6
$\begingroup$

Say if $A$ is an $n \times n$ matrix, why is it that if $A^{T}A$ is positive definite, the matrix $A$ is then invertible? All I know is $A^{T}A$ gives a symmetric matrix but what does $A^{T}A$ is positive definite tell or imply or hint about the matrix $A$ itself that leads to the fact that it will be invertible?

  • 3
    Positive definite implies positive eigenvalues.2011-08-06
  • 3
    determinants is product of eigenvalues2011-08-06
  • 0
    But if $K=A^{T}A$ is a positive definite, that is only the matrix $K$ that has positive eigenvalues and does not imply that $A$ will also have positive eigenvalues, am I right? Wouldn't some singular matrices give a positive eigenvalues too?2011-08-06
  • 1
    $A^tA$ will be diagonalizable with positive entries which you can scale to get something like $SPA^tAP^{-1}=I$ to get $P^{-1}SPA^tA=I$ an explicit inverse for $A$2011-08-06
  • 0
    @xEnOn: Yes, but a singular matrix necessarily has $0$ as an eigenvalue. Therefore...?2011-08-06
  • 7
    ${\rm det}(A^{T}A) = {\rm det}(A)^{2}$ when $A$ is square.2011-08-06
  • 0
    Use the fact that $K$ is positive definite to invert $K$ and this will enable you to find an inverse for $A$.2011-08-06
  • 0
    Thanks everyone. From all the help, if I'm using determinant to proof, I could say $det(K)=det(A^{T})det(A)$. And since all the eigenvalues are positive, $det(K)>0$. Based on the fact that $det(A^{T})=det(A)$, I could say $det(A)=\sqrt{det(K)}>0$ and therefore, $A$ is invertible. Using the nullspace to prove is also interesting with the one Mark and Jonas had given below. Thanks everyone for the help.2011-08-06

3 Answers 3

9

Actually $\ker(A^T A) = ker(A)$ so one is invertible if and only if the other is invertible. Proof:

If $v \in \ker(A)$ then $Av=0$ so $A^T A v = 0$, hence $v \in \ker(A^T A)$. On the other hand, if $v \in \ker (A^T A)$ then $\left\langle w,A^{T}Av\right\rangle =0$ for all $w \in \mathbb{R}^n$, so $\left\langle Aw,Av\right\rangle =0$ for all $w$, and in particular for $w=v$ you get $Av=0$.

A positive definite matrix is invertible (for instance, because it has positive eigenvalues) so you're done.

5

Hint: Show that $A$ is injective as a linear transformation on $\mathbb R^n$. This implies invertibility by the rank-nullity theorem.

  • 0
    But how to prove $A$ is injective?2013-10-18
  • 0
    @logancolss: A linear transformation $A$ is injective if and only if for all $x\neq 0$, $Ax\neq 0$. If $A^TA$ is positive definite, then for all $x\neq 0$, $x^TA^TAx>0$, which implies $Ax\neq 0$.2013-10-19
3

$A^TA$ is always non-negative and I think you don't have any problem with it. (otherwise see that $x^TA^TAx$ is nothing but $\|Ax\|^2_2 \geq 0$)

Suppose A is square and not invertible hence rank deficient. Due to the homework tag, I will just add that when two matrices multiplied they cannot form a matrix that has a higher rank than the individual matrices, for example if you multiply a vector $x x^T$ the resulting matrix cannot be of rank 2.