Geometry · 7

The little Weyl group \(\WX\)

The cone \(\Vc\) of the torus page is a fundamental domain for a reflection group \(\WX\). The Theorem takes \(\WX\)-invariants, and whether \(\WX\)-orbits stay apart inside \(\As/W\) decides whether the basic function generates.

Sections · arrows: “is used by”

DefinitionWhich reflections of \(\LX\) does the geometry of \(X\) remember?

[SV, §2.1, p. 20; Sak, §2.2] Definition

The little Weyl group \(\WX\) is the finite group generated by reflections of \(\LX\otimes\Q\) that has the cone \(\Vc\) as a fundamental domain. Knop's description [Sak, §2.2, p. 9]: \(W\) acts on the set of \(B\)-orbits of maximal rank, transitively, and the stabilizer of the open orbit \(\Xo\) is

\[W(X)=\WX\ltimes W_{\PX},\]

which acts on \(\Xc(X)\) through its quotient \(\WX\).

Two extreme cases. If \(\Vc\) is all of \(\LX\otimes\Q\), then \(\WX=1\); this happens exactly for \(X\) (\(\GL_2/\SL_2\), \(\Aff^n\smallsetminus\{0\}\)). If \(X\) is symmetric, \(\WX\) is the restricted Weyl group \(N_G(S)/Z_G(S)\) of a maximal \(\theta\)-split torus \(S\).

Spherical rootsWhat are the walls of \(\Vc\)?

[SV, §2.1–2.2; Sak, §6.1] Definition

The spherical roots \(\Sigma_X\subseteq\Xc(X)\) are the primitive elements on the extremal rays of the cone \(\{\chi:\ \langle v,\chi\rangle\le0\ \forall v\in\Vc\}\); so

\[\Vc=\{v\in\LX\otimes\Q:\ \langle v,\gamma\rangle\le0\ \ \forall\gamma\in\Sigma_X\}.\]

The normalized spherical root is the positive multiple of \(\gamma\) that is a root of \(G\) or a sum of two orthogonal roots; the normalized spherical roots \(\Delta_X\) are the simple roots of a root system with Weyl group \(\WX\) [SV, Prop. 2.2.1].

Each spherical root has a type. Over \(\bar F\), for a simple root \(\alpha\) and a \(B\)-orbit \(Y\) of maximal rank, \(YP_\alpha/R(P_\alpha)\) is a homogeneous \(\PGL_2\)-variety of one of four types [Sak, §2.2, p. 9]: \(G\) (\(\PGL_2\backslash\PGL_2\)), \(U\) (\(SU\backslash\PGL_2\)), \(T\) (\(T\backslash\PGL_2\)) or \(N\) (\(N(T)\backslash\PGL_2\)). Type \(N\) is excluded by the Theorem's hypothesis (A4) (Hypotheses).

ExamplesWhat does \(\WX\) do to \(\AXs\)?

Table 1. \(\WX\) and its action on \(\dXh\AXs\).
\(X\)\(\WX\)action
\(\GL_2/\SL_2\), \(\Aff^n\smallsetminus\{0\}\)\(1\)—
\(\GL_2/T\)\(\Z/2\)\(y\mapsto y^{-1}\); root \(\epsilon_1-\epsilon_2\), type \(T\)
group \(H\) under \(H\times H\)\(W_H\)diagonally on \((y;y^{-1})\)
\(\GL_{2n}/\Sp_{2n}\)\(S_n\)permutes the pairs; restricted roots \(A_{n-1}\)
\(\GL_{p+q'}/(\GL_p\times\GL_{q'})\)\(W(C_p)\)permutations and \(y_i\mapsto y_i^{-1}\)
\(E_6/F_4\)\(S_3\)restricted roots \(A_2\), each of multiplicity \(8\)
A 60-degree wedge of the plane spanned by a omega1 and b omega6; lattice points with a congruent to b mod 3 are blue, other points grey, and the point 3 omega1 is red.
\(E_6/F_4\): \(\Vc\) is a Weyl chamber of the restricted \(A_2\), and \(\LXp=\{a\omega_1^\vee+b\omega_6^\vee:\ a,b\ge0,\ a\equiv b\bmod 3\}\) (blue); the red point is where the basic function fails to generate (computing r).

\(\WX\) and the surjectivity of \(\Res\)Why invariants under \(\WX\) and not under all of \(W\)?

The Theorem compares \(\HK\simeq\C[\As]^W\) with \(\C[\dXh\AXs]^{\WX}\) through restriction \(\Res\) (the Theorem). Rings of invariants of finite groups are coordinate rings of quotients, so \(\Res\) is the pullback along

\[\iota:\ \dXh\AXs/\WX\longrightarrow\As/W .\]

(surjectivity is geometric) Proved here

\(\Res\) is surjective if and only if \(\iota\) is a closed immersion. In particular, if two points of \(\dXh\AXs\) that are not \(\WX\)-conjugate are \(W\)-conjugate, then \(\Res\) is not surjective.

Proof

A morphism of affine varieties is a closed immersion iff the ring map is surjective; \(\Res\) is that ring map. A closed immersion is injective on points; two \(\WX\)-inequivalent points with the same image in \(\As/W\) violate injectivity.

Sakellaridis's Example 8.0.1 is this situation: \(\WX\) equals the normalizer of \(\AXs\) in \(W\), yet “\(\WX\)-nonconjugate points of \(\AXs\) … can be conjugate under \(W\)” [Sak, p. 64]. For \(E_6/F_4\), two such points are \(y_1\) and \(\sigma y_1\) (computing r). Injectivity on points is necessary, not sufficient: a closed immersion must also be injective on tangent vectors.