[Math] Properties of the pseudoinverse of a matrix

linear algebra

Let $A $ be a $n \times m $ and matrix and let $B = (A^T A)^{-1} A^T $ be its pseudo inverse.
If $A$ and $C$ satisfies
$CAC = C$, $ACA = A$, $(CA)^T = CA$, and $(AC)^T = AC $, then can we conclude that $C$ is the pseudo inverse of $A$ ??

From $ACA = C$ we obtain $A = C (CA)^{-1} $ and so $ACA = C (CA)^{-1}$ taking transposes we obtain $(ACA)^T = (C (CA)^{-1} )^T \implies A^T AC = (CA)^{-T} C^T$. But I get stuck here. I dont see any way to express $C$ by itself. Any help would be greatly appreciated.

Best Answer

For this to work, we need to assume, that $A^TA$ is invertible of course. That is the case, if the columns of $A$ are linearly independent. If that is the case, then:

$(A^TA)C = A^T(AC) = (A^T(AC))^{TT} =((AC)^TA^{TT})^T =(ACA)^T = A^T$

So: $C = (A^TA)^{-1}A^T$.

You can also prove, that: $C = A^T(AA^T)^{-1}$

The other two conditions you gave are not needed to prove that. However, they make $C$ the unique matrix with these 4 properties. For if $C'$ is another matrix satisfying said properties, then:

$AC = (AC)^T = C^TA^T = C^T(AC'A)^T = C^TA^TC'^TA^T = (AC)^T(AC')^T = ACAC' = AC'$ and similarly $CA = C'A$. So: $C = CAC = CAC' = C'AC' = C'$.

Does $C$ satisfy those 4 properties, if $C = (A^TA)^{-1}A^T$? Yes, it does:

We have: $CA = (A^TA)^{-1}A^T A = I$. Hence:

$CAC = C$

$ACA = A$

$(CA)^T = CA$

and finally:

$(AC)^T = C^TA^T = ((A^TA)^{-1}A^T)^TA^T = (A^{TT}((A^TA)^{-1})^T)A^T = A(A^TA)^{-1}A^T = AC$

(Hetebrij already pointed this out in the comments. I'm not that fast ;) ).

Related Question