2
$\begingroup$

revisiting

singular values of "rotated" matrix

Maybe I've since forgotten - or maybe I never bothered to ask myself - but

Since $A^T$ has the same singular values as $A$ and $K$ is orthogonal, conclude that [ $A^\top K$ ] indeed has the same singular values.

Why does this hold?

Why can we say that multiplying by an orthogonal matrix retains the singular values?

$\endgroup$
  • 1
    $\begingroup$ The (nonzero) singular values of $A$ are related to the eigenvalues of $A^TA$ which are in turn related to the eigenvalues of $AA^T$. Now for $B=A^TK$ you have $BB^T=A^TA$. $\endgroup$ – A.Γ. Aug 21 '17 at 8:11
3
$\begingroup$

We calculate the singular values of $A$ by decomposing it into $A=USV^T$ with diagonal $S$ and orthogonal $U$ and $V$. If we have another orthogonal $K$, we have $AK= USV^TK = U S (V^TK)$ which is the SVD of $AK$.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ I see, and $V^\top K $ is orthogonal because $V^\top K (V^\top K)^\top = V^\top K K^\top V = I $. Thank you. $\endgroup$ – User1291 Aug 21 '17 at 8:19
2
$\begingroup$

Let $A = UDV^T$ be the singular value decomposition.

If you premultiply it with $K$, $KA=(KU)DV^T$

$KU$ is an orthogonal matrix as well. In fact $(KU)DV^T$ is a singular value decomposition for $KA$. The $D$ matrix is not changed.

Similar reasoning for post-multiplication.

| cite | improve this answer | |
$\endgroup$

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.