0
$\begingroup$

As you know, a Matrix $A^{+}\in \mathbb{R}^{m\times n}$ is called a a pseudoinverse of $A\in \mathbb{R}^{n\times m}$ if $\Vert b-A A^{+} b \Vert_2\leq\Vert b- A y\Vert_2 \forall b\in \mathbb{R}^{n} \forall y\in \mathbb{R}^m$ and $A^{+}b$ is the smallest vector by norm to do so, which is equally characterized by $\langle b- A A^{+}b,Ay\rangle=0$, since $A A^{+}$ is the orthogonal projection of $\mathbb{R}^{n}$ onto $R(A)$.

Because of it being a orthogonal projection the penrose axioms ${AA^{+}}^T=AA^{+}$ and $A A^{+}A=A$ follow. What I would like to do now is proof the other two axioms $A^{+}A A^{+}=A^{+}$ and ${A^{+} A}^T= A^{+}A$ given a Matrix and its pseudoinverse, by showing that ${(A^{+})}^{+}=A$ or in other words, that $A^{+}A$ is a projection onto $R(A^{+})=N(A)^{\perp}$. On the other hand, if those two penrose-axioms could be proven another way ${(A^{+})}^{+}=A$ would follow by the penrose-axioms.

I would appreciate any proof - especially one that is not using SVD.

$\endgroup$
0
$\begingroup$

$A(b-A^{+}Ab)=0\Rightarrow b-A^{+}Ab\in $N(A)$\Rightarrow \langle b-A^{+}Ab,A^{+}y\rangle=0$ since $A^{+}y \in R(A^{+})={N(A)}^{\perp}$

$\endgroup$

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.