I am currently facing a Sylvester equation

$AX+XB = C$

where $A$, $B$, $C$ are all symmetric and a special constraint here is that $X$ should be orthogonal. The Sylvester equation itself may not hold perfectly and hence the least-square solution is also ok.

  • 2
    $\begingroup$ Try a numerical solution first, e.g., $\min_{X^TX=I} \|AX+XB-C\|_F$ using manopt.org $\endgroup$
    – Suvrit
    Oct 24 '19 at 15:19
  • $\begingroup$ If you relax $X^T X = I$ into $X^T X \preceq I$, you have a semidefinite program. $\endgroup$ Oct 24 '19 at 18:48
  • $\begingroup$ Thank you for the information. I should have made my expression more clear. That is, I prefer an analytical solution rather than a numerical one. And regarding to relax it into a semidefinite program question, may I say that the solution is still not analytical? $\endgroup$
    – lisi
    Oct 25 '19 at 3:16
  • 1
    $\begingroup$ @lisi Analytical solutions are a luxury and a rarity. $\endgroup$ Oct 25 '19 at 6:10

Let me assume that $A$ and $-B$ have no common eigenvalues. Without the orthogonality constraint there is then a unique solution $Y$ of the Sylvester equation $AY+YB=C$, which you can find using known methods. Let $Y$ have the singular value decomposition $Y=U\Sigma V^T$. Then $X=UV^T$ is the orthogonal matrix that minimizes $\sum_{ij}(X_{ij}-Y_{ij})^2$. (For a proof, see here.)

  • $\begingroup$ I got it! Thank you! $\endgroup$
    – lisi
    Oct 24 '19 at 8:30

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.