4
$\begingroup$

Doing practice problems for my qualifying exam and am a bit stumped by the following:

Let $n \ge 1$, $F$ be a field, and let $M$ be a matrix in $M_n(F)$. Show that there exists an $N \in M_n(F)$ such that $M = MNM$.

Here is my attempt at a solution:

We first define the vector space $V = F^n$ on which the matrices in $M_n(F)$ act. Now take $M \in M_n(F)$ and by the rank-nullity theorem we have that $$ker(M) \oplus Im(M) = V$$ Now, first note that if $M$ is invertible, then there exists $N = M^{-1}$ such that $NM = I$ and moreover $$M = MNM$$ giving us the result. We now assume that $M$ is not invertible (ker$(M) \neq 0$). We have by the First isomorphism theorem that $$V / ker(M) \cong Im(M)$$ I want to use the above fact with the property that ideals of fields are trivial to give us that $Im(M) = 0$ and therefore $M$ is just the zero map or something to this effect?

$\endgroup$
4
  • 1
    $\begingroup$ I don't think what you want to show can be true. There are matrices over fields, such as $$\begin{pmatrix}1&0\\0&0\end{pmatrix},$$which are not invertible but are also not the zero map. You're definitely on the right track, though -- I'd try thinking more about using the kernel to construct $N$. $\endgroup$ Commented Jan 4, 2023 at 5:21
  • 2
    $\begingroup$ A matrix-theoretical proof is probably simpler here: let $M=UV^T$ be a rank factorisation, $X$ be a right inverse of $V^T$ and $Y^T$ be a left inverse of $U$. Then $N=XY^T$ will do the job. $\endgroup$
    – user1551
    Commented Jan 4, 2023 at 5:45
  • $\begingroup$ @user1551 can we safely assume that the left and right factorization matrices are invertible? $\endgroup$ Commented Jan 4, 2023 at 15:28
  • 1
    $\begingroup$ Yes, because $U$ has full column rank and $V^T$ has full row rank. $\endgroup$
    – user1551
    Commented Jan 4, 2023 at 18:36

1 Answer 1

4
$\begingroup$

Many times, if you need to define a map, you can think about what it does on a basis. We also know that if $M$ were invertible then we could use $M^{-1}$ so maybe we need to get some sort of inverse $M' : \operatorname{im}(M) \to F^n$.

In terms of a basis, I will let $e_1, \dots, e_n$ be a basis for $F^n$ such that $e_{m + 1}, \dots, e_n$ is a basis for $\ker M$ and $f_i = Me_i$ for $i = 1, \dots, m$ is a basis for $\operatorname{im} M$.

The $M'$ I was talking about above is given by $M'f_i = e_i$. If you play around with this a bit, you should be able to construct $N$ (complete $f_1, \dots, f_m$ to a basis for $F^n$ and define $Nf_i = e_i$).

In terms of matrices, factoring $M$ through the projection $P : F^n \to \operatorname{span}\{e_1,\dots,e_m\}$ gives a rank factorization $M = QP$ where $Qe_i = Me_i$ and $P$ is $m \times n$ and $Q$ is $n \times m$. This was talked about in the comments.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .