10

For every non-square matrix prove that $AA^t$ or/and $A^tA$ is singular.

Like the title, I want to prove this and I tried to think of ways to prove it but I couldn't think of some..

I know by this answer that $AA^t$ is symmetric but I cant make the connection.

If someone has at least a hint to that It'll be great if you could write it in the comment section so I could give it a shot!

Thanks in advance.

Community
  • 1
Zik332
  • 161
  • 3
    Use the fact that $rank(AB) \leq rank(A)$ for any matrices $A,B$ for which the product is defined. Additionally, rank($A$) = rank($A^t$). Finally, look at the dimensions of $AA^t$ and $A^tA$ and you will see that at least one must be rank deficient, i.e. singular. – Ken Duna Oct 24 '16 at 21:54
  • 1
    Hint: If $A$ is an $n\times m$ matrix with $n < m$, then $A$ must not be one-to-one. I.e., there is a $v \neq 0$ such that $Av = 0$. Then the same is true of $A^t A$ (just try applying it to $v$ as before). – Alex G. Oct 24 '16 at 21:59
  • @KenDuna So because $rank(A^tA)\leq rank(A)$ (and $AA^t$) I need to prove that at least one of them is less than $rank(A)$, because Invertible matrix has full rank, correct? – Zik332 Oct 24 '16 at 22:13
  • @AlexG. How can I leverage the fact that $A^tA$ is not one-to-one? I mean, how can that help me prove the question? – Zik332 Oct 24 '16 at 22:16
  • 1
    @Zik332 A square matrix is singular iff it is not one-to-one iff it is not onto. – Alex G. Oct 24 '16 at 22:23
  • @AlexG. Sorry but I cant seem to figure out how to use your hint :\ how come $A^tA$ is not one-to-one? it is nXn, so I cant use $n<m$ like you did – Zik332 Oct 24 '16 at 22:39
  • 1
    @Zik332 As I pointed out, $v$ is in its kernel. So it is not one-to-one. – Alex G. Oct 24 '16 at 22:40
  • @AlexG. Yeah I can see that on $A$ but for $A^tA$ and $AA^t$ I cant seem to get why they are not one-to-one, I am really rusty and I truly sorry I am trying not to ask a lot of question and google a lot. – Zik332 Oct 24 '16 at 22:42

1 Answers1

6

To elaborate on my hint, suppose $A$ is an $n \times m$ matrix and $n \neq m$. It must be that $rank(A^t) = rank(A) \leq \min(n,m) < \max(n,m)$.

Using the fact that $rank(AB) \leq rank(A)$ for any $A,B$ for which the product is defined, we have that:

$$rank(AA^t) \leq rank(A) < \max(n,m)$$ $$rank(A^tA) \leq rank(A^t) < \max(n,m).$$

But it must be the case that the dimensions of $AA^t$ or $A^tA$ is $\max(n,m)$. Therefore at least one of them does not have full rank. For square matrices, not having full rank is equivalent to being singular.

Ken Duna
  • 5,746