K-theory of Springer varieties

By the naturality of Chern character we have the following commuting diagram:

The vertical arrows are monomorphisms since the K-groups K(\mathcal F), K(\mathcal F_\lambda) are free abelian groups and \textrm{ch}\otimes \mathbb Q is an isomorphism.

Now, since \textrm{ch}_{\mathcal{F}_{\lambda}} is injective, \iota_{\lambda}^! is surjective and \iota_\lambda^* and \textrm{ch}_\mathcal F are \mathfrak{S}_n-equivariant, the commutative diagram (Proof) implies that the image \textrm{ch}_{\mathcal{F}_{\lambda}}(K(\mathcal F_\lambda)) in H^*(\mathcal{F}_{\lambda};\mathbb{Q}) is stable under the \mathfrak{S}_n-action. This further implies that K(\mathcal F_\lambda) admits an action of \mathfrak{S}_n such that \textrm{ch}_{\mathcal F_\lambda} and \iota_\lambda^! are equivariant. This can be seen more explicitly as follows.

Let x\in K(\mathcal{F}_{\lambda}). Since \iota_{\lambda}^! is surjective, x=\iota_{\lambda}^!(y) for some y\in K(\mathcal{F}). We define the action of \mathfrak{S}_n on (the right of) K(\mathcal{F}_{\lambda}) as follows: x\cdot\sigma:=\iota_{\lambda}^!(y\cdot\sigma). Indeed from (Proof) we have

Moreover, since \iota_{\lambda}^* and \textrm{ch}_{\mathcal{F}} are \mathfrak{S}_n-equivariant this implies

Again by the commutativity of (Proof) we get

Thus by definition \textrm{ch}_{\mathcal{F}_{\lambda}}(x\cdot\sigma)= \textrm{ch}_{\mathcal{F}_{\lambda}}(x)\cdot\sigma. Since \textrm{ch}_{\mathcal{F}_{\lambda}} is injective and the \mathfrak{S}_n action on H^*(\mathcal{F}_{\lambda};\mathbb{Q}) is well defined, it follows that the \mathfrak{S}_n action on K(\mathcal{F}_{\lambda}) is well defined. It also follows that \textrm{ch}_{\mathcal{F}_{\lambda}} and \iota^{!}_{\lambda} are \mathfrak{S}_n-equivariant.

Fix s\le n. Since the action of \mathfrak{S}_n on K(\mathcal F_\lambda) permutes the L_j, ~1\le j\le n, we need only consider the case where i_j=j, ~~\forall ~~1\leq j\leq s.

We replace N by a conjugate gN g^{-1} so that {\rm Im}\, (gNg^{-1})^{n-k}=U_{p_{\lambda^\vee}(k)}=\mathbb C^{p_{\lambda^\vee}(k)} for k\ge 1. We may then choose the refinement \underline{U}\in \mathcal F to be the standard flag 0\subset \mathbb C\subset \cdots\subset\mathbb C^n. Thus \mathbb C^q\subset V_s for any \underline{V}\in \mathcal F_{gNg^{-1}}. Let \iota_g:\mathcal F\to \mathcal F be the translation by g: \underline {V}\mapsto g\underline {V}=gV_0=0\subset gV_1\subset \cdots \subset gV_n=\mathbb C^n. Since GL(n,\mathbb C) is connected, the composition \mathcal F_{N}\stackrel{\iota_\lambda}{\hookrightarrow} \mathcal F\stackrel{\iota_g}{\longrightarrow }\mathcal F, denoted \iota_{\lambda,g} is homotopic to \iota_\lambda and maps \mathcal F_{N} onto \mathcal F_{gNg^{-1}}\subset \mathcal F. It follows that \iota_\lambda and \iota_{\lambda_g} induce the same map in K-theory and singular cohomology. In particular, \iota_{\lambda,g}^*(\mathcal L_j)=L_j~\forall ~1\leq j\le n.

Let G_{n,s}=G_s(\mathbb C^n) denote the Grassmann variety of s-planes in \mathbb C^n. One has a projection \pi_s: \mathcal F_{N}\to G_{n,s} defined as \underline{V}\mapsto V_s.

Let Y_q\subset G_{n,s} denote the subvariety \{U\in G_{n,s}\mid U\supset U_q\}, 1\le q<s. Then Y_q is isomorphic to a Grassmann variety G_{n-q,s-q}. A specific isomorphism Y_q\cong G_{s-q}(\mathbb C^n/U_q) is obtained by sending U\in Y_q to U/U_q. The tautological complex vector bundle \gamma_{n,s} is of rank s, whose fibre over A\in G_{n,s} is the vector space A. When restricted to Y_q, \gamma_{n,s} has a trivial subbundle q\epsilon of rank q. Indeed we have a commuting diagram

where the vertical arrows are bundle projections. Therefore

where \omega is the complex vector bundle over Y_q whose fibre over A\in Y_q is the complex vector space A':=A/\mathbb C^q.

From Proposition ?, the image of the composition

denoted \pi_{\lambda,s}, is contained in Y_q. Therefore, we have a commuting diagram

Now \pi_s^*(\gamma_{n,s})=\mathcal V_s=\mathcal L_1\oplus \cdots \oplus \mathcal L_s by (?). Therefore L_1\oplus \cdots\oplus L_s=\iota_{\lambda,g}^*(\mathcal L_1\oplus\cdots\oplus \mathcal L_s) =\iota_{\lambda,g}^*\circ \pi_s^*(\gamma_{n,s})=\pi_{\lambda,s}^*(\gamma_{n,s}|_{Y_q})=\pi_{\lambda,s}^*(\omega)\oplus q\epsilon, from (?).

Set \displaystyle\mathcal E=\bigoplus_{1\le j\le s} L_{i_j}. Then, by Proposition ?, \displaystyle\mathcal E\cong \xi+q\epsilon where rank of \xi equals s-q. (Here q=p_{\lambda^\vee}(s) as in the Propositon.) It follows that \displaystyle \gamma^d([\mathcal E]-s)=\gamma^d([\xi]-s+q)= \lambda^{d}([\xi]+d-s+q-1)=0 for all d\ge \textrm{rank}(\xi)+1=s+1-q. The last equation is equivalent to the vanishing of \lambda^d([\mathcal E]-q) for all d\ge s+1-q (see (?)).

In view of (?), it is clear that \mathcal I_\lambda\subset \ker(\psi_\lambda). Also, by Proposition ?, \psi_{\lambda} is surjective. Since \textrm{rank}(K(\mathcal F_\lambda))=\textrm{rank}(H^*(\mathcal F_\lambda;\mathbb Z))={n\choose\lambda}, it suffices to show that R/\mathcal I_\lambda is a free abelian group of rank {n\choose \lambda}. We will show that R/\mathcal I_\lambda has a filtration such that the associated graded ring is isomorphic as an abelian group to S/I_\lambda\cong H^*(\mathcal F_\lambda;\mathbb Z).

Let R^d\subset R denote the abelian group generated by elements of the form P(u)-\theta(P(u))\in \ker(\theta), where P(u) is a polynomial in u_1,\ldots, u_n of degree at most d. Note that R^0=0 and R^d\subset R^{d+1} and R^i\cdot R^j\subset R^{i+j} ~~\forall~~ i, j \ge 1. Clearly h_d(u_\mathbf i)-\overline h_d(u_\mathbf i)\in R^{d-1}.

Let k\ge 1. and let \mathbf s=(s_1,\ldots,s_k) where 1\le s_1\le \cdots \le s_k\le n. We denote by \mathcal S_k the set of all such k-tuples \mathbf s. For \mathbf s\in \mathcal S_k we define the following sets

For a given \mathbf d\in \mathcal D_{\mathbf s} and an integer \displaystyle d\ge \sum_{j=1}^k d_j, we set

An element of \mathcal W_{\mathbf s} will be denoted \iota=( \mathbf i(1),\ldots,\mathbf i(k)).

Given \mathbf s\in \mathcal S_k, ~\iota\in \mathcal W_\mathbf s, ~{\mathbf d}\in \mathcal D_\mathbf s, ~\mathbf c\in \mathcal C_{d, \mathbf d}, we set

where

Let \mathcal H_{d,k} be the set of all elements h_{\iota, \mathbf d,\mathbf c}(u), as we vary \mathbf s\in\mathcal S_k, \iota\in \mathcal W_\mathbf s, \mathbf d\in \mathcal D_\mathbf s, \mathbf c\in \mathcal C_{d, \mathbf d}. Let

Note that the leading term of h_{\iota, \mathbf {d},\mathbf c}(u), when expressed as a polynomial in u_1,\ldots, u_n, equals that of e_{\iota, \mathbf d, \mathbf c}(u):=e_{d_1}(u_{\mathbf i(1)})^{c_1}\cdots e_{d_k}(u_{\mathbf i(k)})^{c_k}. Since \theta (e_{\iota,\mathbf d,\mathbf c}(u))\ne 0, the element e_{\iota,\mathbf d,\mathbf c}(u) is not in R^d. But evidently e_{\iota,\mathbf d,\mathbf c}(u)-\theta(e_{\iota,\mathbf d,\mathbf c}(u)) belongs to R^d. Setting v_j=u_j-1, 1\le j\le n, we have e_{\iota, \mathbf d,\mathbf c}(v)\equiv e_{\iota,\mathbf d,\mathbf c}(u)-\theta( e_{\iota,\mathbf d,\mathbf c}(u))\equiv h_{\iota, \mathbf d, \mathbf c}(u) \mod R^{d-1}.

Recall, from Definition ?, that I_\lambda\subset S=\mathbb Z[y_1,\ldots, y_n] is the Tanisaki ideal generated by the homogeneous polynomials

(see Definition ?). Let S_k denote the homogeneous polynomials in S of degree k where \deg(y_j)=1 for all j\le n and \displaystyle I_{\lambda,k}:=I_\lambda\cap S_k. Writing \displaystyle S^k=\bigoplus_{1\le j\le k}S_j and \displaystyle I^k_\lambda=\bigoplus_{1\le j\le k}I_{\lambda,j}, we see that under the isomorphism \alpha: R\to S defined by u_j\mapsto y_j for 1\le j\le n, we have \alpha(R^k)=S^k for all k\ge 1. Also \alpha(\mathcal I^k_\lambda)\equiv I^k_\lambda\mod S^{k-1} for all k\ge 0. Therefore \alpha induces an isomorphism of abelian groups

Since by Theorem ?, for every k we have

we conclude that, as abelian groups,

This completes the proof.