5
$\begingroup$

I am wondering if there is a nice categorical description of matrix similarity, i.e. the equivalence relation on matrices given by $A\sim B \iff A=QBQ^{-1}$, for some invertible $Q$.

In particular, I am considering the matrices as linear maps from a vector space into itself, which in turn forms a category $\mathrm{End}(V)$ with one object $V$ and arrows linear maps. We can then form the functor category $\mathrm{End}(V)^{\mathbb2},$ where $\mathbb2$ is the category with $2$ objects and $1$ arrow between them, i.e. this is the 2-category of linear maps and pairs of maps between them making the following commute:

\begin{array}{ccccccccc} V & \xrightarrow{A} & V \\ \downarrow & & \downarrow \\ V & \xrightarrow{B} & V \end{array}

So similarity between $A$ and $B$ is stronger than isomorphism in this category, as $(C,D):A\rightarrow B$ is an isomorphism if and only if $C$ and $D$ are linear isomorphisms, whereas similarity would require $C=D$. Do similar matrices in this category obey some characterising or otherwise interesting property, or is isomorphism the strongest we can get?

$\endgroup$
5
$\begingroup$

$\DeclareMathOperator\rank{rank}$If you allow $C\neq D$, an isomorphism $A\cong B$ in the category you describe is equivalent to $\rank(A)=\rank(B)$. You can of course define a category where as morphisms you only allow commutative squares with $C=D$, then an isomorphism $A\cong B$ is indeed equivalent to $A$ and $B$ being similar.

Another way you might want to look at this is in the setting of $K[X]$-modules, where $K$ is the underlying field. A $K[X]$-module amounts to a $K$-vector space $V$ together with a $K$-linear map $f\colon V\to V$ that describes the action of $X$, that is $f(v) = X\cdot v$. In this setting, matrices $A$ and $B$ yield $K[X]$-modules and those are isomorphic if and only if $A$ and $B$ are similar.

$\endgroup$

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.