This paper proposes an enhancement to the monoidal Koszul duality for the Hecke category related to GL2 by adding an extra grading, supported by evidence in the context of HOMFLYPT link homology.
Contribution
It introduces an enhanced grading structure to the monoidal Koszul duality for the Hecke category, specifically for GL2, connecting to HOMFLYPT link homology conjectures.
Findings
01
Evidence supporting the enhanced duality for GL2.
02
Connections established between the duality and HOMFLYPT link homology.
03
Framework developed for future generalizations.
Abstract
The Hecke category participates in an equivalence called monoidal Koszul duality, which exchanges it with the category of (Langlands-dual) "free-monodromic tilting sheaves." Motivated by a recent conjecture of Gorsky and the first-named author on HOMFLYPT link homology, we propose to enhance this duality with an additional grading. We provide evidence for this enhancement in the case of GL2, working in the language of the second-named author's joint work with Achar, Riche, and Williamson.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Algebra and Geometry · Algebraic Geometry and Number Theory · Algebraic structures and combinatorial models
Full text
Ext-enhanced monoidal Koszul duality for GL2
Matthew Hogancamp
Department of Mathematics, University of Southern California, Los Angeles, CA 90089, U.S.A.
The Hecke category participates in an equivalence called monoidal Koszul duality, which exchanges it with the category of (Langlands-dual) “free-monodromic tilting sheaves.” Motivated by a recent conjecture of Gorsky and the first-named author on HOMFLYPT link homology, we propose to enhance this duality with an additional grading. We provide evidence for this enhancement in the case of GL2, working in the language of the second-named author’s joint work with Achar, Riche, and Williamson.
Key words and phrases:
diagrammatic Hecke category, Koszul duality, HOMFLYPT link homology
2010 Mathematics Subject Classification:
20F55, 20C08, 20G05
1. Introduction
In the recent preprint [10], E. Gorsky and the first-named author introduced “y-ified GLn Soergel bimodules,” a certain deformation of complexes of GLn Soergel bimodules, and used them to define a deformation of triply-graded link homology that they call “y-ified homology.” They conjectured that y-ified homology restores a missing q-t symmetry in the triply-graded HOMFLYPT link homology of Khovanov–Rozansky [15, 14], and that this symmetry comes from a monoidal triangulated autoequivalence of the category of y-ified GLn Soergel bimodules.
Soergel bimodules are an algebraic incarnation of the Hecke category, a monoidal category which plays a central role in geometric representation theory. The Hecke category participates in a monoidal triangulated equivalence known as monoidal Koszul duality, which in characteristic [math] is due to Bezrukavnikov–Yun [4] and exchanges the Hecke category associated to a reductive group G with “free-monodromic tilting sheaves” associated to the Langlands dual group G∨.
Monoidal Koszul duality involves bigraded categories. On the other hand, HOMFLYPT homology has a third “Hochschild” grading (in Khovanov’s construction) coming from Ext groups between Soergel bimodules computed in the abelian category of graded bimodules. The aim of this paper is to suggest that monoidal Koszul duality should similarly admit an Ext-enhancement to an equivalence of triply-graded categories.
In recent work [1, 2], P. N. Achar, S. Riche, G. Williamson, and the second-named author proposed a new construction of free-monodromic tilting sheaves that also makes sense in positive characteristic, then used this to establish a positive characteristic monoidal Koszul duality. In this paper, we work in the language of [1, 2] to provide some evidence for an Ext-enhanced monoidal Koszul duality for GL2. In particular, we introduce an Ext-enhancement of the GL2 diagrammatic Hecke category of Elias–Khovanov [8], then start with this category to define similarly Ext-enhanced versions of the categories of [1]. Assuming that the exchange law of a monoidal category continues to hold for Ext-enhanced free-monodromic tilting sheaves, our main result (Theorem 5.1) constructs an Ext-enhanced monoidal Koszul duality functor.
1.1. Acknowledgements
The first-named author was supported by NSF grant DMS-1702274. Initial computations for this work were done while the second-named author was in residence at the Mathematical Sciences Research Institute during Spring 2018, supported by NSF grant DMS-1440140.
2. Preliminaries
2.1. Graded categories
In this paper we consider Z2- and Z3-graded monoidal categories. Let k be a commutative ring, and let C be a k-linear category. Let Γ be an abelian group which acts strictly on C. That is to say, for each γ∈Γ there is an autoequivalence Σγ:C→C, and these satisfy the relations
[TABLE]
We let HomΓ(X,Y):=⨁γ∈ΓHomC(Σγ(X),Y) denote the Γ-graded k-module of homs. We let enΓ(C) denote the category with the same objects as C, but with hom spaces HomCΓ.
This defines a 2-functor from categories with a strict Γ-action to categories enriched in Γ-graded k-modules.
Remark 2.1*.*
Sometimes it is conventional to let (γ) to denote the downward shift by γ, i.e. X(γ):=Σ−γ(X). There is a canonical isomorphism
[TABLE]
There is a 2-functor the other direction. Suppose D is a category enriched in Γ-graded k-modules. Define un(D) to be the category whose objects are pairs (X,γ) consisting of an object X∈D and an element γ∈Γ. The hom spaces in un(D) are
[TABLE]
Remark 2.2*.*
If D=en(C) for some category C with a strict Γ action, then the objects (X,γ) and (Σγ(X),0) are isomorphic in un(D), via the isomorphism
[TABLE]
Thus the functor C→un(en(C)) sending X↦(X,0) is an equivalence of categories.
Example 2.3*.*
If A is a Γ-graded k-algebra, then we let C:=A-gmod denote the category of Γ-graded A-modules with degree zero morphisms. We let Σγ denote the grading shift functor; it acts on objects by Σγ(X)γ′:=Xγ′−γ.
2.2. Graded and super-monoidal categories
Suppose C is a k-linear monoidal category with a strict Γ-action. We typically want the Γ-action and the monoidal structure to be compatible. This compatibility takes the form of a pair of natural isomorphisms
[TABLE]
for all X∈C, subject to certain coherence conditions (see [11] for more details).
Most importantly, we require that Σγ(\mathbbm1), equipped with the braiding morphismsβγ,X, has the structure of an object in the Drinfeld center of C. Furthermore, if X=Σγ(\mathbbm1), then βγ,X equals (−1)⟨γ,γ⟩idΣγ(X)⊗Σγ(X). The sign here is determined by a symmetric bilinear pairing ⟨,⟩:Γ×Γ→Z/2, called the parity form.
Remark 2.4*.*
Typically the parity form is determined by a group homomorphism p:Γ→Z/2 via ⟨γ,γ′⟩=p(γ)p(γ′). More generally, the parity form may also be determined by a group homomorphism p:Γ→(Z/2)r via
[TABLE]
where p(γ)=(p1(γ),…,pr(γ)).
We will say that C is equipped with a strict monoidal Γ-action or that C has the structure of a Γ-monoidal category if it is equipped with a strict Γ-action and natural transformations α, β as above.
Example 2.5*.*
Let C=Ch(k-mod) be the category of Z-graded complexes of k-modules with differentials of degree 1. Then C is Z-monoidal with tensor product ⊗k, extended to complexes via the usual sign rule, and the parity form is determined by the nontrivial homomorphism Z↠Z/2.
Now, if C has the structure of a Γ-monoidal category, then we can consider the enriched category en(C) defined in the previous section. This category inherits a tensor product from C, but the tensor product of morphisms satisfies the graded exchange law
[TABLE]
The enriched category en(C) is not, strictly speaking, a monoidal category. Rather, it is a super-monoidal category, because of the above exchange law.
Remark 2.6*.*
Traditionally, the prefex “super” indicates Z/2-graded categories. We prefer a less restricted use, which instead refers to arbitrary Γ-graded categories equipped with a parity form Γ×Γ→Z/2.
Remark 2.7*.*
The graded ring of endomorphisms EndCΓ(\mathbbm1) in a Γ-graded super-mon-oidal category C is super-commutative with respect to the given parity form ⟨,⟩.
Remark 2.8*.*
There are two natural ways to set up a graphical calculus for graded monoidal categories. In the first, one considers only degree zero morphisms. In this setup, the calculus works in exactly the usual way. The central objects Σγ(\mathbbm1) can be used to encode morphisms of nonzero degree, and all the various signs are captured entirely by the signs involved in braiding Σγ(\mathbbm1) past itself.
In the second setup, one allows morphisms of arbitrary degree. In this case, exchanging the heights of two distant morphisms introduces a sign given by the parity form. This setup will be referred to as the super-calculus for graded monoidal categories. The super-calculus is more compact, but the signs are conventional and occasionally mysterious. Both calculi are preferable in various instances.
3. Ext-enhanced Bott–Samelson bimodules
In the rest of this paper, we fix a field k and let
[TABLE]
where (y1,y2) and (x1,x2) are dual bases of the k-vector spaces V and V∗. This is the GL2 realization (in the sense of Elias–Williamson [9, §3.1]) of the type A1 Coxeter system S2={id,s}, acting on V via s(y1)=y2 and s(y2)=y1.
Let us recall the associated monoidal category of Bott–Samelson bimodules. Let R=Sym(V∗), viewed as a Z-graded k-algebra with degV∗=2. Consider the Z-graded R-bimodule Bsbim=R⊗RsR(1), where Rs⊂R denotes the s-invariants, and (1) shifts the bimodule degree down by 1. Given an expression w=(s,…,s), define the Bott–Samelson bimodule
[TABLE]
For the empty word, B∅bim=R. Let R-gmod-R be the category of Z-graded R-bimodules and bimodule homomorphisms of degree [math]. Then BS2 is defined to be the full subcategory of R-gmod-R whose objects are Bwbim(j) for expressions w and j∈Z.
Now, consider the bounded derived category Db(R-gmod-R). We denote the cohomological shift by ⌈1⌋ (since [1] will be reserved for another shift later on).
Definition 3.1**.**
The category of Ext-enhanced GL2 Bott–Samelson bimodulesBS2Ext is the smallest full subcategory of Db(R-gmod-R) containing BS2 (viewed as complexes supported in cohomological degree [math]) and closed under cohomological shift.
In other words, objects of BS2Ext are of the form B(m)⌈n⌋, where B is a Bott–Samelson bimodule and m,n∈Z. For two such objects,
[TABLE]
Instead of the cohomological shift, we will work primarily with the combined shift
[TABLE]
The category BS2Ext is thus equipped with two grading shifts (1),[[1]]. Given two objects B,B′∈BS2Ext, define the bigraded k-module Hom(B,B′) by
[TABLE]
Let us discuss the monoidal structure on these categories. The tensor product ⊗R makes BS2 into a monoidal category with a grading shift (1). The monoidal structure on Db(R-gmod-R) is defined by the derived tensor product over R. However, since each bimodule in BS2 is free as a left and right R-module, the derived tensor product of such bimodules coincides with the usual tensor product, and the monoidal structure is defined by the ordinary tensor product.
To describe morphisms in BS2Ext, one chooses a resolution of each Bott–Samelson bimodule by graded free R-bimodules and considers morphisms in the homotopy category Kb(R-gmod-R) between these resolutions. The monoidal structure on morphisms in BS2Ext then corresponds to the usual tensor product of morphisms between these resolutions. The discussion of §2 shows that BS2Ext can be viewed either as a genuine monoidal category with two grading shifts (1) and [[1]], or as a Z2-graded super-monoidal category with parity function Z2→Z/2 sending (j,k)↦kmod2.
3.1. Diagrammatic presentation
A diagrammatic monoidal presentation for BS2 (and more generally for the GL-realization of any Sn) was given by Elias–Khovanov [8]: they defined a k-linear strict monoidal category D2 by generators and relations, together with a monoidal equivalence
[TABLE]
Subsequent work of Elias [7] and Elias–Williamson [9] defined the diagrammatic Hecke category for more general Coxeter systems and realizations. This presentation of the Hecke category was crucial in [2], which constructs the monoidal Koszul duality functor by generators and relations.
We follow the same approach for the Ext-enhancement. For this, we need a presentation for BS2Ext. The category we construct will be a Z2-graded super-monoidal category with respect to the parity function p:Z2→Z/2 sending (j,k)↦k mod 2.
Definition 3.2**.**
The Ext-enhanced GL2 diagrammatic Hecke categoryD2Ext is the k-linear Z2-graded strict super-monoidal category defined by the diagrammatic presentation below.
The objects of D2Ext are the same as those of D2; they are indexed by expressions, and the object corresponding to w is denoted by Bw:
[TABLE]
As in D2, a morphism Bw→Bv in D2Ext is a k-linear combination of diagrams in a planar strip, where each diagram has bottom boundary w, top boundary v, and is made up of local pieces given by a list of generating morphisms.
Each generating morphism in D2 of degree m is also a generating morphism in D2Ext of bidegree (m,0):
[TABLE]
Recall that R=k[x1,x2]. Then the bigraded algebra of self Ext’s of R is canonically isomorphic to RExt:=R⊗ΛExt∨, where ΛExt∨=Λ(V[[1]]). We write ξ1,ξ2,ξs for the elements of V[[1]] corresponding to y1,y2,αs∨∈V, so that ΛExt∨ can be identified with the exterior algebra in the two generators ξ1,ξ2 of degrees (0,1). In other words
[TABLE]
In addition to the generators (3.2a), D2Ext has the following generators:
[TABLE]
Here, f is a homogeneous element in RExt, and degf denotes its bidegree. We also define the shorthands
[TABLE]
The morphisms of D2Ext satisfy the defining relations of D2, plus the following additional relations:
[TABLE]
This concludes the definition of D2Ext.
We will sometimes use ⋆ rather than juxtaposition for the monoidal product in D2Ext. We also identify elements of RExt=R⊗kΛExt∨ with k-linear combinations of products of the corresponding boxes.
It is a straightforward exercise in diagrammatics to derive the following further relations from the ones above:
[TABLE]
Below, the bigraded morphism spaces in D2Ext will be denoted by Hom(−,−), with End(B):=Hom(B,B). The following computation of morphism spaces is again straightforward, using the known (thanks to the equivalence (3.1)) morphism spaces in D2:
[TABLE]
In the rest of this paper, we work with D2Ext rather than BS2Ext.
Remark 3.3*.*
One can show that there exists a k-linear monoidal equivalence FExt:D2Ext→∼BS2Ext extending the equivalence (3.1). Full details will appear in future work in a more general setting, but most of the work for this GL2 case is already contained in [10, §3.5] (where D2Ext is denoted by D2). In the notation of [10, §3.5], FExt is defined by generators and relations by sending
[TABLE]
which imply \leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-4.66794ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt4.26794pt\pgfsys@curveto1.42264pt5.05365pt0.7857pt5.69058pt0.0pt5.69058pt\pgfsys@curveto-0.7857pt5.69058pt-1.42264pt5.05365pt-1.42264pt4.26794pt\pgfsys@curveto-1.42264pt3.48224pt-0.7857pt2.8453pt0.0pt2.8453pt\pgfsys@curveto0.7857pt2.8453pt1.42264pt3.48224pt1.42264pt4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture↦ω, \leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-6.09058ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt-4.26794pt\pgfsys@curveto1.42264pt-3.48224pt0.7857pt-2.8453pt0.0pt-2.8453pt\pgfsys@curveto-0.7857pt-2.8453pt-1.42264pt-3.48224pt-1.42264pt-4.26794pt\pgfsys@curveto-1.42264pt-5.05365pt-0.7857pt-5.69058pt0.0pt-5.69058pt\pgfsys@curveto0.7857pt-5.69058pt1.42264pt-5.05365pt1.42264pt-4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt-4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt-4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture↦ω∗. Then [10, Proposition 3.30] verifies many of the necessary relations among these images, and also computes many of the hom spaces needed to show fully faithfulness.
4. Ext-enhanced free-monodromic tilting sheaves
Starting from the diagrammatic Hecke category D2, [1] defined a category FM2 of “free-monodromic complexes” and a full subcategory Tilt2 of “free-monodromic tilting sheaves.” This construction can be repeated with D2 replaced by D2Ext to yield categories FM2Ext of Ext-enhanced free-monodromic complexes and Tilt2Ext of Ext-enhanced free-monodromic tilting sheaves.
We assume familiarity with [1], and only recall those details of these constructions that are relevant for the calculations that follow. See, however, the remark at the end of §4.1 for differences in grading and sign convention.
4.1. Diagram sequences
We need to clarify how the extra grading in D2Ext is treated when dealing with complexes and, soon, free-monodromic complexes.
Let (D2Ext)⊕ denote the additive envelope of D2Ext, obtained by formally adjoining finite direct sums. Let D′:=(D2Ext)⊕,(1),[[1]] denote the envelope of D2Ext in which we adjoin not just formal direct sums, but also formal grading shifts B(m)[[n]] with B∈(D2Ext)⊕ and m,n∈Z. The morphism spaces in this envelope are by definition the bigraded hom spaces
[TABLE]
By construction, D′ is a Z2-graded super-monoidal category with parity form
[TABLE]
We want to consider formal complexes in which the “chain groups” are objects in D′. As a preliminary, we first construct the category which later will play the role of “complexes with zero differential.”
Let Seq(D′) be the category of Z-graded sequences of objects in D′. An object of Seq(D′) is a Z-indexed sequence F=(Fi)i∈Z with Fi∈D′. The morphism spaces in Seq(D′) are the Z3-graded k-modules HomSeq(D′)Z×Z×Z(F,G) with
[TABLE]
Morphisms of degree (i,j,k) are said to have cohomological degreei, Soergel degreej, and Hochschild degreek.
Let Seqb(D′)⊂Seq(D′) denote the full subcategory consisting of finite sequences F, for which Fi=0 for all but finitely many i∈Z. The category Seqb(D′) inherits an operation ⋆ defined on objects by
[TABLE]
If f∈HomSeq(D′)i,j,k(X,X′) and g∈HomSeq(D′)i′,j′,k′(Y,Y′) are two homogeneous morphisms, then we define a morphism f⋆g∈HomSeq(D′)i+i′,j+j′,k+k′(X⋆Y,X′⋆Y′) by its restrictions
[TABLE]
It is an exercise to show that this gives Seqb(D′) the structure of a Z3-graded super-monoidal category with the associated parity form
[TABLE]
In other words, the composition of morphisms satisfies the super-exchange law
[TABLE]
Remark 4.1*.*
It is important to note that if f and g are morphisms in Seqb(D′) with degf=(2i,j,k) and degg=(i′,j′,2k′), then (f⋆id)∘(id⋆g)=(id⋆g)∘(f⋆id) with no sign. In this way the parities associated to the cohomological and Hochschild degrees are independent.
Let [1]:Seq(D′)→Seq(D′) denote the downward grading shift functor F[1]p=Fp+1. On morphisms [1] acts by a conventional sign:
[TABLE]
where ∣f∣∈Z denotes the first component of deg(f)∈Z3. The sign here guarantees that the functor F↦\mathbbm1[1]⋆F is naturally isomorphic to F↦F[1].
Altogether, Seq(D′) is equipped with three grading shift functors [i],(j),[[k]], defined on objects by
[TABLE]
and satisfying
[TABLE]
We will also use the combined shift
[TABLE]
Remark 4.2*.*
The Hochschild degree [math] part of our construction recovers the categories of [1], with two differences in grading and sign convention. First, degree (i,j) in [1] corresponds to degree (i−j,j,0) in this paper. Second, [1] defined an operation ⋆ on Seqb((D2)⊕,(1)) by a careful choice of signs so that Seqb((D2)⊕,(1)) became Z2-graded super-monoidal with parity function (i,j)↦i. One could just as well have used the induced product ⋆ as above and parity function (i,j)↦i+j (corresponding to (i,j,0)↦i in our convention).
4.2. Free-monodromic complexes
Below we will consider differential Z3-graded categories with differentials of degree (1,0,0) in which the Leibniz rule takes the form
[TABLE]
where ∣f∣ is the cohomological degree of f. Such categories will be called dggg categories. For instance Seq(D′) defined in §4.1 is a dggg category with zero differential.
Definition 4.3**.**
Let BEExt,dg:=Chb(D′) denote the dggg category of formal finite complexes over D′. Objects of this category are pairs (F,δ) where F is an object of Seqb(D′) and δ∈EndSeq(D′)1,0,0(F) satisfies δ∘δ=0. The morphism spaces in BEExt,dg are the complexes
[TABLE]
with differential
[TABLE]
Let BEExt denote the cohomology category of BEExt,dg; it has the same objects, but morphism spaces are the degree (0,0,0) chain maps modulo homotopy.
We define some Z3-graded k-algebras in preparation for our definition of FMExt,dg. Regard R=Sym(V∗(−2)) and ΛExt∨=Λ(V[[1]]) now as being Z3-graded, concentrated in cohomological degree [math]. Also define
[TABLE]
We write ν1,ν2,νs for the elements of V∗[1](−2) corresponding to x1,x2,αs∈V∗. Thus
[TABLE]
with degrees
[TABLE]
Consider the differential Z3-graded algebra
[TABLE]
with differential κ determined by κ(xi)=0=κ(yi) and κ(νi)=xi together with the Leibniz rule with respect to the cohomological degree. Let K⊗RSeq(D′) denote the category with the same objects as Seq(D′), but morphism spaces given by
[TABLE]
with composition
[TABLE]
Hom spaces in K⊗RSeq(D′) inherit a differential from K, which we will continue to denote by κ. Thus, K⊗RSeq(D′) is a dggg category.
Definition 4.4**.**
For each F∈Seq(D′), let ΘF∈EndK⊗RSeq(D′)2,0,0(F) denote the closed endomorphism
[TABLE]
Let FMExt,dg denote the dggg category whose objects are pairs (F,δ) where F∈Seq(D′) and δ∈EndK⊗RSeq(D′)1,0,0(F) is an element such that
[TABLE]
The hom spaces in FMExt,dg are by definition the complexes
[TABLE]
with differential
[TABLE]
Let FMExt denote the cohomology category of FMExt,dg. Objects of FMExt,dg or FMExt are called (Ext-enhanced) free-monodromic complexes.
We adopt the usual terminology of dg categories. A homogeneous morphism f∈HomFMExti,j,k((F,δ),(G,δ′)) is closed if κ(f)+δ′∘f−(−1)if∘δ=0 and exact if f=κ(h)+δ′∘h+(−1)ih∘δ for some h∈HomFMExti−1,j,k((F,δ),(G,δ′)). Closed morphisms of degree zero are called chain maps, and exact morphisms of degree zero are called nullhomotopic chain maps.
Morphisms in FMExt are degree zero chain maps modulo homotopy.
Remark 4.5*.*
The endomorphisms ΘF define a closed degree (2,0,0) element Θ of the center of K⊗RSeq(D′). The centrality of Θ is used in the proof that the definition above actually defines a dggg category.
Let us recall the free-monodromic complexes T∅ and Ts defined in [1, §5.3.1] and [1, §5.3.2], respectively. Define the following elements of K:
[TABLE]
The following equations are easily checked:
[TABLE]
Both θ and θs are degree (1,0,0) elements of the (graded) center: for w∈{id,s},
[TABLE]
The underlying sequence of T∅ consists of B∅ in position [math], and δT∅=θ.
The underlying sequence of Ts is (…,0,B∅(−1),Bs,B∅(1),0,…), where the non-zero terms are in positions −1 through 1. This sequence can also be denoted by B∅(−1)[1]⊕Bs[0]⊕B∅(1)[−1], and
[TABLE]
These objects can be depicted by the following pictures:
[TABLE]
Much of [1] was guided by the dream that the category of free-monodromic complexes (or at least a large subcategory thereof) should also be monoidal. To this end, [1, §6] defined an operation ⋆ (“free-monodromic convolution”) for a certain class of free-monodromic complexes called “convolutive” and morphisms between them. For an expression w=(s,…,s), let
[TABLE]
and let Tilt2 be the full subcategory of FM2 consisting of Tw⟨n⟩ for expressions w and n∈Z. As a particular case of the main result [1, Theorem 11.4.2], Tilt2 admits a monoidal structure with the operation ⋆ and identity T∅. The hardest part of this result was to show that ⋆ is a bifunctor, i.e. that morphisms in Tilt2 satisfy the exchange law
[TABLE]
In FM2Ext, the operation ⋆ can be defined by the same formula (but replacing ⋆ by ⋆, see the Remark at the end of §4.1) for a similarly defined class of convolutive complexes and morphisms between them. Let Tilt2Ext be the full subcategory of FM2Ext consisting of Tw⟨n⟩[[m]] for expressions w and n,m∈Z.
Conjecture 4.6**.**
(Tilt2Ext,⋆,T∅)* admits a monoidal structure extending that on (Tilt2,⋆,T∅).*
This again reduces to the exchange law (4.4), but even with the various simplifications possible in this GL2 case, it does not seem clear how to adapt the proof in [1]. We will return to this conjecture in future work.
In this paper, we content ourselves with defining an Ext-enhanced monoidal Koszul duality functor ΦsdExt:D2Ext→Tilt2Ext (see Theorem 5.1) assuming Conjecture 4.6.
4.3. A canonical free-monodromic morphism
In preparation for Theorem 5.1, we now define and study an endomorphism ϕs of Ts that will be the image under ΦsdExt of the degree (−2,1) endomorphism of Bs introduced in (3.2b).
Recall that Ts has underlying diagram sequence B∅(−1)[1]⊕Bs(0)[0]⊕B∅(1)[−1]. Below, we define morphisms involving Ts via matrices. Define the endomorphism
[TABLE]
which can be depicted as follows:
[TABLE]
We will check that ϕs is closed in the course of the proof of the following lemma, which is the main goal of this subsection.
Lemma 4.7**.**
We have
[TABLE]
The proof of Lemma 4.7 occupies the rest of this subsection.
Let f∈EndFMExt−2,2,1(Ts). We claim that for degree reasons, f must be of the form
[TABLE]
for some r1,r2,r3∈Λ[ν1,ν2], where r1,r3 are linear and r2 is quadratic, and some linear ξ∈V[[−1]]⊂ΛExt∨.
Let us comment briefly on this claim. For instance, consider the component f31 of f. This component lives in the hom space
[TABLE]
To obtain the degree (−4,4,1) we try to solve the equation
[TABLE]
Given that degxi=(0,2,0), degyi=(2,−2,0), degνi=(−1,2,0), and degξi=(0,0,1), we see that 2b−c=−4, which forces c≥4. This implies that f31=0 since any degree 3 and higher expression in the odd variables ν1,ν2 is zero.
The rest of the claim above may be checked by repeating a similar degree argument for each component using (3.5).
We consider the equation for f to be closed:
[TABLE]
Compute:
[TABLE]
and
[TABLE]
To obtain all the correct signs above, recall that the composition of morphisms in FMExt,dg satisfies (r⊗g)∘(r′⊗g′)=(−1)∣g∣∣r′∣(rr′)⊗(g∘g′). As a useful rule of thumb, remember that each component of δ (resp. f) has odd (resp. even) cohomological degree. Consider for example the (1,2) component f12=r3⊗\leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-4.66794ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt4.26794pt\pgfsys@curveto1.42264pt5.05365pt0.7857pt5.69058pt0.0pt5.69058pt\pgfsys@curveto-0.7857pt5.69058pt-1.42264pt5.05365pt-1.42264pt4.26794pt\pgfsys@curveto-1.42264pt3.48224pt-0.7857pt2.8453pt0.0pt2.8453pt\pgfsys@curveto0.7857pt2.8453pt1.42264pt3.48224pt1.42264pt4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture. Since r3 is linear in ν1,ν2, it has odd cohomological degree. Thus, in this context should also be regarded as being odd, since f12 is even.
Now, viewing (4.6) as a matrix equation and taking the (1,1), (3,3) components (where κ(f) is zero) yields
[TABLE]
Thus we must have ξ=cξs and r1=−r3=−cνs for some c∈k. For these choices for r1,r3, the (2,1) component of (4.6) yields
[TABLE]
and \leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-6.09058ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt-4.26794pt\pgfsys@curveto1.42264pt-3.48224pt0.7857pt-2.8453pt0.0pt-2.8453pt\pgfsys@curveto-0.7857pt-2.8453pt-1.42264pt-3.48224pt-1.42264pt-4.26794pt\pgfsys@curveto-1.42264pt-5.05365pt-0.7857pt-5.69058pt0.0pt-5.69058pt\pgfsys@curveto0.7857pt-5.69058pt1.42264pt-5.05365pt1.42264pt-4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt-4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt-4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture=0 by (3.5), so r2=0. Thus f=−cϕs.
It remains to show that ϕs is closed. We have
[TABLE]
Combined with the computations above, we obtain that κ(ϕs)+δTs∘ϕs−ϕs∘δTs equals
[TABLE]
The components in positions (1,1), (2,1), and (3,3) are zero by the discussion above. It is an exercise to show that the remaining entries are also zero using (4.2) and the defining relations in D2Ext.
Finally, degree considerations entirely similarly to the ones earlier show that
[TABLE]
so ϕs is not nullhomotopic. This completes the proof of Lemma 4.7.
4.4. Free-monodromic “Hochschild unit and counit” morphisms
As in [1, §5.3.4], define morphisms ηs∈HomFMExt1,−1,0(T∅,Ts) and ϵs∈HomFMExt1,−1,0(Ts,T∅) by the matrices
[TABLE]
These morphisms can be depicted by the following pictures:
[TABLE]
Remark 4.8*.*
It is an exercise (see also [1, §5.3.4]) to show that these morphisms are closed; they can be viewed as degree zero chain maps T∅⟨−1⟩→Ts and Ts→T∅⟨1⟩.
Now, define the (closed) morphisms
[TABLE]
given in terms of components by
[TABLE]
which can be depicted as follows:
[TABLE]
The following is the free-monodromic analogue of (3.3x).
Lemma 4.9**.**
Let ξ∈V[[−1]]⊂ΛExt∨. We have
[TABLE]
as morphisms Ts→Ts[[1]] in FM2Ext.
Proof.
According to the definition of ⋆, we have idTs⋆ξ=idTs⋆ξ and s(ξ)⋆idTs=s(ξ)⋆idTs since these morphisms do not involve nontrivial Λ or R∨ components. Using this, the equality ξ−s(ξ)=αs(ξ)ξs, and the explicit chain maps representing ηExts and ϵs described above, one computes that
[TABLE]
may be represented as
[TABLE]
This chain map is nullhomotopic with nullhomotopy given by αs(ξ)h, where h=[0000000\leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-6.09058ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt-4.26794pt\pgfsys@curveto1.42264pt-3.48224pt0.7857pt-2.8453pt0.0pt-2.8453pt\pgfsys@curveto-0.7857pt-2.8453pt-1.42264pt-3.48224pt-1.42264pt-4.26794pt\pgfsys@curveto-1.42264pt-5.05365pt-0.7857pt-5.69058pt0.0pt-5.69058pt\pgfsys@curveto0.7857pt-5.69058pt1.42264pt-5.05365pt1.42264pt-4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt-4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt-4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture0]. Indeed, κ(h)=0 and
[TABLE]
∎
5. Ext-enhanced monoidal Koszul duality
Let φ:V∗→∼V be the isomorphism φ(xi)=yi. Then φ identifies the GL2 realization of S2 with its dual (in particular, φ is S2-equivariant and φ(αs)=αs∨), hence induces a monoidal equivalence between D2 and the diagrammatic Hecke category associated to the dual realization. Composing this with the monoidal Koszul duality functor Φ of [2, Theorem 4.1], we obtain a k-linear monoidal equivalence
[TABLE]
satisfying Φsd∘(1)=⟨1⟩∘Φsd. (Here, sd stands for “self-dual.”) Extend φ multiplicatively to a k-algebra isomorphism φ:R→∼R∨. The functor Φsd is defined by generators and relations: on objects by Φsd(Bs)=Ts, and on morphisms by
[TABLE]
Here, μT∅ is the left monodromy action defined in [1, Theorem 5.2.2]. The morphisms ηs,ϵs were defined explicitly (see (4.7)), whereas b1,b2 were defined more indirectly in [2, §4.2.3]. See [2, §4.2] for more details.
Theorem 5.1**.**
If Conjecture 4.6 holds, then there exists a k-linear monoidal functor
[TABLE]
extending Φ. In particular, Φ(Bw)=Tw for any expression w, and ΦsdExt∘(1)=⟨1⟩∘ΦsdExt. Moreover, ΦsdExt∘[[1]]=[[1]]∘ΦsdExt.
In fact, we expect ΦsdExt to be an equivalence. One reason to believe in this Koszul duality relating Ext-enhanced categories D2Ext and Tilt2Ext is that it is closely related to—and in a certain sense would explain—the “mirror symmetry” of triply-graded Khovanov–Rozansky homology of knots [5], which has been verified in all available computations [6, 12, 16, 13].
The proof of Theorem 5.1 occupies the rest of this paper. As in [2], we construct ΦsdExt by generators and relations: we specify the images of the generating morphisms of D2Ext, and check that these images satisfy the defining relations of D2Ext.
The images of the generating morphisms of D2 are the same as for Φsd, as given in (5.1). For the new generators (3.2b), define
[TABLE]
Then ΦsdExt(\leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-6.09058ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt-4.26794pt\pgfsys@curveto1.42264pt-3.48224pt0.7857pt-2.8453pt0.0pt-2.8453pt\pgfsys@curveto-0.7857pt-2.8453pt-1.42264pt-3.48224pt-1.42264pt-4.26794pt\pgfsys@curveto-1.42264pt-5.05365pt-0.7857pt-5.69058pt0.0pt-5.69058pt\pgfsys@curveto0.7857pt-5.69058pt1.42264pt-5.05365pt1.42264pt-4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt-4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt-4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture)=ηExts and ΦsdExt(\leavevmode\set@color\leavevmodeto3.65pt\vboxto10.76pt\pgfpicture\makeatletter\lower-4.66794ptto0.0pt\pgfsys@beginscope\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@setlinewidth0.4pt\pgfsys@invoke\ignorespaces\nullfontto0.0pt\pgfsys@beginscope\pgfsys@invoke\ignorespaces\pgfsys@setlinewidth0.8pt\pgfsys@invoke\ignorespaces\pgfsys@moveto0.0pt-4.26794pt\pgfsys@lineto0.0pt4.26794pt\pgfsys@stroke\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\ignorespaces\ignorespaces\pgfsys@beginscope\pgfsys@invoke\definecolor[named]pgffillcolorrgb1,1,1\pgfsys@color@gray@fill1\pgfsys@invoke\pgfsys@moveto1.42264pt4.26794pt\pgfsys@curveto1.42264pt5.05365pt0.7857pt5.69058pt0.0pt5.69058pt\pgfsys@curveto-0.7857pt5.69058pt-1.42264pt5.05365pt-1.42264pt4.26794pt\pgfsys@curveto-1.42264pt3.48224pt-0.7857pt2.8453pt0.0pt2.8453pt\pgfsys@curveto0.7857pt2.8453pt1.42264pt3.48224pt1.42264pt4.26794pt\pgfsys@closepath\pgfsys@moveto0.0pt4.26794pt\pgfsys@fillstroke\pgfsys@invoke\ignorespaces\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\pgfsys@beginscope\pgfsys@invoke\pgfsys@transformcm1.00.00.01.00.0pt4.26794pt\pgfsys@invoke\definecolorpgfstrokecolorrgb0,0,0\pgfsys@color@rgb@stroke000\pgfsys@invoke\pgfsys@color@rgb@fill000\pgfsys@invoke\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\ignorespaces\ignorespaces\ignorespaces\hss\pgfsys@discardpath\pgfsys@invoke\lxSVG@closescope\pgfsys@endscope\hss\lxSVG@closescope\endpgfpicture)=ϵExts by (3.2c) and (4.8).
Let us verify the defining relations of D2Ext for these images. By [2, Theorem 4.1], the relations of D2 hold. In particular, we will use below the Frobenius unit relations
[TABLE]
It remains to verify the new relations from §3.1. The relations (3.3l) and (3.3n), which say ϵs∘ϕs∘ηs=ξs and ϕs∘ϕs=0, follow by direct computation via the explicit chain maps for ϕs,ϵs,ηs. The relation (3.3t) is clear, and (3.3x) was verified in (4.9).
Only the relations (3.3g) remain. First, let us show that
[TABLE]
(Here and in what follows, we omit shifts on morphisms from the notation.) We prove the first equality; the second equality is similar. By Lemma 4.7, b2∘(ηExts⋆idTs)=cϕs for some c∈k. Then
Diagrammatically, (5.4) says that the images of ΦsdExt satisfy
[TABLE]
Now, (3.3g) can be deduced from (5.5) and Frobenius associativity in D2:
[TABLE]
and similarly for the vertical reflections.
We have thus shown that ΦsdExt is a well-defined monoidal functor. It is clear from the construction that ΦsdExt commutes with [[1]]. This completes the proof of Theorem 5.1.
Bibliography16
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] Achar, P. N., Makisumi, S., Riche, S. and Williamson, G., Free-monodromic mixed tilting sheaves on flag varieties, preprint ar Xiv:1703.05843.
2[2] Achar, P. N., Makisumi, S., Riche, S. and Williamson, G., Koszul duality for Kac–Moody groups and characters of tilting modules, J. Amer. Math. Soc. , 32 (2019), no. 1, 261–310.
3[3] Achar, P. N. and Riche, S., Modular perverse sheaves on flag varieties II: Koszul duality and formality, Duke Math. J. , 165 (2016), 161–215.
4[4] Bezrukavnikov, R. and Yun, Z., On Koszul duality for Kac-Moody groups. Represent. Theory , 17 (2013), 1–98.
5[5] Dunfield, N. M., Gukov, S. and Rasmussen, J., The superpolynomial for knot homologies, Experiment. Math. , 15 (2006), no. 2, 129–159.
6[6] Elias, B. and Hogancamp, M., On the computation of torus link homology, Compos. Math. , 155 (2019), no. 1, 164–205.