The Iwahori–Matsumoto basis and its multiplication table
The precise statement: the indicator functions $T_w$ of the double cosets $IwI$ form a basis of $\cH$, and three rules determine every product. The proof is seven steps of counting cosets, each with its displayed equation. Then the consequences the proof workspace needs: $\one_K$, the idempotent $\eK$, and the two characters.
Notation recap Definition
| symbol | meaning | defined |
|---|---|---|
| $F=\Fq((t))$, $\cO$, $q$ | the local field, its integers, the size of the residue field | Groups §1 |
| $G\supset B\supset T$, $K=G(\cO)$, $I=\ev^{-1}(B(\Fq))$ | split reductive group over $\cO$, Borel, torus; hyperspecial and Iwahori subgroups | Groups §1–2 |
| $\Wt=\Xs\rtimes W=\Waff\rtimes\Om$, $\Saff$, $\len$ | extended affine Weyl group, its Coxeter part and length-zero part, simple affine reflections, length | Groups §3–4 |
| $\cH=\Cc(I\backslash\GF/I)$, $\ast$, $\vol(I)=1$ | the Iwahori–Hecke algebra and its convolution | Hecke algebras §1–2 |
| $\ind(w)=|IwI/I|$, $\ind:\cH\to\C$ | the index, and the integral as an algebra homomorphism | Hecke algebras §3 |
| $\Hzero$, $\one_K$, $\eK$, $\HK$ | finite Hecke algebra, indicator of $K$, spherical idempotent, spherical Hecke algebra | Hecke algebras §5, §5 below |
Sources
Cited facts, with exact references Proved (cited)
| fact | reference | checked for $\SL_2$ | |
|---|---|---|---|
| (F1) | $\GF=\bigsqcup_{w\in\Wt}IwI$ | [IM65, Thm 2.16] | SL₂ I §3 |
| (F2) | $\ind(s)=q$ for $s\in\Saff$ | [IM65, Prop 2.6, §3.1 p. 275] | SL₂ I §4 |
| (F3) | $IsI\cdot IwI=IswI$ if $\len(sw)=\len(w)+1$; $IsI\cdot IwI=IwI\cup IswI$ if $\len(sw)=\len(w)-1$ | [IM65, Prop 2.8] | SL₂ I §4 (for $w=s$), SL₂ II §3 |
| (F4) | $\sigma\in\Om$ normalizes $I$: $I\sigma I=\sigma I=I\sigma$ | [IM65, Prop 2.10, Cor 2.19] | SL₂ III §2 |
| (F5) | $\len(sw)=\len(w)\pm1$; $\len(w^{-1})=\len(w)$; $\len(\sigma w\sigma')=\len(w)$ for $\sigma,\sigma'\in\Om$; $\Waff$ is a Coxeter group on $\Saff$ and $\Wt=\Waff\rtimes\Om$ | [IM65, Lemma 1.5, §1.4, §1.7, Cor 1.16] | Groups §5 |
Two facts about the algebra are proved on the background page and used freely: the structure constants are the counts $c^{z}_{x,y}=\#\{uI\subseteq IxI:u^{-1}z\in IyI\}$ (Hecke algebras, Lemma 2.2), and $\ind$ is an algebra homomorphism (Hecke algebras, Lemma 3.2).
The definition and the Concept Proved here
(the Iwahori–Matsumoto basis) [HKP10, §1.2]
$$T_w:=\one_{IwI}\in\cH\qquad(w\in\Wt),\qquad\text{in particular}\qquad T_e=\one_I,\quad T_s=\one_{IsI}\ (s\in\Saff),\quad T_\sigma=\one_{\sigma I}\ (\sigma\in\Om).$$(the Iwahori–Matsumoto basis and presentation) [IM65, Thm 2.16, Prop 3.2, Prop 3.4, Thm 3.5, Prop 3.8] Proved here
and the same formulas with $T_s$ on the right. In particular $T_s^2=q\,T_e+(q-1)\,T_s$.
The Concept answers the question: what is the multiplication table of the algebra of $I$-double cosets? Answer: it is the group algebra of $\Wt$ deformed in one place, the square of a simple reflection, by the single number $q$. The proof workspace uses (a) and the quadratic relation through the consequences in §5–§6.
Proof
Proof, in seven steps Proved here
Throughout, $s\in\Saff$, $\sigma\in\Om$, $w\in\Wt$, and $c^{z}_{x,y}$ are the structure constants of Hecke algebras, Lemma 2.2. Statement (a) up to the index formula is (F1) together with Hecke algebras, Definition 1.1.
(Step 1: support) Proved here
$$T_sT_w\in\begin{cases}\C\,T_{sw}&\text{if }\len(sw)=\len(w)+1,\\ \C\,T_w+\C\,T_{sw}&\text{if }\len(sw)=\len(w)-1,\end{cases}\qquad (IsI)^{-1}=IsI .$$Proof
$T_sT_w$ is supported on $IsI\cdot IwI$ and is a combination of the $T_z$ (Hecke algebras, Lemma 2.2(ii)); (F3) lists the double cosets in that product. Since $s^2=e$ in $\Wt$, $\dot s^2\in T(\cO)\subset I$, so $\dot s^{-1}\in\dot sI$ and $(I\dot sI)^{-1}=I\dot s^{-1}I=I\dot sI$.
(Step 2: the length-increasing case) Proved here
If $\len(sw)=\len(w)+1$, then
Proof
- By Step 1, $T_sT_w=c\,T_{sw}$ with $c=c^{\dot s\dot w}_{s,w}=\#\{uI\subseteq IsI:\ u^{-1}\dot s\dot w\in IwI\}$. The coset $u=\dot s$ counts, so $c\ge1$.
- Let $uI\subseteq IsI$ with $uI\neq\dot sI$; write $u=x\dot s$ with $x\in I$ (every element of $I\dot sI$ lies in $I\dot s\,I$, and the right factor can be dropped from the coset). Then $y:=\dot s^{-1}x^{-1}\dot s\notin I$, since otherwise $uI=x\dot sI=\dot s(\dot s^{-1}x\dot s)I=\dot sI$. But $y\in\dot s^{-1}I\dot s\subseteq IsI\cdot IsI\subseteq I\cup IsI$ by (F3) applied to $w=s$ (where $\len(ss)=0=\len(s)-1$). Hence $y\in IsI$.
- If this $u$ counted, then $u^{-1}\dot s\dot w=y\dot w\in IwI$. Also $y\dot w\in IsI\cdot IwI=IswI$ by (F3). So $IwI\cap IswI\neq\varnothing$, contradicting (F1) since $sw\neq w$. Therefore only $u=\dot s$ counts: $c=1$.
- Apply the homomorphism $\ind$ to $T_sT_w=T_{sw}$: $\ind(s)\ind(w)=\ind(sw)$, and $\ind(s)=q$ by (F2).
(Step 3: the index formula) [IM65, Prop 3.2] Proved here
Proof
- For $w\in\Waff$, induct on $\len(w)$: $\ind(e)=|I/I|=1$; if $\len(w)\ge1$ choose a reduced expression and write $w=sw'$ with $\len(w')=\len(w)-1$; Step 2 gives $\ind(w)=q\,\ind(w')=q\cdot q^{\len(w)-1}$.
- For $w=w'\sigma$ with $w'\in\Waff$, $\sigma\in\Om$: by (F4), $Iw'\sigma I=Iw'I\sigma$, and $zI\mapsto z\sigma I=zI\sigma$ is a bijection of $\GF/I$ carrying $Iw'I/I$ onto $Iw'\sigma I/I$. So $\ind(w'\sigma)=\ind(w')=q^{\len(w')}=q^{\len(w)}$ by (F5).
(Step 4: the quadratic relation) [IM65, Prop 3.4(i)] Proved here
Proof
- By Step 1 with $w=s$, $T_s^2=a\,T_e+b\,T_s$ with $a,b\in\N$.
- $a=c^{e}_{s,s}=\#\{uI\subseteq IsI:\ u^{-1}\in IsI\}=|IsI/I|=q$, because $(IsI)^{-1}=IsI$ (Step 1) and (F2).
- Apply $\ind$: $q^2=a+bq=q+bq$, so $b=q-1$.
(Step 5: the length-decreasing case) [IM65, Cor 3.6] Proved here
If $\len(sw)=\len(w)-1$, then
Proof
Put $w'=sw$; then $w=sw'$ with $\len(w)=\len(w')+1$, so $T_w=T_sT_{w'}$ by Step 2. Hence, by Step 4,
$$T_sT_w=T_s^2T_{w'}=q\,T_{w'}+(q-1)\,T_sT_{w'}=q\,T_{sw}+(q-1)\,T_w .$$(Step 6: the length-zero elements) [IM65, p. 276] Proved here
Proof
By (F4), $I\sigma I=\dot\sigma I$ is a single coset. So $(T_\sigma T_w)(z)=\#\{uI\subseteq\dot\sigma I:u^{-1}z\in IwI\}=\one_{\dot\sigma IwI}(z)$, and $\dot\sigma IwI=I\dot\sigma\dot wI$ since $\dot\sigma I=I\dot\sigma$. For the right-hand formula, $(T_wT_\sigma)(z)=\#\{uI\subseteq IwI:\ u^{-1}z\in\dot\sigma I\}=\#\{uI\subseteq IwI:\ uI=z\dot\sigma^{-1}I\}=\one_{IwI\dot\sigma}(z)=\one_{Iw\sigma I}(z)$. The last two formulas follow: $\sigma\sigma'\in\Om$, and $T_\sigma T_s=T_{\sigma s}=T_{(\sigma s\sigma^{-1})\sigma}=T_{\sigma s\sigma^{-1}}T_\sigma$ with $\sigma s\sigma^{-1}\in\Saff$ by (F5).
(Step 7: generation, braid relations, presentation, right-hand versions) [IM65, Thm 3.3, Prop 3.4(ii), Thm 3.5, Prop 3.8] Proved (cited)
Statement (c) of the Concept, and the right-hand versions of (b).
Proof
- Generation. For $w=s_{i_1}\cdots s_{i_r}\sigma$ reduced, each suffix $s_{i_k}\cdots s_{i_r}$ has length $r-k+1$, so Step 2 applies $r$ times and Step 6 once: $T_w=T_{s_{i_1}}\cdots T_{s_{i_r}}T_\sigma$. This is [IM65, Thm 3.3].
- Braid relations. For $s\neq t$ in $\Saff$ with $(st)^{m}=1$, $m<\infty$, the two alternating words of length $m$ are reduced expressions of one element $w_{st}\in\Waff$ (F5), so by step 1 both $T_sT_tT_s\cdots$ and $T_tT_sT_t\cdots$ equal $T_{w_{st}}$ [IM65, Prop 3.4(ii)].
- Presentation. Let $\mathcal A$ be the algebra generated by symbols $\Delta_s$ subject to the quadratic and braid relations; $\Delta_s\mapsto T_s$ is a surjection $\mathcal A\to\Haff$ by step 1. For each $w\in\Waff$ put $\Delta_w=\Delta_{s_{i_1}}\cdots\Delta_{s_{i_r}}$ along a reduced expression; by Matsumoto's theorem (any two reduced expressions are joined by braid moves) this does not depend on the expression [IM65, Prop 1.15]. The relations give $\Delta_s\Delta_w\in\C\Delta_{sw}+\C\Delta_w$ in both length cases, so the $\Delta_w$ span $\mathcal A$; they map onto the basis $T_w$, hence $\mathcal A\simeq\Haff$ [IM65, Thm 3.5]. The decomposition $\cH=\bigoplus_\sigma\Haff T_\sigma$ is (a) and Step 6 [IM65, Prop 3.8].
- Right-hand versions. $\iota(f)(x)=f(x^{-1})$ is an anti-automorphism of $\cH$ ($\iota(f\ast g)=\iota(g)\ast\iota(f)$, by the substitution $y\mapsto x^{-1}y^{-1}$ in the convolution) with $\iota(T_w)=T_{w^{-1}}$, since $(IwI)^{-1}=Iw^{-1}I$. Applying $\iota$ to the formulas for $T_sT_{w^{-1}}$ and using $\len(w^{-1}s)=\len(sw^{-1})$... precisely $\len((sw^{-1})^{-1})=\len(ws)$ by (F5), gives $T_wT_s=T_{ws}$ or $q\,T_{ws}+(q-1)T_w$ according as $\len(ws)=\len(w)\pm1$.
History: 1965 · For $\SL_2$ every step is a count of edges on SL₂ II.
Consequences Proved here
($\one_K$, $[K:I]$ and the spherical idempotent) Proved here
and $\eK\ast\varphi=\varphi$ for every left $K$-invariant function $\varphi$ on a $\GF$-set. $\eK$ is the unit of $\HK=\eK\cH\eK$, and $\eK\neq T_e$ unless $W=1$.
Proof
$K=\bigsqcup_{w\in W}IwI$ (Hecke algebras, Lemma 5.1) gives the first formula; applying $\ind$ and Step 3 gives the second, which agrees with Groups, Lemma 2.2. For left $K$-invariant $\varphi$, $(\one_K\ast\varphi)(x)=\sum_{y\in K/I}\varphi(y^{-1}x)=[K:I]\varphi(x)$; with $\varphi=\one_K$ this is the third formula, and dividing by $[K:I]$ gives the rest. $\HK=\eK\cH\eK$ is the source page, Lemma 4.1. If $W\neq1$ then $\one_K\neq T_e$ since $K\neq I$.
(every $T_w$ is invertible) Proved here
$$T_s^{-1}=q^{-1}\big(T_s-(q-1)T_e\big),\qquad T_\sigma^{-1}=T_{\sigma^{-1}},\qquad T_w^{-1}=T_\sigma^{-1}T_{s_{i_r}}^{-1}\cdots T_{s_{i_1}}^{-1}.$$Proof
The quadratic relation reads $T_s\big(T_s-(q-1)T_e\big)=q\,T_e$; Step 6 gives $T_\sigma T_{\sigma^{-1}}=T_e$; Step 7 writes $T_w$ as a product.
Further consequences, cited. The finite Hecke algebra $\Hzero$ is the subalgebra spanned by $T_w$, $w\in W$ (Hecke algebras §5). The Bernstein presentation $\cH\simeq\C[\Xs]\otimes\Hzero$ and the centre $\Zc=\C[\Xs]^W$ are on the source page §3 [HKP10, Lemma 1.7.1, Lemma 2.3.1], and the Satake isomorphism $\HK\simeq\Zc$ with $h=\eK z$ on §4 [HKP10, §4.3, (4.6.1)].
The two one-dimensional modules: index and sign Proved here
(index and sign characters) [IM65, p. 276 and §3.2, p. 279] Proved here
There are algebra homomorphisms $\ind,\eps:\cH\to\C$ determined by
On $\one_K$ they take the values
Proof
- $\ind$ is the integral, an algebra homomorphism by Hecke algebras, Lemma 3.2, with $\ind(T_w)=q^{\len(w)}$ by Step 3. (Directly: $q^2=q+(q-1)q$ and $q^m=q^m$ verify the relations.)
- $\eps$: by the presentation (Step 7) it suffices to check the relations on the generators. Quadratic: $(-1)^2=1=q+(q-1)(-1)$. Braid: both sides are $(-1)^m$. Twist: $\eps(T_\sigma T_sT_\sigma^{-1})=-1=\eps(T_{\sigma s\sigma^{-1}})$. Then $\eps(T_w)=(-1)^{r}$ for $w=s_{i_1}\cdots s_{i_r}\sigma$ by step 1 of Step 7. This is the remark of [IM65, §3.2].
- $\eps(\one_K)=\sum_{w\in W}(-1)^{\len(w)}$ by . If $W\neq1$, choose a simple reflection $s\in S$; $w\mapsto ws$ is an involution of $W$ without fixed points and changes $\len$ by $\pm1$ (F5), so it pairs each even-length element with an odd-length one and the sum vanishes. If $W=1$ then $\one_K=T_e$.
Where it is used: in the group case $X=H$ of the proof workspace, the Iwahori-generated part of $\cH$ is the two-sided ideal of $\eK$; the ideal lies in $\ker\eps$, which does not contain the unit $T_e$ when $W\neq1$. At the spherical level the same $\one_K$ is the unit of $\HK$, and no character of $\HK$ can kill it. That contrast is the group case at the Iwahori level [Draft].
Dictionary to the proof workspace's wording Context
| workspace wording | here |
|---|---|
| $\mathcal H_I:=C_c(I\backslash G(F)/I)$, $T_w:=\mathbf 1_{IwI}$ ($w\in W_{\mathrm{ext}}$) | $\cH$, $T_w$ ($w\in\Wt$); the workspace writes $W_{\mathrm{ext}}$ for $\Wt$ |
| $\mathbf 1_K=\sum_{w\in W}T_w$, $[K:I]=|G(\mathbb F_q)/B(\mathbb F_q)|=\sum_{w\in W}q^{\ell(w)}$ | |
| $e_K:=\operatorname{vol}(K)^{-1}\mathbf 1_K=[K:I]^{-1}\mathbf 1_K$, $e_K\ast e_K=e_K$, $e_K\ast\varphi=\varphi$ for $K$-invariant $\varphi$ | |
| “the Iwahori–Matsumoto presentation: generated by $T_s$ and $T_\omega$ with the braid relations and $(T_s+1)(T_s-q)=0$” | the Concept (b)–(c); $(T_s+1)(T_s-q)=T_s^2-(q-1)T_s-q$ |
| “$\varepsilon(T_s)=-1$, $\varepsilon(T_\omega)=1$ defines an algebra homomorphism with $\varepsilon(T_w)=(-1)^{\ell(w)}$” | |
| “$\varepsilon(\mathbf 1_{K_H})=\sum_{w\in W}(-1)^{\ell(w)}=0$ because $W\neq1$ has as many even as odd elements” | , step 3 |
Status and scope Context
- Proved here Steps 1–6, the consequences and the characters are proved on this page from (F1)–(F5).
- Proved (cited) (F1)–(F5) are proved in [IM65] for Chevalley groups and checked by hand for $\SL_2$ and $\PGL_2$ on the example pages. For a split reductive group over $\cO$ that is not of this form (for instance $\GL_n$, or a group whose derived group is not simply connected and not adjoint) the same facts are standard and are used as known in [HKP10, §1.2, §7.1]; this site has not re-derived them from [BT72].
- Proved (cited) The presentation (Step 7, step 3) relies on Matsumoto's theorem on reduced words [IM65, Prop 1.15], not reproved here.
- No gaps are known in the statement or proof. What the Concept does not give: the structure of $\cH$ as a module over its centre and the classification of its simple modules, which need the Bernstein presentation (source page).