0
$\begingroup$

Consider a matrix question as $Q_{m \times m}X_{m \times n} S_{n \times n} = Z_{m \times n}$ where we know that $Q$ and $S$ are symmetric positive semi-definite parametric matrices. Does there exist any $X_{m \times n}$ matrix satisfying this equation (at least one)?

I don't even know how to start. Can someone give some hints?

Thanks

$\endgroup$
1
$\begingroup$

$U^+$ denotes the Moore-Penrose inverse of $U$. Let $Y=XS$. Then Part 1: $QY=Z$; if $QQ^+Z\not= Z$, then no solutions. Otherwise $Y=Q^+Z+(I-Q^+Q)W$ where $W$ is an arbitrary $m\times n$ matrix.

Part 2. One has $Y=XS$. If $YS^+S\not= Y$, then no solutions. Otherwise $X=YS^++V(I-SS^+)$, where $V$ is an arbitrary $m\times n$ matrix.

EDIT. Answer to m0_as. We can also write the equation $(Q\otimes S^T)X=Z$ where we stack the matrix row by row. cf. https://en.wikipedia.org/wiki/Kronecker_product

In the same way as above, using $(Q\otimes S^T)^+$, we can solve the equation.

Remark 1. $rank(Q\otimes S^T)=rank(Q)rank(S)$.

Remark 2. If your matrices are parametric, then make this clear in your question. It is more difficult. Use a formal calculation.

$\endgroup$
  • $\begingroup$ Thanks for your answer. But can you tell me how I can check whether there exists any $Q^+$ that $Q Q^+ Z =Z$? My Q matrix is a parametric matrix and not a numerical one. $\endgroup$ – m0_as Oct 16 '15 at 4:52

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.