0
$\begingroup$

Dear Matrix Calculus Experts,

Let $n\geq p$ and $x\in\mathbb{R}\rightarrow A(x)\in G_{n,p}(\mathbb{C)}$ be a $C^1$ function where $G_{n,p}$ is the subset of $M_{n,p}$ constituted by the matrices with rank $p$.

Can we differentiate the following function $f:x\rightarrow A(x)(A(x)^*A(x))^{-1} A(x)^*$ ?

So, we need to obtain (Jacobian matrix?), i.e., $$ \frac{\partial}{\partial x} \left\{ A(x) \left(A(x)^*A(x)\right)^{-1} A(x)^* \right\}$$ in terms of $ \frac{\partial}{\partial x} A(x)$.

Thank you very much in advance.

$\endgroup$
  • $\begingroup$ You could make the effort to say what $A(x)$ is. $\endgroup$ – loup blanc May 12 '18 at 21:02
  • $\begingroup$ @loupblanc it is not given $\endgroup$ – user550103 May 13 '18 at 5:57
1
$\begingroup$

Let $M$ denote the Moore-Penrose inverse of $A$, and for ease of typing let $$dA = \frac{dA}{dx}$$ The function you are interested in can be written $$f = AM$$ The derivative of $M$ (in a direction of constant rank) is a well known result $$dM = -M(dA)M + MM^*(dA^*)(I-AM) + (I-MA)(dA^*)M^*M$$ Therefore the derivative of your function is $$\eqalign{ df &= d(AM) \cr &= (dA)M + A(dM) \cr &= (dA)M -AM(dA)M + M^*(dA^*)(I-AM) + 0 \cr &= (I-AM)(dA)M + \Big((I-AM)(dA)M\Big)^* \cr }$$

$\endgroup$
  • $\begingroup$ Thank you very much. When you say "well known" result, then can you give me one reference please? $\endgroup$ – user550103 May 13 '18 at 16:52

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.