In the algebraic group $G = \operatorname{PGL}_4(\mathbb{C})$, let $E$ denote the subgroup of elements of order dividing 2 in the diagonal maximal torus; it is generated by the images of the three matrices $$ \begin{bmatrix} 1 & 0 & 0 & 0\\ 0 & -1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & -1 \\ \end{bmatrix},\quad \begin{bmatrix} 1 & 0 & 0\\ 0 & -1 & 0 \\ 0 & 0 & I_2 \end{bmatrix},\quad \begin{bmatrix} -1 & 0 \\ 0 & I_3 \end{bmatrix}. $$ Direct computation shows that $N_{G}(E)/C_{G}(E) \cong S_4\,$, which is the Weyl group of $G$. Analogous result can be obtained for $n = 3, 5, 6, 7, 8, 9, 10$ in $\operatorname{PGL}_{n}(\mathbb{C})$ by direct calculations. Is this observation true for any $n$? Is there any reference relevant?


1 Answer 1


$\newcommand{\ZZ}{{\mathcal Z}_G} \newcommand{\NN}{{\mathcal N}_G} \newcommand{\zz}{{\mathfrak z}_G} \newcommand{\Lie}{{\rm Lie\,}} \renewcommand{\tt}{{\mathfrak t}} \renewcommand{\gg}{{\mathfrak g}} \newcommand{\X}{{\sf X}} \newcommand{\Z}{{\Bbb Z}}$ Yes, this is true for any $n\ge 3$.

Let $G$ be a semisimple group of adjoint type over an algebraically closed field $k$ of characteristic 0. Let $T\subset G$ be a maximal torus. Write $$E=T^{(2)}:= \{t\in T(k)\ |\ t^2=1\}.$$ We wish to compute the centralizer $\ZZ(E)$ and the normalizer $\NN(E)$.

Observe that $\NN(E)\supseteq \ZZ(E)\supseteq T$. We compute $\zz(E):=\Lie \ZZ(E)$.

Lemma 1. $\zz(E)=\tt:=\Lie T$.

Proof. Since $\ZZ(E)\supseteq T$, we have $\zz(E)\supseteq\tt$. Write the root decomposition $$ \Lie G=\tt\oplus\bigoplus_{\beta\in R}\gg_\beta\,,$$ where $R=R(G,T)\subset \X^*(T)$ is the root system. Since $\zz(E)\supseteq\tt$, we have $$ \zz(E)=\tt\oplus\bigoplus_{\beta\in M}\gg_\beta$$ for some subset $M\subseteq R$. Here $$M=\{\beta\in R\ |\ \beta(t)=1\ \forall t\in E\}.$$ Let $S\subset R$ be a system of simple roots (a basis of $R$). Since $G$ is of adjoint type, the set $S\subset R\subset \X^*(T)$ is a basis of the character group $\X^*(T)$ of $T$. It follows that for any simple root $\alpha\in S$, there exists $t\in E=T^{(2)}$ such that $\alpha(t)=-1$. Let $W=W(G,T)=\NN(T)/T$ be the Weyl group. The group $W$ acts on $E$ and on $R$, and $W\cdot S=R$. Therefore, for any root $\beta\in R$, there exists $t\in E$ such that $\beta(t)=-1$, and therefore $M=\varnothing$ and $\zz(E)=\tt$, as required.

We compute $\NN(E)$. Since ${\rm char}(k)=0$, it follows from Lemma 1 that the identity component $\ZZ(E)^0$ of $\ZZ(E)$ is $T$. We see that $\NN(E)$ normalizes $\ZZ(E)$, and hence it normalizes $\ZZ(E)^0=T$. It follows that $\NN(E)\subseteq \NN(T)$. On the other hand, $\NN(T)$ normalizes $T$, and therefore, it normalizes $E=T^{(2)}$, whence $\NN(T)\subseteq \NN(E)$. Thus $\NN(E)=\NN(T)$.

We wish to compute $\ZZ(E)$. Consider the Weyl group $W=\NN(T)/T$. Set $$W_E=\ZZ(E)/T\subseteq \NN(T)/T=W.$$ Then $W_E$ is a finite group, the kernel of the natural homomorphism $W\to{\rm Aut\,} E$, and $\ZZ(E)$ is the preimage of $W_E$ in $\NN(T)$.

Lemma 2. Let $W'=S_n$ (the permutation group on $n$ symbols) naturally acting on the set $$E'=\ker\, \Sigma\colon (\Z/2\Z)^n\to \Z/2\Z,$$ where $\Sigma(x_1,x_2,\dots,x_n)=x_1+x_2+\cdots+x_n\,.$ If $n\ge 3$, then this action is effective (has trivial kernel).

Proof. Left to OP and the reader.

Observe that Lemma 2 is false for $n=2$, when ${\rm Aut\,} E'=\{1\}$ whereas $S_2\neq \{1\}$.

Theorem. Let $G={\rm PGL}_n$ with $n\ge 3$. Then $\NN(E)/\ZZ(E)=\NN(T)/T=W$.

Proof. We have seen that $\NN(E)=\NN(T)$. By Lemma 2, if $n\ge 3$, then $W_E=\{1\}$, whence $\ZZ(E)=T$, and the theorem follows.

Edit. Observe that the analogues of Lemma 2 and the theorem do not hold for the adjoint group ${\rm SO}(2n+1)$ of type ${\sf B}_n$ for $n\ge 1$. Indeed, then $$W\simeq(\Z/2\Z)^n\rtimes S_n\,,$$ and the normal subgroup $(\Z/2\Z)^n\subseteq W$ acts on $E=T^{(2)}\cong (\pm1)^n$ trivially. Therefore, $\ZZ(E)\neq T$.

  • $\begingroup$ Thank you for the answer. I got confused by Lemma 2. What is the map $\Sigma$? $\endgroup$
    – user488802
    Jul 31, 2022 at 1:02
  • $\begingroup$ I have added the definition of the map $\Sigma$. $\endgroup$ Jul 31, 2022 at 5:51
  • $\begingroup$ Is the proof that $C_{G}(E)^{\circ} = T$ a must? Thank you, $\endgroup$
    – user488802
    Jul 31, 2022 at 6:47
  • $\begingroup$ Yes. The argument goes as follows: ${\mathcal N}_G(E)$ normalizes ${\mathcal Z}_G(E)$, and therefore, it normalizes ${\mathcal Z}_G(E)^0=T$. We conclude that ${\mathcal N}_G(E)\subseteq {\mathcal N}_G(T)$. $\endgroup$ Jul 31, 2022 at 7:44
  • 1
    $\begingroup$ @LSpice: In positive characteristic, I cannot write two lines without mistakes. So I suggest that you write, say, in comments, your proof that ${\mathcal Z}_G(E)^0=T$ also in positive characteristic. $\endgroup$ Aug 1, 2022 at 18:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.