Suppose we know the SVD of matrix $\bf A$, and $\bf B$ is a slight perturbation of $A$ (e.g. $\|{\bf B}-{\bf A}\|_{\text F}$ is relatively small), then is there any method that can efficiently compute the SVD of $\bf B$? That is, can the knowledge of SVD of $\bf A$ be helpful for SVD of $\bf B$?

I searched a little bit and found there are some papers on the bound of perturbation, e.g. Perturbation Theory for the Singular Value Decomposition, but I currently have no luck in finding a method to compute SVD of perturbation taking advantage of the SVD of the original matrix.

Any help or reference will be very much appreciated!



Browse other questions tagged or ask your own question.