I came across the following equation while reading a paper. Can anyone please explain me the how is the first step obtained ? Here, $v_{k+1}$ is the leading eigenvector computed for the matrix $M$ using the power iteration method. Any hints would be really helpful. Thanks.

enter image description here


Essentially you are asking about the derivative of $\frac{y}{\|y\|}$. By chain rule, $$ d\frac{y}{\|y\|}=\frac{dy}{\|y\|}-\frac{y(y^\top dy)}{\|y\|^3}=\frac{I-\frac{yy^T}{\|y\|^2}}{\|y\|}dy $$ Now set $y=Mv_k$ and use that $v_{k+1}=\frac{Mv_k}{\|Mv_k\|}$

  • $\begingroup$ Thanks for the answer. That makes it more clear. Can you please expand on how is d||y|| = (y^T dy) / ||y|| ? $\endgroup$ – user2125722 May 27 at 22:01
  • 1
    $\begingroup$ This of course only holds for the Euclidean norm, $\|y\|=\sqrt{y^⊤ y}$, where the outer derivative is $\frac1{2\sqrt{..}}$ and the inner derivative is $2y^⊤dy$ $\endgroup$ – Dr. Lutz Lehmann May 27 at 22:12

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.