensor Isomorphism).

Proof

technique · direct
1.1L1L2givenalgebra

Specialize a=0. By [L1] each local factorization has d2=a⋅(linear form)⋅id and setting a=0 makes d2=0 and kills the a-linear entries. By [L2] the arc row (a, x1−x2) becomes the Z/2-graded complex with even part S, odd part S{−1,1}, even-to-odd map 0 and odd-to-even map x1−x2; the wide-edge tensor of rows (a, β) and (0, γ) becomes the tensor product of the two rows (0,β) and (0,γ).

1.2L2L4givenalgebra

Identify each local a=0 complex with the folding of the Koszul complex of its relations. The arc of labels x1,x2 gives, with f=x1−x2, the folding of K(f;S): K1=S{−1,1}→fK0=S has even part S and odd part S{−1,1} with the same maps. The wide edge gives the folding of K(β,γ;S): place the Koszul symbols for β and γ in the shifted bidegrees (−1,1) and (−1,3), so that K0=S, K1=S{−1,1}⊕S{−1,3}, K2=S{−2,4} and every differential has bidegree (1,1); the tensor product of the two rows (0,β),(0,γ) has even part S⊕S{−2,4}, odd part S{−1,1}⊕S{−1,3} and differentials matching those of K(β,γ) up to a global sign on the odd part.

2.1L1L3step 1.2algebra

Tensor over the layers. The a=0 specialization of a tensor product of factorizations is the tensor product of the a=0 specializations, because the product differential is the sum over the factors of the local differentials tensored with the other factors; the Koszul signs on the product are the Koszul signs of the total complex. By 1.2 the result is the tensor product of the foldings of the local Koszul complexes, which is the folding of the tensor product of those Koszul complexes.

3.1L4L5step 2.1algebra

Apply Koszul concatenation. Take the local relations in the layer order, at layer i first the two wide-edge relations βi,γi and then the differences xi,j−xi−1,j for j∉{s,s+1}, and put the closure differences last. By [L5] the tensor product of the local Koszul complexes over the shared polynomial ring is isomorphic, with Koszul signs, to the Koszul complex of the concatenated sequence, which is K(D). Combining with step 2.1 gives an isomorphism of Z/2-graded S-complexes from C(D)∣a=0 to the folding of K(D), up to the global sign on the odd part that a change of basis absorbs.

4.1L1L2L3step 3.1algebra∎

Read off the shifts and the closure hypothesis. In the arc row the odd generator 1⊗1 of S{−1,1} has bidegree (−1,1), and in the wide edge the two odd generators have bidegrees (−1,1) and (−1,3), exactly the shifts displayed in [L2] for the middle terms; the differential S{−1,1}→S has bidegree (1,1) because x1−x2 has bidegree (0,2), and likewise γi:S{−1,3}→S has bidegree (1,1) because x1x2−x3x4 has bidegree (0,4). For the closed graph the potential vanishes by [L3] and there are no boundary points, so no boundary variable remains and the sequence has exactly (r+1)m elements; the isomorphism of step 3.1 is therefore an identification of the a=0 specialization of C(D) with the folding of K(D), carrying the displayed shifts.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)Open item page →

The first-layer relations of a closed MOY resolution form a regular sequence

Statement

Keep the closed marked MOY resolution D, the ring R~=Q[xi,j:0≤i≤r, 1≤j≤m] and the (r+1)m-element sequence of Setting a to zero in a closed KR factorization gives the layer-by-layer Koszul complex, written in the layer order: at layer 1≤i≤r, with the i-th wide edge at positions s,s+1, the two relations βi=xi,s+xi,s+1−xi−1,s−xi−1,s+1,γi=xi,sxi,s+1−xi−1,sxi−1,s+1, then the m−2 differences xi,j−xi−1,j for j∉{s,s+1}; the last m elements are the closure differences x0,j−xr,j (1≤j≤m).

Then the first rm elements, taken in the layer order i=1,…,r with the displayed order inside each layer, form a regular sequence on R~ (Regular Sequence On A Module), and the Koszul complex of these rm elements is a free resolution of the quotient R~/(first rm)≅B′(D)=⨂j=1rBsj′, the unreduced tensor product of Unreduced type-A Soergel bimodules and the trivial polynomial factor, identifying the quotient with the balanced tensor over the shared strand variables. The last m closure differences are excluded from this sequence; they are not a regular sequence in general, and their Koszul homology is computed by the diagonal Hochschild complex on the remaining closure elements.

Facts & Assumptions

Given: a closed marked MOY resolution D with r wide edges and m strands, the ring R~, and the layer-ordered sequence of the statement.

[L1]

The sequence and its layer structure are those of Setting a to zero in a closed KR factorization gives the layer-by-layer Koszul complex: at layer i the wide edge occupies positions s,s+1, and the relations βi,γi together with the differences xi,j−xi−1,j (j∉{s,s+1}) are precisely the m relations of that layer, the closure differences being listed last.

[L2]

A sequence u1,…,uN in a commutative unital ring A is A-regular when A/(u1,…,uk−1)≠0 and multiplication by uk is injective on it for every k, and A/(u1,…,uN)≠0 (Regular Sequence On A Module).

[L3]

A sequence of elements of A[x1,…,xN] that can be matched bijectively with the variables so that the k-th element is monic of positive degree in the k-th variable after the previous elements have been divided out is a regular sequence: quotienting by a monic polynomial in one variable exhibits the quotient as a free module over the remaining ring, and a nonzerodivisor in A stays a nonzerodivisor in A[x]; the resulting quotient is nonzero and free over Q in the cases below.

[L4]

In the two-variable polynomial ring A=Q[c,d] with symmetric subring B=Q[c+d,cd], the assignment y↦1⊗d induces an isomorphism of graded A-modules Q[c,d][y]/(y2−(c+d)y+cd)≅A⊗BA, both sides being free of rank two over A on the classes of 1,y respectively on 1⊗1,1⊗d.

[L5]

If u is A-regular then K(u;A) is a finite free resolution of A/(u) (Regular Sequences Give Acyclic Koszul Complexes, Koszul Complex Resolves A Regular Quotient), and K(u;A) is the Koszul complex of Koszul Complex Of A Sequence With Coefficients.

Proof

technique · direct
1.1L1givenalgebra

Set A0:=Q[x0,1,…,x0,m], so that R~=A0[x1,1,…,x1,m][x2,1,…,x2,m]⋯[xr,1,…,xr,m], and let Qi be the quotient of the truncated polynomial ring A0[xk,j:1≤k≤i] by the elements of the first i layers. At every induction stage the unused later variables are polynomial extensions of this ring. We prove by induction on i that the concatenation of the first i layers is a regular sequence and that Qi is a free Q-module, nonzero.

2.1L2L3step 1.1algebra

Assume Qi−1 computed and nonzero, and consider layer i in the ring Qi−1[xi,1,…,xi,m]. Write c:=xi−1,s, d:=xi−1,s+1. The first relation βi=xi,s+xi,s+1−c−d is monic of degree one in xi,s with coefficient 1, so it is a nonzerodivisor and its quotient is free over Qi−1[xi,s+1,xi,j(j≠s,s+1)] with basis {1}. In that quotient, substituting xi,s=c+d−xi,s+1 turns the second relation into −(xi,s+12−(c+d)xi,s+1+cd), which is monic of degree two in xi,s+1 with leading coefficient −1; it is a nonzerodivisor. Each remaining difference xi,j−xi−1,j is monic of degree one in the corresponding new variable xi,j and so is a nonzerodivisor on the successive quotients. Thus the m elements of layer i form a regular sequence on Qi−1[xi,1,…,xi,m] with nonzero quotient free over Qi−1. Concatenating with the induction hypothesis, the first i layers form a regular sequence on R~, and Qi is nonzero and free over Q.

3.1L4step 2.1algebra

Identify Qr with B′(D). At layer i, the quotient is the base change along Q[c,d]→Qi−1 of Q[c,d][y]/(y2−(c+d)y+cd), with y=xi,s+1. By [L4] it is the corresponding base change of A⊗BA; explicitly xi,s↦1⊗c and xi,s+1↦1⊗d, while the previous-layer variables act on the left. Both maps are inverse because xi,s=c+d−y and both modules have basis 1,y; the differences at the other positions identify xi,j with the corresponding variable of the previous layer without changing the ring. Iterating over the layers, the surviving ring is generated by all layer variables subject to the displayed invariant-balancing relations and is the balanced tensor product over the shared strand variables of one two-variable balanced tensor A⊗BA per wide edge; each such tensor is the two-variable case of the unreduced simple-reflection bimodule Bs′ of Unreduced type-A Soergel bimodules and the trivial polynomial factor, and the iteration over layers is the balanced tensor product over the shared variables defining B′(D)=⨂j=1rBsj′.

4.1L2L5step 2.1step 3.1∎

Conclude the resolution statement. By steps 1.1 and 2.1 the first rm elements are R~-regular and their quotient is B′(D), which is nonzero; since K(first rm; R~) is a finite free complex by its definition and a regular sequence has acyclic positive Koszul homology, [L5] exhibits it as a finite free resolution of the quotient B′(D). The closure differences are not included in the sequence and nothing is claimed about their regularity.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)Open item page →

The remaining closure Koszul complex is the diagonal Hochschild complex

Statement

Assume the Axiom of Choice (AC), used only through Polynomial Hochschild homology from the diagonal Koszul complex. In the situation of The first-layer relations of a closed MOY resolution form a regular sequence, let R~=Q[xi,j], let the first rm elements of the sequence be as there, and let the remaining m elements be the closure differences x0,j−xr,j (1≤j≤m). Put R′:=Q[x0,1,…,x0,m] and let B′(D) be the quotient R~/(first rm), identified in that lemma with the unreduced tensor product ⨂j=1rBsj′; the ring B′(D) is commutative and carries the R′-bimodule structure whose left action is generated by the classes of x0,j and whose right action is generated by the classes of xr,j. Then:

(a) the classes of x0,1−xr,1,…,x0,m−xr,m in B′(D) are exactly the diagonal elements uj=xjL−xjR of The polynomial diagonal Koszul bimodule complex under the two ring maps R′→B′(D), xj↦x0,j and xj↦xr,j;

(b) the Koszul complex K(x0,1−xr,1,…,x0,m−xr,m;B′(D)) is the diagonal Koszul bimodule complex of B′(D) over R′ and its homology is Tor⁡hR′e(R′,B′(D))≅HHh(R′,B′(D))(h≥0), the isomorphism being the one of Polynomial Hochschild homology from the diagonal Koszul complex for M=B′(D);

(c) the Koszul complex of all (r+1)m elements of the sequence over R~ is quasi-isomorphic to the complex of (b), hence computes HH∙(R′,B′(D)). The isomorphism is natural for maps that preserve the two R′-actions.

The closure differences are not a regular sequence in general: their Koszul homology is HH∙(R′,B′(D)), which is nonzero in several degrees, and no acyclicity is claimed.

Facts & Assumptions

Given: a closed marked MOY resolution D with r wide edges and m strands, the ring R~=Q[xi,j], the layer-ordered sequence of The first-layer relations of a closed MOY resolution form a regular sequence, the quotient B′(D)=R~/(first rm), the ring R′=Q[x0,1,…,x0,m] and AC.

[L1]

The first rm elements form a regular sequence on R~ with quotient R~/(first rm)≅B′(D), the unreduced tensor product ⨂j=1rBsj′, and their Koszul complex is a free resolution of B′(D); the remaining m elements are the closure differences x0,j−xr,j (The first-layer relations of a closed MOY resolution form a regular sequence).

[L2]

The diagonal Koszul bimodule complex of R′=k[x1,…,xm] for k=Q is K(u1,…,um;R′e) with uj=xjL−xjR, augmented by the multiplication μ:R′e→R′; its degree-p term carries the (mp) wedge symbols and the internal degree of θj is 2 when deg⁡xj=2 (The polynomial diagonal Koszul bimodule complex).

[L3]

Under AC, for a field k, R′=k[x1,…,xm] and a k-central R′-bimodule M there is a natural isomorphism HHh(R′,M)≅Hh(K∙(M)) with K∙(M)=K(u1,…,um;R′e)⊗R′eM (Polynomial Hochschild homology from the diagonal Koszul complex); the hypothesis is that the scalar actions agree, which holds for every Q-algebra bimodule.

[L4]

For finite sequences x,y there is a signed chain isomorphism K(x,y;A)≅K(x;A)⊗AK(y;A) (Koszul Complex Concatenation Tensor Isomorphism).

Proof

technique · direct
1.1L1L2givenalgebra

Identify the surviving elements. The variables x0,j and xr,j are elements of the commutative ring B′(D)=R~/(first rm); the left R′-action on B′(D) is the algebra map R′→B′(D), xj↦x0,j, and the right action is xj↦xr,j, both induced by the ring structure. Hence the class of x0,j−xr,j is the image of uj=xjL−xjR under the induced map R′e→B′(D), and its action on B′(D) by multiplication is m↦x0,jm−mxr,j=uj⋅m.

2.1L1L4step 1.1algebra

Reduce the full Koszul complex to the closure complex. By [L4] the Koszul complex of all (r+1)m elements is the tensor product over R~ of the Koszul complex of the first rm elements and that of the last m elements. By [L1] the first factor is a free resolution of B′(D), and its augmentation has acyclic mapping cone. Tensor that cone with the bounded free complex K(last; R~) and filter the total complex by the latter factor's degree. Each associated graded complex is a finite direct sum of shifts of the acyclic cone. The filtration is finite, so induction through its short exact sequences makes the tensor cone acyclic; thus tensoring the augmentation preserves quasi-isomorphism; therefore the Koszul complex of all (r+1)m elements is quasi-isomorphic to B′(D)⊗R~K(last; R~)≅K(last; B′(D)), the Koszul complex of the classes of the closure differences over the commutative ring B′(D).

2.2L2L3L4step 1.1algebra

Identify it with the diagonal complex. Under the identification of the tensor factors, K(last; B′(D)) has degree-p term the p-fold wedge over B′(D) of the elements x0,j−xr,j with the Koszul differential of [L4], and by step 1.1 these elements act by multiplication as uj=xjL−xjR. By [L3] the diagonal Koszul bimodule complex tensored over R′e with B′(D) has degree-p term ⨁∣I∣=pB′(D)θI and differential d(mθI)=∑r(−1)r−1(xirm−mxir)θI∖{ir}, which in the commutative ring B′(D) is multiplication by the classes x0,j−xr,j; both complexes therefore have the same graded terms and the same differential, so K(last; B′(D)) is the diagonal Koszul bimodule complex of B′(D) over R′.

3.1L1L2L3step 2.1step 2.2∎

Compute the homology. The bimodule B′(D) is a Q-algebra bimodule, hence k-central for k=Q, and it is graded with deg⁡xi,j=2. Applying [L3] to M=B′(D) gives the natural isomorphism HHh(R′,B′(D))≅Hh(K(last; B′(D))) and, through steps 2.1 and 2.2, the identification of HH∙(R′,B′(D)) with the homology of the Koszul complex of all (r+1)m elements. This is the only place where AC is used.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule

Statement

Assume the Axiom of Choice (AC), inherited from The remaining closure Koszul complex is the diagonal Hochschild complex. Let D be a closed marked MOY resolution with r wide edges and m≥1 strands, let R~=Q[xi,j], K(D), R′=Q[x0,1,…,x0,m], B′(D)=⨂j=1rBsj′ and B(D)=⨂j=1rBsj be as in The first-layer relations of a closed MOY resolution form a regular sequence, and let R=Q[x1−x2,…,xm−1−xm]⊂R′ be the reduced ring of The reduced type-A polynomial ring and Soergel bimodules for the HHH construction, so that R′=R[t] for t=(x1+⋯+xm)/m by Unreduced type-A Soergel bimodules and the trivial polynomial factor. Then:

(a) for every h≥0 there is an isomorphism HHh(R′,B′(D))≅Hh(K(D)), where K(D) is the Koszul complex of the (r+1)m-element sequence over R~;

(b) with yj:=xj−x1 (2≤j≤m), the diagonal elements uyj=yjL−yjR of the R′-bimodule B′(D) satisfy ut=0, and HHh(R′,B′(D))≅(HHh(R,B(D))⊗QQ[t])⊕(HHh−1(R,B(D)){2}⊗QQ[t]) for every h≥0, with HH−1=0: the homology is the direct sum of two copies of HH∙(R,B(D))⊗QQ[t] with a relative shift of one in the Hochschild degree and two in internal degree (the two-term factor of the trivial coordinate contributes the doubling);

(c) deleting the trivial factor Q[t] from either copy leaves the graded Q-vector space HH∙(R,B(D)), which is the contribution of the resolution D to the reduced Khovanov-Rozansky homology H‾(D) of The reduced Khovanov-Rozansky homology; the trigrading shifts relating the two sides are fixed by the trigrading comparison proved later on this page, and no other normalization is asserted here.

Facts & Assumptions

Given: a closed marked MOY resolution D with r wide edges and m strands, the rings R~, R′, R, the bimodules B′(D), B(D), the Koszul complex K(D), the coordinate t=(x1+⋯+xm)/m, and AC.

[L1]

The first rm elements of the sequence form a regular sequence on R~ with quotient B′(D), their Koszul complex is a free resolution of B′(D), and the remaining m closure differences x0,j−xr,j are the diagonal elements uj=xjL−xjR of the R′-bimodule B′(D); hence the Koszul complex of all (r+1)m elements is quasi-isomorphic to the diagonal Koszul complex of B′(D) and computes HH∙(R′,B′(D)) (The first-layer relations of a closed MOY resolution form a regular sequence, The remaining closure Koszul complex is the diagonal Hochschild complex).

[L2]

B′(D)=B(D)⊗QQ[t] as graded (R′,R′)-bimodules, with R′=R[t] and t central, so the coordinate t acts by the same multiplication on both sides; B(D) is a Q-algebra bimodule over the reduced ring R (Unreduced type-A Soergel bimodules and the trivial polynomial factor).

[L3]

If yi=∑jaijxj and the matrix (aij) is invertible, the induced generator-matrix chain map is an isomorphism K(y;M)≅K(x;M) (Koszul Complex Invariant Under Invertible Generator Change), and K(x,y;M)≅K(x;M)⊗MK(y;M) (Koszul Complex Concatenation Tensor Isomorphism).

[L4]

Under AC, for R′=k[x1,…,xm] and a k-central R′-bimodule M, HHh(R′,M)≅Hh(K(u1,…,um;R′e)⊗R′eM), naturally in M (Polynomial Hochschild homology from the diagonal Koszul complex).

[L5]

The construction of H‾(D) repeats the matrix-factorization construction over the ring of differences Q[a,x2−x1,…,xm−x1]. For a nonempty closed resolution, homogeneous row operations isolate the row (a,0), whose retained cohomology has shift {−1,1}; the remaining rows have first entry zero and compute the resolution homology by their folded Koszul complex (Koszul row operations and variable exclusion preserve homotopy type) (The reduced Khovanov-Rozansky homology, The Khovanov-Rozansky complex and trigraded braid homology).

Proof

technique · direct
1.1L1L4givenalgebra

Compute HH∙(R′,B′(D)) as the homology of K(D). By [L1] the Koszul complex of all (r+1)m elements is quasi-isomorphic to the diagonal Koszul complex of the R′-bimodule B′(D), whose homology is HH∙(R′,B′(D)) by [L4] applied to M=B′(D); this proves (a), with the naturality of [L4] supplying the compatibility with coefficient maps.

1.2L2L3givenalgebra

Change the diagonal generators. In the R′-bimodule B′(D) the diagonal elements ux1,ux2,…,uxm are Q-linear combinations of ut,uy2,…,uym and conversely, the change matrix being the invertible matrix of the coordinate change (x1,…,xm)↔(t,y2,…,ym); by [L3] the two Koszul complexes are isomorphic. The element t, the average of all coordinates, is invariant under every simple reflection, hence balances in every tensor factor and satisfies t⊗1=1⊗t, so ut=0 on B′(D).

2.1L2L3step 1.1step 1.2algebra

Split off the trivial coordinate. By [L3] the Koszul complex of the sequence (ut,uy2,…,uym) over the commutative ring B′(D) is the tensor product of the two-term complex K(ut;B′(D))=[B′(D){2}→0B′(D)] with K(uy2,…,uym;B′(D)). Since the differential of the first factor is zero by step 1.2 and steps 1.1 and 1.2 identify HH∙(R′,B′(D)) with the homology of this Koszul complex, the tensor product is the direct sum of two copies of K(uy2,…,uym;B′(D)), placed in homological degrees 0 and 1, so that HHh(R′,B′(D))≅Hh(K(uy2,…,uym;B′(D)))⊕Hh−1(K(uy2,…,uym;B′(D))){2}, the two summands being the two copies.

3.1L2step 2.1algebra

Identify the reduced complex. By [L2] B′(D)=B(D)⊗QQ[t] and each uyj acts on B(D) through the bimodule structure and trivially on Q[t]; therefore K(uy2,…,uym;B′(D))=K(uy2,…,uym;B(D))⊗QQ[t], and because Q[t] is flat over Q its homology is H∙(K(uy2,…,uym;B(D)))⊗QQ[t].

4.1L3L4step 3.1algebra

Compute the reduced homology. The elements yj=xj−x1 (2≤j≤m) are an invertible linear combination of the standard generators x1−x2,…,xm−1−xm of R (explicitly xj−x1=−(x1−x2)−⋯−(xj−1−xj)), so by [L3] their Koszul complex is isomorphic to the diagonal Koszul complex of the polynomial ring R; by [L4] applied to the polynomial ring R and the Q-central R-bimodule B(D) its homology is HH∙(R,B(D)). Substituting into the two-copy formula of step 2.1 and the tensor decomposition of step 3.1 gives the displayed formula of (b).

5.1L1L2L4L5step 1.2step 2.1step 4.1algebra∎

Recover the original a-theory. In the full closed-graph factorization, the sum of all linear entries is zero: the differences telescope between layers and around the closure, and each wide edge preserves the sum of its two coordinates. The row operation summing the linear rows therefore gives a distinguished row (a,0), with all other first entries zero. Its odd cohomology is Q{−1,1} and its even cohomology is zero; polynomial division in a splits off the contractible pairs, leaving the remaining Koszul complex with this universal shift and parity. When a=0 instead, that distinguished row becomes (0,0) and supplies two copies. After the regular-layer quotient, its zero relation is the closure of the common invariant coordinate t: summing the linear relations gives m(tL−tR)=0. The nonzero scalar m is absorbed by a basis change, so this is exactly the zero diagonal row split in step 2.1. Removing that row and the polynomial coordinate t leaves the reduced diagonal Koszul complex of B(D), whose homology is HH∙(R,B(D)) by step 4.1. Thus the original reduced resolution contribution agrees after accounting for the universal {−1,1} shift with one of the two copies in (b); setting a=0 without removing the zero row retains both copies. This proves (c) and explains the grading correction used by the next lemma. The only use of AC is [L4].

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

The Koszul-Hochschild comparison respects crossing differentials and trigradings

Statement

Assume AC, inherited from the diagonal Koszul comparison in A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule. Let σ be a braid word, let p be one of its crossings at strands s,s+1, and let Γ0 (two arcs) and Γ1 (one wide edge) be the two local resolutions of p. Write χ0 ⁣:C(Γ0)→C(Γ1) and χ1 ⁣:C(Γ1)→C(Γ0) for the wide-edge morphisms of The wide-edge morphisms chi-zero and chi-one and let Cp+=[C(Γ0){0,2}→ χ0 C(Γ1)],Cp−=[C(Γ1){0,−2}→ χ1 C(Γ0){0,−2}] be the positive and the corrected negative crossing complexes of The positive and negative Khovanov-Rozansky crossing complexes. Then:

(a) under the identification of a closed resolution's Koszul complex with the Hochschild homology of its Soergel bimodule and the splitting off of the trivial factor (A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule), the local quotient map for χ0 is rbs ⁣:R{2}→Bs,rbs(1)=ys⊗1+1⊗ys, and the local quotient map for χ1 is brs ⁣:Bs→R,brs(a⊗b)=ab, of Khovanov's generator complexes for the HHH construction, up to multiplication by nonzero rational units (the factor 2 in the composite brsrbs=2ys and the sign of the balanced root). On the reduced resolution homology the induced maps are HHh(R,rbs) and HHh(R,brs), tensored with the other layers. Consequently the two-term complexes agree term by term with the generator complexes F(σs)=[R{2}→Bs] and F(σs−1)=[Bs{−2}→R{−2}] of that item, with matching shifts. The correspondence is uniform in the position s and compatible with the tensor products over the layers, so the resolution-wise identifications intertwine the crossing differentials of the complex computing the termwise Hochschild theory HHH with those of the corrected reduced resolution homology of the Khovanov-Rozansky complex. The full a=0 specialization retains an extra zero Koszul row and has two copies; that row must be removed as specified in the preceding comparison lemma.

(b) write (j,k,l) for the trigrading of The Khovanov-Rozansky complex and trigraded braid homology (cohomological degree, first bigrading, second bigrading) and (c,h,p) for the HHH trigrading (Rouquier degree, Hochschild degree, internal degree). After undoing the source's built-in shift by the global correction (k,l)↦(k+1,l−1) that moves the one-strand class (−1,1,0) to (0,0,0), the two trigradings are related by k=−h,l=p−h,j=c, equivalently a=−h, q=p−h, t=c in the marked variables (a,q,t)=(k,l,j) of the Khovanov-Rozansky theory.

Caveats: the identification is a statement about the reduced theories; the trivial polynomial factor and the source's coordinate x1 are handled by Unreduced type-A Soergel bimodules and the trivial polynomial factor; the constants of (b) are fixed by the bidegrees of the folded Koszul complex and are anchored by the two worked normalizations, the one-strand class of the trivial one-braid and the (2,n) computation; AC enters only through the preceding diagonal Koszul comparison; the local map and grading calculations use no additional choice.

Facts & Assumptions

Given: a braid word σ, a crossing at strands s,s+1 with its two local resolutions Γ0,Γ1, the morphisms χ0,χ1 and the two crossing complexes, the reduced ring R, the bimodules Bs and the maps rbs,brs.

[L1]

χ0 ⁣:C(Γ0)→C(Γ1) has bidegree (0,2) and χ1 ⁣:C(Γ1)→C(Γ0) has bidegree (0,0); in the Koszul standard forms of the two factorizations they are the flip morphisms Id⊗ψ′(x4−x2) and Id⊗ψ(x4−x2), and their composites satisfy χ1χ0=χ0χ1=Id⊗((x4−x2)⋅id) (The wide-edge morphisms chi-zero and chi-one).

[L2]

The positive crossing complex is the cone of χ0 with source shift {0,2} and the corrected negative crossing complex is the cone of χ1 with overall shift {0,−2}; both differentials have bidegree (0,0) as maps of the shifted terms (The positive and negative Khovanov-Rozansky crossing complexes).

[L3]

rbs ⁣:R{2}→Bs and brs ⁣:Bs→R are the well-defined degree-zero maps of graded (R,R)-bimodules with the displayed values, and F(σs)=[R{2}→rbsBs], F(σs−1)=[Bs{−2}→brsR{−2}] are the generator complexes with the B-term in cohomological degree 0 (Khovanov's generator complexes for the HHH construction, Unreduced type-A Soergel bimodules and the trivial polynomial factor).

[L4]

For a closed resolution D with Koszul complex K(D) there is an isomorphism HHh(R′,B′(D))≅Hh(K(D)) natural for coefficient maps, and after splitting off the trivial coordinate the reduced summand is HH∙(R,B(D)) with B(D)=⨂jBsj over the wide edges (A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule).

[L5]

For a layer of a closed marked resolution the first relations form a regular sequence with quotient B′(D), and the Koszul symbols of the folded complex carry the shifts (−1,1) for a linear relation and (−1,3) for the quadratic relation, so that every differential has bidegree (1,1) in the bigrading (deg⁡a,deg⁡x)=(2,0),(0,2) (The first-layer relations of a closed MOY resolution form a regular sequence, Setting a to zero in a closed KR factorization gives the layer-by-layer Koszul complex).

[L6]

The trigraded Khovanov-Rozansky complex of a braid diagram has cohomology H(D)=⨁j,k,lHk,lj(D) with the Euler characteristic ∑(−1)jtkqldim⁡Hk,lj(D), the shifts of the local terms being C0(Γ0)=R⊕R{−2,2}, C1(Γ0)=R{−1,1}⊕R{−1,1}, C0(Γ1)=R⊕R{−2,4}, C1(Γ1)=R{−1,1}⊕R{−1,3} (The Khovanov-Rozansky complex and trigraded braid homology, The wide-edge morphisms chi-zero and chi-one, The reduced Khovanov-Rozansky homology).

[L7]

The reduced instance of the library's Rouquier generator complexes has Bslib=Bs{−1} and differentials the ordinary multiplication map εs(a⊗b)=ab and ηs, with ηs(1)=αs⊗1+1⊗αs and αs=κsys, κs=(−1)s−1 a unit. Thus ηs=κsrbs with the corresponding shifts; the root sign does not multiply the differential εs (The positive and negative Rouquier generator complexes).

Proof

technique · direct
1.1L2L3L4L5givenalgebra

The local quotient modules. At a=0, the two-arc Koszul resolution has quotient the identity bimodule R′, while the wide-edge resolution has quotient Bs′=Q[x1,x2,x3,x4]/(x1+x2−x3−x4, x1x2−x3x4). Choose the left strand coordinates (x4,x3) and right strand coordinates (x1,x2); this order agrees with the arcs x1=x4, x2=x3. Put yL=x4−x3 and yR=x1−x2. The sum relation implies x4−x2=(yL+yR)/2 in Bs′. The common invariant coordinate and the other strand coordinates pass through this local calculation unchanged. The regular-layer quotient and diagonal identification in [L4] carry these coefficient modules to the corresponding termwise Hochschild contributions; removing the universal zero row recovers the original a-theory with its fixed {−1,1} correction.

2.1L1L2L3L7step 1.1algebra

Read the induced maps from the matrices. The degree-zero Koszul homology is the quotient of the even coefficient summand. The even matrices of [L1] have first entries x4−x2 for χ0 and 1 for χ1. Consequently χ0 induces 1↦(yL+yR)/2 from the identity module to Bs′, while χ1 induces the quotient map Bs′→R′ setting yL=yR, namely multiplication. The first map is a bimodule map because (yL−yR)(yL+yR)=0 in the wide-edge quotient and the invariant generators already balance. After removing the common invariant polynomial factor, these are exactly rbs/2 and brs on the reduced modules. In particular rbsbrs is multiplication by yL+yR on Bs, rather than multiplication by the single scalar ys; brsrbs=2ys on R. Rescaling the source of the positive two-term complex by a nonzero rational scalar turns rbs/2 into rbs, and the negative map already equals brs. All shifts are those in [L2] and [L3].

3.1L1L2L3L4step 1.1step 2.1algebra

Assembly over the layers. The local quotient maps of step 2.1 are tensored with the identity on every other layer. The regular-layer augmentations and the diagonal Koszul comparison are natural for these coefficient maps by [L4], so the resolution-wise identifications intertwine each crossing edge. Multiplying a resolution's coefficient module by the product of the source-rescaling constants for its positive arc terms makes all positive maps exactly rbs simultaneously: changing one resolution choice changes precisely the factor belonging to that crossing. The scalars commute, so the two routes around each cube face agree, and the ordinary cube signs are preserved. These isomorphisms therefore assemble into a chain isomorphism between the corrected reduced resolution-homology complex and the termwise Hochschild complex of F(σ). For the universal-row reduction, the normal-form differential is a(e0∧−)+dK, and the crossing matrices and homogeneous row changes contain no a. Comparing the coefficient of a in the chain-map equation forces each crossing map to commute with e0∧−. It therefore preserves the image of this wedge operator, the copy retained by the original (a,0) row after polynomial cancellation. Its induced coefficient map there is the one computed above. The extra zero-row copy of the full a=0 theory is not included.

4.1L4L5step 3.1algebra

The first grading. On the Khovanov-Rozansky side the first bigrading of a class of the folded complex is the negative of the number of Koszul symbols of the closure block that produced it, because each relation of the closure block contributes the shift (−1,1) by [L5], while the layer block contributes nothing on homology: it is a regular sequence whose Koszul complex is a resolution of B′(D). Under the identification of step 1.1 the closure block is exactly the Hochschild block of the termwise complex, whose homology degree is h; hence k=−h after the global correction, which does not involve the first bigrading of the moving classes beyond the fixed shift (1,−1,0).

4.2L2L4L5L6step 3.1algebra

The second and third gradings. By [L5] a relation of internal degree d contributes the shift (−1,d−1) to the folded complex, and the surviving class has internal degree p equal to the sum of the degrees of the relations and of the coefficient bimodules of the resolution; therefore its second bigrading is l=p−h, because each of the h closure factors contributes d to the internal degree p and d−1 to the second bigrading, a difference of exactly h down from p. The cohomological degree j of the assembled complex is the position in the cube of resolutions, which is the same index as the Rouquier degree c of the termwise complex; hence j=c. The three displays combine to k=−h, l=p−h, j=c, that is a=−h, q=p−h, t=c in the variables of the statement.

5.1L6step 4.1step 4.2algebra∎

The global correction and the constants. The source records that the Khovanov-Rozansky trigrading carries a built-in shift by (−1,1,0) coming from the variable a, that both the one-strand classes lie in (0,0,0) after undoing it, and that after the correction the third gradings match, the Hochschild grading equals the Koszul grading with the minus sign, and the second grading equals the x-degree grading minus the Hochschild grading; these are exactly the three displays of steps 4.1 and 4.2, with the correction (k,l)↦(k+1,l−1) applied to the first two. The one-strand class is (−1,1,0) on the Khovanov-Rozansky side and (0,0,0) on the Hochschild side, so the correction is fixed, and the (2,n) computation with its alternating differentials of degrees 2 confirms that no further constant is needed.

TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

HHH is isomorphic to reduced Khovanov-Rozansky homology

Statement

Assume the Axiom of Choice (inherited through the diagonal Koszul comparison of A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule). Let σ be a braid word on m≥1 strands with closure σ^, let HHH(σ) be the termwise Hochschild theory of The termwise Hochschild complex of a Rouquier complex and the groups HHH, and let H‾(σ^) be the reduced Khovanov-Rozansky homology of The reduced Khovanov-Rozansky homology evaluated on the corresponding braid diagram. Then there is an isomorphism of trigraded Q-vector spaces HHHc,h,p(σ)≅H‾c,−h−1,p−h+1(σ^) under the trigrading correspondence of The Koszul-Hochschild comparison respects crossing differentials and trigradings: writing kcorr=k+1 and lcorr=l−1 for the corrected Khovanov-Rozansky grading, kcorr=−h, lcorr=p−h, j=c; equivalently a=−h, q=p−h, t=c in the marked corrected variables (a,q,t)=(kcorr,lcorr,j). The unreduced Khovanov-Rozansky theory is recovered from the reduced one by adjoining the trivial polynomial factor, as recorded in The reduced Khovanov-Rozansky homology.

Caveats: only the reduced theory is compared; the isomorphism is trigrading-preserving only after the global correction, and no absolute normalization beyond it is claimed; the compatibility has to hold over all resolutions, including the corrected negative crossing and the signs of the Koszul total differentials; the Axiom of Choice enters only through the comparison of the bar and diagonal Koszul resolutions inside the cited identification, not through the construction of HHH.

Facts & Assumptions

Given: a braid word σ with N crossings and closure σ^, the cube of resolutions of the braid diagram, the termwise Hochschild complex of The termwise Hochschild complex of a Rouquier complex and the groups HHH and the reduced Khovanov-Rozansky complex of The reduced Khovanov-Rozansky homology, and AC.

[F1]

For the generator complex F(σ) of Khovanov's generator complexes for the HHH construction the groups HHHc,h,p(σ)=Hc(HHh(R,F(σ)∙))p are the cohomology of a bounded complex whose terms are the Hochschild homologies of the resolution terms F(σ)ν=⨂ν(local term) over the crossings, with differentials induced by the maps rbs,brs (The termwise Hochschild complex of a Rouquier complex and the groups HHH).

[F2]

The reduced Khovanov-Rozansky theory is built over the ring of differences with the coefficient a retained; for a fixed resolution Dν the reduced summand of the Koszul homology is HH∙(R,B(Dν)) with B(Dν)=⨂sBs over the wide edges, and the unreduced theory differs by adjoining the trivial polynomial factor (The reduced Khovanov-Rozansky homology, A closed MOY resolution's Koszul complex computes Hochschild homology of its Soergel bimodule).

[F3]

The a=0 specialization of the Khovanov-Rozansky complex is the folded Koszul complex of the resolution sequence, and the crossing complexes are the cones of χ0 and χ1 with the shifts {0,2} and {0,−2} (The Khovanov-Rozansky complex and trigraded braid homology, The positive and negative Khovanov-Rozansky crossing complexes).

[F4]

The local maps induced by χ0 and χ1 on the reduced summands are, up to nonzero rational units, the maps rbs and brs, the identification is compatible with the tensor products over the layers, and the trigradings correspond by k=−h, l=p−h, j=c after the correction (k,l)↦(k+1,l−1) (The Koszul-Hochschild comparison respects crossing differentials and trigradings).

[F5]

AC is the choice-function principle (The Axiom of Choice), used only through the diagonal Koszul identification inside [F2].

Proof

technique · direct
1.1F1F2givenalgebra

Resolution-wise identification. Fix a resolution Dν of the braid diagram, obtained by replacing each crossing by an arc (0-resolution) or a wide edge (1-resolution). By [F2] the reduced Koszul homology of Dν is HH∙(R,B(Dν)), and by [F1] the term of the termwise complex attached to the same choice of resolutions is HH∙ of the corresponding tensor product of local terms, which is the same bimodule B(Dν) when the reduced bimodules Bs are assigned to the wide edges and R to arcs; therefore the resolution-wise contributions coincide up to the fixed shifts, and the identification is natural for bimodule maps.

2.1F3F4step 1.1algebra

Intertwining the differentials. Two adjacent resolutions differ at one crossing, and the differential of the cube complex at that edge is χ0 (positive crossing) or the corrected χ1 (negative crossing) on the Khovanov-Rozansky side, and rbs or brs on the termwise Hochschild side. By [F4] the resolution-wise identifications of step 1.1 carry one into the other up to nonzero rational units and the fixed shifts, and they are compatible with the tensor products over the layers and with the Koszul signs of the total differentials; hence they assemble over the 2N resolutions into a chain isomorphism, up to the overall shift, between the complex computing HHH(σ) and the corrected original reduced Khovanov-Rozansky resolution-homology complex. The complete a=0 specialization has a second copy from its universal zero row; [F4] removes that row and retains the original theory's fixed {−1,1} shift. The consistent vertex rescalings of [F4], not independent arbitrary edge scalars, give the stated chain isomorphism.

3.1F4step 2.1algebra

Cohomology and the trigrading. Taking cohomology of the chain isomorphism of step 2.1 gives a Q-linear isomorphism HHHc,h,p(σ)≅H‾j,k,l(σ^); by [F4] the grading classes correspond by k=−h, l=p−h, j=c after the global correction, and the correction is the one that moves the one-strand class (−1,1,0) in the order (k,l,j) of the Khovanov-Rozansky theory to (0,0,0), matching the one-dimensional class of HHH in (0,0,0) for the trivial one-strand braid.

4.1F2F5step 1.1step 3.1∎

The unreduced theory. The unreduced construction carries the coefficient variable a and the trivial polynomial factor; by [F2] and The reduced Khovanov-Rozansky homology the unreduced groups are obtained from the reduced ones by adjoining that factor, so the isomorphism of step 3.1 is exactly the comparison stated in the source between HHH and the reduced homology. The only use of AC is through [F2], in step 1.1.

CorollaryStatement: Literature-sourcedProof: AI-adaptedprecheck passOpen item page →

The normalized graded Euler series of HHH recovers HOMFLYPT

Statement

Assume the Axiom of Choice, inherited from the comparison HHH is isomorphic to reduced Khovanov-Rozansky homology and the rationality/oriented-link descent of Khovanov-Rozansky homology categorifies the HOMFLYPT polynomial. Let σ be a braid word with nonempty braid closure and braid diagram D on s(D)≥1 strands and closure D^, and form the Euler characteristic of HHH in the corrected trigrading (c,h,p) with the sign (−1)c on the cohomological (Rouquier) degree and the marked variables of the dictionary a=−h, q=p−h, t=c: ⟨HHH(σ)⟩:=∑h,p,c(−1)ct−hqp−hdim⁡QHHHc,h,p(σ). Then the normalized series E(σ):=(tq−1(1−q2))−1α ∣D∣+−∣D∣−−s(D)+1 ⟨HHH(σ)⟩ is the HOMFLYPT invariant of the closure in the v2 normalization of The normalized Khovanov-Rozansky HOMFLYPT Euler series and Khovanov-Rozansky homology categorifies the HOMFLYPT polynomial: E(σ)=F~(D^), the Markov-invariant v2 series with one-strand value α/(1−q−2); here ∣D∣+,∣D∣− are the crossing numbers of D, s(D) its number of strands and α=−t−1q−1. Equivalently ⟨HHH(σ)⟩=tq−1(1−q2)α s(D)−1−∣D∣++∣D∣− F~(D^), so applying the explicit diagram-dependent normalization factor to the HHH Euler series gives the v2 HOMFLYPT series of the closure. That factor is (tq−1(1−q2))−1α ∣D∣+−∣D∣−−s(D)+1; adjoining the trivial factor Q[x], which multiplies the reduced Euler characteristic by (1−q2)−1, recovers the unreduced series of The normalized Khovanov-Rozansky HOMFLYPT Euler series.

Caveats: the formula is stated in the integer grading of the Khovanov-Rozansky theory of The Khovanov-Rozansky complex and trigraded braid homology, in which raw homology changes under stabilization by the specified overall shifts, removed by the displayed series normalization; the sign (−1)c is the one carried by the cohomological (Rouquier) degree, the Hochschild degree entering the weights only through the shifts k=−h and l=p−h, and it is read off the global correction of the comparison and verified on the one-strand braid; the HOMFLYPT normalization is the v2 one of the cited items, namely the series F~ with one-strand value α/(1−q−2)=t−1q/(1−q2); the published function F of the categorification theorem uses the distinct normalization F(unknot)=(t−1−t)/(q−q−1). This corollary identifies HHH with the v2 series only and makes no explicit formula comparison between the two Euler normalizations.

Facts & Assumptions

Given: a braid word σ with diagram D on s(D) strands, its closure D^, the corrected trigrading of the comparison, and AC.

[F1]

The comparison gives HHHc,h,p(σ)≅H‾j,k,l(D^) at raw degrees j=c, k=−h−1, l=p−h+1; applying the fixed correction (k,l)↦(k+1,l−1) gives corrected degrees (−h,p−h) (HHH is isomorphic to reduced Khovanov-Rozansky homology).

[F2]

The reduced Khovanov-Rozansky homology satisfies H(D)≅H‾(D)⊗QQ[x] for the trivial variable x, and the reduced unknot is one-dimensional; the coefficient a is retained in the reduced construction (The reduced Khovanov-Rozansky homology).

[F3]

The Khovanov-Rozansky complex of a braid diagram has trigraded cohomology H(D)=⨁j,k,lHk,lj(D) with the integer-graded Euler characteristic ⟨D⟩=∑j,k,l(−1)jtkqldim⁡QHk,lj(D) (The Khovanov-Rozansky complex and trigraded braid homology).

[F4]

The series F~(D)=α∣D∣+−∣D∣−−s(D)+1⟨D⟩ with α=−t−1q−1 is invariant under all Markov moves and has one-strand value α/(1−q−2); it is the v2 form of the HOMFLYPT function, whose published-normalization version F satisfies F(unknot)=(t−1−t)/(q−q−1) (The normalized Khovanov-Rozansky HOMFLYPT Euler series, Khovanov-Rozansky homology categorifies the HOMFLYPT polynomial).

[F5]

AC is the choice-function principle (The Axiom of Choice), used through [F1] and the rationality/oriented-link descent of [F4].

Proof

technique · direct
1.1F1F3givenalgebra

Read the Euler characteristic through the raw dictionary of [F1]: j=c, k=−h−1, l=p−h+1. Its raw weight is (−1)jtkql=t−1q (−1)ct−hqp−h. Consequently ⟨HHH(σ)⟩=tq−1∑j,k,l(−1)jtkqldim⁡QH‾k,lj(D^)=tq−1⟨H‾(D^)⟩. The constant records precisely the corrected grading used for HHH.

2.1F2F3step 1.1algebra

The trivial factor. By [F2] the unreduced theory is H≅H‾⊗QQ[x], and the trivial variable x has internal degree 2 in the second bigrading of the tower Q[x]{−1,1} of the one-mark circle; the tower therefore contributes ∑m≥0q2m=(1−q2)−1 to the Euler characteristic, so ⟨H‾(D^)⟩=(1−q2)⟨H(D^)⟩. Substituting into step 1.1 gives ⟨HHH(σ)⟩=tq−1(1−q2)⟨H(D^)⟩.

3.1F4F5step 2.1algebra

The invariant normalization. By [F4] and [F3], ⟨H(D^)⟩=α s(D)−1−∣D∣++∣D∣−F~(D^); substituting into step 2.1 gives the displayed identity for ⟨HHH(σ)⟩, and dividing by the explicit factor (tq−1(1−q2))α s(D)−1−∣D∣++∣D∣− shows that the normalized series E(σ) of the statement equals F~(D^), the Markov-invariant v2 HOMFLYPT series of the closure. Adjoining the trivial factor of step 2.1 recovers the unreduced series. AC is used through [F1] and the rationality/oriented-link descent of [F4].

4.1F4step 3.1algebra∎

Verification on the one-strand braid. For the trivial one-strand braid D one has ∣D∣+=∣D∣−=0, s(D)=1, and HHH(σ∗)=Q in (0,0,0) by the base case of the comparison, so ⟨HHH(σ∗)⟩=1 and E(σ∗)=(tq−1(1−q2))−1⋅1⋅α 0=t−1q/(1−q2)=α/(1−q−2), while F~(unknot)=α/(1−q−2)=t−1q/(1−q2) by [F4]; both sides of the identity E(σ∗)=F~(unknot) are therefore equal to t−1q/(1−q2), so the sign (−1)c and the constant tq−1 of the statement are exactly those of the one-strand normalization.

5 · Examples, counterexamples and false statements

None yet.

Sources