Length-zero elements and the lattice chain

How does the rotation supply all the affine-space generators?

1 · Fix the subgroup convention

· The rotation normalizes the Iwahori Proved here

For the upper-triangular Iwahori and the rotation defined on the hub,

$$\rot\Iwa\rot^{-1}=\Iwa,\qquad\rot^n=\unif\,1_n.$$
Proof
  1. Set \(L_0=\OO^n\), \(L_r=\OO^{n-r}\oplus(\unif\OO)^r\) for \(0\le r<n\), and \(L_{nm+r}=\unif^mL_r\).
  2. An integral invertible matrix preserves these lattices precisely when its reduction preserves the flag spanned by the first \(1,2,\ldots,n\) basis vectors; this is the upper-triangular Iwahori.
  3. The rotation sends \(L_j\) onto \(L_{j+1}\) for every \(j\). Conjugating a matrix which fixes every lattice by this rotation still fixes every lattice.
$$\rot\Iwa\rot^{-1}=\Iwa,\qquad\Iwa\rot\Iwa=\rot\Iwa,\qquad\ell(\rot)=0.$$

2 · Compute the lattice

· Every index, including negative indices Proved here

Write \(k=nm+r\), with \(m\in\ZZ\), \(0\le r<n\), using Euclidean division.

$$\boxed{\LLk=\rot^k\OO^n=\unif^m\big(\OO^{n-r}\oplus(\unif\OO)^r\big).}$$ $$L_{k+1}\subset L_k,\qquad\dim_{\resfield}(L_k/L_{k+1})=1.$$
Proof
  1. For \(0\le r<n\), rotate the first \(r\) coordinates to the end, multiplying each by \(\unif\).
  2. Use \(\rot^{nm}=\unif^m1_n\). This is valid for negative \(m\), since the matrices are invertible over \(\FF\).
  3. Exactly one coordinate threshold rises at each step, so the quotient is a one-dimensional residue-field vector space.

3 · Prove the second splitting

· The length-zero complement Proved here

$$\kap|_{\Om}:\Om\xrightarrow{\sim}\ZZ,\qquad\Om=\{\rot^k:k\in\ZZ\},\qquad\Wt=\Wa\rtimes\Om.$$
Proof
  1. The affine Weyl group acts simply transitively on the alcoves in the reduced apartment; consequently \(\Wa\cap\Om=1\). This is the affine Coxeter input of [HR, Lem. 14].
  2. Every extended element takes the base alcove to an alcove. An affine Weyl element carries that alcove back, giving \(\Wt=\Wa\Om\).
  3. The quotient by \(\Wa\) is \(\ZZ\) via \(\kap\). Since \(\kap(\rot)=1\), every class has exactly one representative \(\rot^k\) in \(\Om\).
$$g=(g\rot^{-\kap(g)})\rot^{\kap(g)},\qquad g\rot^{-\kap(g)}\in\Wa.$$

[HR, Lem. 14]

4 · Return to sheaves

Normalization and extension to \(LX\) are supplied by Lemma IIa. In the affine-space application,

$$S_k=\overline{\mathbb Q}_{\ell,L_k}[-k],\qquad G_{\rot^k}=\overline{\mathbb Q}_{\ell,L_k}=S_k[k].$$

Because a length-zero Schubert cell is a point, convolution translates the basic sheaf; IV gives the categorical proof.

← Matrices, translations, and latticesPage 5 of 9The Iwahori–Weyl group in Haines–Rapoport →