This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

On the action of the long cycle on the Kazhdan-Lusztig basis

Fern Gossow\addressmark1 [email protected]       Oded Yacobi\addressmark1 [email protected] Oded Yacobi was partially supported by the ARC grant DP180102563 \addressmark1School of Mathematics and Statistics, University of Sydney, Australia
Abstract

The complex irreducible representations of the symmetric group carry an important canonical basis called the Kazhdan-Lusztig basis. Although it is difficult to express how general permutations act on this basis, some distinguished permutations have beautiful descriptions. In 2010 Rhoades showed that the long cycle (1,2,,n)(1,2,...,n) acts by the jeu-de-taquin promotion operator in the case when the irreducible representation is indexed by a rectangular partition. We prove a generalisation of this theorem in two directions: on the one hand we lift the restriction on the shape of the partition, and on the other hand we enlarge the result to the collection of all separable permutations.

keywords:
Specht modules, Kazhdan-Lusztig basis, promotion, evacuation

1 Main Results

The complex irreducible representations, aka Specht modules, of the symmetric group SnS_{n} are indexed by partitions of nn: for such a partition λn\lambda\vdash n we denote by SλS^{\lambda} the associated representation. These carry a canonical basis called the Kazhdan-Lusztig basis {CTTSYT(λ)}\{C_{T}\mid T\in SYT(\lambda)\}, which is indexed by the set of standard Young tableaux of shape λ\lambda.

Although the Kazhdan-Lusztig basis enjoys many wonderful properties, it is quite complicated to express its transformation under the action of an arbitrary permutation. Nevertheless, owing to a close connection with the RSK correspondence, it is possible for certain distinguished elements.

In the mid 1990s Berenstein-Zelevinsky [1] and Stembridge [10] showed independently that the long element of SnS_{n} acts on the Kazhdan-Lusztig basis (up to sign) by Schützenberger’s evacuation operator. More recently, Rhoades proved in the case of rectangular partitions that the long cycle c=(1,2,,n)c=(1,2,\ldots,n) acts on the Kazhdan-Lusztig basis (up to sign) by the jeu-de-taquin promotion operator, pr\mathrm{pr} [8]. This was the essential new ingredient needed in Rhoades’ cyclic sieving phenomenon regarding the action of promotion on rectangular standard Young tableaux.

Our first main result is a generalisation of Rhoades’ Theorem to arbitrary partitions. Fix a partition λn\lambda\vdash n and number its removable boxes 1,2,,r1,2,\dots,r moving downward. Define the index of TSYT(λ)T\in SYT(\lambda) to be the said number of the removable box containing nn, and denote it idx(T)\operatorname{idx}(T). Choose an ordering of the Kazhdan-Lusztig basis which is weakly increasing along the index. For wSnw\in S_{n} let [w][w] be the matrix of w:SλSλw:S^{\lambda}\to S^{\lambda} in this ordered basis.

Theorem 1.1.

Let λn\lambda\vdash n be an arbitrary partition. Let [c]=QR[c]=QR be the QR decomposition of [c][c] into the product of an orthogonal matrix QQ and an upper triangular matrix RR. Then QQ is a generalised permutation matrix for pr\mathrm{pr} with entries ±1\pm 1, i.e. for any TSYT(λ)T\in SYT(\lambda) we have

QCT=±Cpr(T).QC_{T}=\pm C_{\mathrm{pr}(T)}.

The sign appearing above depends only on the index of TT.

Equivalently, this theorem states that cc acts on the Kazhdan-Lusztig basis by promotion on the leading term, i.e.

cCT=±Cpr(T)+RaRCpr(R),c\cdot C_{T}=\pm C_{\mathrm{pr}(T)}+\sum_{R}a_{R}C_{\mathrm{pr}(R)},

where the sum is over RR with index strictly smaller than the index of TT, and aR𝐙a_{R}\in\mathbf{Z}. In the case of a rectangular partition, the sum is empty and we recover Rhoades’ Theorem.

The key property of rectangular partitions that Rhoades leverages is that the restriction of the corresponding Specht module to Sn1S_{n-1} remains irreducible, and the Kazhdan-Lusztig basis is unchanged if the space is regarded as an Sn1S_{n-1}-module or an SnS_{n}-module. It is therefore not surprising that the key step in our argument involves the interaction between the restriction functor and the Kazhdan-Lusztig basis in a more general setting, which may be interesting in its own right.

For k0k\geq 0 define the subspace Skλ=span{CTTSYT(λ) and idx(T)k}S^{\lambda}_{k}=\mathrm{span}\{C_{T}\mid T\in SYT(\lambda)\text{ and }\operatorname{idx}(T)\leq k\} of SλS^{\lambda}. This gives rise to a filtration 0=S0λS1λS2λSrλ=Sλ0=S^{\lambda}_{0}\subset S^{\lambda}_{1}\subset S^{\lambda}_{2}\subset\cdots\subset S^{\lambda}_{r}=S^{\lambda}, where rr is the number of removable boxes of λ\lambda.

Theorem 1.2.

Each subspace SiλS_{i}^{\lambda} is Sn1S_{n-1}-invariant, and for every 1ir1\leq i\leq r, there is an isomorphism of Sn1S_{n-1}-modules Siλ/Si1λSμiS^{\lambda}_{i}/S^{\lambda}_{i-1}\cong S^{\mu_{i}}, where μin1\mu_{i}\vdash n-1 is the partition obtained from λ\lambda by deleting the ithi^{\text{th}} removable box.

Note that this theorem provides a new proof that the branching law for symmetric groups is multiplicity-free, as

resSn1SnSλi=1rSiλ/Si1λi=1rSμi.\mathrm{res}^{S_{n}}_{S_{n-1}}S^{\lambda}\cong\bigoplus_{i=1}^{r}S_{i}^{\lambda}/S_{i-1}^{\lambda}\cong\bigoplus_{i=1}^{r}S^{\mu_{i}}.

Finally, we mention a generalisation of Theorem 1.1 to the class of all separable permutations, which includes the long element and the long cycle. By definition, separable permutations are those that can be obtained from eS1e\in S_{1} by repeated applications of direct sum and skew sum. Alternatively, these are the permutations that avoid the patterns 24132413 and 31423142 [6].

Fix λn\lambda\vdash n an arbitrary partition. In Section 5, we describe briefly how to associate to each separable permutation wSnw\in S_{n} a bijection φ\varphi of SYT(λ)SYT(\lambda) such that an analogous result to Theorem 1.1 holds. That is, if [w]:SλSλ[w]:S^{\lambda}\to S^{\lambda} is the matrix giving the action of ww on some given ordering of the KL basis, we have:

Theorem 1.3.

Let wSnw\in S_{n} be any separable permutation, and let [w]=QR[w]=QR be the QR decomposition. Then QQ is a (generalised) permutation matrix of φ\varphi.

In specific cases, we can prove this combinatorially but the general case requires results from categorical representation theory and in particular the action of braid groups on triangulated categories [4]. In our forthcoming work we will describe this connection, and use this to also derive variations of these theorems for arbitrary simply-laced semisimple Lie algebras.

2 Background

In this section we briefly recall some basic results we’ll need. We refer the reader to [9] for more details on notions from algebraic combinatorics, and to [2] for the necessary background on Kazhdan-Lusztig theory.

2.1 Combinatorics

Let λn\lambda\vdash n be a partition of nn, which we depict using its associated Young diagram. Recall that a box in λ\lambda is called a removable box if its deletion results in another Young diagram. We label the removable boxes of λ\lambda by 1,2,,r1,2,\dots,r starting from the top, and moving down the tableau. For example, the partition (6,6,3,1)16(6,6,3,1)\vdash 16 has three removable boxes, with labelling given by:

\young(,1,2,3)\young(~{}~{}~{}~{}~{}~{},~{}~{}~{}~{}~{}1,~{}~{}2,3)

Write SYT(λ)SYT(\lambda) for the set of standard Young tableaux of shape λ\lambda. We define the index of TSYT(λ)T\in SYT(\lambda), denoted idx(T)\operatorname{idx}(T), to be the label of the removable box containing nn as above. If RR and TT have the same index, then their largest box occupies the same position. When ordering SYT(λ)SYT(\lambda) according to index, we can break ties by comparing the index values of the tableaux after removing their largest boxes. Repeating this procedure gives the total index ordering on SYT(λ)SYT(\lambda).

Recall that the descent set D(T)D(T) of TT is the set of jj in {1,2,,n1}\{1,2,\dots,n-1\} such that j+1j+1 appears strictly below jj in TT.

The promotion map pr:SYT(λ)SYT(λ)\mathrm{pr}:SYT(\lambda)\to SYT(\lambda) is defined as follows: replace the box with value nn by a dot, and move this dot north-west by repeatedly swapping it with the box above or to the left. If both boxes exist, swap with the larger. Once the box has reached the north-west corner, increment all values in the tableau and replace the dot with 1.

\young(134,25)\young(134,2)\young(14,23)\young(14,23)\young(125,34)\young(134,25)\to\young(134,2\bullet)\to\young(1\bullet 4,23)\to\young(\bullet 14,23)\to\young(125,34)

The evacuation map evn:SYT(λ)SYT(λ)\mathrm{ev}_{n}:SYT(\lambda)\to SYT(\lambda) is defined as follows: denote every box in the tableau ‘fixed’ or ‘unfixed’, with all boxes initially starting unfixed. At each step, perform the inverse of the promotion map on the unfixed boxes, and fix the largest remaining box. Repeat until all boxes are fixed. In the example below fixed boxes are in boldface:

\young(134,25)\young(123,4𝟓)\young(12𝟒,3𝟓)\young(1𝟑𝟒,2𝟓)\young(1𝟑𝟒,𝟐𝟓)\young(𝟏𝟑𝟒,𝟐𝟓)\young(134,25)\to\young(123,4{{\bf 5}})\to\young(12{{\bf 4}},3{{\bf 5}})\to\young(1{{\bf 3}}{{\bf 4}},2{{\bf 5}})\to\young(1{{\bf 3}}{{\bf 4}},{{\bf 2}}{{\bf 5}})\to\young({{\bf 1}}{{\bf 3}}{{\bf 4}},{{\bf 2}}{{\bf 5}})

Both promotion and evacuation are bijections on SYT(λ)SYT(\lambda), and evacuation is in fact an involution. It will also be useful to define the partial evacuation maps. For 1kn1\leq k\leq n, define the involution evk:SYT(λ)SYT(λ)\mathrm{ev}_{k}:SYT(\lambda)\to SYT(\lambda) by fixing all but the smallest kk boxes, and performing evacuation on the remaining unfixed tableau.

Lemma 2.1.

For every TSYT(λ)T\in SYT(\lambda), we have pr(T)=evnevn1(T)\mathrm{pr}(T)=\mathrm{ev}_{n}\mathrm{ev}_{n-1}(T).

The RSK correspondence [7] gives a bijection between wSnw\in S_{n} and ordered pairs (P,Q)(P,Q) of tableaux of some shape λn\lambda\vdash n. We will denote the RSK correspondence by w(P,Q)w\rightsquigarrow(P,Q).

2.2 Representation theory

The symmetric group SnS_{n} is generated by the simple transpositions sj:=(j,j+1)s_{j}:=(j,j+1) for 1jn11\leq j\leq n-1, satisfying the braid relations and sj2=1s_{j}^{2}=1. Given an element wSnw\in S_{n}, we write l(w)l(w) for the length of ww in these generators.

Given permutations v,wSnv,w\in S_{n}, we write v<wv<w in the Bruhat order if ww can be obtained from vv by successively swapping two elements in the permutation’s one-line notation, increasing the length of the intermediate permutations at each swap. In particular, v<wv<w implies l(v)<l(w)l(v)<l(w). Denote by D(w)D(w) the (left) descent set of ww, which consists of the indices jj between 11 and n1n-1 such that sjw<ws_{j}w<w.

Let qq be an indeterminate. The Hecke algebra (of type A) Hn(q)H_{n}(q) is an algebra over 𝐙[q±1/2]\mathbf{Z}[q^{\pm 1/2}] generated by {Ts1,,Tsn1}\{T_{s_{1}},\dots,T_{s_{n-1}}\} satisfying the braid relations and the quadratic relation Tsj2=q+(q1)TsjT_{s_{j}}^{2}=q+(q-1)T_{s_{j}} for all 1jn11\leq j\leq n-1. As a 𝐙[q±1/2]\mathbf{Z}[q^{\pm 1/2}]-module, Hn(q)H_{n}(q) has a basis {Tw}\{T_{w}\} indexed by wSnw\in S_{n}. The quadratic relation shows that each TsjT_{s_{j}}, and hence each TwT_{w}, is invertible. The bar involution ()¯\overline{(\cdot)} of the Hecke algebra is the 𝐙\mathbf{Z}-linear extension of the maps qq1q\mapsto q^{-1} and TwTw11T_{w}\mapsto T_{w^{-1}}^{-1}.

Theorem 2.2 (Kazhdan-Lusztig Basis, [5, Theorem 1.1]).

There is a unique basis {Cw(q)}\{C_{w}(q)\} of Hn(q)H_{n}(q) indexed by SnS_{n}, with elements of the form

Cw(q)=ql(w)/2vSn(1)l(v)l(w)ql(v)Pv,w(q)Tv,C_{w}(q)=q^{-l(w)/2}\sum_{v\in S_{n}}(-1)^{l(v)-l(w)}q^{-l(v)}P_{v,w}(q)T_{v},

such that:

  • Pv,w(q)P_{v,w}(q) is an integer polynomial, and is non-zero only if vwv\leq w,

  • Pv,v(q)=1P_{v,v}(q)=1 for all vSnv\in S_{n},

  • if v<wv<w then deg(Pv,w(q))(1/2)(l(w)l(v)1)deg(P_{v,w}(q))\leq(1/2)(l(w)-l(v)-1), and

  • Cw(q)¯=Cw(q)\overline{C_{w}(q)}=C_{w}(q) for all wSnw\in S_{n}.

The Pv,w(q)P_{v,w}(q) are the well-known Kazhdan-Lusztig (KL) polynomials. The degree bound gives us a statistic on pairs of permutations. For vwv\leq w, we write μ¯(v,w)𝐙\overline{\mu}(v,w)\in\mathbf{Z} for the coefficient of degree (1/2)(l(w)l(v)1)(1/2)(l(w)-l(v)-1) in Pv,w(q)P_{v,w}(q). If l(w)l(v)l(w)-l(v) is even or vv is incomparable with ww in the Bruhat order, this coefficient will be zero. We also define μ¯(w,v):=μ¯(v,w)\overline{\mu}(w,v):=\overline{\mu}(v,w) when w<vw<v, so μ¯\overline{\mu} is symmetric.

Under the specialisation q=1q=1, we have the relation Tsi2=1T_{s_{i}}^{2}=1, and so Hn(1)𝐙SnH_{n}(1)\cong\mathbf{Z}S_{n}. Set Cw=Cw(1)1𝐙Sn𝐙𝐂=𝐂SnC_{w}=C_{w}(1)\otimes 1\in\mathbf{Z}S_{n}\otimes_{\mathbf{Z}}\mathbf{C}=\mathbf{C}S_{n}, so that {CwwSn}\{C_{w}\mid w\in S_{n}\} is a basis for 𝐂Sn\mathbf{C}S_{n}. Define a binary relation on SnS_{n} by setting vLwv\leq_{L}w if, for some jj, CvC_{v} appears with nonzero coefficient in the expansion of sjCws_{j}\cdot C_{w}. Taking the transitive closure gives a preorder on SnS_{n}, called the left KL preorder. By its construction, the span of {CvvLw}\{C_{v}\mid v\leq_{L}w\} for a fixed wSnw\in S_{n} will be a submodule of 𝐂Sn\mathbf{C}S_{n}. The equivalence classes induced by L\leq_{L} are the left cells of SnS_{n}, and we write vLwv\sim_{L}w if vv and ww belong to the same left cell.

For a left cell 𝒞\mathcal{C} of SnS_{n}, we write S𝒞S^{\mathcal{C}} for the linear span of elements {Cvv𝒞}\{C_{v}\mid v\in\mathcal{C}\}. The following defines an irreducible action of SnS_{n} on S𝒞S^{\mathcal{C}} [2, (6.4)]:

sjCv={CvjD(v),Cv+w𝒞,jD(w)μ¯(v,w)CwjD(v).\displaystyle s_{j}\cdot C_{v}=\begin{cases}-C_{v}&j\in D(v),\\ C_{v}+\sum_{w\in\mathcal{C},j\in D(w)}\overline{\mu}(v,w)C_{w}&j\not\in D(v).\end{cases} (2.3)

Left cells in SnS_{n} are determined by the QQ-tableau in the RSK correspondence. That is, vLwv\sim_{L}w if and only if v(P,Q)v\rightsquigarrow(P,Q) and w(P,Q)w\rightsquigarrow(P^{\prime},Q) for some P,P,QSYT(λ)P,P^{\prime},Q\in SYT(\lambda) and λn\lambda\vdash n. Therefore we can uniquely associate each left cell 𝒞\mathcal{C} of SnS_{n} to a tableau QQ with nn boxes, and write 𝒞Q\mathcal{C}_{Q}. Appropriately restricting the RSK correspondence gives a bijection between the basis elements Cv𝒞QC_{v}\in\mathcal{C}_{Q} and TSYT(λ)T\in SYT(\lambda), by identifying CvC_{v} with the PP-tableau of vv. Hence, we can reindex the basis of S𝒞QS^{\mathcal{C}_{Q}} as {CTTSYT(λ)}\{C_{T}\mid T\in SYT(\lambda)\}. We can rewrite the action on the KL basis {CT}\{C_{T}\} in terms of tableaux only:

  1. 1.

    Let vSnv\in S_{n} and suppose v(P,Q)v\rightsquigarrow(P,Q). Then jD(v)j\in D(v) if and only if jD(P)j\in D(P).

  2. 2.

    For λn\lambda\vdash n and tableaux P,T,Q,QSYT(λ)P,T,Q,Q^{\prime}\in SYT(\lambda) we have:

    μ¯((P,Q),(T,Q))=μ¯((P,Q),(T,Q)),\overline{\mu}((P,Q),(T,Q))=\overline{\mu}((P,Q^{\prime}),(T,Q^{\prime})),

    where we identify permutations with their images under the RSK correspondence.

From this we obtain that for tableaux Q,QSYT(λ)Q,Q^{\prime}\in SYT(\lambda), the respective representations S𝒞QS^{\mathcal{C}_{Q}} and S𝒞QS^{\mathcal{C}_{Q^{\prime}}} are equal (not just isomorphic) up to a reindexing of basis elements.

These results have a number of implications. First, we can define the μ¯\overline{\mu} value between tableaux P,TSYT(λ)P,T\in SYT(\lambda) by μ¯(P,T)=μ¯((P,Q),(T,Q))\overline{\mu}(P,T)=\overline{\mu}((P,Q),(T,Q)) for any QSYT(λ)Q\in SYT(\lambda). Furthermore, since S𝒞QS^{\mathcal{C}_{Q}} depends only on the shape of QQ, we can write SλS^{\lambda} for the representation S𝒞QS^{\mathcal{C}_{Q}}, where QQ is any tableau of shape λ\lambda. Finally, we can write the conditions on the descent set of vv in (2.3) in terms of the PP-tableau of vv under RSK. We incorporate these results into the following theorem.

Theorem 2.4 ([2, Theorem 6.5.3]).

Let λn\lambda\vdash n be a partition. The module SλS^{\lambda} is irreducible, and has a basis {CTTSYT(λ)}\{C_{T}\mid T\in SYT(\lambda)\} with the following action:

sjCT={CTjD(T),CT+RSYT(λ),jD(R)μ¯(T,R)CRjD(T).\displaystyle s_{j}\cdot C_{T}=\begin{cases}-C_{T}&j\in D(T),\\ C_{T}+\sum_{R\in SYT(\lambda),j\in D(R)}\overline{\mu}(T,R)C_{R}&j\not\in D(T).\end{cases} (2.5)

The irreducible representations of SnS_{n} are called Specht modules. The above theorem provides a canonical basis for Specht modules, which we refer to as the KL basis.

3 The Branching Rule via the KL Basis

In this section we prove Theorem 1.2. Fix λn\lambda\vdash n, and recall the filtration of SλS^{\lambda} by the subrepresentations SiλS_{i}^{\lambda}, each defined as the subspace spanned by {CT}\{C_{T}\} with idx(T)i\operatorname{idx}(T)\leq i. We will first show that SiλS_{i}^{\lambda} always remains invariant under Sn1S_{n-1}.

Let CSS(λ)CSS(\lambda) be the column super-strict tableau of shape λ\lambda, where the values 1,2,,n1,2,\dots,n are placed column-by-column, reading left-to-right. We take λ=(4,3,1)\lambda=(4,3,1) as an example:

CSS(λ)=\young(1468,257,3)CSS(\lambda)=\young(1468,257,3)

Preimages of the RSK correspondence when Q=CSS(λ)Q=CSS(\lambda) are easy to describe. We simply write the elements of the PP-tableau in column order, reading from bottom-to-top.

85162734(\young(1234,567,8),\young(1468,257,3))85162734\quad\rightsquigarrow\quad\left(\,\young(1234,567,8),\,\young(1468,257,3)\,\right)
Lemma 3.1.

Fix λn\lambda\vdash n and 1ir1\leq i\leq r, where rr is the number of removable boxes of λ\lambda. Then SiλS_{i}^{\lambda} is an Sn1S_{n-1}-submodule of SλS^{\lambda}.

Proof.

If i=ri=r, there is nothing to prove. Otherwise, choose arbitrary tableau T,RT,R in SYT(λ)SYT(\lambda) with CTSiλC_{T}\in S_{i}^{\lambda} and CRSiλC_{R}\not\in S_{i}^{\lambda}, or equivalently, idx(T)i<idx(R)\operatorname{idx}(T)\leq i<\operatorname{idx}(R). Instead of working in SλS^{\lambda}, we work in the module S𝒞CSS(λ)S^{\mathcal{C}_{CSS(\lambda)}}. To this end, define the unique permutations v,wSnv,w\in S_{n} with images (T,CSS(λ))(T,CSS(\lambda)) and (R,CSS(λ))(R,CSS(\lambda)) under RSK. We are required to show that CwC_{w} does not appear in sjCvs_{j}\cdot C_{v} (see (2.5) for the action) when 1jn21\leq j\leq n-2. We consider cases depending on the Bruhat comparibility of vv and ww.

(a) vv and ww are not Bruhat compatible. Then μ¯(v,w)=0\overline{\mu}(v,w)=0 by definition.

(b) w<vw<v. We have a chain w=w1wk=vw=w_{1}\to\cdots\to w_{k}=v of swaps with l(wm)<l(wm+1)l(w_{m})<l(w_{m+1}). At some point, the swap wmwm+1w_{m}\to w_{m+1} moves nn to the right. But this will decrease the length, which is a contradiction.

(c) v<wv<w and l(v,w)>1l(v,w)>1. Suppose CwC_{w} appears in the expansion of sjCvs_{j}\cdot C_{v}. Then jD(w)j\in D(w) and jD(v)j\not\in D(v), so μ¯(v,w)=0\overline{\mu}(v,w)=0 by [2, Proposition 5.1.9].111Björner [2] uses the convention of right descent sets, but the result for left descent sets is analogous.

(d) v<wv<w and l(v,w)=1l(v,w)=1. Again, suppose CwC_{w} appears in the expansion of sjCvs_{j}\cdot C_{v} for some sjSn1s_{j}\in S_{n-1}. Then jD(w)j\in D(w), jD(v)j\not\in D(v) and l(v,w)=1l(v,w)=1, which implies w=sjvw=s_{j}v. But nn occurs at different positions in the one-line notation of vv and ww, so we must have j=n1j=n-1. This is again a contradiction, since sn1Sn1s_{n-1}\not\in S_{n-1}. ∎

Recall that μi\mu_{i} is the partition obtained by deleting the ithi^{\text{th}} removable box from λn\lambda\vdash n. There is a bijection di:{TSYT(λ)idx(T)=i}SYT(μi)d_{i}:\{T\in SYT(\lambda)\mid\operatorname{idx}(T)=i\}\to SYT(\mu_{i}) which removes the largest box of each tableau TT. By taking its linear extension, we have an isomorphism of vector spaces di:Siλ/Si1λSμid_{i}:S_{i}^{\lambda}/S_{i-1}^{\lambda}\to S^{\mu_{i}} for each 1in11\leq i\leq n-1. Our goal is to show that this is in fact an isomorphism of Sn1S_{n-1}-modules. For this, we need to show that the deletion map preserves the function μ¯\overline{\mu}.

We again define a specific recording tableau and work with the permutations explicitly. For a partition λn\lambda\vdash n with rr removable boxes and 1ir1\leq i\leq r, define the tableau CSSi(λ)CSS_{i}(\lambda) by the following procedure:

  • Create a tableau of shape λ\lambda, and place the value nn in removable box ii.

  • Suppose nn is placed in the kthk^{\text{th}} row. For 1mk11\leq m\leq k-1, place the value nmn-m at the end of row kmk-m.

  • Place the remaining values 1,,nk1,\dots,n-k into the tableau column by column.

For example, if λ=(4,3,1)\lambda=(4,3,1) and i=2i=2 we have:

CSSi(λ)=\young(1467,258,3)CSS_{i}(\lambda)=\young(1467,258,3)

The proof that the μ¯\overline{\mu} function is preserved under the deletion map was originally given by Rhoades in the case of rectangular partitions. The following Lemma appears implicitly in the proof:

Lemma 3.2 ([8], Lemma 3.1).

Suppose u=u1un1u=u_{1}\cdots u_{n-1} and v=v1vn1v=v_{1}\cdots v_{n-1} are elements of Sn1S_{n-1}, written in their one-line notation. For a fixed 0kn10\leq k\leq n-1, define the permutations

u\displaystyle u^{\prime} =u1uknuk+1un1,\displaystyle=u_{1}\cdots u_{k}nu_{k+1}\cdots u_{n-1},
v\displaystyle v^{\prime} =v1vknvk+1vn1\displaystyle=v_{1}\cdots v_{k}nv_{k+1}\cdots v_{n-1}

in SnS_{n}. Then we have an equality of μ¯\overline{\mu} values μ¯(u,v)=μ¯(u,v)\overline{\mu}(u,v)=\overline{\mu}(u^{\prime},v^{\prime}).

Proposition 3.3.

Fix λn\lambda\vdash n, and choose tableaux T,RSYT(λ)T,R\in SYT(\lambda), both with index ii. Then

μ¯(T,R)=μ¯(di(T),di(R))\overline{\mu}(T,R)=\overline{\mu}(d_{i}(T),d_{i}(R))
Proof.

Suppose the removable box ii of λ\lambda occurs in row kk. Define the following permutations by their images under the RSK correspondence, using the tableau CSSi(λ)CSS_{i}(\lambda).

u(T,CSSi(λ))u~(di(T),di(CSSi(λ)))v(R,CSSi(λ))v~(di(R),di(CSSi(λ)))\begin{matrix}u\rightsquigarrow(T,CSS_{i}(\lambda))&\quad&\widetilde{u}\rightsquigarrow(d_{i}(T),d_{i}(CSS_{i}(\lambda)))\\ v\rightsquigarrow(R,CSS_{i}(\lambda))&\quad&\widetilde{v}\rightsquigarrow(d_{i}(R),d_{i}(CSS_{i}(\lambda)))\end{matrix}

By the above Lemma, it remains to show that u=u~1u~nknu~nk+1u~n1u=\widetilde{u}_{1}\cdots\widetilde{u}_{n-k}n\widetilde{u}_{n-k+1}\cdots\widetilde{u}_{n-1} and the result for vv follows mutatis mutandis.

The box containing the value nk+1n-k+1 in CSSi(λ)CSS_{i}(\lambda) will be in the top row. Hence, u~nk+1\widetilde{u}_{n-k+1} is inserted directly into the top row when performing RSK on u~\widetilde{u}. In uu, we instead insert nn, which will also insert into the top row. Inserting u~nk+1\widetilde{u}_{n-k+1} will then push nn into the next row. Each following insertion for uu mimics u~\widetilde{u}, except the nn box is constantly pushed down the right-hand side of the insertion tableau. One can verify that this gives the correct tableaux-pair for uu, with an additional nn-box in row kk for each. ∎

For a tableau TSYT(λ)T\in SYT(\lambda) with index ii, write [CT][C_{T}] for the basis element CT+Si1λC_{T}+S_{i-1}^{\lambda} of Siλ/Si1λS_{i}^{\lambda}/S_{i-1}^{\lambda}, where 1in11\leq i\leq n-1 is the unique value such that CTSiλC_{T}\in S_{i}^{\lambda}, but CTSi1λC_{T}\notin S_{i-1}^{\lambda}.

By (2.5), for a tableau TT with index ii we have:

sj[CT]={[CT]jD(T)[CT]+Rμ¯(T,R)[CR]jD(T)s_{j}\cdot[C_{T}]=\begin{cases}-[C_{T}]&j\in D(T)\\ [C_{T}]+\sum_{R}\overline{\mu}(T,R)[C_{R}]&j\not\in D(T)\end{cases}

In the second case, we sum over RSYT(λ)R\in SYT(\lambda) with jD(R)j\in D(R) and idx(R)=idx(T)\operatorname{idx}(R)=\operatorname{idx}(T). Then:

disj[CT]={Cdi(T)jD(T)Cdi(T)+Rμ¯(T,R)Cdi(R)jD(T)d_{i}s_{j}\cdot[C_{T}]=\begin{cases}-C_{d_{i}(T)}&j\in D(T)\\ C_{d_{i}(T)}+\sum_{R}\overline{\mu}(T,R)C_{d_{i}(R)}&j\not\in D(T)\end{cases}

with the sum in the second case as before. If 1jn21\leq j\leq n-2, then jD(T)j\in D(T) if and only if jD(di(T))j\in D(d_{i}(T)), and likewise for RR. Hence:

disj[CT]={Cdi(T)jD(di(T))Cdi(T)+Rμ¯(di(T),di(R))Cdi(R)jD(di(T))d_{i}s_{j}\cdot[C_{T}]=\begin{cases}-C_{d_{i}(T)}&j\in D(d_{i}(T))\\ C_{d_{i}(T)}+\sum_{R}\overline{\mu}(d_{i}(T),d_{i}(R))C_{d_{i}(R)}&j\not\in D(d_{i}(T))\end{cases}

This is the action of sjs_{j} on Cdi(T)C_{d_{i}(T)} in SμiS^{\mu_{i}}. Thus, sjs_{j} and did_{i} commute on every basis element [CT][C_{T}] of Siλ/Si1λS_{i}^{\lambda}/S_{i-1}^{\lambda} for every generator sjs_{j} of Sn1S_{n-1}. This completes the proof of Theorem 1.2.

4 The long cycle action on the KL basis

In this section we prove Theorem 1.1. Let wnSnw_{n}\in S_{n} be the long element, and wn1w_{n-1} be the image of the long element of Sn1S_{n-1} under the standard embedding into SnS_{n}. Our goal is to relate the identity c=wnwn1c=w_{n}w_{n-1} to pr=evnevn1\mathrm{pr}=\mathrm{ev}_{n}\mathrm{ev}_{n-1} from Lemma 2.1.

Note that wn1Sn1w_{n-1}\in S_{n-1} will act on a tableau of size n1n-1 by evn1\mathrm{ev}_{n-1} (up to sign) [10]. Moreover, did_{i} commutes with wn1w_{n-1} on Siλ/Si1λS_{i}^{\lambda}/S_{i-1}^{\lambda}. Therefore

wn1[CT]\displaystyle w_{n-1}\cdot[C_{T}] =di1wn1di[CT]\displaystyle=d_{i}^{-1}w_{n-1}d_{i}\cdot[C_{T}]
=di1wn1Cdi(T)\displaystyle=d_{i}^{-1}w_{n-1}C_{d_{i}(T)}
=±di1Cevn1di(T)=±[Cevn1(T)].\displaystyle=\pm d_{i}^{-1}C_{\mathrm{ev}_{n-1}d_{i}(T)}=\pm[C_{\mathrm{ev}_{n-1}(T)}].

Note that the sign depends only on the shape of di(T)d_{i}(T). Since the index of TT matches the index of evn1(T)\mathrm{ev}_{n-1}(T), we obtain:

Lemma 4.1.

Fix a partition λn\lambda\vdash n. The action of wn1w_{n-1} on the KL basis {CT}\{C_{T}\} of SλS^{\lambda} is

wn1CT=±Cevn1(T)+RaRCR,w_{n-1}\cdot C_{T}=\pm C_{\mathrm{ev}_{n-1}(T)}+\sum_{R}a_{R}C_{R},

with the sum over RSYT(λ)R\in SYT(\lambda) satisfying idx(R)<idx(T)\operatorname{idx}(R)<\operatorname{idx}(T), and aR𝐙a_{R}\in\mathbf{Z} are unknown coefficients. Furthermore, the sign of Cevn1(T)C_{\mathrm{ev}_{n-1}(T)} depends only on idx(T)\operatorname{idx}(T).

Theorem 1.1 follows immediately from:

Theorem 4.2.

Fix a partition λn\lambda\vdash n. The action of cc on the KL basis {CT}\{C_{T}\} of SλS^{\lambda} is

cCT=±Cpr(T)+RbRCpr(R),c\cdot C_{T}=\pm C_{\mathrm{pr}(T)}+\sum_{R}b_{R}C_{\mathrm{pr}(R)},

with the sum over RSYT(λ)R\in SYT(\lambda) satisfying idx(R)<idx(T)\operatorname{idx}(R)<\operatorname{idx}(T), and bR𝐙b_{R}\in\mathbf{Z} are unknown coefficients. Furthermore, the sign of Cpr(T)C_{\mathrm{pr}(T)} depends only on idx(T)\operatorname{idx}(T).

Proof.

Reindexing the sum from the Lemma with Revn1(R)R\mapsto{\mathrm{ev}_{n-1}(R)},

wn1CT=±Cevn1(T)+Raevn1(R)Cevn1(R).w_{n-1}\cdot C_{T}=\pm C_{\mathrm{ev}_{n-1}(T)}+\sum_{R}a_{\mathrm{ev}_{n-1}(R)}C_{\mathrm{ev}_{n-1}(R)}.

We set bR:=aevn1(R)b_{R}:=a_{\mathrm{ev}_{n-1}(R)} and apply the long element to both sides:

wnwn1CT=±Cevnevn1+R±bRCevnevn1(R).w_{n}w_{n-1}\cdot C_{T}=\pm C_{\mathrm{ev}_{n}\mathrm{ev}_{n-1}}+\sum_{R}\pm b_{R}C_{\mathrm{ev}_{n}\mathrm{ev}_{n-1}(R)}.

Replacing wnwn1w_{n}w_{n-1} with cc and evnevn1\mathrm{ev}_{n}\mathrm{ev}_{n-1} with pr\mathrm{pr} gives the result required. ∎

Example 4.3.

Fix λ=(3,1,1)\lambda=(3,1,1). We arrange SYT(λ)SYT(\lambda) using the total index ordering:

\young(145,2,3)\young(135,2,4)\young(125,3,4)\young(134,2,5)\young(124,3,5)\young(123,4,5)\displaystyle\young(145,2,3)\prec\;\young(135,2,4)\prec\;\young(125,3,4)\prec\;\young(134,2,5)\prec\;\young(124,3,5)\prec\;\young(123,4,5)

Using MAGMA [3] we compute [c][c] with respect to this ordered basis and calculate its QR decomposition:

[c]=(000100000010100110000001010101001011)=(000100000010100000000001010000001000)(100110010101001011000100000010000001)[c]=\text{\footnotesize$\begin{pmatrix}0&0&0&1&0&0\\ 0&0&0&0&1&0\\ 1&0&0&-1&1&0\\ 0&0&0&0&0&1\\ 0&1&0&-1&0&1\\ 0&0&1&0&-1&1\end{pmatrix}$}=\text{\footnotesize$\begin{pmatrix}0&0&0&1&0&0\\ 0&0&0&0&1&0\\ 1&0&0&0&0&0\\ 0&0&0&0&0&1\\ 0&1&0&0&0&0\\ 0&0&1&0&0&0\end{pmatrix}\begin{pmatrix}1&0&0&-1&1&0\\ 0&1&0&-1&0&1\\ 0&0&1&0&-1&1\\ 0&0&0&1&0&0\\ 0&0&0&0&1&0\\ 0&0&0&0&0&1\end{pmatrix}$}

One can verify that QQ is precisely the permutation matrix [pr][\mathrm{pr}].

Corollary 4.4.

Let TSYT(λ)T\in SYT(\lambda) and suppose TT has index 11. Then cCT=±Cpr(T)c\cdot C_{T}=\pm C_{\mathrm{pr}(T)}, with the sign depending only on λ\lambda. In particular, when λ\lambda is a rectangular partition, then cc acts by the promotion operator on the KL basis of SλS^{\lambda} up to sign.

5 Other permutations acting on the KL basis

In this section we discuss Theorem 1.3. Details will appear in forthcoming work.

Fix a partition λn\lambda\vdash n. Let II denote the set of generators {s1,,sn1}\{s_{1},\dots,s_{n-1}\} of SnS_{n}. For a non-empty subset JIJ\subseteq I, let SJSnS_{J}\leq S_{n} denote the associated parabolic subgroup, and wJw_{J} its longest element. To each such subset, we can associate a filtration 0=S0λS1λSpλ=Sλ0=S_{0}^{\lambda}\subset S_{1}^{\lambda}\subset\cdots\subset S_{p}^{\lambda}=S^{\lambda} of the Specht module such that each subspace SjλS_{j}^{\lambda} is SJS_{J}-invariant and spanned by KL basis elements, and the quotient representation Sjλ/Sj1λS_{j}^{\lambda}/S_{j-1}^{\lambda} is isomorphic to a simple SJS_{J}-module.

This filtration induces a total preorder J\prec_{J} on SYT(λ)SYT(\lambda), with TJRT\prec_{J}R when CTC_{T} appears strictly before CRC_{R} in this filtration. Using partial evacuation operators, we also associate a bijection φJ\varphi_{J} on SYT(λ)SYT(\lambda) such that the action of wJw_{J} on the KL basis is given by

wJCT=±CφJ(T)+RJTaRCφJ(R)w_{J}\cdot C_{T}=\pm C_{\varphi_{J}(T)}+\sum_{R\prec_{J}T}a_{R}C_{\varphi_{J}(R)}

for some constants aR𝐙a_{R}\in\mathbf{Z}, and a sign which depends only on the equivalence class of TT under J\prec_{J}.

We can extend this result in a natural way to chains of subsets of JJ. Call a permutation descending if it is of the form wJ=wJkwJ1w_{J_{\bullet}}=w_{J_{k}}\cdots w_{J_{1}} for some increasing chain JJ_{\bullet} of subsets J1JkJJ_{1}\subset\cdots\subset J_{k}\subseteq J. It turns out that a permutation is descending if and only if it is separable.222This observation was communicated to the authors by Joel Gibson, who discovered this fact by enumerating the descending permutations computationally in MAGMA. Each descending permutation has an associated bijection φJ=φJkφJ1\varphi_{J_{\bullet}}=\varphi_{J_{k}}\cdots\varphi_{J_{1}} and a total preorder J\prec_{J_{\bullet}} on SYT(λ)SYT(\lambda). The statement is as before, with the action of wJw_{J_{\bullet}} on the KL basis of SλS^{\lambda} given by

wJCT=±CφJ(T)+RJTbRCφJ(R)w_{J_{\bullet}}\cdot C_{T}=\pm C_{\varphi_{J_{\bullet}}(T)}+\sum_{R\prec_{J_{\bullet}}T}b_{R}C_{\varphi_{J_{\bullet}}(R)} (5.1)

for some constants bR𝐙b_{R}\in\mathbf{Z}, and a sign which depends only on the equivalence class of TT under J\prec_{J_{\bullet}}. Theorem 1.3 follows from this formula.

When JIJ\subseteq I is a connected subset {sa+1,,sb}\{s_{a+1},\dots,s_{b}\} for 0ab<n0\leq a\leq b<n, then SJS_{J} is isomorphic to the symmetric group SbaS_{b-a}. Moreover, wJ=wb+1wbawb+1w_{J}=w_{b+1}w_{b-a}w_{b+1}, where wkSnw_{k}\in S_{n} is the permutation reversing {1,,k}\{1,\dots,k\}. Likewise, φJ=evb+1evbaevb+1\varphi_{J}=\mathrm{ev}_{b+1}\mathrm{ev}_{b-a}\mathrm{ev}_{b+1}, where evk\mathrm{ev}_{k} is the partial Schützenberger involution from Section 2.1. Finally, the preorder Ji\prec_{J_{i}} can be defined explicitly by: RJTR\prec_{J}T if and only if evaevn(R)\mathrm{ev}_{a}\mathrm{ev}_{n}(R) precedes evaevn(T)\mathrm{ev}_{a}\mathrm{ev}_{n}(T) in the total index ordering.

In the case of chains of connected subsets of JJ, equation (5.1) can be proved combinatorially. However, in the case of general chains, we use results from categorical representation theory, which will appear in later work by the authors. This will also prove analogues of these results in other types concerning the action of an ADE Weyl group on the Lusztig’s dual canonical basis in the zero weight space of an irreducible representation of the corresponding Lie algebra.

Finally, we note that the result in (5.1) does not hold for arbitrary permutations. For example, take the non-separable permutation w=2413S4w=2413\in S_{4} and λ=(3,1)\lambda=(3,1). Choose any of the 3!3! orderings of the KL basis for SλS^{\lambda}, and compute the QRQR decomposition of [w]:SλSλ[w]:S^{\lambda}\to S^{\lambda} with respect to this basis. One can verify that QQ is not a generalised permutation matrix under any of these orderings.

Acknowledgements.
The results in Sections 1-4 are part of the first author’s University of Sydney Honours Thesis (2021), supervised by the second author. The authors thank the anonymous referees for helpful comments, and Joel Gibson, Ed Heng, Tony Licata and Eloise Little for useful conversations that improved this work.

References

  • [1] A. Berenstein and A. Zelevinsky “Canonical bases for the quantum group of type ArA_{r} and piecewise-linear combinatorics” In Duke Math. J. 82.3, 1996, pp. 473–502 DOI: 10.1215/S0012-7094-96-08221-6
  • [2] A. Bjorner and F. Brenti “Combinatorics of Coxeter Groups”, Graduate Texts in Mathematics Springer Berlin Heidelberg, 2006 URL: https://books.google.com.au/books?id=1TBPz5sd8m0C
  • [3] W. Bosma, J. Cannon and C. Playoust “The Magma algebra system. I. The user language” Computational algebra and number theory (London, 1993) In J. Symbolic Comput. 24.3-4, 1997, pp. 235–265 DOI: 10.1006/jsco.1996.0125
  • [4] I. Halacheva, T. Licata, I. Losev and O. Yacobi “Categorical braid group actions and cactus groups”, 2021 arXiv:2101.05931
  • [5] D. Kazhdan and G. Lusztig “Representations of Coxeter groups and Hecke algebras” In Invent. Math. 53.2, 1979, pp. 165–184 DOI: 10.1007/BF01390031
  • [6] S. Kitaev “Patterns in permutations and words” With a foreword by Jeffrey B. Remmel, Monographs in Theoretical Computer Science. An EATCS Series Springer, Heidelberg, 2011, pp. xxii+494 DOI: 10.1007/978-3-642-17333-2
  • [7] D. E. Knuth “Permutations, matrices, and generalized Young tableaux” In Pacific J. Math. 34, 1970, pp. 709–727 URL: http://projecteuclid.org.ezproxy.library.sydney.edu.au/euclid.pjm/1102971948
  • [8] B. Rhoades “Cyclic sieving, promotion, and representation theory” In J. Combin. Theory Ser. A 117.1, 2010, pp. 38–76 DOI: 10.1016/j.jcta.2009.03.017
  • [9] B. E. Sagan “The cyclic sieving phenomenon: a survey” In Surveys in combinatorics 2011 392, London Math. Soc. Lecture Note Ser. Cambridge Univ. Press, Cambridge, 2011, pp. 183–233
  • [10] J. R. Stembridge “Canonical bases and self-evacuating tableaux” In Duke Math. J. 82.3, 1996, pp. 585–606 DOI: 10.1215/S0012-7094-96-08224-1