Let $A,C\in\mathbb{R}^{m\times n}$, $n\ge m$, $B\in\mathbb{R}^{n\times m}$, and $P$ be a real positive definite $m\times m$ matrix. Denote by $\mathcal{S}^n$ the space of $n\times n$ real symmetric matrices. Suppose that $AB$ is non-singular. Let us define

$$F := (AB)^{-1} CB (AB)^{-1}P(AB)^{-\top} + (AB)^{-1}P(AB)^{-\top}(CB)^\top (AB)^{-\top}.$$

Next, consider the scalar function $$ c(X)=\mathrm{tr}(B^{\top}XBF) $$ where $X\in\mathcal{S}^n$ is such that $B^{\top}XB\ne 0$ and $\mathrm{tr}(\cdot)$ denotes the trace operator.

Is it possible to find a suitable $X$ such that $c(X)$ turns into a quadratic expression, i.e., something of the form $$ c(X)=\mathrm{tr}(L^\top Y L), $$ where $L$ and $Y$ are functions of $A,B,C,P$ and $Y$ is positive definite?

Note. I made several attempts with different $X$'s, but without any success yet. Hence, I wonder whether there is a rationale to follow either to get a positive answer to my question or to prove that this is actually not doable.

  • $\begingroup$ What about $X=0,L=0,Y=I$? $\endgroup$ – Federico Poloni Dec 20 '17 at 19:39
  • $\begingroup$ @FedericoPoloni: Right. I should have added that $B^\top XB\ne 0$ to rule out trivial cases. Just edited now. $\endgroup$ – Ludwig Dec 20 '17 at 20:05

Nope. That would mean $c(X)$ is always nonnegative, but that can't be that way (unless it's always $0$) because it switches sign if you replace $C$ with $-C$.

  • $\begingroup$ Right. Lastly, I wonder whether we could argue that if $c(X)=0$ then $CB$ must have some special form (besides the trivial one $CB=0$). $\endgroup$ – Ludwig Dec 21 '17 at 6:30

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.