arXiv reader

Characteristic polynomials and some combinatorics for finite Coxeter groups

Shoumin Liu and Yuxiang Wang

arXiv
2504.19468v1 · PDF · 28 April 2025
Categories
math.RT
Licence
CC0 1.0 The site shows the text.

Abstract

Let W be a finite Coxeter group with Coxeter generating set S=\{s_1,\ldots,s_n\}, and \rho be a complex finite dimensional representation of W. The characteristic polynomial of \rho is defined as

d(S,\rho)=\det[x_0I+x_1\rho(s_1)+\cdots+x_n\rho(s_n)],

where I is the identity operator. In this paper, we show the existence of a combinatorics structure within W, and thereby prove that for any two complex finite dimensional representations \rho_1 and \rho_2 of W, d(S,\rho_1)=d(S,\rho_2) if and only if \rho_1 \cong \rho_2.

Items 63

Theorem

Theorem 1.1

UOY4

Let W be a finite Coxeter group with S=\{s_1,\ldots,s_n\} as its Coxeter generating set.

  1. (a) There is an order of all conjugacy classes of W: C_1,\ldots,C_r and a sequence \alpha_1,\ldots,\alpha_r\in\mathbb{N}^n such that

    \begin{aligned} &\{g=\phi(w):{\rm sig}(w)=\alpha_i\}\cap C_i\neq\varnothing\quad\text{for}\ i\in[r], \\ &\{g=\phi(w):{\rm sig}(w)=\alpha_i\}\cap C_j=\varnothing\quad\text{for}\ i,j\in[r]\ \text{and}\ j>i. \end{aligned}
  2. (b) For two finite dimensional complex linear representations \rho_1 and \rho_2 for W, d(S,\rho_1)=d(S,\rho_2) if and only if \rho_1\cong\rho_2.

Definition

Definition 2.1

5W7L

Let M=(m_{ij})_{1\leq i,j\leq n} be a symmetric n\times n matrix with entries from \mathbb{N}\cup \{\infty\} such that m_{ii}=1 for all i\in[n] and m_{ij}>1 whenever i\neq j. The Coxeter group of type M is the group

W(M)=\langle\{s_1,\ldots,s_n\}\ |\ \{(s_is_j)^{m_{ij}}=1,\ i,j\in [n],m_{ij}<\infty\}\rangle.

S=\{s_1,\ldots,s_n\} is called the Coxeter generating set. If no confusion is imminent, we write W instead of W(M). The pair (W,S) is called the Coxeter system of type M.

Theorem

Theorem 2.2

NCI1

Let (W,S) be a Coxeter system.

  1. (a) The mapping w\mapsto\sigma_w given by \sigma_w=\sigma_{r_1}\cdots\sigma_{r_q} whenever w=s_{r_1}\cdots s_{r_q} \\ \in W with r_j\in [n] (j\in[q]) defines a linear representation of W on V, which is called the reflection representation (or the geometric representation) of W, denoted by \rho_{\it ref}.

  2. (b) The reflection representation of W is faithful.

Definition

Definition 2.3

MUCZ

The Coxeter matrix M=(m_{ij})_{1\leq i,j\leq n} is often described by a labelled graph \Gamma(M) whose vertex set is [n] and in which two nodes i and j are joined by an edge labelled m_{ij} if m_{ij}>2. If m_{ij}=3, then the label 3 of of the edge \{i,j\} is often omitted. If m_{ij}=4, then instead of the label 4 at the edge \{i,j\} one often draws a double bond. This labelled graph is called the Coxeter diagram of M. We say that the Coxeter system (W,S) is irreducible if the Coxeter diagram \Gamma is connected.

Definition

Definition 2.4

YDBX

Let (W,S) be a Coxeter system, J\subseteq S be any subset and consider the subgroup W_J\coloneqq\langle J\rangle\leqslant W. Then W_J is called a parabolic subgroup of W. We call W_J a proper parabolic subgroup if J\neq S.

Proposition

Proposition 2.5

J9EW

Let Coxeter system (W,S) have Coxeter diagram \Gamma, with connected components \Gamma_1,\ldots,\Gamma_r, and let S_1,\ldots,S_r be the corresponding subsets of S. Then W is the direct product of the parabolic subgroups W_{S_1},\ldots,W_{S_r}, and each Coxeter system (W_{S_i},S_i) is irreducible.

Theorem

Theorem 2.6

OGTU

An irreducible Coxeter group is finite if and only if its Coxeter diagram occurs in Table table1 (see Section 4) or Table table2 (see Section 5), where the types in the former are called classical types while the types in the latter are called exceptional types.

Corollary

Corollary 2.7

OQ9E

A Coxeter group G is finite if and only if G\cong W(M_1)\times W(M_2)\times\cdots\times W(M_k) where each M_i occurs in Table table1 or Table table2.

Theorem

Theorem 3.1

9301

Let G be any finitely generated group with generating set S=\{s_1,\ldots,s_n\}, and \rho_1,\rho_2 \in {\mathcal R}(G). If d(S,\rho_1)=d(S,\rho_2), then for each given \alpha\in\mathbb{N}^n, we have

\sum_{{\rm sig}(w)=\alpha} \chi_{\rho_1}(w)=\sum_{{\rm sig}(w)=\alpha} \chi_{\rho_2}(w)

where \chi_{\rho_i} denotes the character of \rho_i.

Definition

Definition 3.2

TWKG

Let G be any finitely generated group with finite number of conjugacy classes C_1,\ldots,C_r, and S=\{s_1,\ldots,s_n\} be the generating set. Given a signature \alpha =(a_1,\ldots,a_n), E_{\alpha}\coloneqq\{u\in M(S):{\rm sig}(u)=\alpha\}. The signature vector of \alpha is

V_{\alpha}=(|E_{\alpha}\cap \phi^{-1}(C_1)|,\ldots,|E_{\alpha}\cap \phi^{-1}(C_r)|).
Definition

Definition 3.3

3OHN

Let G,S and C_i be as in Definition evdef. A sequence \alpha_1,\ldots,\\\alpha_r of signatures is called an independent signature sequence (abbreviated as ISS) if V_{\alpha_1},\ldots,V_{\alpha_r} is linearly independent. Suppose \alpha_1,\ldots,\alpha_r is an ISS, the corresponding independent signature matrix (abbreviated as ISM) is the matrix

\left( \begin{array}{c} V_{\alpha_1}\\ \vdots\\ V_{\alpha_r}\\ \end{array} \right).
Used by Remark 7.5
Proposition

Proposition 3.4

YFXL

Let G,S and C_i be as in Definition evdef. If G has an ISS, then for any \rho_1,\rho_2\in {\mathcal R}(G), d(S,\rho_1)=d(S,\rho_2) implies \chi_{\rho_1}(g)=\chi_{\rho_2}(g) for each g\in G.

Proof of Proposition 3.4

Let \alpha_1,\ldots,\alpha_r be an ISS of G. By Theorem Tsig, V_{\alpha}\mathcal{C}_1=V_{\alpha}\mathcal{C}_2 for any signature \alpha. Therefore

\left( \begin{array}{c} V_{\alpha_1}\\ \vdots\\ V_{\alpha_r}\\ \end{array} \right) \mathcal{C}_1= \left( \begin{array}{c} V_{\alpha_1}\\ \vdots\\ V_{\alpha_r}\\ \end{array} \right) \mathcal{C}_2.

Since the ISM is invertible, we have \mathcal{C}_1=\mathcal{C}_2, i.e. \chi_{\rho_1}(C_i)=\chi_{\rho_2}(C_i) for each i\in[r]. Thus \chi_{\rho_1}(g)=\chi_{\rho_2}(g) for each g\in G.

Remark

Remark 3.5

T9VY

Suppose G is a finite group. It is well known that for \rho_1,\rho_2\in {\mathcal R}(G), \rho_1\cong\rho_2 if and only if \chi_{\rho_1}(g)=\chi_{\rho_2}(g) for each g\in G. Therefore, if G has an ISS, then d(S,\rho_1)=d(S,\rho_2) implies \rho_1\cong\rho_2.

Lemma

Lemma 3.6

K5GE

Let G=G_1\times G_2 where G_1 and G_2 are any groups. Let \{C_{1,\alpha}:\alpha\in \Omega_1\} and \{C_{2,\beta}:\beta\in \Omega_2\} be the sets of all conjugacy classes of G_1 and G_2, respectively. Then \{C_{1,\alpha}\times C_{2,\beta}:(\alpha,\beta)\in\Omega_1\times\Omega_2\} is the set of all conjugacy classes of G.

Lemma

Lemma 3.7

TB5C

Let G=G_1\times G_2, where G_1 and G_2 are finitely generated groups. Generating sets of G_1,G_2 and G are S,T and S\sqcup T, respectively. Given w=r_1\cdots r_n \in M(S\sqcup T) and r_i\in S\sqcup T, let

r_i^{(1)}= \begin{cases} r_i & r_i\in S\\ \varepsilon & r_i\in T \end{cases} , \quad r_i^{(2)}= \begin{cases} \varepsilon & r_i\in S\\ r_i & r_i\in T \end{cases} .

Let w^{(1)}=r_1^{(1)}r_2^{(1)}\cdots r_n^{(1)}, w^{(2)}=r_1^{(2)}r_2^{(2)}\cdots r_n^{(2)}, and f denote the map w\mapsto w^{(1)}w^{(2)}, then |f^{-1}(w_1w_2)|=|w_1w_2|!/(|w_1|!|w_2|!) for any w_1\in M(S), w_2\in M(T).

Proof of Lemma 3.7

Since |f^{-1}(w_1w_2)| is equal to the number of all the methods for selecting |w_1| positions from |w_1w_2| different positions. we have |f^{-1}(w_1w_2)|=|w_1w_2|!/(|w_1|!|w_2|!).

Theorem

Theorem 3.8

MW96

Let G=G_1\times G_2, where G_1 and G_2 are any finitely generated groups with finite number of conjugacy classes. Generating sets of G_1,G_2 and G are S,T and S\sqcup T respectively.

  1. (a) If both G_1 and G_2 have an ISS, then G has an ISS.

  2. (b) If both G_1 and G_2 have a lower triangular ISM, then G has a lower triangular ISM.

Proof of Theorem 3.8

Let C_{i,1},\ldots,C_{i,r_i} be all conjugacy classes of G_i for i=1,2. Suppose \alpha_1,\ldots,\alpha_{r_1} is an ISS of G_1; \beta_1,\ldots,\beta_{r_2} is an ISS of G_2. Let \alpha_i||\beta_j denote the signature spliced by \alpha_i and \beta_j, for example, (1,2)||(3,4)\coloneqq(1,2,3,4). Consider signature sequence

\alpha_1||\beta_1,\ldots,\alpha_1||\beta_{r_2},\ldots,\alpha_{r_1}||\beta_1,\ldots,\alpha_{r_1}||\beta_{r_2}.

We will prove that it is an ISS of G. For each \alpha_i and \beta_j, we have

E_{\alpha_i||\beta_j}=\bigsqcup_{\substack{u_1\in E_{\alpha_i}\\ u_2\in E_{\beta_j}}}f^{-1}(u_1u_2)

where f is as defined in Lemma countlm. Let V_{\alpha_i}=(x_{i,1},\ldots,x_{i,r_1}) and V_{\beta_j}=(y_{j,1},\ldots,y_{j,r_2}). For each conjugacy class C_{1,k}\times C_{2,l}, it follows that

\begin{aligned} &|E_{\alpha_i||\beta_j}\cap \phi^{-1}(C_{1,k}\times C_{2,l})|\\ =&|\bigsqcup_{\substack{u_1\in E_{\alpha_i}\\ u_2\in E_{\beta_j}}}(f^{-1}(u_1u_2)\cap \phi^{-1}(C_{1,k}\times C_{2,l}))| \\ =&\sum_{\substack{u_1\in E_{\alpha_i}\\ u_2\in E_{\beta_j}}}|f^{-1}(u_1u_2)\cap \phi^{-1}(C_{1,k}\times C_{2,l})|\\ =&\sum_{\substack{u_1\in E_{\alpha_i}\cap \phi^{-1}(C_{1,k})\\ u_2\in E_{\beta_j}\cap \phi^{-1}(C_{2,l})}}|f^{-1}(u_1u_2)| \\ =&\sum_{\substack{u_1\in E_{\alpha_i}\cap \phi^{-1}(C_{1,k})\\ u_2\in E_{\beta_j}\cap \phi^{-1}(C_{2,l})}}\frac{(|\alpha_i|+|\beta_j|)!}{|\alpha_i|!|\beta_j|!}\\ =&|E_{\alpha_i}\cap \phi^{-1}(C_{1,k})|\cdot|E_{\beta_j}\cap \phi^{-1}(C_{2,l})|\frac{(|\alpha_i|+|\beta_j|)!}{|\alpha_i|!|\beta_j|!} \\ =&\frac{(|\alpha_i|+|\beta_j|)!}{|\alpha_i|!|\beta_j|!}x_{i,k}y_{j,l}. \end{aligned}

Arrange all the conjugacy classes in G as

C_{1,1}\times C_{2,1},\ldots,C_{1,1}\times C_{2,r_2},\ldots,C_{1,r_1}\times C_{2,1},\ldots,C_{1,r_1}\times C_{2,r_2}.

Then we have

V_{\alpha_i||\beta_j}= \frac{(|\alpha_i|+|\beta_j|)!}{|\alpha_i|!|\beta_j|!}(x_{i,1}y_{j,1},\ldots,x_{i,1}y_{j,r_2},\ldots,x_{i,r_1}y_{j,1},\ldots,x_{i,r_1}y_{j,r_2}).

Let M_{i,j}\coloneqq(|\alpha_i|+|\beta_j|)!/(|\alpha_i|!|\beta_j|!), we have

\begin{gathered} \begin{pmatrix} V_{\alpha_1||\beta_1}\\ \vdots\\ V_{\alpha_1||\beta_{r_2}}\\ \vdots\\ V_{\alpha_{r_1}||\beta_1}\\ \vdots\\ V_{\alpha_{r_1}||\beta_{r_2}}\\ \end{pmatrix} = \begin{pmatrix} M_{1,1}&~&~&~&~&~&~\\ ~&\ddots&~&~&~&~&~\\ ~&~&M_{1,r_2}&~&~&~&~\\ ~&~&~&\ddots&~&~&~\\ ~&~&~&~&M_{r_1,1}&~&~\\ ~&~&~&~&~&\ddots&~\\ ~&~&~&~&~&~&M_{r_1,r_2}\\ \end{pmatrix} \boldsymbol{\cdot}\\ \begin{pmatrix} V_{\alpha_1}\\ \vdots\\ V_{\alpha_{r_1}}\\ \end{pmatrix} \bigotimes \begin{pmatrix} V_{\beta_1}\\ \vdots\\ V_{\beta_{r_2}}\\ \end{pmatrix}. \end{gathered}

Then

{\rm det} \begin{pmatrix} V_{\alpha_1||\beta_1}\\ \vdots\\ V_{\alpha_{r_1}||\beta_{r_2}}\\ \end{pmatrix} = (\prod_{i=1}^{r_1}\prod_{j=1}^{r_2}M_{i,j}) {\rm det}^{r_2} \begin{pmatrix} V_{\alpha_1}\\ \vdots\\ V_{\alpha_{r_1}}\\ \end{pmatrix} {\rm det}^{r_1} \begin{pmatrix} V_{\beta_1}\\ \vdots\\ V_{\beta_{r_2}}\\ \end{pmatrix} \neq 0.

Hence sequence (Proof) is an ISS of G.

If the corresponding ISMs of \alpha_1,\ldots,\alpha_{r_1} and \beta_1,\ldots,\beta_{r_2} are both lower triangular matrices, then the corresponding ISM of (Proof) is also a lower triangular matrix by equation (Proof).

Corollary

Corollary 3.9

2WNS

Let G=G_1\times\cdots\times G_m, where G_1,\ldots,G_m are any finitely generated groups with finite number of conjugacy classes. Generating sets of G_1,\ldots,G_m and G are S_1,\ldots,S_m and S=\bigsqcup_{i=1}^mS_i respectively.

  1. (a) If each G_i has an ISS, then G has an ISS.

  2. (b) If each G_i has a lower triangular ISM, then G has a lower triangular ISM.

Remark

Remark 3.10

090W

By Proposition keyprop and Corollary dpc, to prove Theorem MT, it suffices to show each irreducible finite Coxeter group has an ISS with a lower triangular ISM.

Lemma

Lemma 4.1

2AYW

Let \alpha_i=(a_{i,1},\ldots,a_{i,n})\in \mathbb{C}^n, where i\in[m] and m\geq n. Let \gamma(\alpha_i)\coloneqq\{j:a_{i,j}\neq0\}. If the following holds:

  1. (a) |\gamma(\alpha_1)|=1,

  2. (b) |\bigcup_{i=1}^{m}\gamma(\alpha_i)|=n,

  3. (c) |\gamma(\alpha_k)\setminus\bigcup_{i=1}^{k-1}\gamma(\alpha_i)|=0\ {\rm or}\ 1 for each k\in\{2,\ldots,m\}.

Then there is a subsequence \alpha_{p_1},\ldots,\alpha_{p_n} that satisfies:

  1. (1) \alpha_{p_1},\ldots,\alpha_{p_n} is linearly independent,

  2. (2) the matrix

    \left(\begin{array}{c} \alpha_{p_1}\\ \vdots\\ \alpha_{p_n}\\ \end{array}\right)

    can be transformed into a lower triangular matrix by permuting the columns.

Proof of Lemma 4.1

Let

A=\{\alpha_k:k\geq2, |\gamma(\alpha_k)\setminus\bigcup_{i=1}^{k-1}\gamma(\alpha_i)|=1\}\cup\{\alpha_1\}.

By (a), (b) and (c), we have |A|=n. Let A=\{\alpha_{p_1},\ldots,\alpha_{p_n}\} where p_1<\cdots<p_n. For each r\in[n], let \gamma(\alpha_{p_r})\setminus\bigcup_{i=1}^{p_r-1}\gamma(\alpha_i)=\{q_r\}, we have a_{p_r,q_r}\neq 0 and a_{p_i,q_r}=0 for each i< r. Thus, by permuting the columns of the matrix (a_{p_i,j})_{i,j\in[n]} according to q_1,\ldots, q_n, we obtain an invertible lower triangular matrix (a_{p_i,q_j})_{i,j\in[n]}. Therefore, for the sequence \alpha_{p_1},\ldots,\alpha_{p_n}, properties (1) and (2) hold.

Theorem

Theorem 4.2

T0BM

Let W be a finite Coxeter group of type {\rm I_2(m)} with S=\{g_1,g_2\} as its Coxeter generating set. Then W has an ISS with a lower triangular ISM.

Proof of Theorem 4.2

We define a partial order on \mathbb{N}^2. Let \alpha,\beta\in \mathbb{N}^2, \alpha=(a_1,a_2), \beta=(b_1,b_2). Let \alpha\leq\beta if (|\alpha|,a_1,a_2)\leq_{\rm lex}(|\beta|,b_1,b_2) where \leq_{\rm lex} is the lexicographic order on \mathbb{N}^3. It can be seen that \leq is a total order. Let C_1,\ldots,C_r be all conjugacy classes of W. For each i\in[r], take a word w_i\in \phi^{-1}(C_i), let \alpha_i={\rm sig}(w_i), let H=\{w_1,\ldots,w_r\}. Let

A=\{\alpha\in \mathbb{N}^2:\exists i\ s.t.\ \alpha\leq\alpha_i\},

Clearly |A|<\infty, so let A=\{\beta_1,\ldots, \beta_N\}, where \beta_1<\cdots<\beta_N. Then we consider \{V_{\beta_1},\ldots,V_{\beta_N}\} where V_{\beta_i} defined in Definition evdef. Let \gamma be as defined in Lemma csslm. Since \beta_1=(0,0), we have |\gamma(V_{\beta_1})|=1. For each conjugacy class C_h, note that w_h\in \phi^{-1}(C_h) and \alpha_h={\rm sig}(w_h)\in A, so h\in\gamma(V_{\alpha_h})\in \bigcup_{i=1}^N\gamma(V_{\beta_i}). For the arbitrariness of h, we have |\bigcup_{i=1}^N\gamma(V_{\beta_i})|=r. For each k\in \{2,\ldots,N\}, let \beta_k=(b_1,b_2). If |b_1-b_2|\geq 2, then for each h\in\gamma(V_{\beta_k}), there exists a w=t_1\cdots t_{|\beta_k|}\in \phi^{-1}(C_h), t_i\in\{g_1,g_2\} such that {\rm sig}(w)=\beta_k and t_j=t_{j+1} for some j. Thus there is a w'=t_1\cdots t_{j-1}t_{j+2}\cdots t_{|\beta_k|} such that \phi(w')=\phi(w) and {\rm sig}(w')<{\rm sig}(w)=\beta_k. So {\rm sig}(w')\in A and h\in\gamma(V_{{\rm sig}(w')})\subseteq\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i}). For the arbitrariness of h, we have |\gamma(V_{\beta_k})\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i})|=0. If |b_1-b_2|=1, suppose b_2=b_1+1. If b_1=0, then |\gamma(V_{\beta_k})|=1 and |\gamma(V_{\beta_k})\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i})|=0\ {\rm or}\ 1. If b_1>0, then g_2g_1\cdots g_2g_1g_2\in E_{\beta_k} and is conjugate to g_1 or g_2. For each w=t_1\cdots t_{|\beta_k|}\in E_{\beta_k}\setminus\{g_2g_1\cdots g_2g_1g_2\} where t_i\in\{g_1,g_2\}, there is a t_j=t_{j+1} for some j. Thus for each h\in\gamma(V_{\beta_k}), there are w,w'\in M(S) such that {\rm sig}(w')<{\rm sig}(w)=\beta_k and \phi(w') is conjugate to \phi(w). Hence h\in\gamma(V_{{\rm sig}(w')})\subseteq\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i}). For the arbitrariness of h, we have |\gamma(V_{\beta_k})\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i})|=0. If b_1=b_2, then g_1g_2\cdots g_1g_2,g_2g_1\cdots g_2g_1\in E_{\beta_k}. Let

\tilde{E}_{\beta_k}\coloneqq E_{\beta_k}\setminus\{g_1g_2\cdots g_1g_2, g_2g_1 \cdots g_2g_1\}.

For each w=t_1t_2\cdots t_{|\beta_k|} \in \tilde{E}_{\beta_k}, there is a t_j=t_{j+1} for some j. Let

V'=(|\tilde{E}_{\beta_k}\cap \phi^{-1}(C_1)|,\ldots,|\tilde{E}_{\beta_k}\cap \phi^{-1}(C_r)|).

By the same argument in the previous cases, we have |\gamma(V')\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i})|=0. Since g_1g_2\cdots g_1g_2 and g_2g_1 \cdots g_2g_1 are in the same conjugacy class, the number of conjugacy classes in E_{\beta_k} is at most one more than the number of conjugacy classes in \tilde{E}_{\beta_k}, so |\gamma(V_{\beta_k})\setminus\gamma(V')|=0\ {\rm or}\ 1. Thus |\gamma(V_{\beta_k})\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\beta_i})|=0\ {\rm or}\ 1. Consequently, by using Lemma csslm, there is a linearly independent subsequence V_{\beta_{p_1}},\ldots,V_{\beta_{p_r}} of V_{\beta_1},\ldots,V_{\beta_N}, hence \beta_{p_1},\ldots,\beta_{p_r} is an ISS of W. Besides, with proper reordering of the conjugacy classes C_1,\ldots,C_r, the ISM of \beta_{p_1},\ldots,\beta_{p_r} can become a lower triangular matrix.

Proposition

Proposition 4.3

T7FQ

Let W be a finite Coxeter group of type {\rm A_n}, {\rm B_n} or {\rm D_{n+1}}, and S=\{g_1,g_2,\cdots\} is a set of Coxeter generators for W. Given a word u\in M(S) and {\rm sig}(u)=(a_1,\ldots,a_n) or (a_1,\ldots,a_{n+1}), let {\rm ct}(u)\coloneqq(|u|,a_1,\ldots,a_{n-1}).

  1. (a) For each word w\in M(S), there exists a word \tilde{w} such that \phi(w) is conjugate to \phi(\tilde{w}) and {\rm ct}(w)\geq_{\rm lex}{\rm ct}(\tilde{w}) where \leq_{\rm lex} is the lexicographic order on \mathbb{N}^n.

  2. (b) Let words w_1,w_2\in M(S), if {\rm sig}(w_1)={\rm sig}(w_2) and {\rm ct}(w_1)={\rm ct}(\tilde{w}_1)={\rm ct}(\tilde{w}_2)={\rm ct}(w_2), then \phi(w_1) is conjugate to \phi(w_2).

Theorem

Theorem 4.4

RI0P

Let W be a finite Coxeter group of type {\rm A_n}, {\rm B_n} or {\rm D_{n+1}} with S as its Coxeter generating set. Then W has an ISS with a lower triangular ISM.

Proof of Theorem 4.4

Let C_1,\ldots,C_r be all conjugacy classes of W. Take a word v_i\in \phi^{-1}(C_i) for each i\in[r], let H=\{v_1,\ldots,v_r\} and J=\{w:\exists v_i\in H\ s.t.\ w\leq v_i\}. Clearly we have |J|<\infty, so (J,\leq) can extend to a totally ordered set (J,\widetilde{\leq}) (i.e. w_i\leq w_j implies w_i\widetilde{\leq}w_j). So let J=\{w_1,\ldots,w_N\}, where w_1\widetilde{<}\cdots\widetilde{<}w_N. Let \alpha_i={\rm sig}(w_i) for each i\in[r]. Since w_1=\varepsilon, it follows that |\gamma(V_{\alpha_1})|=1. For each h\in[r], we have v_h\in \phi^{-1}(C_h) and v_h\in J. Let \alpha_k={\rm sig}(v_h), then h\in\gamma(V_{\alpha_k})\subseteq\bigcup_{i=1}^N\gamma(V_{\alpha_i}). For the arbitrariness of h, we have |\bigcup_{i=1}^N\gamma(V_{\alpha_i})|=r. For each k\in\{2,\ldots,N\}, by Proposition ABDprop part (a), we divide E_{\alpha_k} into two parts:

\hat{E}_{\alpha_k}\coloneqq E_{\alpha_k}\cap\{u:{\rm ct}(\tilde{u})<_{\rm lex}{\rm ct}(u)\}

and

\bar{E}_{\alpha_k}\coloneqq E_{\alpha_k}\cap\{u:{\rm ct}(\tilde{u})={\rm ct}(u)\}.

Let

V'=(|\hat{E}_{\alpha_k}\cap \phi^{-1}(C_1)|,\ldots,|\hat{E}_{\alpha_k}\cap \phi^{-1}(C_r)|).

For each h\in\gamma(V'), there exists u_0\in \hat{E}_{\alpha_k} and u_0\in \phi^{-1}(C_h). Since {\rm ct}(\tilde{u}_0)<_{\rm lex}{\rm ct}(u_0)={\rm ct}(w_k), we have \tilde{u}_0<w_k and thus \tilde{u}_0\in J. Let \tilde{u}_0=w_l, {\rm sig}(\tilde{u}_0)=\alpha_l, then w_l\widetilde{<}w_k and then l<k. Since \phi(\tilde{u}_0) and \phi(u_0) are in the same conjugacy class, h\in \gamma(V_{\alpha_l})\subseteq\bigcup_{i=1}^{k-1}\gamma(V_{\alpha_i}). For the arbitrariness of h, we have \gamma(V')\subseteq\bigcup_{i=1}^{k-1}\gamma(V_{\alpha_i}). Now consider words in \bar{E}_{\alpha_k}. Let u_1,u_2\in \bar{E}_{\alpha_k}, we have {\rm sig}(u_1)={\rm sig}(u_2) and {\rm ct}(u_1)={\rm ct}(\tilde{u}_1)={\rm ct}(\tilde{u}_2)={\rm ct}(u_2), so \phi(u_1) and \phi(u_2) are in the same conjugacy class by Proposition ABDprop part (b). Thus the number of conjugacy classes in E_{\alpha_k} is at most one more number than the number of conjugacy classes in \hat{E}_{\alpha_k}. So we have |\gamma(V_{\alpha_k})\setminus\gamma(V')|=0\ {\rm or}\ 1, and |\gamma(V_{\alpha_k})\setminus\bigcup_{i=1}^{k-1}\gamma(V_{\alpha_i})|=0\ {\rm or}\ 1 since \gamma(V')\subseteq\bigcup_{i=1}^{k-1}\gamma(V_{\alpha_i}). Therefore, by using Lemma csslm, there is a linearly independent subsequence V_{\alpha_{p_1}},\ldots,V_{\alpha_{p_r}} of V_{\alpha_1},\ldots,V_{\alpha_N}, hence \alpha_{p_1},\ldots,\alpha_{p_r} is an ISS of W. Besides, with proper reordering of the conjugacy classes C_1,\ldots,C_r, the ISM of \alpha_{p_1},\ldots,\alpha_{p_r} can become a lower triangular matrix.

Remark

Remark 4.5

I2JI

The idea for the proof of Theorem clsp2 is inspired by the proof in [CST, Section 7], where the authors proved that for types {\rm A,B\ \text{and}\ D}, d(S,\rho_1)=d(S,\rho_2) if and only if \rho_1\cong\rho_2. We refined its proof and obtained a stronger conclusion. Theorem clsp1, clsp2 together with the results in the following several sections form the foundation for proving Theorem MT.

Definition

Definition 5.2

ZMP2

Let signature \alpha\in\mathbb{N}^n. The non-cuspidal signature vector of \alpha is

{\it NV}(\alpha)\coloneqq(|E_{\alpha}\cap \phi^{-1}({\it Ncus}_1)|,\ldots,|E_{\alpha}\cap \phi^{-1}({\it Ncus}_p)|).

The cuspidal signature vector of \alpha is

{\it CV}(\alpha)\coloneqq(|E_{\alpha}\cap \phi^{-1}({\it Cus}_1)|,\ldots,|E_{\alpha}\cap \phi^{-1}({\it Cus}_q)|).
Used by Proof, Proof
Definition

Definition 5.3

762U
  1. (a) A sequence \alpha_1,\ldots,\alpha_p of signatures is called a non-cuspidal signature sequence (abbreviated as NSS) if

    1. 1) {\it NV}(\alpha_1),\ldots,{\it NV}(\alpha_p) is linearly independent.

    2. 2) |E_{\alpha_i}\cap \phi^{-1}({\it Cus}_j)|=0 for each i\in[p],\ j\in[q].

    Suppose \alpha_1,\ldots,\alpha_p is an NSS, the corresponding non-cuspidal signature matrix (abbreviated as NSM) is the matrix

    \left( \begin{array}{c} {\it NV}(\alpha_1)\\ \vdots\\ {\it NV}(\alpha_p)\\ \end{array} \right).
  2. (b) A sequence \alpha_1,\ldots,\alpha_q of signatures is called a cuspidal signature sequence (abbreviated as CSS) if {\it CV}(\alpha_1),\ldots,{\it CV}(\alpha_q) is linearly independent. Suppose \alpha_1,\ldots,\alpha_q is a CSS, the corresponding cuspidal signature matrix (abbreviated as CSM) is the matrix

    \left( \begin{array}{c} {\it CV}(\alpha_1)\\ \vdots\\ {\it CV}(\alpha_q)\\ \end{array} \right).
Proposition

Proposition 5.4

1RT2
  1. (a) If W has an NSS and a CSS, then W has an ISS.

  2. (b) If W has a lower triangular NSM and a lower triangular CSM, then W has a lower triangular ISM.

Proof of Proposition 5.4

Suppose \alpha_1,\ldots, \alpha_p is an NSS of W and \beta_1,\ldots, \beta_q is a CSS of W. Arrange all the conjugacy classes as

{\it Ncus}_1,\ldots,{\it Ncus}_p,{\it Cus}_1,\ldots,{\it Cus}_q.

It follows that

\left( \begin{array}{c} V_{\alpha_1}\\ \vdots\\ V_{\alpha_p}\\ V_{\beta_1}\\ \vdots\\ V_{\beta_q}\\ \end{array} \right) = \begin{pNiceArray}{ccccc|ccccc}[margin] &\Block{3-3}{\rm NSM}&&&&&\Block{3-3}{0}&\\ & & & & & & & & &\\ & & & & & & & & &\\ \hline &\Block{3-3}{\ddots}&&&&&\Block{3-3}{\rm CSM}&\\ & & & & & & & & &\\ & & & & & & & & &\\ \end{pNiceArray}.

It can be verified that V_{\alpha_1},\ldots,V_{\alpha_p},V_{\beta_1},\ldots,V_{\beta_q} is linearly independent and then \alpha_1,\ldots,\alpha_p,\beta_1,\ldots,\beta_q is an ISS of W. Furthermore, if NSM and CSM in equation (Proof) are both lower triangular, then ISM of \alpha_1,\ldots,\alpha_p,\beta_1,\ldots,\beta_q is also lower triangular.

Theorem

Theorem 5.5

SEBV

If each proper parabolic subgroup of W has an ISS with a lower triangular ISM, then W has an NSS with a lower triangular NSM.

Proof of Theorem 5.5

Let U denote the set of all non-cuspidal classes of W. Let W_{J_1},\ldots,W_{J_m} be all proper parabolic subgroups of W. For each proper parabolic subgroups W_{J_k}, by assumption, there is an order of all conjugacy classes of W_{J_k}: C_1^{J_k},\ldots,C_{r_k}^{J_k} and a sequence \alpha_{k,1},\ldots,\alpha_{k,r_k} \\ \in\mathbb{N}^{r_k} such that

\begin{aligned} &E_{\alpha_{k,i}}\cap\phi^{-1}(C_i^{J_k})\neq\varnothing\quad\text{for}\ i\in[r_k], \\ &E_{\alpha_{k,i}}\cap\phi^{-1}(C_j^{J_k})=\varnothing\quad\text{for}\ i,j\in[r_k]\ \text{and}\ j>i. \end{aligned}

Let

\psi: C_i^{J_k}\mapsto \{gxg^{-1}:x\in C_i^{J_k},g\in W\}.

\psi(C_i^{J_k}) is the conjugacy class of W that contains C_i^{J_k}. It is easy to see that each \psi(C_i^{J_k}) is an non-cuspidal class of W. Furthermore, since each non-cuspidal class of W contains at least one word that belongs to some proper parabolic subgroup, we have

\bigcup_{k=1}^m\bigcup_{i=1}^{r_k}\psi(C_i^{J_k})={\it U}.

Suppose J_k=\{s_{k_1},\ldots,s_{k_d}\} where k_1<\cdots<k_d. Given \alpha_{k,i}=(a_1,\ldots,a_d), define \alpha'_{k,i}=(a'_1,\ldots,a'_n) where

a'_x= \begin{cases} a_i &{\rm if}\ x=k_i\ {\rm for\ some}\ i\in[d],\\ 0 &{\rm else}. \end{cases}

for each x\in[n]. In another word, \alpha'_{k,i} is the signature on W induced by \alpha_{k,i}. Let {\it U}=\{{\it Ncus}_1,{\it Ncus}_2,\ldots,{\it Ncus}_p\} where

\min_{\leq_{\rm lex}}\{(k,i):\psi(C_i^{J_k})={\it Ncus}_1\}\leq_{\rm lex}\cdots\leq_{\rm lex}\min_{\leq_{\rm lex}}\{(k,i):\psi(C_i^{J_k})={\it Ncus}_p\}.

Let

\beta_j=\alpha'_{\theta},\ \theta=\min_{\leq_{\rm lex}}\{(k,i):\psi(C_i^{J_k})={\it Ncus}_j\}

The below shows that \beta_1,\ldots,\beta_p is an NSS with a lower triangular NSM of W. For any given \beta_j, let \beta_j=\alpha'_{k,i}. First, since \phi(E_{\beta_j})\subset W_{J_k} and W_{J_k}\cap{\it Cus}_l=\varnothing for each l\in[q], we have E_{\beta_j}\cap\phi^{-1}({\it Cus}_l)=\varnothing for each l\in[q]. Next, by (?) we have E_{\beta_j}\cap \phi^{-1}(C_i^{J_k})\neq\varnothing, by (?) we have \psi(C_i^{J_k})={\it Ncus}_j, thus E_{\beta_j}\cap\phi^{-1}({\it Ncus}_j)\neq\varnothing. Suppose E_{\beta_j}\cap\phi^{-1}({\it Ncus}_h)\neq\varnothing. By (?), we obtain {\it Ncus}_h=\psi(C_{i'}^{J_k}) for some i'\leq i. Let \beta_h=\alpha'_{\tilde{k},\tilde{i}}, then (\tilde{k},\tilde{i})\leq_{\rm lex}(k,i')\leq_{\rm lex}(k,i) by (?), and then h\leq j by (?). Thus, for each h>j, E_{\beta_j}\cap\phi^{-1}({\it Ncus}_h)=\varnothing. Therefore \beta_1,\ldots,\beta_p is an NSS with a lower triangular NSM of W.

Theorem

Theorem 5.7

J3TQ

Each Coxeter group of exceptional types has an ISS with a lower triangular ISM.

Proof of Theorem 5.7

By Proposition cssp and Theorem cmp, it is sufficient to show that each Coxeter group W(M) of exceptional types has an NSS with a lower triangular NSM. Furthermore, by Theorem ucsp, it is sufficient to show that each proper parabolic subgroup of W(M) has an ISS with a lower triangular ISM. Consider all Coxeter groups of exceptional types as the order {\rm H_3},{\rm H_4},{\rm F_4},\\{\rm E_6},{\rm E_7},{\rm E_8}. By observing Table table1 and Table table2, given an exceptional type M,
there are only three possible cases for the proper parabolic subgroups of W(M):

  1. a) classical types,

  2. b) exceptional types which are before it,

  3. c) direct products of some types in the first two cases.

By Theorem clsp1, clsp2 and Corollary dpc, for each exceptional type M, all that is needed is to show that every previous exceptional types have an ISS with a lower triangular ISM. Hence {\rm H_3} is proven firstly, and then {\rm H_4,F_4,E_6,E_7} and {\rm E_8} are proven in turn, so the theorem holds.

Lemma

Lemma 6.1

3AF8

Let {\it Cus}_1,\ldots,{\it Cus}_q be all cuspidal classes of W with l({\it Cus}_1)\leq\cdots\leq l({\it Cus}_q).

  1. (a) If l({\it Cus}_i)\neq l({\it Cus}_j) for each i\neq j, then W has a CSS with a lower triangular CSM.

  2. (b) If

    \begin{gathered} l({\it Cus}_1)<\cdots<l({\it Cus}_{k_1})=l({\it Cus}_{k_1+1})<\cdots\\<l({\it Cus}_{k_h})=l({\it Cus}_{k_h+1})<\cdots<l({\it Cus}_q) \end{gathered}

    and for each j\in[h], there exists a word w_{k_j}\in\phi^{-1}(\it Cus_{k_j}) such that |w_{k_j}|=l({\it Cus}_{k_j}) and |E_{{\rm sig}(w_{k_j})}\cap \phi^{-1}({\it Cus}_{k_j+1})|=0, then W has a CSS with a lower triangular CSM.

Proof of Lemma 6.1
  1. (a) For each i\in[q], take a word w_i\in \phi^{-1}({\it Cus}_i) such that |w_i|=l({\it Cus}_i). Then |E_{{\rm sig}(w_i)}\cap \phi^{-1}({\it Cus}_j)|=0 for all j>i since |w_i|=l({\it Cus}_i)<l({\it Cus}_j). Also, clearly we have |E_{{\rm sig}(w_i)}\cap \phi^{-1}({\it Cus}_i)|\neq0. Hence {\rm sig}(w_1),\ldots,{\rm sig}(w_q) is a CSS with a lower triangular CSM.

  2. (b) For each i\in[q], if i=k_j for some j\in[h], take w_i such that |w_i|=l({\it Cus}_{i}) and |E_{{\rm sig}(w_i)}\cap \phi^{-1}({\it Cus}_{i+1})|=0 by assumption; or else, take w_i\in \phi^{-1}({\it Cus}_i) such that |w_i|=l({\it Cus}_i). Since |E_{{\rm sig}(w_i)}\cap \phi^{-1}({\it Cus}_i)|\neq0 and |E_{{\rm sig}(w_i)}\cap \phi^{-1}({\it Cus}_j)|=0 for all j>i, it follows that {\rm sig}(w_1),\ldots,{\rm sig}(w_q) is a CSS with a lower triangular CSM.

Proposition

Proposition 6.2

560G

Let {\it Cus}_i^{\rm M} denote the ith cuspidal class of W(M) in [GP, Appendix B].

  1. (a) For W=W({\rm H_3}), l({\it Cus}_1^{\rm H_3})<\cdots<l({\it Cus}_4^{\rm H_3}).

  2. (b) For W=W({\rm H_4}),

    l({\it Cus}_1^{\rm H_4})<\cdots<l({\it Cus}_7^{\rm H_4})=l({\it Cus}_8^{\rm H_4})<\cdots<l({\it Cus}_{20}^{\rm H_4}).
  3. (c) For W=W({\rm F_4}),

    l({\it Cus}_1^{\rm F_4})<\cdots<l({\it Cus}_4^{\rm F_4})=l({\it Cus}_5^{\rm F_4})<\cdots<l({\it Cus}_9^{\rm F_4}).
  4. (d) For W=W({\rm E_6}), l({\it Cus}_1^{\rm E_6})<\cdots<l({\it Cus}_5^{\rm E_6}).

  5. (e) For W=W({\rm E_7}), l({\it Cus}_1^{\rm E_7})<\cdots<l({\it Cus}_{12}^{\rm E_7}).

  6. (f) For W=W({\rm E_8}),

    \begin{gathered} l({\it Cus}_1^{\rm E_8})<\cdots<l({\it Cus}_5^{\rm E_8})=l({\it Cus}_6^{\rm E_8})<\cdots<l({\it Cus}_9^{\rm E_8})=l({\it Cus}_{10}^{\rm E_8})<\\ l({\it Cus}_{11}^{\rm E_8})=l({\it Cus}_{12}^{\rm E_8})<l({\it Cus}_{13}^{\rm E_8})=l({\it Cus}_{14}^{\rm E_8})<\cdots\\ <l({\it Cus}_{21}^{\rm E_8})=l({\it Cus}_{22}^{\rm E_8})<l({\it Cus}_{23}^{\rm E_8})=l({\it Cus}_{24}^{\rm E_8})<\cdots<l({\it Cus}_{30}^{\rm E_8}). \end{gathered}
Proposition

Proposition 6.4

TG4F

For W=W({\rm H_3}),W({\rm H_4}),W({\rm E_6}),W({\rm E_7}),\text{or}\ W({\rm E_8}), let C_0 be a cuspidal class of W. Let g\in W, then g\in C_0 if and only if p_g(\lambda)=p_{C_0}(\lambda).

Proof of Proposition 6.4

If g\in C_0, then p_g(\lambda)=p_{C_0}(\lambda). Conversely, if p_g(\lambda)=p_{C_0}(\lambda), it follows that p_g(1)\neq 0 so the conjugacy class of g in W is a cuspidal class by Lemma cuspolylemma. It is known that the characteristic polynomials of any two different cuspidal classes in W are different by [GP, Appendix B]. Hence g is in C_0.

Remark

Remark 6.5

UDFL

For W=W({\rm F_4}), the result in Proposition pgprop is false since p_{{\it Cus}_4^{\rm F_4}}(\lambda)=p_{{\it Cus}_5^{\rm F_4}}(\lambda)=(\lambda+1)(\lambda^3+1).

Remark

Remark 6.6

1O49

By Lemma cuslemma and Proposition cusprop, it remains for us to show that for each two cuspidal classes {\it Cus}_i^M and {\it Cus}_j^M that M={\rm H_4,F_4} or {\rm E_8} and l({\it Cus}_i^M)=l({\it Cus}_j^M), there exists a word w\in\phi^{-1}({\it Cus}_i^M) such that |w|=l({\it Cus}_i^M) and E_{{\rm sig}(w)}\cap \phi^{-1}({\it Cus}_j^M)=\varnothing. It is a natural idea to design an algorithm to check whether E_{{\rm sig}(w)}\cap \phi^{-1}({\it Cus}_j^M)=\varnothing for given {\it Cus}_i^M,{\it Cus}_j^M and word w such that \phi(w)\in {\it Cus}_i^M and |w|=l({\it Cus}_i^M). Specifically, such an algorithm should realize the following functions:

  1. a) Traverse E_{{\rm sig}(w)}.

  2. b) For each u\in E_{{\rm sig}(w)}, check whether \phi(u)\in {\it Cus}_j^M.

However, such E_{{\rm sig}(w)} can be very large, especially in W({\rm E_8}). To be precise, suppose {\rm sig}(w)=(a_1,\ldots,a_n), then |E_{{\rm sig}(w)}|=|{\rm sig}(w)|!/(a_1!\cdots a_n!). Thus the traversal usually cannot be completed within an acceptable time. So we hope to find a small enough subset H\subseteq E_{{\rm sig}(w)} such that

E_{{\rm sig}(w)}\cap\phi^{-1}({\it Cus}_j^M)=\varnothing\quad\text{if and only if}\quad H\cap\phi^{-1}({\it Cus}_j^M)=\varnothing

so that we just need to traverse H instead of E_{{\rm sig}(w)}.

Definition

Definition 6.7

H1RP

Suppose (W,S) is a Coxeter system, let w=t_1t_2\cdots t_n\in M(S), t_1,\ldots,t_n\in S. The cyclic class of w is a subset of M(S) defined as

\{t_1\cdots t_n,t_2\cdots t_nt_1,\ldots,t_nt_1\cdots t_{n-1}\}.
Uses Definition 2.1 · Used by Proof
Proposition

Proposition 6.8

9TMF

Let (W(M),S) be a Coxeter system and M={\rm H_4}\ \text{or}\ {\rm F_4}. Suppose two cuspidal classes {\it Cus}_i^M and {\it Cus}_j^M satisfy l({\it Cus}_i^M)=l({\it Cus}_j^M), and a word w\in\phi^{-1}({\it Cus}_i^M) satisfies |w|=l({\it Cus}_i^M) and let \alpha={\rm sig}(w). Then E_{\alpha}\cap\phi^{-1}({\it Cus}_j^M) =\varnothing if and only if I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})\cap\phi^{-1}({\it Cus}_j^M)=\varnothing.

Proof of Proposition 6.8

Let {\rm sig}(w)=(a_1,a_2,a_3,a_4). It follows that a_1\leq a_2 and a_4\leq a_3, since |w|=l({\it Cus}_i^M). It is obvious that I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})\subseteq E_{\alpha}, hence E_{\alpha}\cap\phi^{-1}({\it Cus}_j^M)=\varnothing implies I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})\cap\phi^{-1}({\it Cus}_j^M)=\varnothing.

Conversely, suppose I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})\cap\phi^{-1}({\it Cus}_j^M)=\varnothing. Since each element in I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}) is conjugate to an element in I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc}), we have I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}})\cap\phi^{-1}({\it Cus}_j^M)=\varnothing. Let u\in E_{\alpha} and C_0 is the conjugacy class of \phi(u). If |u|>l(C_0), then u\notin\phi^{-1}({\it Cus}_j^M) since |u|=l({\it Cus}_j^M). If |u|=l(C_0), we perform the following operations on u. First, for each letter s_1 in u, swap it with letters to its right until a letter non-commuting with s_1 is encountered. If this s_1 reaches the rightmost position, then move it to the leftmost position and repeat the above step. Since |u|=l(C_0), no s_1 will encounter another s_1. Thus every s_1 will eventually encounter an s_2 to its right through this procedure. Next, perform a similar procedure for each s_4 in the newly obtained word, then every s_4 will eventually encounter an s_3 to its right. Therefore, we finally obtain a new word u'\in I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}) and \phi(u') is conjugate to \phi(u). Since I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}})\cap\phi^{-1}({\it Cus}_j^M)=\varnothing, we have u'\notin\phi^{-1}({\it Cus}_j^M) and then u\notin\phi^{-1}({\it Cus}_j^M). Hence E_{\alpha}\cap\phi^{-1}({\it Cus}_j^M)=\varnothing.

Proposition

Proposition 6.9

G9MF

For Coxeter system (W({\rm E_8}),S), suppose two cuspidal classes {\it Cus}_i^{\rm E_8} and {\it Cus}_j^{\rm E_8} satisfy l({\it Cus}_i^{\rm E_8})=l({\it Cus}_j^{\rm E_8}), word w\in\phi^{-1}({\it Cus}_i^{\rm E_8}) satisfies |w|=l({\it Cus}_i^{\rm E_8}) and let \alpha={\rm sig}(w). Then E_{\alpha}\cap\phi^{-1}({\it Cus}_j^{\rm E_8})=\varnothing if and only if I_{87}^{\alpha}I_7^{\alpha}I_5^{\alpha}I_2^{\alpha}I_{13}^{\alpha}I_3^{\alpha}(E_{{\alpha}_{46}}^{\rm cyc})\cap\phi^{-1}({\it Cus}_j^{\rm E_8})=\varnothing.

Proposition

Proposition 6.10

6ZCK

For W=W({\rm F_4}), there exists w\in\phi^{-1}({\it Cus}_5^{\rm F_4}) such that |w|=l({\it Cus}_5^{\rm F_4}) and E_{{\rm sig}(w)}\cap\phi^{-1}({\it Cus}_4^{\rm F_4})=\varnothing.

Proof of Proposition 6.10

Let w=1213213234 which is a representative of {\it Cus}_5^{\rm F_4} listed in [GP, Appendix B]. Let \alpha={\rm sig}(w), then \alpha=(3,3,3,1) and {\alpha}_{23}=(0,3,3,0). Select an arbitrary element from each cyclic class in E_{{\alpha}_{23}} and obtain

E_{\alpha_{23}}^{\rm cyc}=\{222333,223233,232233,232323\},

and then

\begin{aligned} I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})=\{&1212124333,1212123433,1212123343,1212431233,\\&1212312433,1212312343,1243121233,1231212433,\\&1231212343,1243123123,1231243123,1231231243\}. \end{aligned}

Note that \phi(1212)=\phi(21) and the last three words in (Proof) are all in \phi^{-1}({\it Cus}_5^{\rm F_4}), so I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc})\cap\phi^{-1}({\it Cus}_4^{\rm F_4})=\varnothing. By Proposition algprop, we have E_{\alpha}\cap\phi^{-1}({\it Cus}_4^{\rm F_4})=\varnothing.

Algorithm

Algorithm A

GRUU

Given a Coxeter system (W({\rm M}),S) where M={\rm H_4} or {\rm E_8}, two cuspidal classes {\it Cus}_i^M and {\it Cus}_j^M such that l({\it Cus}_i^M)=l({\it Cus}_j^M), a word w\in \phi^{-1}({\it Cus}_i^M) such that |w|=l({\it Cus}_i^M) and let \alpha={\rm sig}(w), check whether E_{\alpha}\cap\phi^{-1}({\it Cus}_j^M)=\varnothing.

  1. A1. Flag \leftarrow 0. Calculate p_{{\it Cus}_j^{\rm H_4}}(\lambda) (resp. p_{{\it Cus}_j^{\rm E_8}}(\lambda)).

  2. A2. Traverse I_4^{\alpha}I_1^{\alpha}(E_{{\alpha}_{23}}^{\rm cyc}) (resp. I_{87}^{\alpha}I_7^{\alpha}I_5^{\alpha}I_2^{\alpha}I_{13}^{\alpha}I_3^{\alpha}(E_{{\alpha}_{46}}^{\rm cyc})). For each v\in I_4^{\alpha}I_1^{\alpha} \\ (E_{{\alpha}_{23}}^{\rm cyc}) (resp. I_{87}^{\alpha}I_7^{\alpha}I_5^{\alpha}I_2^{\alpha}I_{13}^{\alpha}I_3^{\alpha}(E_{{\alpha}_{46}}^{\rm cyc})), calculate p_{\phi(v)}(\lambda).

    1. A2.1. If p_{\phi(v)}(\lambda)=p_{{\it Cus}_j^{\rm H_3}}(\lambda) (resp. p_{\phi(v)}(\lambda)=p_{{\it Cus}_j^{\rm E_8}}(\lambda)), then Flag \leftarrow 1 and the traversal terminates.

  3. A3. Output the value of Flag.

Remark

Remark 6.11

F53O

By Proposition pgprop and algprop, E_{{\rm sig}(w)}\cap\phi^{-1}({\it Cus}_j^M)=\varnothing if and only if running algorithm algA yields {\it Flag}=0.

Remark

Remark 6.12

LPW6

For W({\rm H_4}), let

\begin{aligned} &w=1212132121321234\ {\rm for}\ {\it Cus}_7^{\rm H_4},{\it Cus}_8^{\rm H_4}.& \end{aligned}

For W({\rm E_8}), let

\begin{aligned} &w=1231423454657658\ {\rm for}\ {\it Cus}_5^{\rm E_8},{\it Cus}_6^{\rm E_8};&\\ &w=1234234542345654765876\ {\rm for}\ {\it Cus}_9^{\rm E_8},{\it Cus}_{10}^{\rm E_8};&\\ &w=123142314542345654765876\ {\rm for}\ {\it Cus}_{11}^{\rm E_8},{\it Cus}_{12}^{\rm E_8};&\\ &w=12314231454231456542345678\ {\rm for}\ {\it Cus}_{13}^{\rm E_8},{\it Cus}_{14}^{\rm E_8};&\\ &w=12314231545231436542314354265431765423456878\ {\rm for}\ {\it Cus}_{22}^{\rm E_8},{\it Cus}_{21}^{\rm E_8};&\\ &w=1231423154523165456237654567238765456782345678\ {\rm for}\ {\it Cus}_{23}^{\rm E_8},{\it Cus}_{24}^{\rm E_8}.& \end{aligned}

By running Algorithm algA, we get {\it Flag}=0 in each case. Hence these w satisfy w\in\phi^{-1}({\it Cus}_i^M), |w|=l({\it Cus}_i^M) and E_{{\rm sig}(w)}\cap \phi^{-1}({\it Cus}_j^M)=\varnothing.

Proof (Proof of Theorem cmp) of Remark 6.12

By Lemma cuslemma and Proposition cusprop, it suffices to show that for each two cuspidal classes {\it Cus}_i^M and {\it Cus}_j^M that M={\rm H_4},{\rm F_4},\text{or}\ {\rm E_8} and l({\it Cus}_i^M)=l({\it Cus}_j^M), there exists a word w\in\phi^{-1}({\it Cus}_i^M) such that |w|=l({\it Cus}_i^M) and E_{{\rm sig}(w)}\cap \phi^{-1}({\it Cus}_j^M)=\varnothing. By Proposition f4prop and Remark h4e8rmk, such w exists for each case, so the theorem holds.

Proof

Proof (Proof of Theorem MT)

3FST

By Theorem clsp1, clsp2 and expp, any finite irreducible Coxeter group has an ISS with a lower triangular ISM. Thus any finite Coxeter group has an ISS with a lower triangular ISM by Corollary dpc. Let W be a finite Coxeter group, \rho_1 and \rho_2 are two finite dimensional complex linear representations of W. It is obvious that \rho_1\cong\rho_2 implies d(S,\rho_1)=d(S,\rho_2). Conversely, by Proposition keyprop, d(S,\rho_1)=d(S,\rho_2) implies \rho_1\cong\rho_2.

Corollary

Corollary 7.1

X4OS

Let (W,S) be a finite Coxeter system. Let \rho_1,\ldots,\rho_r be a complete list of irreducible representations of W.

  1. (a) For \rho\in\mathcal{R}(W), \rho\cong\bigoplus_{i=1}^rk_i\rho_i if and only if d(S,\rho)=\prod_{i=1}^{r}d^{k_i}(S,\rho_i).

  2. (b) There is a semigroup isomorphism

    \widehat{\mathcal R}(W)\cong\mathscr{D}(W,S).

    In particular, \mathscr{D}(W,S)=\langle d(S,\rho_1),\ldots,d(S,\rho_r)\rangle is a free commutative semigroup.

Remark

Remark 7.2

IR6W

Let

\tilde{d}(S,\rho)\coloneqq\det[I+x_1\rho(s_1)+\cdots+x_n\rho(s_n)]

and

\widetilde{\mathscr{D}}(W,S)\coloneqq\{\tilde{d}(S,\rho):\rho\in{\mathcal R}(W)\}.

Replacing d and \mathscr{D} by \tilde{d} and \widetilde{\mathscr{D}}, Theorem MT and Corollary MC still hold.

Example

Example 7.3

5E0B

For W=W({\rm H_3}), there are 10 conjugacy classes

\begin{aligned} &C(\varepsilon), C(3), C(23), C(13), C(12), C(123), C(2323),\\ &C(12323), C(123212323), C(123213232132323). \end{aligned}

where C(w)\coloneqq\{g\phi(w)g^{-1}:g\in W\}. Let

\begin{aligned} &\alpha_1=(0,0,0), \alpha_2=(0,0,1), \alpha_3=(0,1,1), \alpha_4=(1,0,1), \alpha_5=(1,1,0),\\ &\alpha_6=(1,1,1), \alpha_7=(0,2,2), \alpha_8=(1,2,2), \alpha_9=(2,4,3), \alpha_{10}=(3,6,6). \end{aligned}

By performing calculations in MATLAB, it follows that

\begin{pmatrix} V_{\alpha_1}\\ V_{\alpha_2}\\ \vdots\\ V_{\alpha_{10}}\\ \end{pmatrix} = \begin{pmatrix} 1&0&0&0&0&0&0&0&0&0\\ 0&1&0&0&0&0&0&0&0&0\\ 0&0&2&0&0&0&0&0&0&0\\ 0&0&0&2&0&0&0&0&0&0\\ 0&0&0&0&2&0&0&0&0&0\\ 0&0&0&0&0&6&0&0&0&0\\ 4&0&0&0&0&0&2&0&0&0\\ 0&20&0&0&0&0&0&10&0&0\\ 0&738&0&0&0&468&0&36&18&0\\ 0&194450&0&0&0&24690&0&192000&9240&40 \end{pmatrix}.

Hence these signatures form an ISS with a lower triangular ISM of W({\rm H_3}).

Example

Example 7.4

PA22

For the symmetric group S_n=W({\rm A_{n-1}}), we display permutation \pi\in S_n by cycle notation

\pi=(i_1,i_2,\ldots,i_l)\cdots(i_m,i_{m+1},\ldots,i_n),\quad i_j\in[n].

Then Coxeter generating set T=\{(1,2),\ldots,(n-1,n)\}. A partition of n is a sequence

\mu=(p_1,\ldots,p_k)

where the p_i are weakly decreasing and \sum_{i=1}^kp_i=n, written by \mu\vdash n. Suppose \pi consists of k cycles whose lengths are q_1,\ldots,q_k respectively where q_1\geq\cdots\geq q_k, the type of \pi is defined as (q_1,\ldots,q_k), which is a partition of n. Let

C^{\mu}\coloneqq\{\pi:{\it type}\ \pi=\mu\}.

Let \mu_1,\ldots,\mu_r be all partition of n, then all the conjugacy classes of S_n are C^{\mu_1},\ldots,C^{\mu_r}. For any \mu_j=(p_{j1},\ldots,p_{jk_j})\vdash n, signature

\alpha(\mu_j)\coloneqq(\underbrace{1,\ldots,1}_{p_{j1}-1},0,\underbrace{1,\ldots,1}_{p_{j2}-1},0,\ldots,0,\underbrace{1,\ldots,1}_{p_{jk_j}-1}).

It follows that E_{\alpha(\mu_j)}\subseteq\phi^{-1}(C^{\mu_j}) and |E_{\alpha(\mu_j)}|=(n-k_j)!. Hence \alpha(\mu_1),\ldots, \\ \alpha(\mu_r) is an ISS with a lower triangular ISM

diag((n-k_1)!,\ldots,(n-k_r)!).

Each partition \mu\vdash n corresponds to a irreducible representation \rho^{\mu} named Young’s natural representation, see [S]. All the \rho^{\mu} form a complete list of irreducible representations of S_n. The characters and the characteristic polynomials of Young’s natural representations for S_3,S_4 are given in the following tables.

The paper has a table here, which this page does not show.
The paper has a table here, which this page does not show.

As an application of Theorem MT, the characteristic polynomials of representations can help us calculate the decomposition of the restricted representation \rho\downarrow_{S_m}^{S_n} for m<n. Let \rho\in\mathcal{R}(S_n) and T_n be the Coxeter generating set of S_n. It follows that

d(T_m,\rho\big\downarrow_{S_m}^{S_n})=d(T_n,\rho)(x_0,x_1,\ldots,x_{m-1},0,\ldots,0).

So the decomposition of \rho\downarrow_{S_m}^{S_n} corresponds to the decomposition of the right side of (?). For instance,

\begin{aligned} &d(T_3,\rho^{(3,1)}\big\downarrow_{S_3}^{S_4})\\ =&d(T_4,\rho^{(3,1)})(x_0,x_1,x_2,0)\\ =&x_0^3-x_1^3-x_2^3+x_0^2x_1+x_0^2x_2-x_0x_1^2-x_0x_2^2+x_0x_1x_2\\ =&(x_0+x_1+x_2)(x_0^2-x_1^2-x_2^2+x_1x_2)\\ =&d(T_3,\rho^{(3)})d(T_3,\rho^{(2,1)}). \end{aligned}

So we have \rho^{(3,1)}\downarrow_{S_3}^{S_4}\cong\rho^{(3)}\oplus\rho^{(2,1)}, which can also be obtained by Branching Rule, see [S].

Remark

Remark 7.5

4E14

In this paper, by studying the existence of the independent signature sequence, we generalize the main result of [CST] to any finite Coxeter groups. A semigroup isomorphism between representations and their characteristic polynomials for finite Coxeter groups is established. We conjecture such sequence may exist for a larger category of finitely generated groups so that the result can be further generalized. On the other hand, there are many interesting topics about these polynomials, such as whether the characteristic polynomial for an irreducible representation is irreducible, how to calculate the characters from the polynomials directly. Therefore, we need more efforts to work on these topics.

Source and licence

The text on this page is from arXiv:2504.19468v1 by Shoumin Liu and Yuxiang Wang, under the licence CC0 1.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2504/2504.19468.jsonl.