0
$\begingroup$

I wanted to prove/disprove whether for every matrix $A\in \operatorname{Mat}_{n \times n}$

$\operatorname{rank} (A^2) = \operatorname{rank}(A)$ for any matrix $A\in \operatorname{Mat}_{n \times n}$

I know that for two matrices $A, B$ (it actually doesn't matters what is the size of $B$ as long as $AB$ is defined)

$$\operatorname{rank}(AB) \leq \min \{ \operatorname{rank}(A), \operatorname{rank}(B)\}$$

Therefore, $$\operatorname{rank}(A^2) \leq \operatorname{rank} (A)$$

In order to prove that:

$$\operatorname{rank}(A^2) = \operatorname{rank}(A)$$

you'd have to prove that: $$\operatorname{rank}(A^2) \geq \operatorname{rank}(A)$$ which I don't know how to prove, or finding a different way to prove directly both directions (meaning; one direction is $\leq$ and the other is $\geq$)

|cite|improve this question
$\endgroup$
  • $\begingroup$ What about nilpotent matrices? $\endgroup$ – Davide Morgante Aug 5 '18 at 15:16
  • $\begingroup$ If $A \in \operatorname{Mat}_{m\times n}$ then $A^2$ exists only if $n=m. \qquad$ $\endgroup$ – Michael Hardy Aug 5 '18 at 15:22
  • $\begingroup$ In general, for a square matrix $A$, $\text{rank}(A^2)=\text{rank}(A)$ if and only if the column space of $A$ intersects the nullspace of $A$ trivially. $\endgroup$ – Batominovski Aug 5 '18 at 16:50
4
$\begingroup$

It is not true that the square of a matrix has the same rank as the matrix itself. For instance, the $2\times 2$ matrix $A= \left[ {\begin{array}{cc} 0 & 1 \\ 0 & 0 \\ \end{array} } \right]$ has rank $1$ and satisfies $A^2=0$.

|cite|improve this answer
$\endgroup$
  • $\begingroup$ but here's what I'm confused about - for a squared matrix, $\text rank (A) = \text rank (A^T)$, so $\text rank (AA^T) = \text rank (A^2)$ $\endgroup$ – Jneven Aug 5 '18 at 15:20
  • 2
    $\begingroup$ @Jneven This is a totally separate problem, in general $A^T$ has nothing to do with powers of $A$. I suggest you look at this then: math.stackexchange.com/questions/2315/… $\endgroup$ – Suzet Aug 5 '18 at 15:21
  • $\begingroup$ but $\text rank (A) \leq \text {min}$ {${sp\{R(A)\}}, {sp\{C(A)\}}$ $\endgroup$ – Jneven Aug 5 '18 at 15:32
  • $\begingroup$ @Jneven The rank is not multiplicative. It is not true that $\operatorname{rank}(AB)=\operatorname{rank}(A)\operatorname{rank}(B)$ as you seem to suggest. The quantities $\operatorname{rank}(AA^T)$ and $\operatorname{rank}(A^2)$ are not equal. The matrix $A$ I exhibed in my answer is a counter example for it. I let you compute $AA^T$ and see that it is not zero, whereas $A^2=0$ ! $\endgroup$ – Suzet Aug 5 '18 at 15:38
  • 1
    $\begingroup$ Yes, and this is true. However, $\operatorname{rank}(AA^T)=\operatorname{rank}(A^2)$ is not true. All you can say, using your inequality, is $\operatorname{rank}(AA^T)\leq \operatorname{rank}(A)$ (and also $\operatorname{rank}(A^2)\leq \operatorname{rank}(A)$). $\endgroup$ – Suzet Aug 5 '18 at 15:46
4
$\begingroup$

First, it makes no sense to write $A^2$ if $A$ is not a square matrix. So you need $m=n$.

This being said, there is a class of matrices called nilpotent, from which it is trivial to build counterexamples:

$$\operatorname{rk}\left(\matrix{0&1\\0&0}\right)=1$$$$\operatorname{rk}\left(\matrix{0&1\\0&0}\right)^2=0$$

|cite|improve this answer
$\endgroup$
  • $\begingroup$ I would have to disagree about the statement that "it would make no sense". in fact, I think writing $AA$ would just be weird. $\endgroup$ – Jneven Aug 5 '18 at 15:18
  • 2
    $\begingroup$ @Jneven No, seriously, it would make no mathematical sense. A product of matrices is defined when the number of columns of the left factor is the number of rows of the right factor. $\endgroup$ – Arnaud Mortier Aug 5 '18 at 15:19
  • $\begingroup$ you're right, I understood my confusion. $\endgroup$ – Jneven Aug 5 '18 at 15:22

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.