---
title: Kazhdan–Lusztig Varieties
url: https://www.emergentmind.com/topics/kazhdan-lusztig-varieties
type: topic
---

# Kazhdan–Lusztig Varieties

Kazhdan–Lusztig varieties are the local intersection spaces that model singularities of Schubert varieties inside flag varieties. In the Weyl-group/flag-variety setting, for \(v\le w\) in Bruhat order, they are the intersections
\[
X_{vw}:=C_v\cap \overline{V_w},
\]
where \(V_w\) is a Schubert cell and \(C_v\) is an opposite Schubert cell; equivalently, in type \(A\), they appear as local patches \(X_w\cap \Omega_v^\circ\) around the torus-fixed point \(e_v\). They are the prototype for a broader geometric formalism in which point-counting, intersection cohomology, Gröbner degenerations, and atlas constructions are all organized around transverse slices to a stratified space [1712.01250].

## 1. Classical definition and local-slice interpretation

Let \(G\) be a split reductive algebraic group over a finite field \(F\), with opposite Borels \(B,B^*\subset G\) and common maximal torus \(T=B\cap B^*\). The Weyl group is
\[
W:=N(T)/T,
\]
and the flag variety is
\[
Y:=G/B.
\]
For each \(w\in W\), the Schubert cell and opposite Schubert cell are
\[
V_w := \{gB\mid g\in BwB\},\qquad
C_w:= \{gB\mid g\in B^*wB\}.
\]
The Bruhat order is determined by
\[
v\le w \iff V_v\subset \overline{V_w},
\]
and the weak rank function is
\[
r_{vw}=\ell(w)-\ell(v).
\]
For \(v\le w\), one sets
\[
U_{vw}:=C_v\cap V_w,\qquad X_{vw}:=C_v\cap \overline{V_w}.
\]
These \(X_{vw}\) are exactly the local pieces usually called Kazhdan–Lusztig varieties in the flag variety setting: intersections of an opposite Schubert cell with a Schubert variety. They are transverse slices to Schubert strata, and they capture the local singularity of \(\overline{V_w}\) at the point \(e_v=wB\) [1712.01250].

In the type \(A\) matrix-patch language, the same local geometry is expressed by the opposite Schubert cell
\[
\Omega_v^\circ=B_-vB/B
\]
and the local model
\[
X_w\cap \Omega_v^\circ.
\]
Kazhdan–Lusztig’s lemma gives
\[
X_w\cap \Omega_v^\circ \cong \mathcal N_{v,w}\times \mathbb A^{\ell(v)},
\]
where
\[
\mathcal N_{v,w}=X_w\cap \Omega_v
\]
is the Kazhdan–Lusztig variety. Thus, after removing a trivial affine factor, \(\mathcal N_{v,w}\) is the essential local model of \(X_w\) at \(e_v\) [1001.3437].

A persistent terminological issue is that atlas papers often use “Kazhdan–Lusztig variety” for an intersection of an opposite Schubert cell with a Schubert variety in a Kac–Moody flag manifold, while local Schubert-geometry papers use it for the finite-type slice \(X_w\cap \Omega_v^\circ\). These are compatible usages: both refer to the same basic construction of a Schubert variety cut by an opposite cell [1910.13017].

## 2. Intersection cohomology, point-counting, and Kazhdan–Lusztig polynomials

The reason Kazhdan–Lusztig varieties are central is that they compute local intersection cohomology. In the general framework of a stratified variety
\[
Y=\bigsqcup_{x\in P}V_x,
\]
one chooses cone-like slices \(C_x\) and forms
\[
U_{xy}:=C_x\cap V_y,\qquad X_{xy}:=C_x\cap \overline{V_y}.
\]
The key IC-restriction property is that the restriction of \(IC_{\overline{V_y}}\) to \(C_x\) is \(IC_{X_{xy}}\). In the flag-variety case this yields
\[
IH^*_{e_v}(\overline{V_w}) \cong IH^*(X_{vw}),
\]
so the local intersection cohomology of a Schubert variety is computed by the Kazhdan–Lusztig variety \(X_{vw}\) [1712.01250].

Point-counting on the open stratum \(U_{vw}=C_v\cap V_w\) produces the classical \(R\)-polynomials:
\[
R_{vw}(q)=|U_{vw}(F)|.
\]
The main abstract theorem in the geometric KLS formalism then identifies the right KLS-polynomials with stalk IC Poincaré polynomials. In the Bruhat case this gives
\[
P_{v,w}(t)=\sum_{i\ge 0} t^i \dim IH^{2i}_{e_v}(\overline{V_w})
       =\sum_{i\ge 0} t^i \dim IH^{2i}(X_{vw}),
\]
together with the classical degree bound
\[
\deg P_{v,w}(t)<\frac{\ell(w)-\ell(v)}{2}.
\]
The same formalism also produces the left KLS-polynomials through opposite Schubert geometry and identifies \(Z\)-polynomials with global IC Poincaré polynomials of Richardson varieties
\[
\overline{C_x}\cap \overline{V_w}
\]
via
\[
Z_{xw}(t)=\sum_{i\ge 0}t^i \dim IH^{2i}(\overline{C_x}\cap \overline{V_w}) .
\]
In this sense, Kazhdan–Lusztig polynomials are local IC invariants of Kazhdan–Lusztig varieties, while \(Z\)-polynomials are global IC invariants of Richardson varieties [1712.01250].

A distinct but related local invariant is the \(h\)-polynomial \(H_{v,w}(q)\) of the associated graded local ring at \(e_v\). For Schubert varieties,
\[
\operatorname{Hilb}\!\left(\operatorname{gr}_{\mathfrak m_{e_v}}\mathcal O_{e_v,X_w},q\right)
=\frac{H_{v,w}(q)}{(1-q)^{\ell(w)}},
\qquad
\operatorname{mult}_{e_v}(X_w)=H_{v,w}(1).
\]
For covexillary \(w\), \(H_{v,w}(q)\) has a positive tableau formula and satisfies
\[
P_{v,w}(q)\preceq H_{v,w}(q),\qquad \deg P_{v,w}(q)=\deg H_{v,w}(q),
\]
but this comparison is genuinely special to the covexillary setting: the paper explicitly notes that \(P_{v,w}(q)\preceq H_{v,w}(q)\) is false in general [1006.1887].

## 3. Explicit equations and Gröbner-theoretic models

One of the most developed aspects of the subject is the passage from geometric slices to explicit defining ideals. In the complete flag variety \(GL_n/B\), the opposite cell around \(e_v\) is coordinatized by a specialized generic matrix \(Z^{(v)}\), and the Kazhdan–Lusztig ideal
\[
I_{v,w}\subseteq \mathbb C[z^{(v)}]
\]
is generated by all minors of size
\[
1+r_{ij}^w
\]
of the southwest submatrices \(Z^{(v)}_{ij}\). The corresponding affine scheme is
\[
\mathcal N_{v,w}\cong \operatorname{Spec}\bigl(\mathbb C[z^{(v)}]/I_{v,w}\bigr),
\]
so Kazhdan–Lusztig ideals are the explicit equations for Kazhdan–Lusztig varieties in local coordinates [1001.3437].

For covexillary \(w\), the geometry is especially well behaved. The essential minors form a Gröbner basis for a generalized antidiagonal term order; the initial ideal is squarefree and equidimensional; and the associated Stanley–Reisner complex is a vertex decomposable shellable ball or sphere. The initial ideal admits a pipe-dream prime decomposition, and multiplicity is counted by facets of that complex, equivalently by flagged semistandard Young tableaux of shape \(\lambda(w)\) with flagging \(\mathbf b(\Theta_{v,w})\) [1001.3437].

Affine type \(A\) admits a parallel but technically more intricate picture. In the affine flag variety \(Fl(V)\cong G(\mathcal K)/I\), opposite Schubert conditions are expressed using infinite periodic matrices, while a preferred reduced word \(Q(w)\) gives Bott–Samelson coordinates in which the Schubert cell \(\mathcal X_\circ^w\) is linearly parametrized. Pulling back the opposite Schubert equations yields explicit generators \(EQ_{w,v}\) for the affine Kazhdan–Lusztig ideal, and these generators form a Gröbner basis. The initial ideal is the Stanley–Reisner ideal of the subword complex \(\Delta(Q(w),v)\), giving a flat degeneration of the affine Kazhdan–Lusztig variety to a combinatorial model [1911.07760].

Type \(C\) has a different explicit realization on a large class of patches. In the symplectic flag variety, for \(v\in C_n\) satisfying
\[
v\ge v_\square \quad\Longleftrightarrow\quad v \text{ is }123\text{-avoiding},
\]
the opposite cell admits coordinates by a partial symmetric matrix, and the type \(C\) Kazhdan–Lusztig variety
\[
\mathcal N_{v,w}=X_w\cap \Omega_v^\circ
\]
is defined by essential southwest minors of that partially filled symmetric matrix. The central theorem states that these essential minors form a Gröbner basis with respect to any diagonal term order. The resulting initial ideals are squarefree Stanley–Reisner ideals of subword complexes, and the associated multidegrees and \(K\)-polynomials are described by type \(C\) pipe dreams. The paper also emphasizes that the \(123\)-avoiding or “small-patch” condition is not merely technical: beyond that regime, the natural essential minors need not form a Gröbner basis [2104.09589].

## 4. Combinatorial and homological invariants of special families

Kazhdan–Lusztig varieties support a broad range of singularity invariants beyond intersection cohomology. In the covexillary case, Gröbner degenerations of Kazhdan–Lusztig ideals provide a positive combinatorial rule for Hilbert–Samuel multiplicity and a formula for the Hilbert series of the local ring. If \(w\) is covexillary, then
\[
\mathrm{mult}_{e_v}(X_w)
\]
equals the number of flagged semistandard Young tableaux of shape \(\lambda(w)\) with row bounds \(\mathbf b(\Theta_{v,w})\), and also admits the determinantal formula
\[
\mathrm{mult}_{e_v}(X_w) = \det\!\left[ \binom{b_i+\lambda_i-i+j-1}{\lambda_i-i+j} \right]_{1\le i,j\le \ell(\lambda)}.
\]
The Hilbert series of the local ring is expressed through flagged set-valued tableaux [1001.3437].

For covexillary Schubert varieties, the local \(h\)-polynomial
\[
H_{v,w}(q)
\]
admits a positive formula by flagged semistandard tableaux and lower-saturated set-valued tableaux, and the paper proves nonnegativity and upper semicontinuity in Bruhat order. It also gives a drift-configuration rule for \(P_{v,w}(q)\), establishing a parallel between local intersection cohomology and the tangent-cone Hilbert series. At \(q=1\), this yields
\[
P_{v,w}(1)\le \operatorname{mult}_{e_v}(X_w)
\]
in the covexillary setting [1006.1887].

A newer homological invariant is Castelnuovo–Mumford regularity. For \(321\)-avoiding permutations \(v\ge w\), the coordinate ring
\[
\mathbb C[\mathbf z^v]/J_{v,w}
\]
is homogeneous, and the paper gives a combinatorial algorithm computing
\[
\operatorname{reg}\big(\mathbb C[\mathbf z^v]/J_{v,w}\big)
\]
from skew excited Young diagrams. The main formula is
\[
\operatorname{reg}\big(\mathbb C[\mathbf z^v]/J_{v,w}\big) = \#D_{\tt zip}^K(v,w)-\ell(w),
\]
equivalently,
\[
\operatorname{reg}\big(\mathbb C[\mathbf z^v]/J_{v,w}\big)
=
\sum_{q\in[m]}\sum_{\mathbf b\in \mathrm{TopCR}(C_q)} \mathrm{room}_{v,w}(\mathbf b).
\]
In the ladder specialization, this becomes a count of unforced elbows in a distinguished zipped lattice-path family. The paper stresses that the \(321\)-avoiding hypothesis is essential: it guarantees homogeneity and the skew-diagram combinatorics on which the algorithm rests [2308.14208].

## 5. Atlases and local models beyond finite type

The concept of a Kazhdan–Lusztig variety also appears as a local model in broader stratified geometry. A stratified variety \((M,\mathcal Y)\) has a Kazhdan–Lusztig atlas with modelling Kac–Moody flag variety \(H/B_H\) if there is a ranked poset injection into \(W_H\) and charts \(U_f\) around minimal strata such that
\[
U_f \xrightarrow{\sim} X^{v(f)}_\circ\cap X_{v(M)}.
\]
In this language, the partial flag manifold \(G/P\) with the projected Richardson stratification admits a Kazhdan–Lusztig atlas: each standard chart
\[
w_1U_-^PP/P
\]
is stratified-isomorphic to an affine Kazhdan–Lusztig variety in the affine flag manifold,
\[
\mathcal X_\circ^{\,w_1t^\lambda w_{0,P}w_0w_2}\cap \mathcal X_{\,t^\lambda w_{0,P}w_0},
\]
and the strata correspond by
\[
w_1U_-^PP/P\cap \Pi_v^w \cong \mathcal X_\circ^{\,w_1t^\lambda w_{0,P}w_0w_2}\cap \mathcal X_{\,vt^\lambda w^{-1}} .
\]
Thus projected Richardson varieties in \(G/P\) are locally modeled by affine Kazhdan–Lusztig varieties [1910.13017].

A second atlas theory concerns toric surfaces. A Kazhdan–Lusztig atlas on a stratified toric surface \(V\) consists of affine neighborhoods of torus-fixed points stratified-isomorphic to
\[
X_o^{w(f)}\cap X_{w(V)}
\]
in the flag variety of a Kac–Moody group, together with a \(T\)-equivariant degeneration
\[
V \rightsquigarrow \bigcup_{f\in V^T} X^{w(f)}\cap X_{w(V)}.
\]
For toric surfaces this reduces to decomposing the moment polygon into quadrilateral moment polytopes of Richardson surfaces, the “pizza” construction. The paper proves that the only toric surfaces admitting equivariant Bruhat atlases are \(CP^2\) and \(CP^1\times CP^1\), while in the Kazhdan–Lusztig-atlas setting there are \(19\) or \(20\) broken toric surfaces admitting simply-laced atlases and at most \(7543\) broken toric surfaces where \(H\) is any Kac-Moody group [1610.04667].

These atlas results clarify a common misconception: a Kazhdan–Lusztig variety need not only be viewed as an isolated local slice inside a finite-dimensional flag variety. In atlas theory it functions as a universal local model for larger stratified spaces.

## 6. Extensions, identifications, and related constructions

Several recent works show that Kazhdan–Lusztig varieties arise in settings that initially look unrelated to Schubert patches. In classical types, the decisive geometric input is an isomorphism of Kazhdan–Lusztig varieties due to Anderson–Ikeda–Jeon–Kawago. For a covexillary Schubert variety \(X_{w_0w}\) and a torus-fixed point \(p_{w_0v}\), there exist inverse Grassmannian elements \(w_{\nu^{-1}},v_{\xi^{-1}}\) in a doubled-rank classical group such that
\[
X_{w_0w}\cap \Omega^\circ_{w_0v}
\cong
X_{w_{\nu^{-1}}}\cap \Omega^\circ_{v_{\xi^{-1}}}.
\]
Consequently,
\[
P_{w_0v,\;w_0w}(q)=P_{v_{\xi^{-1}},\;w_{\nu^{-1}}}(q),
\]
and, after standard symmetries, the covexillary Kazhdan–Lusztig polynomial is identified with a Grassmannian one. This uses the Kazhdan–Lusztig variety as the local singularity model that transfers known formulas from Grassmannian to covexillary Schubert geometry in types \(A,B,C,D\) [2112.06337].

A more dramatic identification appears in quiver geometry. For a type \(A\) quiver \(Q\) with dimension vector \(\mathbf d\), the generalized Zelevinsky map
\[
\zeta_Q: \operatorname{rep}_Q(\mathbf d)\to B^-v_QP_Q/P_Q
\]
restricts, for each rank parameter \(\mathbf r\), to a scheme-theoretic isomorphism
\[
\zeta_Q:\overline{\mathcal O_{\mathbf r}} \xrightarrow{\sim}
\overline{B\,w_Q(\mathbf r)\,P_Q/P_Q}\cap B^-v_QP_Q/P_Q
=
Y^{v_Q}_{w_Q(\mathbf r)}.
\]
In other words, every type \(A\) quiver locus is a Kazhdan–Lusztig variety. Because Kazhdan–Lusztig varieties are known to be normal, Cohen–Macaulay, and to have rational singularities (in characteristic \(0\)), the same follows immediately for type \(A\) quiver loci [2304.10798].

Finally, some adjacent literature uses different global geometric objects whose cohomology realizes characters of Kazhdan–Lusztig basis elements. The varieties \(Y_w(X)\) and their parabolic analogues \(Y_{z,J}(X)\) are Lusztig varieties, not classical Kazhdan–Lusztig varieties; their intersection cohomology realizes characters of KL basis elements, and the paper explicitly distinguishes them from the usual opposite-cell slices [2205.14835, 2212.13497]. This suggests a useful boundary of the term: in the narrow Schubert-geometric sense, Kazhdan–Lusztig varieties remain the local intersections \(X_w\cap \Omega_v^\circ\) or \(C_v\cap\overline{V_w}\); in broader representation-theoretic usage, they sit inside a larger ecosystem of geometric models governed by the same Hecke-theoretic and intersection-cohomological structures.

Source: https://www.emergentmind.com/topics/kazhdan-lusztig-varieties