2
$\begingroup$

I have two matrices, both are $n \times m$ where $m < n$. These two matrices are $A$ and $B$.

I also know the singular value decomposition of $AB^{\top} = U \Sigma V^{\top}$.

Is there anything I can say about the SVD of $A$ or $B$? (for example, can I say that the left singular vectors of $A$ are $U$?) Are there any conditions under which I could say something like that?

Thanks.

$\endgroup$
  • $\begingroup$ Nothing else special about either matrix? $\endgroup$ – J. M. is a poor mathematician Jul 23 '12 at 18:27
  • $\begingroup$ nothing really, other than perhaps full rank ($m$), if that can help... but I prefer to avoid making that assumption. $\endgroup$ – kloop Jul 23 '12 at 18:29
  • $\begingroup$ but I would like to hear if there are any assumptions under which some properties of the SVD of $A$ could be identified. $\endgroup$ – kloop Jul 23 '12 at 18:43
0
$\begingroup$

Let $V_A$ be the right singular vectors of $A$, and $U_B$ be the left singular vectors of $B^T$, we have $$AV_A=U_A\Sigma_A, \quad B^TV_B=U_B\Sigma_B$$

If $V_A= U_B$, then $$AB^TV_B=AU_B\Sigma_B=AV_A\Sigma_B=U_A\Sigma_A\Sigma_B$$

Also, because the the left singular vectors of $B^T$ are right singular vectors of $B$, we can see that, if the right singular vectors of $B$ are right singular vectors of $A$, then we have that the right singular vectors of $B^T$ is the right singular vectors of $AB^T$, and the left singular vectors of $A$ is the left singular vectors of $AB^T$.

$\endgroup$

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.