Let $A \in \mathbb R^{m\times n}$ and its singular values be denoted by $\sigma_1 \geq \sigma_2 \geq \ldots \sigma_n \geq 0$. Then $$ \sigma_i(A) = \min\{\|B \|_2\colon B \in \mathbb R^{m \times n}, \mathrm{rank}(A-B) \leq i-1 \}. $$ How is this characterization of singular values obtained, is there a reference about it?

  • 1
    $\begingroup$ This is a consequence of the Eckart-Young theorem. I'm not sure where to find a better reference for this fact, but I think "Topics in Matrix Analysis" by Horn and Johnson would be a good bet. $\endgroup$ – Omnomnomnom Feb 6 at 14:10
  • 1
    $\begingroup$ I find the second proof given here particularly nice $\endgroup$ – Omnomnomnom Feb 6 at 14:19
  • 1
    $\begingroup$ Does that answer your question sufficiently, or did you want a more thorough explanation? $\endgroup$ – Omnomnomnom Feb 6 at 19:00
  • $\begingroup$ Clear, thanks for the reference! I got it through the first proof (optimality of the best rank-$i$ approximation by truncation of the SVD). $\endgroup$ – G. Gare Feb 7 at 12:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.