Resolutions of standard modules over KLR algebras of type A
Doeke Buursma
Department of Mathematics
University of Oregon
Eugene
OR 97403, USA
[email protected]
,
Alexander Kleshchev
Department of Mathematics
University of Oregon
Eugene
OR 97403, USA
[email protected]
and
David J. Steinberg
Department of Mathematics
University of Oregon
Eugene
OR 97403, USA
[email protected]
Abstract.
Khovanov-Lauda-Rouquier algebras Rθ of finite Lie type are affine quasihereditary with standard modules Δ(π) labeled by Kostant partitions π of θ. In type A, we construct explicit projective resolutions of standard modules Δ(π).
2010 Mathematics Subject Classification:
16E05, 16G99, 17B37
The second author was supported by the NSF grants DMS-1161094, DMS-1700905, the Max-Planck-Institut, the Fulbright Foundation, and the DFG Mercator program through the University of Stuttgart.
1. Introduction
Let Rθ,F be a Khovanov-Lauda-Rouquier (KLR) algebra of finite Lie type over a field F corresponding to θ∈Q+.
It is known that Rθ,F is affine quasihereditary [4, 1, 7, 8]. In particular, it has finite global dimension and comes with a family of standard modules {Δ(π)F∣π∈KP(θ)} and proper standard modules {Δˉ(π)F∣π∈KP(θ)}, where KP(θ) denotes the set of the Kostant partitions of θ. The (proper) standard modules have well-understood formal characters.
Moreover, L(π)F:=headΔ(π)F≅headΔˉ(π)F is irreducible, and {L(π)F∣π∈KP(θ)} is a complete set of irreducible Rθ,F-modules up to isomorphism and degree shift. Finally, the projective cover P(π)F of L(π)F has a finite Δ-filtration and the (graded) decomposition number dσ,πF:=[Δˉ(σ)F:L(π)F]q equals the (well-defined) multiplicity (P(π)F:Δ(σ)F)q, see [1, Corollary 3.14].
The KLR algebra is defined over the integers, so we have a Z-algebra Rθ=Rθ,Z with Rθ,F=Rθ⊗ZF. The standard modules have natural integral forms Δ(π) with Δ(π)F≅Δ(π)⊗ZF, see [9, §4.2]. Moreover, as illustrated in [9, §4], understanding p-torsion in the Z-module ExtRθm(Δ(π),Δ(σ)) is relevant for comparing decomposition numbers dσ,πC and dσ,πF for a field F of characteristic p. This motivates our interest in
ExtRθm(Δ(π),Δ(σ)) and projective resolutions of (integral forms of) standard modules.
The problem of constructing a projective resolution of Δ(π) reduces easily to the semicuspidal case. To be more precise, let us fix a convex total order ⪯ on the set Φ+ of positive roots. A Kostant partition of θ is a sequence π=(β1m1,…,βtmt) such that m1,…,mt∈Z>0,
β1≻⋯≻βt are positive roots, and
m1β1+⋯+mtβt=θ. Then
Δ(π)≅Δ(β1m1)∘⋯∘Δ(βtmt) where ‘∘’ stands for induction product. Since induction product preserves projective modules, it is enough to resolve the standard modules of the form Δ(βsms), which are exactly the semicuspidal standard modules.
Moreover, the case where βs is a simple root is easy: the algebra Rmαi is the nil-Hecke algebra of rank m and the standard module Δ(αim) is the projective indecomposable module P(αim)=Rmαi1i(m) for an explicit primitive idempotent 1i(m)∈Rmαi, see [5, §2.2].
Let us now specialize to Lie type A∞. Then every non-simple positive root is of the form α=αa+αa+1+⋯+αb+1 for integers a≤b, and we will work with the lexicographic convex order on the positive roots. Let θ:=mα. We consider the set of compositions
[TABLE]
Let Λ(n):={λ∈Λ∣λa+⋯+λb=n}.
In Section 3, for each λ∈Λ, we define explicit idempotents eλ∈Rθ as concatenations of ‘divided power idempotents’ of the form 1i(m) mentioned above. We define projective modules Pλ:=qsλRθeλ, where qsλ stands for the grading shift by an explicitly defined integer sλ. For every μ∈Λ(n+1) and μ∈Λ(n) we then define explicit elements dnμ,λ∈eμRθeλ so that
[TABLE]
is an Rθ-module homomorphism. Taking direct sum over all such gives us a homomorhism
[TABLE]
We note that Pn=0 for n>m(b−a+1) and that dnμ,λ=0 unless the compositions μ and λ differ in just one part. We also have a natural map
[TABLE]
where vαm is the standard generator of Δ(αm) of weight am(a+1)m⋯(b+1)m, see (3.33).
Theorem A**.**
We have that
[TABLE]
is a projective resolution of the standard module Δ(αm).
When m=1 this is a version of the resolution constructed in [1, Theorem 4.12], and our resolution can be thought of as the ‘thick calculus generalization’ (cf. [6]) of that resolution.
For an arbitrary θ, we denote Δ:=⨁π∈KP(θ)Δ(π).
In [2], we will use the resolution from Theorem A to describe the algebra ExtRθ∗(Δ,Δ) in two special cases: (1) θ is a positive root in type A and (2) Lie type is A2, i.e. θ is of the form rα1+sα2.
We now describe the contents of the paper and the main idea of the proof. After reviewing KLR algebras and standard modules in Section 2, we introduce the necessary combinatorial notation and define the resolution P∙ in Section 3.
In order to prove that P∙ is a resolution of Δ(αm), we want to show that it is a direct summand of a resolution Q∙ of qm(m−1)/2Δ(α)∘m introduced in §3.3. The resolution Q∙ is obtained by taking the mth induced power of the resolution constructed in [1, Theorem 4.12].
To check that
P∙ is a direct summand of Q∙, in §3.4
we construct what will end up being a pair of chain maps
f:P∙→Q∙ and g:Q∙→P∙ with g∘f=id. The main difficulty is to verify that f and g are indeed chain maps. This verification occupies Sections 4 and 5. Modulo the fact that f and g are chain maps, Theorem A is proved in §3.7.
2. Preliminaries
2.1. Basic notation
Throughout, we work over an arbitrary commutative unital ground k (since everything is defined over Z, one could just consider the case k=Z).
For r,s∈Z, we use the segment notation
[r,s]:={t∈Z∣r≤t≤s}, [r,s):={t∈Z∣r≤t<s}, etc.
Let q be a variable, and Z((q)) be the ring of Laurent series. For a non-negative integer n we define
[n]:=(qn−q−n)/(q−q−1) and
[n]!:=[1][2]⋯[n].
We denote by Sd the symmetric group on d letters.
It is a Coxeter group with generators {sr:=(r,r+1)∣1≤r<d} and the corresponding length function ℓ. The longest element of Sd is denoted w0 or w0,d. An element w∈Sd is called fully commutative if it is possible to go between any two reduced decompositions of w using only the relations of the form srst=stsr for ∣r−t∣>1.
By definition, Sd acts on [1,d] on the left.
For a set I the d-tuples from Id are written as words i=i1⋯id. The group Sd acts on Id via place permutations as w⋅i=iw−1(1)⋯iw−1(d).
Given a composition λ=(λ1,…,λk) of d, we have the corresponding standard parabolic subgroup
Sλ:=Sλ1×⋯×Sλk≤Sd.
For compositions λ and μ of d, we denote by Dλ (resp. μD, resp. μDλ) the set of the shortest coset representatives for
Sd/Sλ (resp.
Sμ\Sd, resp.
Sμ\Sd/Sλ). The following is well-known and can be deduced for example from [3, Lemma 1.6]:
Lemma 2.1**.**
Let λ,μ be compositions of d and
w∈μD. Then there exist unique elements x∈μDλ and y∈Sλ such that w=xy and ℓ(w)=ℓ(x)+ℓ(y).
2.2. KLR Algebras
From now on, we set I:=Z. If i,j∈I with ∣i−j∣=1 we set εi,j:=j−i∈{1,−1}.
For i,j∈I, we set
[TABLE]
We identify I with the set of vertices of the Dynkin diagram of type A∞ so that the numbers ci,j are the entries of the corresponding Cartan matrix.
The corresponding simple roots are denoted {αi∣i∈I} and set of the positive roots is Φ+:={αr+⋯+αs∣r≤s}. The root lattice is Q:=⨁i∈IZ⋅αi, and we set Q_{+}:=\{\sum_{i}m_{i}\alpha_{i}\in Q\mid m_{i}\in\mathbb{Z}_{\geq 0}\ \text{for all i}\}. For θ=∑imiαi∈Q+, we define its height ht(θ):=∑imi. For θ∈Q+ of height d, we define
Iθ:={i=i1⋯id∈Id∣αi1+⋯+αid=θ}. If i∈Iθ and j∈Iη then the concatenation of words ij is an element of Iθ+η.
Let θ∈Q+ be of height d. The KLR algebra [5, 10] is the unital k-algebra Rθ (with identity denoted 1θ)
with generators
[TABLE]
and defining relations
[TABLE]
The algebra Rθ is graded with deg1i=0; deg(ys)=2; deg(ψr1i)=−cir,ir+1.
We will use the Khovanov-Lauda [5] diagrammatic notation for elements of Rθ. In particular, for i=i1⋯id∈Iθ, 1≤r<d and 1≤s≤d, we denote
[TABLE]
For each element w∈Sn, fix a reduced expression w=sr1⋯srl which determines an element ψw=ψr1⋯ψrl (depending on the choice of a reduced expression).
Theorem 2.2**.**
[5, Theorem 2.5]**, [10, Theorem 3.7]
Let θ∈Q+ and d=ht(θ). Then the following sets are k-bases of Rθ:
[TABLE]
2.3. Parabolic subalgebras and divided power idempotents
Let θ1,…,θt∈Q+ and set θ:=θ1+⋯+θt. Set
[TABLE]
Then we have an algebra embedding
[TABLE]
obtained by horizontal concatenation of the Khovanov-Lauda diagrams. For r1∈Rθ1,…,rt∈Rθt, when there is no confusion, we often write
[TABLE]
For example,
[TABLE]
We fix for the moment i∈I, d∈Z>0 and take θ=dαi, in which case Rdαi is isomorphic to the rank d nil-Hecke algebra NHd.
Following [5], we consider the following elements of Rdαi:
[TABLE]
The following is well-known, see for example [5, §2.2]:
Lemma 2.5**.**
In Rdαi, the elements 1i(d) and 1i(d)′ are idempotents. Moreover,
- (i)
ψw0,dfψw0,d=0* for any polynomial f in y1,…,yd of degree less than d(d−1)/2.*
2. (ii)
ψw0,dy0,dψw0,d=ψw0,d.
Now, let θ∈Q+ be arbitrary. We define Idivθ to be the set of all expressions of the form
i1(d1)⋯ir(dr) with
d1,…,dr∈Z≥0, i1,…,ir∈I
and d1αi1+⋯+drαir=θ. We refer to such expressions as divided power words. We identify Iθ with the subset of Idivθ which consists of all divided power words as above with all dk=1.
We use the same notation for concatenation of divided power words as for concatenation of words.
For i=i1(d1)⋯ir(dr)∈Idivθ, we define the divided power idempotent
[TABLE]
Then we have the following analogue of (2.4):
[TABLE]
To be used as part of the Khovanov-Lauda diagrammatics, we denote
[TABLE]
For example, if d=3, we have
[TABLE]
More generally, we denote
[TABLE]
For any 1≤r<t≤d, we denote the cycle (t,t−1,…,r)∈Sd by (t→r). Note that (t→r)=st−1st−2⋯sr is a unique reduced decomposition. So
[TABLE]
In terms of the diagrammatic notation we have
[TABLE]
Lemma 2.8**.**
In the algebra Rdαi, we have
- (i)
If r1+⋯+rt=d then
1i(r1)⋯i(rt)ψw0,d=ψw0,d and
1i(r1)⋯i(rt)1i(d)=1i(d).
2. (ii)
1i(d)ψd→11ii(d−1)=ψd→11ii(d−1).
Proof.
(i) Write ψw0,d=(ψw0,r1∘⋯∘ψw0,rt)ψu
for some u∈Sd and use Lemma 2.5.
(ii) We have
[TABLE]
where we have used Lemma 2.5(ii) for the third equality.
∎
The following lemma easily follows from the defining relations of Rθ:
Lemma 2.9**.**
Let 1≤r<s≤d, t∈(r,s), u∈[r,s), and i∈Iθ. Then in Rθ we have:
- (i)
1iψs→rψt=1iψt−1ψs→r* unless is=it−1=it±1.*
2. (ii)
1iψs→ryu+1=1iyuψs→r* unless is=iu.*
2.4. Modules over Rθ
Let θ∈Q+. We denote by Rθ-Mod the category of graded left Rθ-modules. The morphisms in this category are all homogeneous degree zero Rθ-homomorphisms, which we denote homRθ(−,−).
For V∈Rθ-Mod, let qdV denote its grading shift by d, so if Vm is the degree m component of V, then (qdV)m=Vm−d. More generally, for a Laurent series a=a(q)=∑nanqn∈Z[q,q−1] with non-negative coefficients, we set aV:=⨁n(qnV)⊕an.
For U,V∈Rθ-Mod,
we set
HomRθ(U,V):=⨁d∈ZHomRθ(U,V)d,
where
[TABLE]
We define
ExtRθm(U,V) similarly in terms of extRθm(U,V).
For θ1,…,θt∈Q+ and θ:=θ1+⋯+θt, recalling (2.3), we have a functor
[TABLE]
For V1∈Rθ1-Mod,…,Vt∈Rθt-Mod, we denote by V1⊠⋯⊠Vt the k-module V1⊗⋯⊗Vt, considered naturally as an (Rθ1⊗⋯⊗Rθt)-module, and set
[TABLE]
For v1∈V1,…,vt∈Vt, we denote
[TABLE]
Since Rθ1θ1,…,θt is a free Rθ1⊗⋯⊗Rθt-module of finite rank by Theorem 2.2, we get the following well-known properties:
Lemma 2.10**.**
The functor Indθ1,…,θt is exact and sends finitely generated projectives to finitely generated projectives.
The following lemma is easy to check using the fact that 1k is an idempotent in Rχ for k∈Idivχ:
Lemma 2.11**.**
For i∈Idivθ and j∈Idivη, we have
Rθ1i∘Rη1j≅Rθ+η1ij.
2.5. Standard modules
The algebra Rθ is affine quasihereditary in the sense of [7]. In particular, it comes with an important class of standard modules, which we now describe explicitly, referring to [1, §3]. We are going to work with the lexicographic convex order ⪯ on Φ+, i.e. for α=αr+⋯+αs∈Φ+ and α′=αr′+⋯+αs′∈Φ+, we have α≺α′ if and only if either r<r′ or r=r′ and s<s′.
Fix α=αr+⋯+αs∈Φ+ of height l:=s−r+1, and set
[TABLE]
We define the Rα-module Δ(α) to be the cyclic Rα-module generated by a vector vα of degree [math] with defining relations
1ivα=δi,iαvα for all i∈Iα;
ψtvα=0 for all 1≤t<l;
ytvα=yuvα for all 1≤t,u≤l.
By [1, Corollary 3.5], Δ(α) is indeed the standard module corresponding to α. In fact, if k=Zp, by [9, (4.11)], Δ(α) is a universal k-form of the standard module in the sense of [9, §4.2].
The module Δ(α) can be considered as an (Rα,k[x])-bimodule with the right action given by vαx:=y1vα. Then it is easy to check that there is an isomorphism of graded k-modules k[x]→Δ(α), f↦vαf.
Fix an integer m∈Z>0.
Let NHm be the rank m nil-Hecke algebra with standard generators x1,…,xm, τ1,…,τm−1. For any i∈I, there is an isomorphism φ:Rmαi⟶∼NHm, yt↦xt, ψu↦τu, and we have the idempotent
[TABLE]
The Rmα-module Δ(α)∘m is cyclicly generated by vα∘m.
As explained in [1, §3.2], NHm acts on Δ(α)∘m on the right so that
[TABLE]
where z is the longest element of D(l,l).
Define
[TABLE]
As in [1, Lemma 3.10], we have an isomorphism of Rmα-modules
[TABLE]
Given θ∈Q+ and a Kostant partition π=(β1m1,…,βtmt)∈KP(θ) as in the Introduction, we define the corresponding standard module
[TABLE]
If k is a field, the modules {Δ(π)∣π∈KP(θ)} are the standard modules for an affine quasihereditary structure on the algebra Rθ, see [1, 7]. If k=Z or Zp, they can be thought of as integral forms of the standard modules, see [9, §4].
3. Semicuspidal resolution
Throughout the subsection, we fix m∈Z>0, a,b∈Z with a≤b, and set
[TABLE]
We denote
[TABLE]
and θ:=mα. Note that l=ht(α) and d=ht(θ).
Our goal is to construct a resolution P∙=P∙αm of the semicuspidal standard module Δ(αm).
3.1. Combinatorics
We consider the set of compositions
[TABLE]
For λ∈Λ, we denote ∣λ∣:=λa+⋯+λb, and for n∈Z≥0, we set
[TABLE]
Let a≤i≤b. We set
[TABLE]
with 1 in the ith position.
Let λ∈Λ. Set
[TABLE]
We also associate to λ a composition ωλ of d with 2n+1 non-negative parts:
[TABLE]
Let i∈[a,b]. We denote
[TABLE]
where ra−1−(λ) is interpreted as [math], and rb+1+(λ) is interpreted as d−∑s=abλs.
Moreover, denote
[TABLE]
Define
[TABLE]
Observe that for all j∈[a,b+1], we have
[TABLE]
We also consider the sets of multicompositions
[TABLE]
Note that by definition all δi(r)∈{0,1}.
For 1≤r≤m and a≤i≤b, we define eir∈Λ(1) to be the multicomposition whose rth component is ei and whose other components are zero.
For δ∈Λ, denote
[TABLE]
Fix δ=(δ(1),…,δ(m))∈Λ. Define
[TABLE]
For i∈[a,b] we define
[TABLE]
Observe that for any j∈[a,b+1], we have
[TABLE]
For λ∈Λ, δ∈Λ, and i∈[a,b], we define some signs:
[TABLE]
(τλ,τδ are not to be confused with τw∈NHd which will not be used again).
Lemma 3.1**.**
Let δ=(δ(1),…,δ(m))∈Λ, i∈[a,b] and r∈[1,m]. If δi(r)=0, then
[TABLE]
Proof.
Let λ:=λδ and γ:=δ+eir.
Writing ‘≡’ for ‘≡ (mod 2)’, we have to prove
[TABLE]
Note that
[TABLE]
so the required comparison boils down to
[TABLE]
which is easy to see.
∎
Lemma 3.3**.**
Let γ=(γ(1),…,γ(m))∈Λ, i∈[a,b] and r∈[1,m]. If γi(r)=1, then
[TABLE]
Proof.
Let μ:=λγ, δ:=γ−eir, and λ:=λδ. Writing ‘≡’ for ‘≡ (mod 2)’, we have to prove the comparison
[TABLE]
Note that
[TABLE]
So, using also (3.2), the required comparison boils down to
[TABLE]
which is easy to see.
∎
3.2. The resolutions P∙αm and P∙π
Let λ∈Λ. Recalling the divided power word iλ∈Idivθ and the integer sλ from §\refSSComb, we set
[TABLE]
In particular, Pλ is a projective left Rθ-module.
Note that 1jλeλ=eλ1jλ=eλ.
Further, set for any n∈Z≥0:
[TABLE]
Note that Pn=0 for n>d−m. The projective resolution P∙=P∙αm of Δ(αm) will be of the form
[TABLE]
To describe the boundary maps dn, we first consider a more general situation. Suppose we are given two sets of idempotents {ea∣a∈A} and {fb∣b∈B} in an algebra R. An A×B matrix D:=(da,b)a∈A,b∈B with every da,b∈eaRfb then yields the homomorphism between the projective R-modules
[TABLE]
which we refer to as the right multiplication with D.
We now define a Λ(n+1)×Λ(n) matrix Dn with entries dnμ,λ∈eμRθeλ. Let λ∈Λ(n) and a≤i≤b be such that λi<m. Recalling (2.7), define
[TABLE]
Note that ψλ;i1jλ=1jλ+eiψλ;i.
Recalling the sign sgnλ;i from §3.1, we now set
[TABLE]
Diagrammatically, for μ=λ+ei as above, we have
[TABLE]
We now set the boundary map dn to be the right multiplication with Dn:
[TABLE]
Example 3.5**.**
Let a=b=1 and m=2. Then the resolution P∙ is
[TABLE]
where d1 is a right multiplication with
[TABLE]
and d0 is a right multiplication with
[TABLE]
It is far from clear that kerdn=imdn+1 but at least the following is easy to see:
Lemma 3.6**.**
The homomorphisms dn are homogeneous of degree [math] for all n.
Proof.
Let λ∈Λ(n) be such that λi<m for some a≤i≤b, so that μ:=λ+ei∈Λ(n+1).
The homomorphism dn=ρDn is a right multiplication with the matrix Dn. Its (μ,λ)-component is a homomorphism
Pμ→Pλ obtained by the right multiplication with ±eμψλ;ieλ. Recall that Pμ=qsμRθeμ and Pλ=qsλRθeλ. So we just need to show that sμ=sλ+deg(eμψλ;ieλ). This is an easy computation using the fact that by definition we have
deg(eμψλ;ieλ)=m−2λi.
∎
Let η∈Q+ be arbitrary and π=(β1m1,…,βtmt)∈KP(η). We will define the resolution P∙π of Δ(π)=Δ(β1m1)∘⋯∘Δ(βtmt) using the general notion of the induced product of chain complexes.
Let (C∙,dC) and (D∙,dD) be chain complexes of left Rη′-modules and Rη′′-modules, respectively. We define a complex of Rη′+η′′-modules
[TABLE]
with differential given by
[TABLE]
Lemma 3.8**.**
If C∙ and D∙ are projective resolutions of modules M and N, respectively, then C∙∘D∙ is a projective resolution of M∘N.
Proof.
This follows from Lemma 2.10 and [11, Lemma 2.7.3].
∎
We now define
[TABLE]
which, by Lemma 3.8, will turn out to be a projective resolution of Δ(π) in view of Theorem 3.34.
3.3. The resolution Q∙
In order to check that P∙ is a resolution of Δ(αm), we show that it is a direct summand of a known resolution Q∙ of qm(m−1)/2Δ(α)∘m. To describe the latter resolution, let us first consider the special case m=1.
Lemma 3.10**.**
We have that P∙α is a resolution of Δ(α).
Proof.
This is a special case of [1, Theorem 4.12], corresponding to the standard choice of (αi+1+⋯+αj,αi) as the minimal pair for an arbitrary positive root αi+⋯+αj in the definition of iα,σ, see [1, §4.5].
∎
Let Q∙ be the resolution qm(m−1)/2(P∙α)∘m, cf. (3.7). To describe Q∙ more explicitly,
let n∈Z≥0 and δ=(δ(1),…,δ(m))∈Λ(n).
Recalling the definitions of §3.1, we set
[TABLE]
For example taking each δ(r) to be 0∈Λ(0), we get
[TABLE]
and
e0=1(a(a+1)⋯(b+1))m.
Further for any n∈Z≥0, we set
[TABLE]
The projective resolution Q∙ is
[TABLE]
with the augmentation map
[TABLE]
see §2.5, and cn being right multiplication with the Λ(n+1)×Λ(n) matrix C=(cnγ,δ) defined as follows.
If δ+eir∈Λ for some r∈[1,m] and i∈[a,b], i.e. δi(r)=0, we set
[TABLE]
Recalling the signs defined in §3.1, for δ∈Λ(n) and γ∈Λ(n+1), we now define
[TABLE]
The fact that Q∙ is indeed isomorphic to the resolution qm(m−1)/2(P∙α)∘m is easily checked using the isomorphism Rθeδ≅Rθ1jδ(1)∘⋯∘Rθ1jδ(m), which comes from Lemma 2.11.
Example 3.14**.**
Let a=b=1 and m=2. Then the resolution Q∙ is
[TABLE]
where c1 is a right multiplication with the matrix
(−12121ψ3 12121ψ1),
and c0 is a right multiplication with the matrix
(12112ψ111221ψ3).
3.4. Comparison maps
We now construct what will end up being a pair of chain maps
f:P∙→Q∙ and g:Q∙→P∙ with g∘f=id. As usual, fn and gn will be given as right multiplications with certain matrices Fn and Gn, respectively.
Let λ∈Λ. Recall the definitions of §3.1. We denote by w0λ the longest element of the parabolic subgroup Sωλ≤Sd. We also denote
[TABLE]
Let δ=(δ(1),…,δ(m))∈Λ.
We define u(δ)∈Sd as follows: for all i=a,…,b, the permutation u(δ) maps:
- ∙
the elements of Ui±(λδ) increasingly to the elements of Ui±(δ);
2. ∙
the elements of Ub+1(λδ) increasingly to the elements of Ub+1(δ).
Set w(δ):=u(δ)−1.
Then w(δ) can also be characterized as the element of Sd which for all i=a,…,b, maps
the elements of Ui(±)(δ) increasingly to the elements of Ui(±)(λδ) and the elements of Ub+1(δ) increasingly to the elements of Ub+1(λδ).
Recall the signs σδ and τδ defined in §3.1.
We now define Fn as the Λ(n)×Λ(n)-matrix with the entries fnλ,δ defined for any λ∈Λ(n),δ∈Λ(n) as follows:
[TABLE]
We define Gn as the Λ(n)×Λ(n)-matrix with the entries gnδ,λ defined for any δ∈Λ(n),λ∈Λ(n) as follows:
[TABLE]
Example 3.15**.**
Let m=d=2 as in Examples 3.5 and 3.14. Then:
[TABLE]
Lemma 3.16**.**
Let δ∈Λ(n) and λ=λδ. Then:
- (i)
deg(ψw(δ)eδ)=2m(m−1)(l−1)−mn+∑i=abλi2,
2. (ii)
deg(fnλ,δ)=−2m(m−1)(l+1)+mn−∑i=abλi2,
3. (iii)
deg(gnδ,λ)=2m(m−1)(l+1)−mn+∑i=abλi2.
Proof.
(i) We prove this by induction on m. Denote the right hand side by R(m) and the left hand side by L(m). If m=1 then w(δ)=1, so L(1)=0. Moreover,
[TABLE]
Let m>1. It suffices to prove that R(m)−R(m−1)=L(m)−L(m−1).
Let δ(m)=(εa,…,εb).
Then, since all εi are [math] or 1, we have
[TABLE]
On the other hand, consider the Khovanov-Lauda diagram of ψw(δ)eδ. The bottom positions of the diagram correspond to to the letters if the word jδ, and so the rightmost l bottom positions of this diagram correspond to the letters of jδ(m). In other words, counting from the right, the sequence of colors of these positions is aεa,…,bεb,b+1,b1−εb,…,a1−εa.
Note that the strings which originate in these positions do not intersect each other, so L(m)−L(m−1) equals the sum of the degrees of the intersections of these strings with the other strings of the diagram, i.e.
[TABLE]
which is easily seen to equal the expression for R(m)−R(m−1) obtained above.
(ii) This follows from (i) since deg(fnλ,δ)=deg(ψw(δ)ed)+deg(eλψw0λ) and
[TABLE]
(iii) This follows from (i) since
[TABLE]
∎
Corollary 3.17**.**
The homomorphisms fn and gn are homogeneous of degree [math] for all n.
Proof.
Let δ∈Λ(n) and λ=λδ.
The homomorphism fn is a right multiplication with the matrix Fn. Its (λ,δ)-component is a homomorphism
Pλ→Qδ obtained by the right multiplication with fnλ,δ. Recall that Pλ=qsλRθeλ and Qδ=qn+m(m−1)/2Rθeδ. So we just need to show that
sλ=n+m(m−1)/2+deg(fnλ,δ), which easily follows from Lemma 3.16(ii).
The homomorphism gn is a right multiplication with the matrix Gn. Its (δ,λ)-component is a homomorphism
Qδ→Pλ obtained by the right multiplication with gnδ,λ. So we just need to show that
n+m(m−1)/2=sλ+deg(gnδ,λ), which easily follows from Lemma 3.16(iii).
∎
Corollary 3.18**.**
Suppose δ,ε∈Λ(n) are such that λδ=λε. Then deg(ψw(δ)eδ)=deg(ψw(ε)eε).
3.5. Independence of reduced decompositions
Throughout this subsection we fix δ∈Λ(n) and set λ:=λδ.
Recall that in general the element ψw∈Rθ depends on a choice of a reduced decomposition of w∈Sd.
While it is clear from the form of the braid relations in the KLR algebra that eλψw0λ does not depend on a choice of a reduced decomposition of w0λ, it is not obvious that a similar statement is true for ψw(δ)eδ and eδψu(δ). So a priori the elements fnλ,δ=±eλψw0λψw(δ)eδ and gnδ,λ=±eδψu(δ)yλeλ might depend on choices of reduced decompositions of w(δ) and u(δ). In this subsection we will prove that this is not the case, and so in a sense the maps fn and gn are canonical.
Recall the composition ωλ and the words jλ,jδ from §3.1.
Lemma 3.19**.**
The element w(δ) is the unique element of ωλD(lm) with w(δ)⋅jδ=jλ.
Proof.
That w(δ)∈ωλD(lm) and w(δ)⋅jδ=jλ follows from the definitions. To prove the uniqueness statement, let w∈ωλD(lm) and w⋅jδ=jλ.
Since w∈ωλD, it maps the elements of Ub+1(δ) increasingly to the elements of Ub+1(λ).
By definition, jδ=jδ(1)…jδ(m). For every r∈[1,m], the entries of jδ(r) have the following properties: (1) each i∈[a,b+1] appears among them exactly once; (2) the entries that precede b+1 appear in the increasing order; (3) the entries that succeed b+1 appear in the decreasing order.
Since w∈D(lm), it maps the positions corresponding to the entries in (2) to the positions which are to the left of the positions occupied with b+1 in jλ, and it maps the positions corresponding to the entries in (3) to the positions which are to the right of the positions occupied with b+1 in jλ.
In other words, for all i∈[a,b], the permutation w maps the elements of Ui±(δ) to the elements of Ui±(λ).
As w∈ωλD, it now follows that for every i∈[a,b], the permutation w maps the elements of Ui±(δ) to the elements of Ui±(λ) increasingly. We have shown that w=w(δ).
∎
Lemma 3.20**.**
Let ε,δ∈Λα, and jε=w⋅jδ for some w∈Sl. Then either ε=δ and w=1, or deg(ψw1jδ)>0.
Proof.
Since every i∈[a,b+1] appears in jδ exactly once, ε=δ implies w=1. On the other hand, if ε=δ, let i be maximal with εi=δi.
Then the strings colored i and i+1 in the Khovanov-Lauda diagram D for ψw1jδ intersect (for any choice of a reduced decomposition of w), which contributes a degree 1 crossing into D. On the other hand, since every j∈[a,b+1] appears in jδ exactly once, D has no same color crossings, which are the only possible crossings of negative degree. The lemma follows.
∎
Lemma 3.21**.**
Suppose that w∈(lm)Dωλ and w⋅jλ is of the form i(1)…i(m) with i(1),…,i(m)∈Iα. Then w⋅jλ=jε for some ε∈Λ with λε=λ.
Proof.
Ler r∈[1,m]. By assumption, the entries i1(r),…,il(r) of i(r) have the following properties: (1) each i∈[a,b+1] appears among them exactly once; (2) the entries that precede b+1 appear in the increasing order; (3) the entries that succeed b+1 appear in the decreasing order. The result follows.
∎
Lemma 3.22**.**
Let P={w∈ωλD∣w⋅jδ=jλ}. Then w(δ)∈P and deg(ψw(δ)eδ)<deg(ψweδ) for any w∈P∖{w(δ)}.
Proof.
It is clear that w(δ)∈P. On the other hand, by Lemma 2.1, an arbitrary w∈P can be written uniquely in the form w=xy with x∈ωλD(lm), y∈S(lm) and ℓ(xy)=ℓ(x)+ℓ(y).
Since xy⋅jδ=jλ, we have y⋅jδ=x−1⋅jλ.
As x−1∈(lm)Dωλ, it follows from Lemma 3.21, that y⋅jδ=x−1⋅jλ is of the form jε for some ε∈Λ with λε=λ. By Lemma 3.19, x=w(ε). If w=w(δ), then y=1 and we have
[TABLE]
where we have used Corollary 3.18 for the second equality and Lemma 3.20 for the inequality.
∎
Lemma 3.23**.**
The element fnλ,δ=σδeλψw0λψw(δ)eδ is independent of the choice of reduced expressions for w0λ and w(δ).
Proof.
It is clear from the form of the braid relations in the KLR algebra that eλψw0λ is independent of the choice of a reduced expression for w0λ. On the other hand, if ψw(δ)eδ and ψw(δ)′eδ correspond to different reduced expressions of w(δ), it follows from the defining relations of the KLR algebra and Theorem 2.2 that ψw(δ)eδ−ψw(δ)′eδ is a linear combination of elements of the form ψuyeδ with u∈Sd, y∈k[y1,…,yd] such that, deg(ψuyeδ)=deg(ψw(δ)eδ) and ujδ=jλ. We have to prove eλψw0λψuyeδ=0. Suppose otherwise.
Since we are using any preferred reduced decompositions for u, we may assume in addition that u∈ωλD, since otherwise eλψw0λψu=0. Now by Lemma 3.22, deg(ψueδ)>deg(ψw(δ)eδ), whence deg(ψuyeδ)>deg(ψw(δ)eδ), giving a contradiction.
∎
Lemma 3.24**.**
The element gnδ,λ=τδeδψu(δ)yλeλ is independent of the choice of a reduced expression for u(δ).
Proof.
The argument is similar to that of the previous lemma.
If eδψu(δ) and eδψu(δ)′ correspond to different reduced expressions of u(δ), then eδψu(δ)−eδψu(δ)′ is a linear combination of elements of the form eδyψw with w∈Sd, y∈k[y1,…,yd] such that deg(eδyψw)=deg(eδψu(δ)) and w−1jδ=jλ. Moreover, in the KL dialgram of ψw1jλ the strings colored b+1 do not cross each other, since this was the case for the KL diagram of ψu(δ)1jλ.
Hence, if eδyψwyλeλ=0, we may assume in addition that w−1∈ωλD. Now, using Lemma 3.22, we conclude that deg(eδyψw)>deg(eδψu(δ)), getting a contradiction.
∎
3.6. Splitting
In this subsection, we aim to show that g∘f=id.
We fix λ∈Λ(n) throughout the subsection. We need to prove
∑δ∈Λ(n)fnλ,δgnδ,λ=eλ. Since fnλ,δ=0 unless λδ=λ, this is equivalent to
[TABLE]
Let δ∈Λ(n) with λδ=λ. We say that δ is initial if a preceeds a+1 in jδ(r) for r∈[1,m−λa] and a succeeds a+1 in jδ(r) for r∈(m−λa,m]. In other words, δ is initial if δa(r)=0 for r∈[1,m−λa] and δa(r)=1 for r∈(m−λa,m].
Let w∈Sd and 1≤r,s≤d. We say that (r,s) is an inversion pair for w if r<s, w(r)>w(s), and jsλ−jrλ=±1.
Lemma 3.25**.**
Let δ∈Λ(n) be initial with λδ=λ. Set αˉ=αa+1+⋯+αb, θˉ=mαˉ, λˉ=(λa+1,…,λb), nˉ:=λa+1+⋯+λb, and δˉ=(δˉ(1),…,δˉ(m)), where δˉ(r)=(δa+1(r),…,δb(r)) for all r∈[1,m]. Then
[TABLE]
Proof.
By definition,
[TABLE]
Throughout the proof, ‘inversion pair’ means ‘inversion pair for w(δ)’. Recall that w(δ)=u(δ)−1.
Since δ is initial, in the Khovanov-Lauda diagram for ψw(δ) (for any choice of reduced expression) no strings of color a cross each other.
We want to apply quadratic relations on pairs of strings one of which has color a and the other has color a+1. These correspond to inversion pairs (r,s) with r∈Ua−(λ),s∈Ua−(λ) or s∈Ua+(λ),r∈Ua+(λ).
Note that there are exactly r−1 inversion pairs of the form (r,s) when r∈Ua−(λ) and d−s inversion pairs of the form (r,s) when s∈Ua+(λ). Applying the corresponding quadratic relations, we see that fnλ,δgnδ,λ equals
[TABLE]
where (∗) a sum of elements of the form
[TABLE]
with X∈Rθˉ, Y± a polynomial in the variables yr with r∈Ua±(λ), and degY−+degY+<degy0,m−λa+degy0,λa′.
By Lemma 2.5(i), we have (∗)=0. So by Lemma 2.5(ii), the expression (3.27) equals the right hand side of (3.26).
∎
Define δλ=(δλ(1),⋯,δλ(m)) to be the unique element of Λ(n) such that for each a≤i≤b we have:
i precedes b+1 in jδλ(r) for 1≤r≤m−λi;
i succeeds b+1 in jδλ(r) for m−λi<r≤m.
Note that λδλ=λ but δλδ in general differs from δ.
Lemma 3.28**.**
Let δ∈Λ satisfy λδ=λ. Then
[TABLE]
Proof.
If δ=δλ, the result follows by induction on ht(α) using from Lemma 3.25. If δ=δλ, we may assume using Lemma 3.25 that δ is not initial. This implies that for some r∈[1,m), we have δa(r)=1 and δa(r+1)=0, i.e. the last entry of the word jδ(r) and the first entry of the word jδ(r+1) are both equal to a. It follows that
a$$a
is a sub-diagram of a Khovanov-Lauda diagram for ψw(δ)ψu(δ)yλeλ, so
fnλ,δgnδ,λ=±eλψw0λψw(δ)ψu(δ)yλeλ=0.
∎
Corollary 3.29**.**
For any n∈Z≥0, we have gn∘fn=id.
3.7. Proof of Theorem A, assuming f and g are chain maps
In Sections 4 and 5, we will prove that f and g constructed above are chain maps. The goal of this subsection is to demonstrate that this is sufficient to establish our main result.
Let
[TABLE]
We also define W0
to be the longest element of D(lm).
For 1≤r<m, we have a fully commutative element
[TABLE]
In fact, if w0,m=sr1…srN is a reduced decomposition in Sm, then W0=zr1…zrN in Sd with ℓ(W0)=ℓ(zr1)+⋯+ℓ(zrN).
Recalling (3.11) we have
[TABLE]
Then the following is easy to check:
Lemma 3.30**.**
Let P={w∈D(lm)∣w⋅j0=j0}.
Then W0∈P and deg(ψW0e0)<deg(ψwe0) for any w∈P∖{W0}
Lemma 3.31**.**
Let ψW0 and ψW0′ in Rθ correspond to different reduced expressions of W0∈Sd. Then
ψW0vα∘m=ψW0′vα∘m in
Δ(α)∘m.
Proof.
It follows from the defining relations of the KLR algebra and Theorem 2.2 that ψW0e0−ψW0′e0 is a linear combination of elements of the form ψwe0 where w⋅j0=j0, w=W0, and deg(ψwe0)=deg(ψW0e0).
Moreover, for each such w, by Lemma 2.1, there exist (unique) elements x∈D(lm) and y∈S(lm) such that w=xy and ℓ(w)=ℓ(x)+ℓ(y).
Since ψyvα∘m=0 if y=1, we may assume w∈D(lm).
We now have w∈D(lm), w⋅j0=j0, and w=W0.
By Lemma 3.30, no such element w can exist, so (ψW0e0−ψW0′e0)vα∘m=0.
∎
Lemma 3.32**.**
There is a choice of reduced expression of W0 for which g00,0f00,0=e0Y0ψW0e0.
Proof.
Note that ℓ(u(0)w00w(0))=ℓ(u(0))+ℓ(w00)+ℓ(w(0)) and u(0)w00w(0)=W0. So for an appropriate choice of reduced expression for W0, we have ψu(0)ψw00ψw(0)=ψW0. We now compute:
[TABLE]
where the first equality is by definition of g00,0 and f00,0, the second follows from Lemma 2.5(ii), the third by the defining relations of Rθ and the observation that Y0=u(0)⋅y0, and the fourth from the remark at the beginning of the proof.
∎
Note that
[TABLE]
where g is the longest element of Sd with g⋅(iαm)=am(a+1)m⋯(b+1)m. Let
[TABLE]
We refer to vαm as the standard generator of Δ(αm). It can be checked directly but also follows from Theorem 3.34 that it does generate Δ(αm). Note that e0vαm=vαm, so there is a homomorphism
[TABLE]
where vαm is the standard generator of Δ(αm), see (3.33).
Theorem 3.34**.**
If f and g are chain maps then P∙=P∙αm is a projective resolution of Δ(αm), with the augmentation map p.
Proof.
The modules Pn are projective by construction. By Corollary 3.29, P∙ is a complex, isomorphic to a direct summand of the complex Q∙. Since Q∙ is a resolution of qm(m−1)/2Δ(α)∘m,
it follows from the assumptions that P∙ is exact in degrees >0 and its [math]th cohomology is given by qf0g0(Q0) where q:Q0→qm(m−1)/2Δ(α)∘m is the augmentation map (3.12).
Recalling the (Rθ,NHm)-bimodule structure of qm(m−1)/2Δ(α)∘m from §2.5 and the idempotent em∈NHm defined in (2.12), using Lemma 3.31, we have
[TABLE]
where w0,m=sr1⋯srN is a reduced decomposition in Sm.
We now have:
[TABLE]
where the first equality follows from Lemma 3.32, the second from the definition of q, the third from (3.35), and the last two from the definitions.
It remains to note that q(f0(e0))=vαm. For,
note that
q(f0(e0))=f0vα∘m, and q(f0(e0))∈q(f0(P0))=q(f0(g0(Q0)))=Δ(αm). So
f0vα∘m=f0vα∘mem=vαm.
∎
4. Verification that f is a chain map
We continue with the running assumptions of the previous section.
In addition, throughout the section we fix
[TABLE]
4.1. Special reduced expressions
Recall the notation of §3.1.
Let λ∈Λαm and δ∈Λα be such that λˉ:=λ−δ∈Λαm−1. For i∈[a,b+1], we denote by pi=pi(δ)∈[1,l] the position occupied by i in jδ, and set
[TABLE]
Let
[TABLE]
Note that {pa,…,pb+1}=[1,l].
Define xδλ∈Sd to be the permutation which maps pi to qi for all i∈[a,b+1], and maps the elements of [l+1,d] increasingly to the elements of [1,d]∖Q. It is easy to see that xδλ is fully commutative, so ψxδλ is well defined, and 1jλψxδλ=ψxδλ1jδjλˉ.
Now let λ=λδ, and r∈[1,m]. Define
[TABLE]
Define x(δ,1):=xδ(1)λ∈Sd. More generally, define for all r=1,…,m−1, the permutations x(δ,r)∈Sd so that x(δ,r) is the image of
[TABLE]
under the natural embedding S(r−1)l×S(m−r+1)l↪Sd.
Recall the element w(δ)∈Sd defined in §3.4. The following lemma follows from definitions:
Lemma 4.2**.**
We have w(δ)=x(δ,1)⋯x(δ,m−1) and ℓ(w(δ))=ℓ(x(δ,1))+⋯+ℓ(x(δ,m−1)).
In view of the lemma, when convenient, we will always choose reduced decompositions so that
[TABLE]
By Lemma 3.23, we then have
[TABLE]
Example 4.5**.**
Let a=1, b=2, λ=(2,2), and {\bm{\delta}}=\big{(}(1,0),(1,1),(0,1),(0,0)\big{)}, so that m=4 and n=4. Then fnλ,δ is
[TABLE]
4.2. A commutation lemma
It will be convenient to use the following notation. Let λ∈Λ(n). Consider the parabolic (non-unital) subalgebra
[TABLE]
We have
[TABLE]
The natural (unital) embeddings of the algebras
[TABLE]
into Rλ, allow us to consider them as (non-unital) subalgebras of Rθ. We denote the corresponding (non-unital) algebra embeddings by
[TABLE]
For example, setting
[TABLE]
for all i∈[a,b], can write ψw0λ∈Rθ as a commuting product
[TABLE]
Lemma 4.7**.**
Let i∈[a,b] with μi>0, and λ:=μ−ei. Then
[TABLE]
Proof.
We have
[TABLE]
where we have used (4.6) for the second equality, Lemma 2.9(i) for the third equality, and the definition of ψλ,i as an explicit cycle element for the fourth equality.
∎
Example 4.8**.**
We illustrate Lemma 4.7 diagramatically. Let a=1, b=2, m=4 and μ=(2,3). Then, Lemma 4.7 claims the following equality:
[TABLE]
4.3. f is a chain map
Let λ=λδ. For i∈[a,b+1], let
qi=qi(λ,δ(1)) be defined as in (4.1), and consider the cycle
[TABLE]
Let c be the commuting product of cycles:
[TABLE]
Recall also the elements defined in (3.13).
Lemma 4.9**.**
Suppose i∈[a,b] is such that μi>0 and λ:=μ−ei=λδ. Set θˉ=(m−1)α, μˉ:=μ−δ(1), and λˉ:=λ−δ(1). Then 1jμψλ;ieλψw0λψx(δ,1) equals
[TABLE]
Proof.
By Lemma 2.5(ii), we have eλψw0λ=1jλψw0λ. So using also Lemma 4.7, we get
[TABLE]
As usual, we denote by pj the position occupied by j in δ(1) and qj be defined as in (4.1).
By Lemma 4.2, the permutation x(δ,1) maps pj to qj for all j∈[a,b+1], and the elements of [l+1,d] increasingly to the elements of [1,d]∖Q.
Case 1: δi(1)=0. In this case we have qi=li−(λ). So the KLR diagram D of 1jμψli+(μ)→li−(λ)ψx(δ,1) has an i-string S from the position pi in the bottom to the position li+(μ) in the top, and the only (i,i)-crossings in D will be the crossings of the string S with the m−μi strings which originate in the positions Li−(μ) in the top.
The (i+1)-string T in D originating in position pi+1 in the bottom is to the right of all (i,i) crossings. Pulling T to the left produces error terms, which arise from opening (i,i)-crossings, but all of them, except the last one, amount to zero, when multiplied on the left by 1jμψw0μ. The last error term is equal to ψxδ(1)μια,θˉ(1α⊗ψli+(μˉ)→li−(λˉ))), and the result of pulling T past all (i,i)-crossings gives ψx(δ+ei1,1)ψδ;1,i. Multiplying on the left by 1jμψw0μ, gives
[TABLE]
and it remains to observe using Lemma 2.9 that
[TABLE]
Case 2: δi(1)=1.
Let D be the KLR diagram of 1jμψli+(μ)→li−(λ)ψx(δ,1).
Let S be the i-string originating in the position li+(μ) in the top row of D, and T be the (i+1)-string originating in the position pi+1 in the bottom row of D. The quadratic relation on these strings produces a difference of two terms, one with a dot on S and the other with a dot on T. The term with a dot on T equals [math] after multiplying on the left by 1jμψw0μ. The term with a dot on S, when multiplied on the left by 1jμψw0μ, yields
[TABLE]
where we have used Lemma 2.9 for the last equality.
∎
Example 4.10**.**
We illustrate Lemma 4.9 diagramatically.
Let a=1, b=2, m=4, μ=(2,3) and δ=((1,0),(0,1),(1,1),(0,1)).
First, take i=2. Then δ1(2)=0 and Lemma 4.9 claims the following:
[TABLE]
On the other hand, if i=1, we have δ1(1)=1, and Lemma 4.9 posits the following equality.
[TABLE]
Corollary 4.11**.**
If μi>0 for some i∈[a,b] and λ:=μ−ei=λδ, then
[TABLE]
Proof.
The proof is by induction on m, the induction base m=1 being obvious. Let δˉ:=(δ(2),…,δ(m)), θˉ=(m−1)α, μˉ:=μ−δ(1), and λˉ:=λ−δ(1).
By (4.3), we have
[TABLE]
Now we apply Lemma 4.9. We consider the case δi(1)=0, the case δi(1)=1 being similar. Then we get the following expression for the right hand side of (4.13):
[TABLE]
Opening parentheses, we get two summands S1+S2.
Note that
[TABLE]
Moreover, using the inductive assumption for the third equality below, we get that S2 equals
[TABLE]
As δi(1)=0, we have ∑s=2r−1δi(s)=∑s=1r−1δi(s). So
[TABLE]
Moreover,
[TABLE]
So S2 equals
[TABLE]
where we have used Lemma 2.9(i) to see the first equality.
Thus S1+S2 equals the right hand side of (4.12).
∎
The following statement means that f is chain map:
Proposition 4.14**.**
Let μ∈Λ(n+1) and δ∈Λ(n). Then
[TABLE]
Proof.
By definition, dnμ,λ=0 unless λ=μ−ei for some i∈[a,b], and fnλ,δ=0 unless λ=λδ. On the other hand, fnμ,γ=0 unless μ=λγ,
and
cnγ,δ=0, unless δ=γ−eir for some i∈[a,b] and r∈[1,m]. So we may assume that μ=λδ+eir for some i∈[a,b] and r∈[1,m] such that δi(r)=0. In this case, letting λ:=λδ, we have to prove
[TABLE]
By definition of the elements involved, this means
[TABLE]
Equivalently, we need to prove
[TABLE]
which, in view of Corollary 4.11, is equivalent to the statement that
[TABLE]
for all r∈[1,m] such that δi(r)=0. But this is Lemma 3.1.
∎
5. Verification that g is a chain map
We continue with the running assumptions of Section 3.
In addition, throughout the section we fix n∈Z≥0, λ∈Λ(n) and γ=(γ(1),…,γ(m))∈Λ(n+1).
5.1. Special reduced expressions and a commutation lemma
Recall the notation of §3.1.
Let μ∈Λαm and γ∈Λα be such that μˉ:=μ−γ∈Λαm−1. For i∈[a,b+1], we denote
[TABLE]
Let Q:={qa,…,qb+1}.
Note that {pa,…,pb+1}=(d−l,d].
Define zμγ∈Sd to be the permutation which maps qi to pi for all i∈[a,b+1], and maps the elements of
[1,d]∖Q increasingly to the elements of [1,d−l]. It is easy to see that zμγ is fully commutative, so ψzμγ is well defined, and ψzμγ1jμ=1jμˉjγψzμγ.
Now let μ:=λγ, and r∈[1,m]. Define
[TABLE]
Define z(γ,m):=zμγ(m)∈Sd. More generally, define for all r=2,…,m, the permutations
[TABLE]
Recall the element u(γ)∈Sd defined in §3.4. The following lemma follows from definitions:
Lemma 5.6**.**
We have u(γ)=z(γ,2)⋯z(γ,m) and ℓ(u(γ))=ℓ(z(γ,1))+⋯+ℓ(z(γ,m)).
In view of the lemma, when convenient, we will always choose reduced decompositions so that
[TABLE]
In view of Lemma 3.23, we have
[TABLE]
Example 5.9**.**
Let a=1, b=2, μ=(2,2), and {\bm{\gamma}}=\big{(}(1,0),(1,1),(0,1),(0,0)\big{)}, so that m=4 and n=4. Then gnλ,γ is
[TABLE]
Lemma 5.10**.**
Let i∈[a,b] with λi<m, and μ:=λ+ei. Then
[TABLE]
Proof.
Recall the notation of §4.2, we have that yμeμψλ;ieλ equals
[TABLE]
where we have used Lemma 2.9 for the second equality and Lemma 2.8 for the last equality.
∎
Example 5.11**.**
We illustrate Lemma 5.10 diagramatically. Let a=1, b=2, m=4, μ=(2,3) and λ=(1,3) so that i=1 using the notation of Lemma 5.10. Then, the lemma claims the following equality.
[TABLE]
5.2. g is a chain map
Recall that we have fixed λ∈Λ(n) and γ∈Λ(n+1).
Lemma 5.12**.**
Suppose that i∈[a,b] is such that λi<m and μ:=λ+ei=λγ. Set θˉ=(m−1)α, μˉ:=μ−γ(m), and λˉ:=λ−γ(m). Then ψz(γ,m)yμeμψλ;ieλ equals
[TABLE]
Proof.
Using Lemma 5.10, we get
[TABLE]
Recalling (5.1),(5.5), the permutation z(γ,m) maps qj to pj for all j∈[a,b+1], and the elements of
[1,d]∖Q increasingly to the elements of [1,d−l].
Case 1: γi(m)=1. In this case we have qi=ri+(λ). So the KLR diagram D of ψz(γ,m)ψλ;i1jλ has an i-string S from the position ri−(λ) in the bottom to the position pi in the top, and the only (i,i)-crossings in D will be the crossings of the string S with the λi strings which originate in the positions Ri+(λ) in the bottom.
The (i+1)-string T in D originating in the position pi+1 in the top is to the left of all (i,i) crossings. Pulling T to the right produces error terms, which arise from opening (i,i)-crossings, but all of them, except the last one, amount to zero, when multiplied on the right by yλeλ. The last error term is equal to
−ιθˉ,α(ψλˉ;i⊗1α)ψzλγ(m),
and the result of pulling T past all (i,i)-crossings gives ψγ−eim;m,iψz(γ−eim,m).
Multiplying on the left by yλeλ, gives
[TABLE]
and it remains to observe using Lemma 2.9 that
[TABLE]
Case 2: γi(m)=0.
Let D be the KLR diagram of ψz(γ,m)ψλ;i1jλ.
Let S be the i-string originating in the position ri−(λ) in the bottom row of D, and T be the (i+1)-string originating in the position pi+1 in the top row of D. The quadratic relation on these strings produces a linear combination of two diagrams, one with a dot on S and the other with a dot on T. The term containing a dot on T produces a term which is zero after multiplying on the right by yλeλ. The term containing a dot on S, when multiplied on the right by yλeλ, yields
[TABLE]
where we have used Lemmas 2.9 and 2.8 to deduce the last equality.
∎
Example 5.13**.**
We illustrate Lemma 5.12 diagramatically. Using the notation of the lemma, let a=1, b=2, μ=(2,2), λ=(1,2), m=4 and i=1. Then, if {\bm{\gamma}}=\big{(}(1,0),(1,1),(0,1),(0,0)\big{)}, we are in the γi(m)=0 case and the lemma claims the following equality.
[TABLE]
If instead we let {\bm{\gamma}}=\big{(}(0,0),(1,1),(0,1),(1,0)\big{)}, we are in the γi(m)=1 case and the lemma claims the following equality.
[TABLE]
Corollary 5.14**.**
Suppose that i∈[a,b] is such that λi<m and μ:=λ+ei=λγ. Then
[TABLE]
Proof.
The proof is by induction on m, the induction base m=1 being obvious. Let γˉ:=(γ(1),…,γ(m−1)), θˉ=(m−1)α, μˉ:=μ−γ(m), and λˉ:=λ−γ(m).
By (5.7), we have
[TABLE]
Now we apply Lemma 5.12. We consider the case γi(m)=1, the case γi(m)=0 being similar. Then we get the following expression for the right hand side of (5.15):
[TABLE]
Opening parentheses, we get two summands S1+S2.
Note that
[TABLE]
Moreover,
using the inductive assumption,
for the third equality below,
S2 equals
[TABLE]
Thus S1+S2 equals the right hand side of the expression in the corollary.
∎
The following statement means that g is chain map:
Proposition 5.16**.**
Let λ∈Λ(n) and γ∈Λ(n+1). Then
[TABLE]
Proof.
By definition, dnμ,λ=0 unless μ=λ+ei for some i∈[a,b], and gn+1γ,μ=0 unless μ=λγ. On the other hand, gnδ,λ=0 unless λ=λδ,
and
cnγ,δ=0, unless δ=γ−eir for some i∈[a,b] and r∈[1,m]. So we may assume that there exists i∈[a,b] such that λγ=λ+ei, in which case, setting μ:=λγ, we have to prove
[TABLE]
By definition of the elements involved, this means
[TABLE]
Equivalently, we need to prove
[TABLE]
which, in view of Corollary 5.14, is equivalent to the statement that
[TABLE]
for all r∈[1,m] such that γi(r)=1. But this is Lemma 3.3.
∎