The inverse of a transpose matrix to "cancel" the transpose?

5k Views Asked by At

When it comes to solving and equation containing matrices I don't always understand some of the rules involved. In particular, I am trying to figure out the derivation of the Gauss-Newton algorithm. In this particular case I have to solve for $z$:

$A^TAz = A^Tb$

According to wikipedia, the solution to this is:

$z = (A^TA)^{-1}A^Tb$

I understand what they are doing here, but why can't you do the following?:

$(A^T)^{-1}A^TAz = (A^T)^{-1}A^Tb$

$Az = b$

$z = A^{-1}b$

I have had this question for some time while reading various derivations. What is the underlying rule that prevents me from doing it the second way?