[Math] Normalizer of $S_n$ in $GL_n(K)$

abstract-algebrafinite-groupsgroup-theory

In the exercises on direct product of groups of Dummit & Foote, I proved that the symmetric group $S_n$ is isomorphic to a subgroup of $GL_n(K)$, called the permutation matrices with one 1 in each row and each column.

My question is how can I find the normalizer of this subgroup in $GL_n(K)$?

Best Answer

Edit: I've revised the answer to make it more elementary, and to fix the error YACP pointed out (thank you).

Suppose $X\in N_{GL_n(K)}(S_n)$. Then for every permutation matrix $P\in S_n$ we have $XPX^{-1}\in S_n$, so conjugation by $X$ is an automorphism of $S_n$. If $n\ne 2, 6$, then as YACP noted it must be an inner automorphism, i.e. we have some $P'\in S_n$ such that for every $P\in S_n$, $XPX^{-1}=P'P{P'}^{-1}$. Thus $(X^{-1}P')P(X^{-1}P')^{-1}=P$, so $X^{-1}P'\in C_{GL_n(K)}(S_n)$ (the centralizer of $S_n$). Thus $X\in C_{GL_n(K)}(S_n)\cdot S_n$, so all we have to do is find $C_{GL_n(K)}(S_n)$, as $C_{GL_n(K)}(S_n)\cdot S_n\subseteq N_{GL_n(K)}(S_n)$ holds trivially.

Let $\mathcal C$ denote of all matrices (including non-invertible ones) $X$ such that $PXP^{-1}=X$ for all $P\in S_n$. Note that conjugation is linear, i.e. $A(X+Y)A^{-1}=AXA^{-1}+AYA^{-1}$ for any $A,X,Y\in M_{n\times n}(K)$, so $\mathcal C$ is closed under addition. Conjugation also respects scalar multiplication, i.e. $AcXA^{-1}=cAXA^{-1}$, so $\mathcal C$ is closed under scalar multiplication. Recall that $M_{n\times n}(K)$ is a vector space over $K$, so this makes $\mathcal C$ a subspace of $M_{n\times n}$. The use of $\mathcal C$ is that $C_{GL_n(K)}(S_n)=\mathcal C\cap GL_n(K)$, yet unlike $C_{GL_n(K)}(S_n)$ it is a vector subspace, and vector subspaces are nice to work with.

It is easy to see that $\mathcal C$ contains diagonal matrices $D$ with constant diagonal, as well as all matrices $M$ such that the entries $m_{ij}$ are the same for all $i,j$. Since $\mathcal C$ is a vector subspace, this means it contains all sums of these matrices as well. We want to show that every matrix in $\mathcal C$ can be written as $D+M$ where $D$ and $M$ are as above. If $X\in \mathcal C$ then we can subtract a diagonal matrix $D$ and a matrix $M$ of the second kind to get the upper left and right entries to be $0$: $$X-D-M=\begin{pmatrix} 0 & x_{12} & \cdots & x_{1n-1} & 0\\ x_{21} & x_{22} & \cdots & x_{2n-1} & x_{2n}\\ \vdots & \vdots & \ddots & \vdots & \vdots\\ x_{n1} & x_{n2} & \cdots & x_{nn-1} & x_{nn}\\ \end{pmatrix}$$ Call this matrix $X'$; we wish to show $X'=0$. Exchanging the second and last column must be the same as exchanging the second and last row, and since the first action switches $x_{12}$ and $x_{1n}$ while the second leaves the first row unchanged we have $x_{12}=x_{1n}=0$. Continuing in this manner we see that the whole first row is $0$. Exchanging the first and second row is the same as exchanging the first and second column, so the whole second row must be $0$ as well. Continuing in this manner we get that $X'=0$ as desired. Thus $\mathcal C$ is the set of matrices of the form $D+M$, i.e. with $a$ on the diagonal and $b$ off the diagonal.

$C_{GL_n(K)}(S_n)$ is the set of such matrices with nonzero determinant. Let $X\in \mathcal C$ have entries $a$ on the diagonal and $b$ off it. Clearly if $a=b$ then the determinant is $0$, so suppose $a\ne b$. Then we can write $X=(a-b)(I_n+cr)$ where $c$ is a column consisting entirely of $1$'s and $r$ is a row consisting entirely of entries $\frac{b}{a-b}$. By Sylvester's Determinant Theorem the determinant of this is $(a-b)^n(1+rc)$, and $rc=\frac{nb}{a-b}$, which gives us $\det(X)=(a-b)^{n-1}(a-b+nb)$. Thus for any $X\in \mathcal C$, $\det(X)=0$ iff either $a=b$ or $a=(1-n)b$.

Putting this all together, we get that $$N_{GL_n(K)}(S_n)=\left\{\begin{pmatrix} a & b & \cdots & b \\ b & a & \ddots &\vdots \\ \vdots &\ddots & \ddots & b\\ b & \cdots & b & a\\ \end{pmatrix}P: a\neq b, a\neq (1-n)b, P\in S_n \right\}$$