Is the direction of inequality preserved when multiplying the linear matrix inequality (LMI) by a nonsingular square matrix left and right?

129 Views Asked by At

Let $Y$ be a square and nonsingular matrix (invertible).

Given $X \prec 0$, if we multiply $Y^*$ on the left and $Y$ on the right to the $X$ (assuming the dimension matches),

I have a claim that

$$ X \prec 0 \iff Y^*XY \prec 0 $$

Question: I can get an intuitive understanding with a scalar analogy but I am not sure how I can prove this.

Attempt:

I first tried transforming this into the quadratic form (making it scalar).

For instance, multiply them by the vectors $z^*$ and $z$, left and right, respectively:

$$ X \prec 0 \implies \forall z \neq 0, \quad z^*Xz < 0. $$ By letting $z = Yw$ for non zero $w$ ($z^{*} = w^{*}Y^{*}$), $$ \implies \forall w \neq 0, \quad w^*Y^*XYw < 0 \implies Y^*XY \prec 0.$$

For the opposite direction, we have:

$$ \forall w \neq 0, \quad w^*Y^*XYw < 0, $$

Let $z = Yw$ and since $Y^{-1}$ exsits by the assumption, we have $w = Y^{-1}z$.

By substituting $w$ into the above inequality:

$$ \forall z \neq 0, \quad z^*Xz < 0 \implies X \prec 0.$$

Is my proof correct?

Some side question I have:

Is the assumption of $Y$ being square and nonsingular essential? I obviously used it in my attempt above. If I don't have it, is it possible still to get the same result?

1

There are 1 best solutions below

0
On BEST ANSWER

Your proof is right. For the side question, you need $Y$ to be square and invertible to show that the relation is iff.

Example for $\implies$: let $X=-1$ and $Y=[1 \ 1]$, then you have $X<0$, but $$Y^\top X Y=\begin{bmatrix} -1 & -1 \\ -1 & -1\end{bmatrix}\leq 0$$

Example for $\Longleftarrow$: let $X=\begin{bmatrix} -1 & -1 \\ -1 & -1\end{bmatrix}$ and $Y=\begin{bmatrix} 1\\ 1 \end{bmatrix}$, then $Y^\top X Y=-4<0$, but $X\leq 0$.