Equivariant K-theory and Tangent spaces to Schubert varieties

Let $G$ be a semisimple algebraic group defined over an algebraically closed field $k$ of characteristic 0. Let $P\supseteq B\supseteq T$ be a parabolic subgroup, Borel subgroup, and maximal torus of $G$ respectively. We denote the set of weights of a representation $E$ of $T$ by $\Phi(E)$. Let $W$ be the Weyl group of $(G,T)$, and $S$ the set of simple reflections in $W$ relative to $B$.

Fix $w\leq x\in W$. Let $X^{w}$ be the Schubert variety $\overline{B^{-}wB}$, and $Y_{x}^{w}$ the Kazhdan-Lusztig variety $BxB\cap\overline{B^{-}wB}$, in $G/B$. The Kazhdan-Lusztig variety $Y_{x}^{w}$ (and thus its tangent space at $x$, $T_{x}Y_{x}^{w}$) is an affine subvariety of an ambient space $V$ in $G/B$ with weights $\Phi(V)=I(x^{-1})$, the inversion set of $x^{-1}$. If $\mathbf{s}=(s_{1},\ldots,s_{l})$, $s_{i}\in S$, is a reduced expression for $x$, then the elements of $I(x^{-1})$ are given explicitly by the formula $\gamma_{i}=s_{1}\cdots s_{i-1}(\alpha_{i})$, $i=1,\ldots,l$, where $\alpha_{i}$ is the simple root corresponding to $s_{i}$.

Our main result is the following theorem (see Theorem 5.8):

This theorem, which applies to Kazhdan-Lusztig varieties in $G/B$, extends to Schubert varieties and to $G/P$. Moreover, when $x$ is a cominuscule Weyl group element of $W$, all $\gamma_{j}$ are integrally indecomposable in $I(x^{-1})$, so Theorem A recovers all weights of the tangent space.

Our proof of Theorem A uses equivariant $K$-theory. Let us fix notation and give some basic definitions and properties. If $T$ acts on a smooth scheme $M$, the Grothendieck group of $T$-equivariant coherent sheaves (or vector bundles) on $M$ is denoted by $K_{T}(M)$. If $N$ is a $T$-stable subscheme of $M$, then the class in $K_{T}(M)$ of the pushfoward of the structure sheaf $\mathcal{O}_{N}$ of $N$ is denoted by $[\mathcal{O}_{N}]_{M}$, or sometimes just $[\mathcal{O}_{N}]$. A $T$-equivariant vector bundle on a point is a representation of $T$, so $K_{T}(\{\text{point}\})$ can be identified with $R(T)$, the representation ring of $T$. The inclusion $i_{m}:\{m\}\to M$ of a $T$-fixed point induces a pullback $i_{m}^{*}:K_{T}(M)\to K_{T}(\{m\})=R(T)$.

Consider for the moment a more general situation than that of the previous subsection: $V$ a representation of $T$ such that all weights of $V$ lie in an open half-space and have multiplicity one, $Y\subseteq V$ a $T$-stable subscheme, and $x\in Y$ a $T$-fixed point. The structure sheaf $\mathcal{O}_{Y}$ defines a class $[\mathcal{O}_{Y}]\in K_{T}(V)$. We show that the factors of $i_{x}^{*}[\mathcal{O}_{Y}]\in R(T)$ contain information about the tangent space $T_{x}Y$. Let us say that $1-e^{-\theta}$ is a simple factor of $i_{x}^{*}[\mathcal{O}_{Y}]$ if $i_{x}^{*}[\mathcal{O}_{Y}]=(1-e^{-\theta})Q$ for some $Q\in R(T)$ which is a polynomial in $e^{-\lambda}$, $\lambda\in\Phi(V)\setminus\{\theta\}$. We prove (see Proposition 3.5)

Now set $V$ and $x$ as in Subsection 1.1 and set $Y$ to be the Kazhdan-Lusztig variety $Y_{x}^{w}$. For $\theta\in\Phi(V)=I(x^{-1})$, we have $\theta=\gamma_{j}$ for some $j$. Proposition B then becomes

This characterization of $\Phi(T_{x}Y_{x}^{w})$ would appear to suffer from a computational difficulty: determining whether $1-e^{-\gamma_{j}}$ is a factor of $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$, let alone whether it is a simple factor, seems to be a nontrivial problem. It requires some sort of division algorithm in $R(T)$. One approach would be to search for an expression for $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$ as a sum of terms in which $1-e^{-\gamma_{j}}$ appears explicitly as a factor of each summand. To rule out the possibility that $1-e^{-\gamma_{j}}$ is a factor of $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$, then, one would need to show that no such expression exists. This would presumably require knowledge of all possible expressions for $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$.

We show a way around this apparent computational difficulty. When $\gamma_{j}$ is integrally indecomposable in $I(x^{-1})$, the question of whether or not $1-e^{-\gamma_{j}}$ is a simple factor of $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$ can be answered by using a single expression for $i_{x}^{*}[\mathcal{O}_{Y_{x}^{w}}]$ due to Graham [Gra02] and Willems [Wil06]:

where $\mathcal{T}_{w,\mathbf{s}}$ is the set of sequences $\mathbf{t}=(i_{1},\ldots,i_{m})$, $1\leq i_{1}<\cdots<i_{m}\leq l$, such that $H_{s_{i_{1}}}\cdots H_{s_{i_{m}}}=H_{w}$ in the 0-Hecke algebra, and $e(\mathbf{t})=m-\ell(w)$. More precisely, we prove (see Theorem 5.6):

We note that one direction of this theorem follows immediately from (1.1). Combining Proposition C and Theorem D yields

Now the equivalence of (i) and (ii) of Theorem A is essentially a reformulation of Theorem E, using some properties of 0-Hecke algebras. The equivalence of (ii) and (iii) is due to Knutson-Miller [KM04, Lemma 3.4 (1)].

The paper is organized as follows. In Section 2, we recall definitions and properties of equivariant $K$-theory and weights of tangent spaces to schemes with $T$-actions. In Section 3 we prove Proposition B. In Section 4, we give a corollary to a theorem by Knutson-Miller on subword complexes [KM04], [KM05]. Our proof of Theorem D relies on this corollary. In Section 5, we apply the material of the previous sections to Kazhdan-Lusztig varieties in order to prove Proposition C and Theorems D and E. In Section 6, we show how to extend these results to $G/P$ and discuss the case of cominuscule Weyl group elements and cominuscule $G/P$.

The related paper [GK21] examines rationally indecomposable weights of the ambient space $V$. Rational indecomposability is a stricter condition than integral indecomposability, so the set of rationally indecomposable weights is contained in the set of integrally indecomposable weights. For this smaller set of weights, [GK21] obtains stronger results. For example, it is shown that the elements of $\Phi(T_{x}Y_{x}^{w})$ which are rationally indecomposable in $I(x^{-1})$ lie in $\Phi(TE_{x}Y_{x}^{w})$. Several results of [GK21] rely on those of this paper.

Let $T=(k^{*})^{n}$ be a torus, and let $\widehat{T}=\operatorname{Hom}(T,k^{*})$ be the character group of $T$. The mapping $\lambda\mapsto d\lambda$ from a character to its differential at $1\in T$ embeds $\widehat{T}$ in the dual ${\mathfrak{t}}^{*}$ of the Lie algebra of $T$. We will usually view $\widehat{T}$ as a subset of ${\mathfrak{t}}^{*}$ under this embedding and express the group operation additively. If $\lambda$ denotes an element of $\widehat{T}$ viewed as an element of ${\mathfrak{t}}^{*}$, then the corresponding homomorphism $T\to k^{*}$ is written as $e^{\lambda}$. The representation ring $R(T)$ is defined to be the free $\mathbb{Z}$-module with basis $e^{\lambda}$, $\lambda\in\widehat{T}$, with multiplication given by $e^{\lambda}e^{\mu}=e^{\lambda+\mu}$.

Let $V$ be a finite dimensional representation of $T$ such that all weights of $V$ have multiplicity one and lie in an open half-space in the real span of the characters of $T$. Denote the set of weights of $T$ on $V$ by $\Phi(V)$ and the set of nonnegative integer linear combinations of elements of $\Phi(V)$ in ${\mathfrak{t}}^{*}$ by $\operatorname{Cone}_{\mathbb{Z}}\Phi(V)$. The dual representation $V^{*}$ has weights $-\Phi(V)$, and the coordinate ring $k[V]=\operatorname{Sym}(V^{*})$ of $V$ has weights $-\operatorname{Cone}_{\mathbb{Z}}\Phi(V)$. Denote $\prod_{\lambda\in\Phi(V)}(1-e^{-\lambda})$ by $\lambda_{-1}(V^{*})$. For $\Phi_{A}\subseteq\Phi(V)$, let $\mathbb{Z}[e^{-\lambda},\lambda\in{\Phi_{A}}]$ denote the subring of $R(T)$ generated over $\mathbb{Z}$ by $e^{-\lambda}$, $\lambda\in\Phi_{A}$.

The map $i_{0}^{*}:K_{T}(V)\to R(T)$ is an isomorphism, which we denote by $i_{V}^{*}$.

The coordinate ring of $V$ is $\operatorname{Sym}(V^{*})=k[x_{\lambda},\lambda\in\Phi(V)]$, a polynomial ring, where $x_{\lambda}$ denotes a vector of $V^{*}$ of weight $-\lambda$. Observe that this polynomial ring is graded by the coordinates $x_{\lambda}$ and also by the weights of the $T$ action. The weights of any $T$-stable subspace of $V$ form a subset of $\Phi(V)$, whose corresponding weight vectors span the subspace. Thus there is a bijection between the $T$-stable subspaces of $V$ and the subsets of $\Phi(V)$. The coordinate ring of a $T$-stable subspace $Z$ is $\operatorname{Sym}(Z^{*})=k[x_{\lambda},\lambda\in\Phi(Z)]$.

Let $Y\to V$ be a $T$-equivariant closed immersion, with $T$-fixed point $x\in Y$ mapping to $0\in V$. Let ${\mathfrak{m}}$ be the maximal ideal of the local ring of $Y$ at $x$. The tangent space to $Y$ at $x$, denoted ${T_{x}Y}$, is defined to be $({\mathfrak{m}}/{\mathfrak{m}}^{2})^{*}$, a vector space over $k$. It embeds naturally in the tangent space to $V$ at $0$, which is isomorphic to $V$ [GW10, (6.2), (6.3)].

Let

Note that the degree one components of $B$ and $C$, which are denoted by $B_{1}$ and $C_{1}$ respectively, are both equal to ${\mathfrak{m}}/{\mathfrak{m}}^{2}$. We will often identify ${T_{x}Y}$ with the affine scheme $\operatorname{Spec}(B)$. The tangent cone to $Y$ at $x$, denoted $TC_{x}Y$, is defined to be $\operatorname{Spec}(C)$. The projection $B\twoheadrightarrow C$ induces an inclusion ${TC_{x}Y}\hookrightarrow{T_{x}Y}$.

Both ${T_{x}Y}$ and ${TC_{x}Y}$ are $T$-stable, and ${TC_{x}Y}\hookrightarrow{T_{x}Y}$ is $T$-equivariant. The coordinate ring $B$ of ${T_{x}Y}$ is equal to $k[x_{\lambda},\lambda\in\Phi({T_{x}Y})]$, with character

This lives in $\widehat{R}(T)$, the set of expressions of the form $\sum_{\lambda\in\widehat{T}}c_{\lambda}e^{\lambda}$. Similarly, we have a formula for the character of $C$.

In this subsection we give a brief review of simplicial complexes and their Euler characteristics.

Recall that an (abstract) simplicial complex on a finite set $A$ is a set $\Delta$ of subsets of $A$, called faces, with the property that if $F\in\Delta$ and $G\subseteq F$ then $G\in\Delta$. The dimension of a face $F$ is $\#F-1$, and the dimension of $\Delta$ is the maximum dimension of a face. A maximal face of $\Delta$ is called a facet. Note that $\Delta=\emptyset$ and $\Delta=\{\emptyset\}$ are distinct simplicial complexes, called the void complex and irrelevant complex respectively. If $\Delta\neq\emptyset$, then $\emptyset$ must be a face of $\Delta$.

The reduced Euler characteristic of $\Delta$ is defined to be $\widetilde{\chi}(\Delta)=\sum_{F\in\Delta}(-1)^{\dim F}$. If $\Delta\neq\emptyset$, so that $\emptyset\in\Delta$, then $\emptyset$ contributes a summand of $-1$ to $\widetilde{\chi}(\Delta)$. From this we see, for example, that $\widetilde{\chi}(\{\emptyset\})=-1$, but $\widetilde{\chi}(\emptyset)=0$.

Suppose that $\Delta\neq\emptyset$ or $\{\emptyset\}$. Denoting the elements of $A$ by $x_{1},\ldots,x_{m}$, the set $A$ can be embedded in $\mathbb{R}^{m}$ by mapping $x_{i}$ to the $i$th standard basis vector of $\mathbb{R}^{m}$. For any face $F$ of $\Delta$, let $|F|$ be the convex hull of its vertices in $\mathbb{R}^{m}$. The geometric realization of $\Delta$ is then defined to be $|\Delta|=\bigcup_{F\in\Delta}|F|$, a topological subspace of $\mathbb{R}^{m}$. If $|\Delta|$ is homeomorphic to a topological space $Y$, then $\Delta$ is called a triangulation of $Y$. In this case, the reduced Euler characteristic of $\Delta$ is equal to the topological reduced Euler characteristic of $Y$. If $Y$ is a manifold with boundary and its boundary $\partial Y$ is nonempty, then there exists a subcomplex of $\Delta$ which is a triangulation of $\partial Y$ [Mau80, Proposition 5.4.4]. This subcomplex is called the boundary of $\Delta$ and denoted by $\partial\Delta$.

For $m\geq 0$, let $B^{m}=\{x\in\mathbb{R}^{m}:\|x\|\leq 1\}$ and $S^{m}=\{x\in\mathbb{R}^{m+1}:\|x\|=1\}$, the $m$-ball and $m$-sphere respectively. Both can be triangulated. In the sequel, when we refer to an $m$-ball or $m$-sphere or their notations, $m\geq 0$, we will mean a triangulation of the object. When we refer to the sphere $S^{-1}$, we will mean the irrelevant complex $\{\emptyset\}$. With these conventions, for $m\geq 0$, $\widetilde{\chi}(B^{m})=0$, $\widetilde{\chi}(S^{m-1})=(-1)^{m-1}$, $\partial B^{m}=S^{m-1}$, and $\partial S^{m-1}=\emptyset$. (Observe that $\partial B^{0}$ is the irrelevant complex, but $\partial S^{m-1}$ is the void complex.) For $\Delta=B^{m}$ or $S^{m}$, define $\widetilde{\chi}^{\circ}(\Delta)=\widetilde{\chi}(\Delta)-\widetilde{\chi}(\partial\Delta)$.

Let $G$ be a semisimple algebraic group, $B$ a Borel subgroup, $B^{-}$ the opposite Borel subgroup, and $T=B\cap B^{-}$ a maximal torus. Let $W=N_{G}(T)/T$, the Weyl group of $G$. Let $S$ be the set of simple reflections of $W$ relative to $B$. The 0-Hecke algebra $\mathcal{H}$ associated to $(W,S)$ over a commutative ring $R$ is the associative $R$-algebra generated by $H_{u}$, $u\in W$, and subject to the following relations: $H_{1}$ is the identity element, and if $u\in W$ and $s\in S$, then $H_{u}H_{s}=H_{us}$ if $\ell(us)>\ell(u)$ and $H_{u}H_{s}=H_{u}$ if $\ell(us)<\ell(u)$. If $\mathbf{q}=(q_{1},\ldots,q_{l})$ is any sequence of elements of $S$, define the Demazure product $\delta(\mathbf{q})\in W$ by the equation $H_{q_{1}}\cdots H_{q_{l}}=H_{\delta(\mathbf{q})}$. Define $\ell(\mathbf{q})=l$ and $e(\mathbf{q})=\ell(\mathbf{q})-\ell(\delta(\mathbf{q}))$.

Let $\mathbf{s}=(s_{1},\ldots,s_{l})$ be a sequence of elements of $S$ and $w\in W$. The subword complex $\Delta(\mathbf{s},w)$ is defined to be the set of subsequences $\mathbf{r}=(s_{i_{1}},\ldots,s_{i_{t}})$, $1\leq i_{1}<\cdots<i_{t}\leq l$, whose complementary subsequence $\mathbf{s}\setminus\mathbf{r}$ contains a reduced expression for $w$. One checks that $\Delta(\mathbf{s},w)$ is a simplicial complex. Subword complexes were introduced in [KM04], [KM05]. We will require that $\mathbf{s}$ contains a reduced expression for $w$.

The following theorem is [KM04, Theorem 3.7]: