Abstract
An explicit rule is given for the product of the degree two class with an arbitrary Schubert class in the torus-equivariant homology of the affine Grassmannian. In addition a Pieri rule (the Schubert expansion of the product of a special Schubert class with an arbitrary one) is established for the equivariant homology of the affine Grassmannians of SL n and a similar formula is conjectured for Sp 2n and SO 2n+1. For SL n the formula is explicit and positive. By a theorem of Peterson these compute certain products of Schubert classes in the torus-equivariant quantum cohomology of flag varieties. The SL n Pieri rule is used in our recent definition of k-double Schur functions and affine double Schur functions.
Similar content being viewed by others
1 Introduction
Let G be a semisimple algebraic group over ℂ with a Borel subgroup B and maximal torus T. Let Gr G =G(ℂ((t)))/G(ℂ[[t]]) be the affine Grassmannian of G. The T-equivariant homology H T (Gr G ) and cohomology H T(Gr G ) are dual Hopf algebras over S=H T(pt) with Pontryagin and cup products, respectively. Let \(W_{\mathrm{af}}^{0}\) be the minimal length cosets in W af/W where W af and W are the affine and finite Weyl groups. Let \(\{\xi_{w}\mid w\in W_{\mathrm{af}}^{0}\}\) be the Schubert basis of H T (Gr G ). Define the equivariant Schubert homology structure constants \(d_{uv}^{w}\in S\) by
where \(u,v\in W_{\mathrm{af}}^{0}\). One interest in the polynomials \(d^{w}_{uv}\) is the fact that they are precisely the Schubert structure constants for the T-equivariant quantum cohomology rings QH T(G/B) [9, 13]. Due to a result of Mihalcea [12], they have the positivity property
Our first main result (Theorem 6) is an “equivariant homology Chevalley formula”, which describes \(d_{r_{0},v}^{w}\) for an arbitrary affine Grassmannian. Our second main result (Theorem 20) is an “equivariant homology Pieri formula” for G=SL n , which is a manifestly positive formula for \(d_{\sigma_{m},v}^{w}\) where the homology classes \(\{\xi_{\sigma_{m}}\mid1\le m\le n-1\}\) are the special classes that generate \(H_{T}(\mathrm{Gr}_{SL_{n}})\). In a separate work [10] we use this Pieri formula to define new symmetric functions, called k-double Schur functions and affine double Schur functions, which represent the equivariant Schubert homology and cohomology classes for \(\mathrm{Gr}_{SL_{n}}\).
2 The equivariant homology of Gr G
We recall Peterson’s construction [13] of the equivariant Schubert basis \(\{j_{w}\mid w\in W_{\mathrm{af}}^{0}\}\) of H T (Gr G ) using the level-zero variant of the Kostant and Kumar (graded) nilHecke ring [6]. We also describe the equivariant localizations of Schubert cohomology classes for the affine flag ind-scheme in terms of the nilHecke ring; these are an important ingredient in our equivariant Chevalley and Pieri rules.
2.1 Peterson’s level-zero affine nilHecke ring
Let I and I af=I∪{0} be the finite and affine Dynkin node sets and (a ij ∣i,j∈I af) the affine Cartan matrix.
Let \(P_{\mathrm{af}}= \mathbb{Z}\delta\oplus\bigoplus_{i\in I_{\mathrm{af}}} \mathbb{Z}\varLambda_{i}\) be the affine weight lattice, with δ the null root and Λ i the affine fundamental weight. The dual lattice \(P_{\mathrm{af}}^{*}=\mathrm{Hom}_{\mathbb{Z}}(P_{\mathrm{af}},\mathbb{Z})\) has dual basis \(\{d\} \cup\{\alpha_{i}^{\vee}\mid i\in I_{\mathrm{af}}\}\) where d is the degree generator and \(\alpha_{i}^{\vee}\) is a simple coroot. The simple roots {α i ∣i∈I af}⊂P af are defined by \(\alpha_{j} = \delta_{j0} \delta+ \sum_{i\in I_{\mathrm{af}}} a_{ij}\varLambda_{i}\) for j∈I af where (a ij ∣i,j∈I af) is the affine Cartan matrix. Then \(a_{ij}=\langle\alpha_{i}^{\vee},\alpha_{j}\rangle\) for all i,j∈I af. Let (a i ∣i∈I af) (resp. \((a_{i}^{\vee}\mid i\in I_{\mathrm{af}})\)) be the tuple of relatively prime positive integers giving a relation among the columns (resp. rows) of the affine Cartan matrix. Then \(\delta= \sum_{i\in I_{\mathrm {af}}} a_{i}\alpha_{i}\). Let \(c=\sum_{i\in I_{\mathrm{af}}} a_{i}^{\vee}\alpha_{i}^{\vee}\in P_{\mathrm{af}}^{*}\) be the canonical central element. The level of a weight λ∈P af is defined by 〈c,λ〉.
There is a canonical projection P af→P where P is the finite weight lattice, with kernel ℤδ⊕ℤΛ 0. There is a section P→P af of this projection whose image lies in the sublattice of \(\bigoplus_{i\in I_{\mathrm{af}}} \mathbb {Z}\varLambda_{i}\) consisting of level-zero weights. We regard P⊂P af via this section.
Let W and W af denote the finite and affine Weyl groups. Denote by {r i ∣i∈I af} the simple generators of W af. W af acts on P af by \(r_{i} \cdot\lambda=\lambda- \langle\alpha_{i}^{\vee},\lambda \rangle \alpha_{i}\) for i∈I af and λ∈P af. W af acts on \(P_{\mathrm{af}}^{*}\) by \(r_{i}\cdot\mu= \mu- \langle\mu,\alpha_{i}\rangle \alpha_{i}^{\vee}\) for i∈I af and \(\mu\in P_{\mathrm{af}}^{*}\). There is an isomorphism W af≅W⋉Q ∨ where \(Q^{\vee}= \bigoplus_{i\in I} \mathbb{Z}\alpha_{i}^{\vee}\subset P_{\mathrm{af}}^{*}\) is the finite coroot lattice. The embedding Q ∨→W af is denoted μ↦t μ . The set of real affine roots is W af⋅{α i ∣i∈I af}. For a real affine root α=w⋅α i , the associated coroot is well-defined by \(\alpha^{\vee}= w\cdot\alpha_{i}^{\vee}\).
Let S=Sym(P) be the symmetric algebra, and Q=Frac(S) the fraction field. W af≅W⋉Q ∨ acts on P (and therefore on S and on Q) by the level-zero action:
Since \(t_{-\theta^{\vee}} = r_{\theta}r_{0}\) we have
Finally, we have δ=α 0+θ where θ∈P is the highest root. So under the projection P af→P, α 0↦−θ.
Let \(Q_{W_{\mathrm{af}}} = \bigoplus_{w\in W_{\mathrm{af}}} Q w\) be the skew group ring, the Q-vector space Q⊗ℚℚ[W af] with Q-basis W af and product (p⊗v)(q⊗w)=p(v⋅q)⊗vw for p,q∈Q and v,w∈W af. \(Q_{W_{\mathrm{af}}}\) acts on Q: q∈Q acts by left multiplication and W af acts as above.
For i∈I af define the element \(A_{i}\in Q_{W_{\mathrm {af}}}\) by
A i acts on S since
The A i satisfy \(A_{i}^{2}=0\) and
where
For w∈W af we define A w by
The level-zero graded affine nilHecke ring \(\mathbb{A}\) (Peterson’s [13] variant of the nilHecke ring of Kostant and Kumar [6] for an affine root system) is the subring of \(Q_{W_{\mathrm{af}}}\) generated by S and {A i ∣i∈I af}. In \(\mathbb{A}\) we have the commutation relation
In particular
2.2 Localizations of equivariant cohomology classes
Using the relation
w∈W af may be regarded as an element of \(\mathbb{A}\). For v,w∈W af define the elements ξ v(w)∈S by
Using a reduced decomposition (9) for w and substituting (12) for its simple reflections, one obtains the formula [1] [2]
where the sum runs over b such that \(\prod_{b_{j}=1} r_{i_{j}} = v\) is reduced and the product over j is an ordered left-to-right product of operators. Each b encodes a way to obtain a reduced word for v as an embedded subword of the given reduced word of w: if b j =1 then the reflection \(r_{i_{j}}\) is included in the reduced word for v. Given a fixed b and an index j such that b j =1, the root associated to the reflection \(r_{i_{j}}\) is by definition \(r_{i_{1}}r_{i_{2}}\dotsm r_{i_{j-1}} \cdot\alpha_{i_{j}}\). The summand for b is the product of the roots associated to reflections in the given embedded subword.
It is immediate that
The element ξ v(w)∈S has the following geometric interpretation. Let X af=G af/B af be the Kac–Moody flag ind-variety of affine type [7]. For every v∈W af there is a T-equivariant cohomology class [X v ]∈H T(X af) and for each w∈W af there is an associated T-fixed point (denoted w) in X af and a localization map \(i_{w}^{*}:H^{T}(X_{\mathrm{af}})\to H^{T}(w) \simeq H^{T}(\mathrm{pt})\) [7]. Then \(\xi^{v}(w) = i_{w}^{*}([X_{v}])\). Moreover, the map H T(X af)→H T(W af)≅Fun(W af,S) given by restriction of a class to the T-fixed subset W af⊂X af, is an injective S-algebra homomorphism where Fun(W af,S) is the S-algebra of functions W af→S with pointwise product. The function ξ v∈Fun(W af,S) is the image of [X v ]. The image Φ of H T(X af) in Fun(W af,S) satisfies the GKM condition [3] [6]: For f∈Φ we haveFootnote 1
Lemma 1
Suppose u,v∈W af with ℓ(uv)=ℓ(u)+ℓ(v). Then
Lemma 2
Suppose v,w∈W af. Then
2.3 Peterson subalgebra and Schubert homology basis
Let K⊂G denote the maximal compact subgroup of G. The homotopy equivalence between Gr G and the based loop space ΩK endows the equivariant homology H T (Gr G ) and cohomology H T(Gr G ) with the structure of dual Hopf algebras. The Pontryagin multiplication in the homology H T (Gr G ) is induced by the group structure of ΩK. We let {ξ w } denote the equivariant Schubert basis of H T (Gr G ), dual (via the cap product) to the basis {ξ w} of H T(Gr G ).
The Peterson subalgebra of \(\mathbb{A}\) is the centralizer subalgebra \(\mathbb{P}=Z_{\mathbb{A}}(S)\) of S in \(\mathbb{A}\).
Theorem 3
[13] There is an isomorphism H T (Gr G )→ℙ of commutative Hopf algebras over S. For \(w\in W_{\mathrm{af}}^{0}\) let j w denote the image of ξ w in ℙ. Then j w is the unique element of ℙ with the property that \(j_{w}^{w}=1\) and \(j_{w}^{x}=0\) for any \(x\in W_{\mathrm{af}}^{0} \setminus\{w\}\) where \(j_{w}^{x}\in S\) are defined by
Moreover, if \(j^{x}_{w}\ne0\) then ℓ(x)≥ℓ(w) and \(j^{x}_{w}\) is a polynomial of degree ℓ(x)−ℓ(w).
The Schubert structure constants for H T (Gr G ) are obtained as coefficients of the elements j w .
Proposition 4
([13]) Let \(u,v,w\in W_{\mathrm {af}}^{0}\). Then
Due to the fact [9, 13] that the collections of Schubert structure constants for H T (Gr G ) and QH T(G/B) are the same and Mihalcea’s positivity theorem for equivariant quantum Schubert structure constants, we have the positivity property
Proposition 5
\(j_{w}^{x} \in\mathbb{Z}_{\ge0}[\alpha_{i}\mid i\in I]\) for all \(w\in W_{\mathrm{af}}^{0}\) and x∈W af.
Given \(u\in W_{\mathrm{af}}^{0}\) let t u=t λ where λ∈Q ∨ is such that t λ W=uW.
Since the translation elements act trivially on S and \(W_{\mathrm{af}}\subset\mathbb{A}\) via (12), we have t λ ∈ℙ for all λ∈Q ∨, so that \(t_{\lambda}\in \bigoplus_{v\in W_{\mathrm{af}}^{0}} S j_{v}\). For any \(w\in W_{\mathrm{af}}^{0}\), we have
by the definitions and Lemma 1.
Define the \(W_{\mathrm{af}}^{0} \times W_{\mathrm{af}}^{0}\)-matrices
The matrix A is lower triangular by (15) and has nonzero diagonal terms, and is hence invertible over Q=Frac(S). We have
Taking the coefficient of A x for x∈W af, we have
Note that if \(\varOmega\subset W_{\mathrm{af}}^{0}\) is any order ideal (downwardly closed subset) then the restriction A| Ω×Ω is invertible. In the sequel we choose certain such order ideals and find a formula for the inverse of this submatrix. Since the values of ξ x are given by (14) we obtain an explicit formula for \(j_{v}^{x}\) for v∈Ω and all x∈W af.
3 Equivariant homology Chevalley rule
Theorem 6
For every x∈W af∖{id}, \(\xi^{x^{-1}}(r_{\theta})\in\theta S\) and
Proof
For x≠id, the GKM condition (17) and (15) implies that \(\xi^{x^{-1}}(r_{\theta})\in\theta S\). \(\varOmega= \{\mathrm{id}, r_{0}\}\subset W_{\mathrm{af}}^{0}\) is an order ideal. The matrix A| Ω×Ω and its inverse are given by
Since id=t id and \(t_{\theta^{\vee}} = t^{r_{0}}\) (as \(t_{\theta ^{\vee}}=r_{0}r_{\theta}\)), we have
By the length condition in Theorem 3 we have
By (15) \(j_{r_{0}}^{y} = 0\) unless \(y\le t_{\theta^{\vee}} =r_{0}r_{\theta}\). So assume this.
Suppose r 0 y<y. Write y=r 0 x. Then
If r 0 y>y then we write y=x≤r θ and
as required. □
The formula (14) shows that \(\xi^{x^{-1}}(r_{\theta}) \in \mathbb{Z}_{\ge0}[\alpha_{i}\mid i\in I]\). The same holds for \(\theta^{-1} \xi^{x^{-1}}(r_{\theta})\). Indeed,
Lemma 7
α −1 ξ x(r α )∈ℤ≥0[α i ∣i∈I] for any positive root α.
Proof
The reflection r α has a reduced word i=i 1 i 2⋯i r−1 i r i r−1⋯i 1 which is symmetric. Consider the different embeddings j of reduced words of x into i, as in (14). If j uses the letter i r , then the corresponding term in (14) has θ as a factor. Otherwise, j uses i s but not i s+1,…,i r , for some s. But then there is another embedding of j′ of the same reduced word of x into i, which uses the other copy of the letter i s in i. The two terms in (14) which correspond to j and j′ contribute A(β−r α ⋅β)=A(〈α ∨,β〉α) where A∈ℤ≥0[α i ∣i∈I], and β is an inversion of r α . It follows that 〈α ∨,β〉>0. The lemma follows. □
Remark 8
The polynomials \(\xi^{x^{-1}}(r_{\theta})\) appearing in (25) may be computed entirely in the finite Weyl group and finite weight lattice.
Remark 9
In [8, Proposition 2.17], we gave an expression for the non-equivariant part of \(j_{r_{0}}\), consisting of the terms \(j_{r_{0}}^{x}A_{x}\) where ℓ(x)=1=ℓ(r 0). This follows easily from Theorem 6 and the fact [6] that \(\xi^{r_{i}}(w) = \omega_{i}- w\cdot\omega_{i}\), where ω i is the ith fundamental weight.
3.1 Application to quantum cohomology
The equivariant homology Chevalley rule (Theorem 6) may be used to obtain a new formula for some Gromov–Witten invariants for QH T(G/P) where P⊆̷G is a parabolic subgroup.Footnote 2
For this subsection we adopt the notation of [9], some of which we recall presently. Our goal is Proposition 10, which is the equivariant generalization of [9, Prop. 11.2].
Consider the Levi factor of P. It has Dynkin node subset I P ⊂I, Weyl group W P ⊂W, coroot lattice \(Q_{P}^{\vee}\subset Q^{\vee}\), root system R P ⊂R and positive roots \(R_{P}^{+}\). Denote the natural projection Q af→Q by \(\beta\mapsto\overline{\beta}\). Define
Every element w∈W af has a unique expression w=w 1 w 2 with w 1∈(W P)af and w 2∈(W P )af; denote by π P :W af↦(W P)af the map that sends w↦w 1.
Recall that the ring H T (Gr G ) has an S-basis \(\{\xi_{x}\mid x\in W_{\mathrm{af}}^{-}\}\). It has an ideal
The set \(\mathcal{T} = \{\xi_{\pi_{P}(t_{\lambda})} \mid\lambda\in \tilde{Q}\}\) is multiplicatively closed, where \(\tilde{Q}=\{\lambda\in Q^{\vee}\mid\langle\lambda,\alpha_{i}\rangle \le0 \text{for all $i\in I$}\}\) is the set of antidominant elements of Q ∨. Finally let \(\eta_{P}:Q^{\vee}\to Q^{\vee}/Q_{P}^{\vee}\) be the natural projection. Then by [9, Thm. 10.16] there is an isomorphism
where (q) denotes localization at the quantum parameters. For \(x\in W_{\mathrm{af}}^{-}\cap(W^{P})_{\mathrm{af}}\) with x=wt λ for w∈W and λ∈Q ∨, we have w∈W P and \(\lambda\in\tilde{Q}\). Then \(\varPsi_{P}(\xi_{x}) = q_{\eta_{P}(\lambda)} \sigma_{P}^{w}\) where \(\sigma_{P}^{w}\) is the quantum Schubert class in QH T(G/P) associated with w∈W P.
Proposition 10
Let w∈W P. Then
Proof
Choose λ∈Q ∨ such that 〈λ,α i 〉=0 for i∈I P and 〈λ,α i 〉≪0 for i∈I∖I P . Then 〈λ,α〉=0 for α∈R P and 〈λ,α〉≪0 for \(\alpha\in R^{+}\setminus R_{P}^{+}\).
We have \(x=wt_{\lambda}\in W_{\mathrm{af}}^{-} \cap(W^{P})_{\mathrm{af}}\) by [9, Lemmata 3.3, 10.1]. Define the set
Using the characterization of the Schubert basis in Theorem 3, for \(z\in W_{\mathrm{af}}^{-}\) the coefficient of j z in \(j_{r_{0}} j_{x}\) is given by the coefficient of A z in \(j_{r_{0}} A_{x}\). We obtain
where χ(true)=1 and χ(false)=0. We shall apply the map Ψ P to the above expression. First it is desirable to factor out the dependence of the right hand side on λ.
Suppose u∈W (which holds for u≤r θ ∈W). We claim that \(u\in\mathcal{A}_{x}\) if and only if ℓ(uw)=ℓ(w)−ℓ(u). Suppose \(u\in\mathcal{A}_{x}\). Since \(ux\in W_{\mathrm{af}}^{-}\) we have ℓ(ux)=ℓ(uwt λ )=ℓ(t λ )−ℓ(uw) and ℓ(u)+ℓ(x)=ℓ(u)+ℓ(t λ )−ℓ(w). Since ℓ(ux)=ℓ(u)+ℓ(x) it follows that ℓ(uw)=ℓ(w)−ℓ(u). Conversely suppose ℓ(uw)=ℓ(w)−ℓ(u). Since w∈W P it follows that uw∈W P. In particular \(uw t_{\lambda}\in W_{\mathrm{af}}^{-}\). Therefore ℓ(ux)=ℓ(u)+ℓ(x) and \(u\in \mathcal{A}_{x}\).
Let us fix the assumption that u∈W and ℓ(uw)=ℓ(w)−ℓ(u). Then \(u\in\mathcal{A}_{x}\) and ux∈(W P)af since uw∈W P. One may show that:
-
(1)
r 0 ux>ux if and only if (uw)−1⋅θ∈R + and (ux)−1⋅α 0∈ℤ>0 δ−(uw)−1⋅θ.
-
(2)
r 0 ux∉(W P)af if and only if \((uw)^{-1}\cdot \theta \in R_{P}^{+}\).
-
(3)
\(r_{0} ux \notin W_{\mathrm{af}}^{-}\) if and only if uxα i =α 0 for some i∈I.
It follows that under the assumption on u, \((uw)^{-1}\theta\in R^{+}\setminus R_{P}^{+}\) if and only if r 0 ux>ux, \(r_{0}ux \in W_{\mathrm{af}}^{-}\), and r 0 ux∈(W P)af.
We now apply the map Ψ P . By [9, Remark 10.1] \(r_{0}\in W_{\mathrm{af}}^{-}\cap (W^{P})_{\mathrm{af}}\). Since \(r_{0}= r_{\theta}t_{-\theta^{\vee}}\) we have \(\varPsi_{P}(\xi_{r_{0}}) = q_{\eta_{P}(-\theta^{\vee})} \sigma_{P}^{\pi _{P}(r_{\theta})}\).
By [9, Prop. 10.5, 10.8] π P (w)=w, π P (t λ )=t λ and π P (x)=x. Therefore \(\varPsi_{P}(\xi_{x}) = q_{\eta_{P}(\lambda)}\sigma_{P}^{w}\).
Let 1≠u≤r θ and \(u\in\mathcal{A}_{x}\). It follows that uw∈W P and ux=uwt λ ∈(W P)af. Then \(\varPsi_{P}(\xi_{ux}) = q_{\eta_{P}(\lambda)} \sigma_{P}^{uw}\).
Finally let 1≠u≤r θ be such that \(u\in\mathcal{A}_{x}\), \(r_{0}\in \mathcal{A}_{ux}\), and r 0 ux∈(W P)af. We have \(r_{0} ux = r_{\theta}t_{-\theta^{\vee}} u w t_{\lambda}= r_{\theta}uwt_{\lambda- (uw)^{-1}\theta^{\vee}}\). Therefore \(\varPsi_{P}(r_{0} ux) = q_{\eta_{P}(\lambda-(uw)^{-1}\theta^{\vee})} \sigma_{P}^{\pi_{P}(r_{\theta}uw)}\). Applying Ψ P to (27) yields the required equation. □
4 Alternating equivariant Pieri rule in classical types
We first establish some notation for G=SL n , Sp 2n , and SO 2n+1. Our root system conventions follow [5].
4.1 Special classes
We give explicit generating classes for H T (Gr G ).
4.1.1 \(H_{T}(\mathrm{Gr}_{SL_{n}})\)
Define the elements
So \(\ell(\hat{\sigma}_{p})=p-1\) and ℓ(σ p )=p. These elements have associated translations
4.1.2 \(H_{T}(\mathrm{Gr}_{Sp_{2n}})\)
For 1≤p≤2n−1 we define the elements \(\hat{\sigma}_{p}\in W\) by
For 1≤p≤2n−1 define \(\sigma_{p}\in W_{\mathrm{af}}^{0}\) and t p−1∈W af by
4.1.3 \(H_{T}(\mathrm{Gr}_{SO_{2n+1}})\)
For 1≤p≤2n−1 we define the elements \(\hat{\sigma}_{p}\in W_{\mathrm{af}}^{0}\) by
For 1≤p≤2n−1 define \(\sigma_{p}\in W_{\mathrm{af}}^{0}\) by
For 1≤p≤2n−2 define t p−1∈W af by
For 1≤p≤2n−1 let \(\sigma'_{p}\) be σ p but with every r 0 replaced by r 1. Then define
Then we conjecture that
where B is defined in (23). The sign is − for q≤2n−2 and + for q=2n−1.
4.1.4 Special classes generate
Let k′=n−1 for G=SL n and k′=2n−1 for G=Sp 2n or G=SO 2n+1. Let \(\hat{\mathbb{P}}:=S[[j_{\sigma_{m}}\mid1\le m\le k']]\) be the completion of ℙ≅H T (Gr G ) generated over S by series in the special classes. It inherits the Hopf structure from ℙ. The Hopf structure on ℙ is determined by the coproduct on the special classes.
Proposition 11
For G=SL n ,Sp 2n ,SO 2n+1, \(\mathbb{Q}\otimes_{\mathbb{Z}}\mathbb{P}\subset\mathbb{Q}\otimes_{\mathbb{Z}}\hat{\mathbb{P}}\).
Proof
It is known that the special classes generate the homology H ∗(Gr G ) non-equivariantly for G=SL n ,Sp 2n ,SO 2n+1 see [11, 14]. Furthermore, the equivariant homology Schubert structure constant \(d_{uv}^{w}\) is a polynomial in the simple roots of degree ℓ(w)−ℓ(u)−ℓ(v), and when ℓ(w)=ℓ(u)+ℓ(v), it is equal to the non-equivariant homology Schubert structure constant. It follows easily from this that each equivariant Schubert class can be expressed as a formal power series in the equivariant special classes. □
Remark 12
For G=SL n and G=Sp 2n the special classes generate H ∗(Gr G ) over ℤ.
4.2 The alternating equivariant affine Pieri rule
Let k=n−1 for G=SL n , k=2n−1 for G=Sp 2n , and k=2n−2 for G=SO 2n+1. Our goal is to compute \(j_{\sigma_{m}}^{x}\) for 1≤m≤k; note that for G=SO 2n+1, the element σ 2n−1 has been treated in (35). For this purpose consider the Bruhat order ideal \(\varOmega= \{ \mathrm{id}= \sigma_{0},\sigma_{1},\dotsc,\sigma_{k} \}\) in \(W_{\mathrm{af}}^{0}\). Since j 0=id, to compute \(j_{\sigma_{p}}^{x}\) for p≥1 we may assume x≠id by length considerations. It suffices to invert the matrix A given in (22) over Ω∖{id}×Ω∖{id}.
Define the matrices \(M_{pm}= (-1)^{m} \xi^{\sigma_{m}}(\sigma_{p})\) for 1≤p,m≤k, \(N_{mq}=\xi^{\hat{\sigma}_{m} r_{\theta}}(\hat{\sigma}_{q} r_{\theta})\) for 1≤m,q≤k, and the diagonal matrix \(D_{pq} = \delta_{pq} \,\xi^{t_{p-1}}(t_{p-1})\) for 1≤p,q≤k.
Conjecture 13
Conjecture 14
For 1≤m≤k and x≠id we have
In particular \(j_{\sigma_{m}}^{x}=0\) unless ℓ(x)≥m and x≤t q for some 0≤q≤m−1.
Conjecture 14 follows immediately from Conjecture 13: we have M −1=ND −1, and (37) follows from (24).
Theorem 15
Conjecture 14 holds for G=SL n .
The proof appears in Appendix A. Examples of (36) appear in Appendix B.
5 Effective Pieri rule for \(H_{T}(\mathrm{Gr}_{SL_{n}})\)
The goal of this section is to prove a formula for \(j_{\sigma_{m}}^{x}\) that is manifestly positive. In this section we work with G=SL n , W=S n , and \(W_{\mathrm{af}}=\tilde{S}_{n}\). We first establish some notation. For a≤b write
for upward and downward sequences of reflections and for sums of consecutive roots. In particular we have \(\theta= \alpha_{1}+\alpha_{2}+\dotsm+\alpha_{n-1} =\alpha_{1}^{n-1}\).
5.1 V’s and Λ’s
The support Supp(b) of a word b is the set of letters appearing in the word. For a permutation w, Supp(w) is the support of any reduced word of w. A V is a reduced word (for some permutation) that decreases to a minimum and increases thereafter. Special cases of V’s include the empty word, any increasing word and any decreasing word. A Λ is a reduced word that increases to a maximum and decreases thereafter. A (reverse) N is a reduced word consisting of a V followed by a Λ, such that the support of the V is contained in the support of the Λ. For example, the words 32012, 23521, and 32012453 are a V, Λ, and N, respectively.
By abuse of language, we say a permutation is a V if it admits a reduced word that is a V. We use similar terminology for Λ’s and N’s.
A permutation is connected if its support is connected (that is, is a subinterval of the integers). The following basic facts are left as an exercise.
Lemma 16
A permutation that is a V, admits a unique reduced word that is a V. Similarly for a connected Λ or a connected N.
Lemma 17
A connected permutation is a V if and only if it is a Λ, if and only if it is an N.
5.2 t q -factorizations
For 0≤q≤n−2, we call
the standard reduced word for t q . Since this word is an N it follows that any x≤t q is an N. We call the subwords \(q(q-1)\dotsm1\), \(12\dotsm(n-2)\) and \((n-2) \dotsm (q+1)\) the left, middle, and right branches.
Lemma 18
If \(x\in\tilde{S}_{n}\) admits a reduced word in which i+1 precedes i for some i∈ℤ/nℤ then \(x\not\le t_{i}\).
Proof
Suppose x≤t i . Since the standard reduced word of t i has all occurrences of i preceding all occurrences of i+1, it follows that x has a reduced word with that property. But this property is invariant under the braid relation and the commuting relation, which connect all reduced words of x. □
Let c(x) denote the number of connected components of Supp(x). If J and J′ are subsets of integers then we write J<J′−1 if max(J)<min(J′)−1. The following result follows easily from the definitions.
Lemma 19
Suppose x≤t q . Then x has a unique factorization \(x = v_{1}\cdots v_{r} y_{1}\* y_{2} \cdots y_{s}\), called the q-factorization, where each v i ,y i has connected support such that
-
(1)
Supp(v i )<Supp(v i+1)−1 and Supp(y i )<Supp(y i+1)−1
-
(2)
Supp(v 1⋯v r )⊂[0,q]
-
(3)
Supp(y 1⋯y s )⊂[q+1,n−1]
-
(4)
Each v i is a V
-
(5)
Each y i is a Λ.
We say that v r and y 1 touch if q∈Supp(v r ) and q+1∈Supp(y 1). We denote
Note that ϵ(x,q) depends only on Supp(x) and q.
Each k in the q-factorization of x≤t q , is (S1) in the left branch of some v i , or (S2) in the right branch of some v i , or (S3) at the bottom of a v i , or (S1′) in the left branch of some y i , or (S2′) in the right branch of some y i , or (S3′) at the top of a y i . We call these sets S1, S2, S3, S1′, S2′, and S3′. Note that k can belong to both S1 and S2, or both S1′ and S2′.
For each x and each q such that x≤t q , we define the polynomials
We also define \(R(x,q,m)= \prod_{k\in S2'\cap[m,n-1]}(-\alpha_{q+1}^{k})\).
5.3 The equivariant Pieri rule
Let
and
be the root associated with the reflection \(r_{\beta_{i}}\) that exchanges the numbers 1+q i and 1+q i+1. For a root β and f∈S define
Theorem 20
We have
where \(Y(x,m) = (\alpha_{0}^{q_{1}})^{c(x)-1}R(x,q_{1},m)\).
The proof of Theorem 20 is given in Sect. 6.
5.4 Positive formula
Define \(\tilde{S2}'=S2'\cap[m,n-1]\), and let \(K = \tilde{S2}' \cup\{n-1,\ldots,n-1\} = \{k_{1} \geq k_{2} \geq\cdots\geq k_{d}\}\) be the multiset where the element (n−1) is added to \(\tilde{S2}'\) (c(x)−1) times.
Theorem 21
where s(i,R)=#{r∈R∣i<r}+1.
The proof of Theorem 21 is given in Sect. 6.
Example 22
Let n=8, m=4, and x=r 0 r 4 r 5 r 7 r 4 r 2 r 1. The components of Supp(x) are [0,2], [4,5], and [7] so that c(x)=3. We have p=3 with (q 1,q 2,q 3)=(0,2,3), v 1=r 0, y 1=r 2 r 1, y 2=r 4 r 5 r 4, y 3=r 7, ϵ(x,q 1)=1, S1=S2=∅, S3={0}, S1′={4}, S2′={1,4}, S3′={2,5,7}, S2′∩[m,n−1]={4}. Thus K={7,7,4}. Then writing \(\alpha_{a}^{b} = x_{a}-x_{b+1}\), and noting that \(\alpha_{0}^{n-1} = 0\), Theorem 20 yields
agreeing with Theorem 21.
6 Proof of Theorems 20 and 21
6.1 Simplifying (37)
Let 0≤q≤m−1. By (14) and Lemma 2 we have
We also have
Define
so that by Theorem 15,
Explicitly we have
6.2 Evaluation at t q
Proposition 23
If x≤t q , then
Proof
We compute ξ x(t q ) using (14) by computing all embeddings of reduced words of x into the standard reduced word (41) of t q . We refer to the q-factorization of x. Each k∈S1 must embed into the left branch of the N, and has associated root \(\alpha_{k}^{q}\). Each k∈S2 embeds into the middle branch of the N and has associated root \(\alpha_{0}^{k-1}\). Each k∈S1′ embeds into the middle branch of the N and has associated root \(\alpha_{0}^{k}\). Each k∈S2′ embeds into the right branch of the N and has associated root \(-\alpha_{q+1}^{k}\). Each k∈S3 is either 0 and has associated root \(\alpha_{0}^{q}\), or can be embedded into the left or middle branch of the N, and the sum of the two associated roots for these positions is \(\alpha_{k}^{q}+\alpha_{0}^{k-1}=\alpha_{0}^{q}\). Each k∈S3′ is either n−1, which has associated root \(-\alpha_{q+1}^{n-1} = \alpha_{0}^{q}\), or can be embedded into the middle or right branch of the N, and the sum of associated roots is \(\alpha_{0}^{k}-\alpha_{q+1}^{k}=\alpha_{0}^{q}\). Since all the various choices for embeddings of elements of S3 and S3′ can be varied independently, the value of ξ x(t q ) is the product of the above contributions. Each minimum of a v i and maximum of a y j contributes \(\alpha_{0}^{q}\). If there is a component of x which contains both q and q+1 (that is, if v r and y 1 touch) then it is unique and contributes two copies of \(\alpha_{0}^{q}\). All this yields (50). □
6.3 Rotations
We now relate ξ x(t q ) with ξ x(t q′). Let r p,q denote the transposition that exchanges the integers p and q.
Proposition 24
Let x≤t q and consider the q-factorization of x. Let a be such that this reduced word of x contains the decreasing subword \((q+a)(q+a-1)\dotsm(q+1)\) but not \((q+a+1)(q+a)\dotsm(q+1)\). If q+1∉Supp(x), then set a=1. Then
and
Let y ↑ denote y with every r i changed to r i+1. The following lemma follows easily by induction.
Lemma 25
Let y be increasing with support in [b,a−1]. Then
Proof of Proposition 24
We assume that q+1∈Supp(x), for otherwise the claim is easy.
By Lemma 18 we have \(x\not\le t_{q+i}\) for 1≤i≤a−1. Equation (51) follows from (15). We now prove (52). The first goal is to compute the q+a-factorization of x. Since x≤t q we may consider the q-factorization of x. The decreasing word \((q+a-1)\dotsm(q+2)(q+1)\) must embed into the right hand branch, that is, [q+1,q+a−1]⊂S2′. The hypotheses imply that \(q+a\not\in S2'\). There are two cases: either q+a∈S1′ or q+a∈S3′ (so that \(q+a+1\not\in\mathrm{Supp}(x)\)). We treat the former case, as the latter is similar: the two cases correspond to the touching and nontouching cases for the q+a-factorization of x, whose existence we now demonstrate.
Suppose q+a∈S1′. Then there is a \(y_{1}'\) with \(\mathrm{Supp}(y_{1}')\subset[q+a+1,n-1]\) and a y with an increasing reduced word such that Supp(y)⊂[q+1,q+a−1] and \(y_{1} = y r_{q+a} y_{1}' d^{q+a-1}_{q+1}= y d^{q+a}_{q+1} y_{1}'\). Suppose v r and y 1 touch. Then \(v_{r}':=v_{r} y d^{q+a}_{q+1}\) is an N and therefore a V. Moreover x≤t q+a since x has a q+a-factorization given by the q-factorization of x but with v r and y 1 replaced by \(v_{r}'\) and \(y_{1}'\), respectively. To verify that \(v_{r}'\) is a V, by the touching assumption, q∈Supp(v r ) and we have \(v_{r}' = v_{r} y d^{q+a}_{q+1} = v_{r} d^{q+a}_{q+1} y^{\uparrow}=d^{q+a}_{q+2} v_{r} r_{q+1} y^{\uparrow}\) which expresses \(v_{r}'\) in a V.
Suppose v r and y 1 do not touch, that is, q∉Supp(v r ). We have the V given by \(v'_{r+1} = y d^{q+a}_{q+1} = d^{q+a}_{q+1}y^{\uparrow}\). Then x≤t q+a , as x has the q+a factorization given by the q-factorization of x except that there is a new V, namely, \(v'_{r+1}\) and the first y is \(y_{1}'\) instead of y 1.
In every case we calculate that
The calculation for L and R follows from the fact that [q+2,q+a]⊂S1 q+a , but \([q+1,q+a-1]\subset S2'_{q}\). The calculation for M follows from the fact that Supp(y)⊂S2 q and Supp(y ↑)⊂S2 q+a , together with the following boundary cases:
If q+a+1∈Supp(x) then \(q+a \in S1_{q+a}\cap S1'_{q}\). Thus q+a contributes a factor of \(\alpha_{0}^{q+a}\) to M(x,q). This factor appears in M(x,q+a) as the factor \((\alpha_{0}^{q+a})^{\epsilon(x,q+a)}\), since ϵ(x,q+a)=1.
If q∈Supp(x) one has ϵ(x,q)=1 and q+1∈S2 q+a contributes a factor of \(\alpha_{0}^{q}\) to M(x,q+a). This factor appears in M(x,q) as the factor \((\alpha_{0}^{q})^{\epsilon(x,q)} = \alpha_{0}^{q}\).
Using that \(d^{q+a}_{q+1} \alpha_{0}^{q} = \alpha_{0}^{q+a}\), \(d^{q+a}_{q+1}(-\alpha_{q+1}^{q+a})=\alpha_{q+a}\), and \(r_{1+q,1+q+a}\alpha_{q+1}^{q+a} = -\alpha_{q+1}^{q+a}\), the above relations between M(x,q), L(x,q), R(x,q) and their counterparts for q+a, together with Proposition 23, yield
To obtain (52), since \(r_{1+q,1+q+a} = d^{q+a}_{q+1}u_{q+2}^{q+a}\), it suffices to show that
However, it is clear that \(\alpha_{0}^{q}\) and L(x,q) are invariant, and the only part of R(x,q) that must be checked is the product ∏ k∈S2′∩[q+1,q+a](−α q+1,k ). However, we have S2′∩[q+1,q+a]=[q+1,q+a−1], and indeed the product \(\prod_{k=q+1}^{q+a} (-\alpha_{q+1}^{k})\) is invariant under \(u_{q+2}^{q+a}\), as required. □
Recall the definition of q j from (43). In light of the proof of Proposition 24, we write
Recall the definition of β i from (44). For i≤j we also define
Let
so that \(Y_{i}(x,m) = r_{\beta_{i-1}} Y_{i-1}(x,m)\).
Recall the definitions of D(q,m) and Y i (x,m) from (46).
Lemma 26
Proof
The proof proceeds by induction on j. Let D j be the denominator of the right hand side. Suppose first that j=1. Consider the embedding of x into \(t_{q_{1}}\). By the definition of q 1, it follows that \(L(x,q_{1}) \alpha_{0}^{q_{1}} = \xi^{\sigma _{q_{1}+1}}(\sigma_{q_{1}+1})\). By the definition of the q j , we also have \(S2' \cap[q_{1}+1,m-1] =[q_{1}+1,m-1] \setminus\{q_{2},q_{3},\dotsc,q_{p}\}\). These considerations and Proposition 23 imply that
This proves the result for j=1. Suppose the result holds for 1≤j≤p−1. We show it holds for j+1. By induction we have
Proposition 24 yields
It remains to show
We have \(D(q_{j},m)=\prod_{k=0}^{q_{j}} \alpha_{k}^{q_{j}} \prod_{k=q_{j}+1}^{m-1} \alpha_{q_{j}+1}^{k}\). For k∈[0,q j ] we have \(r_{\beta_{j}} \alpha_{k}^{q_{j}} = \alpha_{k}^{q_{j+1}}\). For k∈[q j +1,q j+1−1] we have \(r_{\beta_{j}} \alpha_{q_{j}+1}^{k} = -\alpha_{k+1}^{q_{j+1}}\), \(r_{\beta_{j}} \alpha_{q_{j}+1}^{q_{j+1}} = -\alpha_{q_{j}+1}^{q_{j+1}}\), and for k∈[q j+1+1,m−1] we have \(r_{\beta_{j}} \alpha_{q_{j}+1}^{k} =\alpha_{q_{j+1}+1}^{k}\). Therefore
We also have \(r_{\beta_{j}} \beta_{j-1}^{i} = \beta_{j}^{i}\) for 1≤i≤j−1 and \(r_{\beta_{j}} \beta_{j}^{i} = \beta_{j+1}^{i}\) for j+1≤i≤p−1. Therefore
□
The following result is immediate from the definitions.
Lemma 27
\(r_{\beta_{j}} Y_{i}(x,m) = Y_{i}(x,m)\) for j≥i+2.
6.4 Proof of Theorem 20
Note that if \(r_{\beta_{j+1}} Y = Y\) and i≤j then
So using Lemma 27 we have
Thus
by (47), as required. □
6.5 Proof of Theorem 21
We first count the gratuitous negative signs in M(x)=M(x,q 1) and Y(x,m). Letting q=q 1, using the q 1-factorization of x, and recalling that \(\tilde{S2}'=S2'\cap[m,n-1]\), this number is
Therefore all signs cancel and we have
Let x i be the standard basis of the finite weight lattice ℤn with α i =x i −x i+1. Then \(r_{\beta_{j}}\) acts by exchanging \(x_{q_{j}+1}\) and \(x_{q_{j+1}+1}\). Let us write
where \(n-1 \ge k_{1} \ge k_{2} \ge\dotsm\ge k_{d} \ge m\). Note that q j +1≤q p +1≤m. Since
and since ∂ i 1=0, we have
So \(\partial_{\beta_{1}}\) can act on any factor (giving the answer 1 and thus effectively removing the factor), and to the left each variable \(x_{q_{1}+1}\) is reflected to \(x_{q_{2}+1}\). Next we apply \(\partial_{\beta_{2}}\). It kills any factor \(x_{q_{1}+1}-x_{k_{i}+1}\). Therefore we may assume it acts on a factor of the form \(x_{q_{2}+1}-x_{k_{i}+1}\) which is to the left of the factor removed by \(\partial_{\beta_{1}}\). Continuing in this manner we see that \(\partial_{\beta_{p-1}}\dotsm \partial_{\beta_{1}} Z\) is the sum of products of positive roots, where a given summand corresponds to the selection of p−1 of the factors, which are removed, and between the rth and r+1th removed factor from the right, an original factor \(x_{q_{1}+1}-x_{k_{i}+1}\) is changed to \(x_{q_{r+1}+1}-x_{k_{i}+1}\).
It follows that Theorem 20 yields Theorem 21. □
Notes
Using equivariance for the maximal torus T af⊂G af, the GKM condition characterizes the image of localization to torus fixed points. However, after forgetting equivariance down to the smaller torus T, elements of Φ are characterized by additional conditions, which were determined in [4].
This notation for P will be used only in this subsection and should not cause confusion for the reader with its previous use as the weight lattice of G.
References
Andersen, H.H., Jantzen, J.C., Soergel, W.: Representations of quantum groups at a pth root of unity and of semisimple groups in characteristic p: independence of p. Astérisque 220, 321 (1994)
Billey, S.: Kostant polynomials and the cohomology ring for G/B. Duke Math. J. 96, 205–224 (1999)
Goresky, M., Kottwitz, R., MacPherson, R.: Equivariant cohomology, Koszul duality, and the localization theorem. Invent. Math. 131(1), 25–83 (1998)
Goresky, M., Kottwitz, R., MacPherson, R.: Homology of affine Springer fibers in the unramified case. Duke Math. J. 121, 509–561 (2004)
Kac, V.: Infinite Dimensional Lie Algebras, 3rd edn. Cambridge University Press, Cambridge (1990)
Kostant, B., Kumar, S.: The nil Hecke ring and cohomology of G/P for a Kac–Moody group G. Adv. Math. 62(3), 187–237 (1986)
Kumar, S.: Kac-Moody Groups, Their Flag Varieties and Representation Theory. Progress in Mathematics, vol. 204. Birkhäuser, Boston (2002), pp. xvi+606
Lam, T., Shimozono, M.: Dual graded graphs for Kac-Moody algebras. Algebra Number Theory 1(4), 451–488 (2007)
Lam, T., Shimozono, M.: Quantum cohomology of G/P and homology of affine Grassmannian. Acta Math. 204, 49–90 (2010)
Lam, T., Shimozono, M.: k-Double Schur functions and equivariant (co)homology of the affine Grassmannian. preprint, arXiv:1105.2170
Lam, T., Schilling, A., Shimozono, M.: Schubert polynomials for the affine Grassmannian of the symplectic group. Math. Z. 264(4), 765–811 (2010)
Mihalcea, L.: Positivity in equivariant quantum Schubert calculus. Am. J. Math. 128(3), 787–803 (2006)
Peterson, D.: Lecture Notes at MIT (1997)
Pon, S.: Affine Stanley symmetric functions for classical groups. Ph.D. thesis, University of California, Davis (2010)
Acknowledgements
T.L. was supported by NSF grant DMS-0901111, and by a Sloan Fellowship. M.S. was supported by NSF DMS-0652641 and DMS-0652648.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Proof of Theorem 15
In this section we assume that G=SL n and prove (36).
The matrices M and N are easily seen to be lower triangular. We first check the diagonal:
by (2), (4), and Lemma 1.
It remains to check below the diagonal. Let p>q and p≥k≥q. We have
Note that the second factor is independent of k. We also have
with the second factor independent of k. Therefore, to prove that
it is equivalent to show that
The above identity can be rewritten as
To prove this last identity, let q′ be such that q<q′≤p. It is easy to show by descending induction on q′ that
Then for q′=q+1 the sum is the negative of the k=q summand of (57) as required. □
Appendix B: Examples of (36)
Example 28
G=SL 3 has affine Cartan matrix
The column dependencies give the coefficients of the null root δ=α 0+θ=α 0+α 1+α 2 which is set to zero due to the finite torus equivariance.
p | \(\hat{\sigma}_{p}\) | σ p | t p−1 | \(\hat{\sigma}_{p} r_{\theta}\) |
---|---|---|---|---|
1 | id | r 0 | r 0 r 1 r 2 r 1 | r 1 r 2 r 1 |
2 | r 1 | r 1 r 0 | r 1 r 0 r 1 r 2 | r 2 r 1 |
We compute the matrices
For x=r 1 r 2 we compute the column vector with values (−1)ℓ(x) ξ x(t j ) for j=1,2. Acting on this column vector by ND −1, we obtain the coefficients of A x in j 1 and j 2.
Doing the same thing for x=r 1 r 0 r 2 we have
Example 29
Sp 2n for n=2 has affine Cartan matrix
We have δ=α 0+θ=α 0+2α 1+α 2.
p | \(\hat{\sigma}_{p}\) | σ p | t p−1 | \(\hat{\sigma}_{p} r_{\theta}\) |
---|---|---|---|---|
1 | id | r 0 | r 0 r 1 r 2 r 1 | r 1 r 2 r 1 |
2 | r 1 | r 1 r 0 | r 1 r 0 r 1 r 2 | r 2 r 1 |
3 | r 2 r 1 | r 2 r 1 r 0 | r 2 r 1 r 0 r 1 | r 1 |
We have
Now let x=r 0 r 1 r 2. We have
The matrix ND −1 acting on the above column vector, gives the vector
Now let x=r 1 r 2 r 1. We have
Example 30
SO 2n+1 for n=3 has affine Cartan matrix
We have δ=α 0+θ=α 0+α 1+2α 2+2α 3.
p | \(\hat{\sigma}_{p}\) | σ p | t p−1 | \(\hat{\sigma}_{p} r_{\theta}\) |
---|---|---|---|---|
1 | id | r 0 | r 0 r 2 r 3 r 2 r 1 r 2 r 3 r 2 | r 2 r 3 r 2 r 1 r 2 r 3 r 2 |
2 | r 2 | r 2 r 0 | r 2 r 0 r 2 r 3 r 2 r 1 r 2 r 3 | r 3 r 2 r 1 r 2 r 3 r 2 |
3 | r 3 r 2 | r 3 r 2 r 0 | r 3 r 2 r 0 r 2 r 3 r 2 r 1 r 2 | r 2 r 1 r 2 r 3 r 2 |
4 | r 2 r 3 r 2 | r 2 r 3 r 2 r 0 | r 2 r 3 r 2 r 0 r 2 r 3 r 2 r 1 | r 1 r 2 r 3 r 2 |
5 | r 0 r 2 r 3 r 2 | r 0 r 2 r 3 r 2 r 0 | r 0 r 2 r 3 r 2 r 0 r 1 r 2 r 3 r 2 r 1 | r 2 r 3 r 2 |
To save space let us write α ijk :=iα 1+jα 2+kα 3. We have
D has diagonal entries
One may verify that MN=D.
Rights and permissions
About this article
Cite this article
Lam, T., Shimozono, M. Equivariant Pieri Rule for the homology of the affine Grassmannian. J Algebr Comb 36, 623–648 (2012). https://doi.org/10.1007/s10801-012-0353-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10801-012-0353-5