---
title: Elliptic Elements in Weyl Groups
url: https://www.emergentmind.com/topics/elliptic-element
type: topic
---

# Elliptic Elements in Weyl Groups

An elliptic element, in the representation-theoretic setting of Weyl groups and reductive groups, is an element \(w\) with no nonzero fixed vectors in the reflection representation; equivalently, \(1\) is not among its eigenvalues in the standard reflection representation [1007.5040][1109.5487]. For elliptic conjugacy classes of minimal length, this condition governs a substantial geometric structure: it controls varieties defined by relative position of Borel subgroups, determines associated unipotent classes, produces affine orbit spaces and generalized Steinberg sections, and constrains the order of representatives in semisimple algebraic groups [1007.5040][1012.2074][1103.1769][1109.5487]. The same phrase also appears in other areas, but those meanings are distinct.

## 1. Definition and equivalent formulations

Let \(W\) be the Weyl group of a connected reductive group \(G\), and let \(V\) be the reflection representation of \(W\). For \(w\in W\), the fixed-point dimension is
\[
\dim V^w=\dim\{v\in V\mid wv=v\}.
\]
An element \(w\), or its conjugacy class \(C\subset W\), is called elliptic if
\[
\dim V^w=0.
\]
Thus ellipticity means that \(w\) has no nonzero fixed vectors in the reflection representation [1007.5040].

The same condition is described in equivalent spectral language: an element \(w\) of a Weyl group is elliptic if it has no eigenvalue \(1\) in the standard reflection representation. Equivalently, \(1\) is not among the eigenvalues of \(w\) acting on the real span of the root system [1109.5487]. In the untwisted setting, this is also equivalent to saying that the conjugacy class of \(w\) does not meet any proper parabolic subgroup \(W_J\); in the twisted setting one replaces ordinary conjugacy by \(\epsilon_\Delta\)-conjugacy or \(\bullet\)-conjugacy and requires avoidance of any proper \(\epsilon_\Delta\)-stable or \(\bullet\)-stable parabolic subgroup [1103.1769][1304.4463].

A second datum is the length function \(l:W\to\mathbb N\). For a conjugacy class \(C\), one writes
\[
C_{\min}=\{w\in C\mid l(w)=d_C\},\qquad d_C=\min_{x\in C}l(x).
\]
The minimal-length condition is structurally decisive throughout the theory. Several of the geometric statements below are stated only for \(w\in C_{\min}\), and one paper explicitly notes that the key statements can fail if this hypothesis is dropped [1007.5040][1103.1769].

## 2. Relative position and the basic varieties attached to \(w\)

Let \(\mathcal B\) denote the variety of Borel subgroups of \(G\). For each \(w\in W\), there is a corresponding \(G\)-orbit
\[
O_w\subset \mathcal B\times \mathcal B,
\]
and two Borel subgroups \(B_1,B_2\) are said to be in relative position \(w\) if \((B_1,B_2)\in O_w\). If \(g\in G\) and \(B\in\mathcal B\), the condition relevant to elliptic elements is
\[
(B,gBg^{-1})\in O_w.
\]
This relative-position condition is the basic bridge from Weyl-group combinatorics to the geometry of conjugacy classes in \(G\) [1007.5040].

For a unipotent class \(\gamma\subset G\), one defines
\[
B_w^\gamma=\{(g,B)\in \gamma\times \mathcal B\mid (B,gBg^{-1})\in O_w\}.
\]
The group \(G\) acts on \(B_w^\gamma\) by conjugation:
\[
x\cdot(g,B)=(xgx^{-1},xBx^{-1}).
\]
For fixed \(g\in\gamma\), one also considers
\[
\mathcal B_g=\{B\in\mathcal B\mid (B,gBg^{-1})\in O_w\},
\]
which carries a natural action of the centralizer \(Z(g)\) [1007.5040].

Closely related constructions appear in two companion settings. In the \(q=1\) setting one has
\[
B_w=\{(g,B)\in D\times \mathcal B;\ (B,gBg^{-1})\in \mathcal O_w\},
\]
while for \(q>1\) one has the Deligne–Lusztig-type variety
\[
X_w=\{B\in\mathcal B;\ (B,F(B))\in \mathcal O_w\},
\]
with \(F\) the Frobenius map. The same formalism extends to disconnected groups and twisted conjugacy classes: for a class \(c\subset \Delta\) and \(w\in C_{\min}\),
\[
\mathfrak B_w=\{(g,B)\in c\times \mathcal B\ ;\ (B,gBg^{-1})\in \mathcal O_w\}.
\]
These variants retain the same conceptual role: they encode conjugacy-theoretic data by means of a minimal Bruhat position [1012.2074][1304.4463].

## 3. Homogeneity for elliptic minimal-length classes

A central theorem states that if \(C\in\underline W\) is elliptic, \(w\in C_{\min}\), and \(\gamma=\Phi(C)\) is the unipotent class associated to \(C\) by Lusztig’s map \(\Phi\), then
\[
B_w^\gamma
\]
is a single \(G\)-orbit. Equivalently, \(B_w^\gamma\) is a homogeneous \(G\)-space [1007.5040]. In the same framework, if \(g\in\gamma\), then \(\mathcal B_g\) is a single orbit under the conjugation action of \(Z(g)\). The theorem assumes that \(G\) is reductive over an algebraically closed field whose characteristic is not a bad prime for \(G\) [1007.5040].

This homogeneity statement gives a rigid geometric realization of the passage from elliptic conjugacy classes in \(W\) to unipotent classes in \(G\). It says that once \(w\) is elliptic and minimal in its conjugacy class, all compatible pairs \((g,B)\) arise from one another by global conjugation, and for fixed \(g\) the residual ambiguity is already exhausted by the centralizer action [1007.5040].

The proof is case-by-case. For type \(A\), it is described as easy. For types \(B,C,D\), the argument reduces to \(G=\operatorname{Is}(V)\) and uses explicit adapted bases together with detailed linear-algebraic analysis. For exceptional types, the result is reduced to computation using character tables of Hecke algebras, Green functions, and computer algebra. In the good-characteristic setting, one of the controlling formulas is
\[
(w_i^t,w_j^t)=
\begin{cases}
\operatorname{sgn}(j-i)\displaystyle \binom{|j-i|+\pi-1}{|j-i|-\pi}, & |j-i|\ge \pi,\\[1.5ex]
0, & |j-i|<\pi,
\end{cases}
\]
for the relevant parameter \(\pi\); these pairing identities are used to force the orbit structure into the required homogeneous form [1007.5040].

A parallel homogeneity phenomenon appears in the disconnected and twisted setting. If \(G\) is almost simple and \(|G/G^0|\le 2\), and in exceptional type one assumes additionally that \(G=G^0\) and \(p=0\) or \(p\) is good, then for any distinguished \(G\)-conjugacy class \(c\subset \Delta\) there exists an elliptic class \(C\in W_\Delta\) such that \(C\dashv c\), meaning that for \(w\in C_{\min}\) the variety \(\mathfrak B_w\) is a single \(G^0\)-orbit [1304.4463].

## 4. Affine orbit spaces and generalized Steinberg sections

When \(w\) is \(\bullet\)-elliptic and has minimal length in its \(\bullet\)-conjugacy class, the orbit space of the natural \(G\)-action acquires an affine structure. In the \(q=1\) case, assuming additionally that \(G\) is semisimple, every isotropy group for the \(G\)-action on \(B_w\) is trivial; for the auxiliary finite covering \(\widetilde B_w\), every stabilizer is isomorphic to a subgroup of a finite torus \(T_w\), hence is a finite diagonalizable group. Moreover, \(B_w\) and \(\widetilde B_w\) are affine varieties. The analogous affineness for \(X_w\) and \(\widetilde X_w\) in the \(q>1\) case is also recorded [1012.2074].

In the classical types \(A_n,B_n,D_n\) with \(\bullet=1\), the orbit space admits an explicit form. If \(w\) is elliptic and minimal in its conjugacy class, then
\[
G\backslash B_w \cong k^{\,l(w)},\qquad
G\backslash \widetilde B_w \cong T_w\backslash k^{\,l(w)}
\]
for a natural action of the finite group \(T_w\). Thus in these cases the quotient is not merely affine; it is an affine space, or an affine space modulo a finite diagonalizable group [1012.2074].

A closely related construction generalizes Steinberg’s cross-section. For an elliptic element \(w\in W\) of minimal length in its conjugacy class, the map
\[
\Xi_w: U\times (U\dot w)\to U\dot w U,\qquad (u,z)\mapsto uz\pi(u)^{-1},
\]
is injective, and under the stated hypotheses is bijective. This yields a canonical slice inside \(U\dot w U\), and the orbit space \(U\backslash U\dot w U\) is naturally an affine space of dimension \(l(w)\) [1103.1769].

The corresponding closed subvariety is denoted \(\Sigma\), and in the untwisted case it is realized as
\[
\Sigma=\dot w\,U^w.
\]
It is isomorphic to affine space of dimension \(d=l(w)\). The elliptic conjugacy class \(C\subset W\) determines a unipotent class \(\gamma\) such that
\[
\operatorname{codim}_G(\gamma)=l(w),
\]
and
\[
\Sigma\cap \gamma
\]
is a finite set. In the Coxeter case this recovers Steinberg’s theorem: the section meets the regular unipotent class in exactly one point [1103.1769].

## 5. Representatives in semisimple algebraic groups and the spin invariant

Let \(G\) be a semisimple algebraic group with Weyl group \(W=N_G(T)/T\), and let \(w\in W\) be elliptic of order \(d\). The representative-order problem asks for the order of an element \(g\in N_G(T)\) mapping to \(w\). The basic theorem is that for elliptic \(w\), all representatives in \(N_G(T)\) have the same order, and that order is always either \(d\) or \(2d\) [1109.5487].

This dichotomy is encoded by the spin invariant. One says that \(w\) has spin \(1\) if every representative \(g\) has order \(d\), and spin \(-1\) if every representative \(g\) has order \(2d\). If \(g_0\in N_0\) is a standard representative, then for elliptic \(w\) one has
\[
g_0^d\in T_0,
\]
and \(g_0^d\) is called the spin signature. Since \(T_0\) is abelian and all its elements have order dividing \(2\), the order of \(g_0\) is read off from whether \(g_0^d\) is trivial [1109.5487].

Several general facts are established. If \(w\) is elliptic of odd order, then \(w\) has spin \(1\). If \(w\) is linked to \(-I\) and \(G\) is simple, then \(w\) has spin \(1\). If \(w\) and \(w^r\) are both elliptic, then they have the same spin and spin signature [1109.5487].

The global classification is especially restrictive. If \(G\) is simple and \(w\) is elliptic of order \(d\), then a representative \(g\) of \(w\) has order \(d\) in \(G\) for all types except \(C_n\) and \(F_4\) [1109.5487]. More precisely, every elliptic \(w\in W(C_n)\) has universal spin \(-1\), while in type \(F_4\) exactly one elliptic conjugacy class, the class with Carter diagram \(A_3\times A_1\), has spin \(-1\). In all other simple types, elliptic elements lift without increasing order [1109.5487].

## 6. Twisted ellipticity and distinguished conjugacy classes

For a possibly disconnected reductive algebraic group \(G\) with identity component \(G^0\), a connected component \(\Delta\subset G\) acts on the Weyl group \(W\) of \(G^0\) via an automorphism
\[
\epsilon_\Delta:W\to W
\]
preserving the length function. The relevant conjugacy classes are the \(\epsilon_\Delta\)-conjugacy classes, i.e. the orbits for
\[
w\mapsto x^{-1}w\,\epsilon_\Delta(x).
\]
A class \(C\in W_\Delta\) is elliptic if it does not meet any proper parabolic subgroup stable under \(\epsilon_\Delta\); explicitly, if \(J\subsetneq I\) is proper and \(\epsilon_\Delta(J)=J\), then
\[
C\cap W_J=\varnothing.
\]
This is the twisted analogue of the fixed-point-free condition in the ordinary Weyl-group setting [1304.4463].

On the group side, a \(G\)-conjugacy class \(c\subset \Delta\) is called distinguished if for any \(g\in c\),
\[
Z_G(g)\big/\bigl(G^0\cap Z_G(g)\bigr)
\]
is a unipotent group [1304.4463]. The relation
\[
C\dashv c
\]
is defined by the condition that for \(w\in C_{\min}\), the incidence variety
\[
\mathfrak B_w=\{(g,B)\in c\times \mathcal B\ ;\ (B,gBg^{-1})\in \mathcal O_w\}
\]
is a single \(G^0\)-orbit. Thus the twisted elliptic class controls the geometry of the distinguished class through a minimal Bruhat position [1304.4463].

The resulting theorem states that every distinguished class in the component \(\Delta\) comes from some elliptic twisted Weyl-group class. In the classical groups, the proof proceeds through explicit linear-algebra models: Jordan-block constructions, lines \(L_1,L_2,\dots,L_{\sigma+\kappa}\), transitivity of isometry-group actions, and explicit stabilizer calculations. In exceptional groups, the argument uses the classification of elliptic classes and distinguished classes together with Green-function computations and earlier results [1304.4463]. This suggests a precise parallel between “elliptic” on the Weyl-group side and “distinguished” on the group side.

## 7. Distinct meanings in other areas

The phrase elliptic element is not unique to Weyl-group theory. In the Cremona group \(\mathrm{Bir}(S)\), it is defined dynamically through the action on the Picard–Manin hyperbolic space \((S)\): an element \(f\in\mathrm{Bir}(S)\) is elliptic if the corresponding isometry on \((S)\) is elliptic, equivalently if there exists an ample divisor \(H\) on \(S\) such that the sequence \(\{\deg_H(f^n)\}_{n\in\mathbb N}\) is bounded [1802.08485]. This is unrelated to the reflection-representation definition in Weyl groups.

In free probability, an elliptic element is a non-self-adjoint operator of the form
\[
z=x+iy,
\]
where \(x\) and \(y\) are freely independent semicircular elements, with variances parametrized by \(s\) and \(t\). In that setting the term interpolates between self-adjoint and circular behavior and is tied to the elliptic law for Brown measures [2007.06100].

These alternate usages are mathematically independent of the Weyl-group notion. In the representation theory of reductive groups, however, the term has a sharply defined role: it singles out those Weyl-group elements whose minimal-length representatives control especially rigid orbit geometry, affine quotient structures, generalized cross-sections, and the behavior of lifts to algebraic groups [1007.5040][1012.2074][1103.1769][1109.5487].

Source: https://www.emergentmind.com/topics/elliptic-element