1
$\begingroup$

I have a matrix equation: $$M - EV - (EV)^T = 0$$ which i want to solve against $E$. Is this possible? How to do that?

Remarks: matrices are square, $E$, $M$ are symmetric, $V$ is invertibile.

Regards, Marek

$\endgroup$
  • 1
    $\begingroup$ think first as an equation on $Y = EV$ $\endgroup$ – Daniel Aug 13 at 15:07
2
$\begingroup$

We can write this equivalently as $$ EV + (EV)^T = M. $$ Because the (linear) operator $E \mapsto EV + (EV)^T$ is not invertible, this equation will have infinitely many solutions. In particular, we can always take $E = \frac 12 MV^{-1}$. Indeed, plugging this $E$ in yields $$ EV + (EV)^T = \frac 12 MV^{-1}V + (\frac 12 MV^{-1}V)^T = \frac 12 M + \frac 12 M^T = M $$ As it turns out, every solution to this equation can be written in the form $$ E = \frac 12 (M + S)V^{-1} $$ where $S$ is skew symmetric, which is to say that $S$ satisfies $S^T = -S$.


Now, requiring that $E$ is symmetric amounts to requiring that $$ [(M + S)V^{-1}]^T = (M + S)V^{-1} \implies\\ V^{-T}(M - S) = (M + S)V^{-1} \implies\\ V^{-T}S + SV^{-1} = V^{-T}M - MV^{-1}. $$ In other words: there exists a symmetric solution $E$ if and only if the Sylvester equation (more specifically Lyapunov equation) $V^{-T}S + SV^{-1} = V^{-T}M - MV^{-1}$ has a solution $S$ in which $S$ is also skew-symmetric. I don't see a way of simplifying this condition.

It is notable that if the eigenvalues of $V$ have all positive real part or negative real part, then there exists a unique solution $S$. If the eigenvalues of $V$ have only negative real part, then this unique $S$ can be written explicitly as the integral $$ S = \int_0^\infty e^{V^{-T}\tau}[V^{-T}M - MV^{-1}]e^{V^{-1}\tau} d\tau. $$ This $S$ is necessarily skew-symmetric, which is to say that our original equation has a (unique) symmetric solution $E$ in this case.


Another approach: if we're looking specifically for a symmetric solution $E$, then we can rewrite the original equation as $$ EV + V^TE^T = M \implies V^TE + EV = M. $$ Now, if the eigenvalues of $V$ all have negative real parts, then this equation has a unique solution that can be written as $$ E = \int_0^\infty e^{V^{T}\tau}\,M\,e^{V\tau} d\tau, $$ which is necessarily a symmetric matrix.

$\endgroup$
  • $\begingroup$ Ok, but it should exist only one $E$ which is symmetric, isn't it? It seems the question transforms to: how to find $S$ for which $E$ is symmetric? $\endgroup$ – mrkwjc Aug 13 at 16:05
  • $\begingroup$ I missed that $E$ is supposed to be symmetric. I don't think that we can necessarily guarantee that a symmetric solution $E$ exists, but I'll think about it and add something. $\endgroup$ – Omnomnomnom Aug 13 at 16:08
  • $\begingroup$ See my latest edit. $\endgroup$ – Omnomnomnom Aug 13 at 16:32
  • $\begingroup$ Thanks man, I'm enlightened now :) But... i think that the original equation is also a Lyapunov one and maybe we can write such integral directly for E? $\endgroup$ – mrkwjc Aug 13 at 16:48
  • $\begingroup$ Technically it isn't because the second $E$ is transposed. However, since we're looking for a symmetric solution we can make it one. I'll add something. $\endgroup$ – Omnomnomnom Aug 13 at 17:16
2
$\begingroup$

In one dimensional case $$ M - EV - (EV)^T = M - 2EV = 0 \implies E = \frac{M}{2V}. $$

So, maybe you would like to check if $$ E = \frac{1}{2} MV^{-1} $$ in general case?

$\endgroup$

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.