Linear maps preserving matrices annihilated by a fixed polynomial

TL;DR

This paper characterizes linear maps between matrix algebras that preserve matrices annihilated by a fixed polynomial, providing a standard form under certain conditions, extending results from idempotent preservers to more general polynomial annihilators.

Contribution

It introduces a comprehensive description of linear preservers of matrices annihilated by a fixed polynomial with multiple roots, generalizing known results for idempotent preservers.

Findings

The standard form of such linear maps is characterized under specific conditions.

The form involves tensor products, similarity transformations, and diagonal matrices with zero multipliers.

Special case analysis reduces to the study of linear idempotent preservers.

Abstract

Let ${\bf M}_n(\mathbb{F})$ be the algebra of $n\times n$ matrices over an arbitrary field $\mathbb{F}$. We consider linear maps $\Phi: {\bf M}_n(\mathbb{F}) \rightarrow {\bf M}_r(\mathbb{F})$ preserving matrices annihilated by a fixed polynomial $f(x) = (x-a_1)\cdots (x-a_m)$ with $m\ge 2$ distinct zeroes $a_1, a_2, \ldots, a_m \in \mathbb{F}$; namely, $$ f(\Phi(A)) = 0\quad\text{whenever} \quad f(A) = 0. $$ Suppose that $f(0)=0$, and the zero set $Z(f) =\{a_1, \dots, a_m\}$ is not an additive group. Then $\Phi$ assumes the form \begin{align}\label{eq:standard} A \mapsto S\begin{pmatrix} A \otimes D_1 &&\cr & A^{T} \otimes D_2& \cr && 0_s\cr\end{pmatrix}S^{-1}, \tag{$\dagger$} \end{align} for some invertible matrix $S\in {\bf M}_r(\mathbb{F})$, invertible diagonal matrices $D_1\in {\bf M}_p(\mathbb{F})$ and $D_2\in {\bf M}_q(\mathbb{F})$, where $s=r-np-nq\geq 0$. The diagonal entries $\lambda$ in $D_1$ and $D_2$, as well as $0$ in the zero matrix $0_s$, are zero multipliers of $f(x)$ in the sense that $\lambda Z(f) \subseteq Z(f)$. In general, assume that $Z(f) - a_1$ is not an additive group. If $\Phi(I_n)$ commutes with $\Phi(A)$ for all $A\in {\bf M}_n(\mathbb{F})$, or if $f(x)$ has a unique zero multiplier $\lambda=1$, then $\Phi$ assumes the form \eqref{eq:standard}. The above assertions follow from the special case when $f(x) = x(x-1)=x^2-x$, for which the problem reduces to the study of linear idempotent preservers. It is shown that a linear map $\Phi: {\bf M}_n(\mathbb{F}) \rightarrow {\bf M}_r(\mathbb{F})$ sending disjoint rank one idempotents to disjoint idempotents always assume the above form \eqref{eq:standard} with $D_1=I_p$ and $D_2=I_q$, unless ${\bf M}_n(\mathbb{F}) = {\bf M}_2(\mathbb{Z}_2)$.