If you have two matrices A and B such as through singular value decomposition:


And any invertible matrix T such as

AB = $PS^{1/2}TT^{-1}S^{1/2}V$

Could you prove that B = $T^{-1}S^{1/2}V$

This question might be too out of context, so to situate it here is the page where I encontered the problem, it is from the book Biological Learning and Control by Sandro Mussa Ivaldi:

page 264

  • $\begingroup$ You can't "prove" it- that is just one possible value of the many different A and B whose product will give that result. $\endgroup$ – user247327 Sep 10 '16 at 14:22
  • $\begingroup$ I am not sure to get your meaning. If A and (PS^1/2) are invertible... you could prove it by taking T to be T = (P * S^1/2)^-1 * A.. but then you would need to prove that T is also invertible.. no ? $\endgroup$ – Albert James Teddy Sep 10 '16 at 16:36
  • $\begingroup$ so you want to prove: there exists an invertible matrix $T$ such that $B=T^{-1}S^{1/2}V$? $\endgroup$ – Christiaan Hattingh Sep 10 '16 at 17:50
  • $\begingroup$ What i ultimately wanted to understand is Pr. Ivaldi reasoning in his book (see pic) ... but hmm I am probably missing something $\endgroup$ – Albert James Teddy Sep 10 '16 at 18:27

The rank of $AB$ is less than or equal to the minimum of the rank of $A$ and $B$ (this is known as Sylvester's theorem I think). And the rank of $S^{1/2}$ is the same as the rank of $AB$ (SVD theory: rank of $AB$ equals the number of nonzero singular values). The equation $B=T^{-1}S^{1/2}V$ basically states that $B$ and $S^{1/2}$ are equivalent (matrix equivalence) and hence must have the same rank. But since the rank of $AB$ may very well be less than the rank of $B$ we see that this cannot be true in general. A trivial counter example: let $A=0$ and $B=I$.


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.