Let $W$ be $n$ by $n$ real symmetric positive definite matrix, let $V$ be an invertible $n$ by $n$ square matrix. Then the matrix $V^TWV$ is also symetric and positive definite. Now let $\|V\|$ denote the largest singular value of $V$. That is, the square root of the largest eigenvalue of $V^TV$.

Then, show that: \begin{equation} \min_{\|x\| = 1} x^TVWV^Tx \leq \|V\|^2 \min_{\|y\| = 1} y^TWy \end{equation}


Fix $x$ such that $\|x\| = 1$.

Let $y = V^\top x / \|V^\top x\|$. Note that $\|y\| = 1$ and that $\|V^\top x\| \le \|V^\top\| = \|V\|$. Then $$x^\top V W V^\top x = \|V^\top x\|^2 \cdot y^\top W y \le \|V\|^2 \cdot y^\top W y.$$

  • $\begingroup$ Why does $\|V^{T}x\| \leq \|V^T\|$ Hold? The second is the largest e value of $V^TV$ not of $V^T$ $\endgroup$ – jmsac Nov 23 '18 at 3:06
  • $\begingroup$ Sorry the second quantity is the square root of the largest e-value of $V^TV$ $\endgroup$ – jmsac Nov 23 '18 at 3:20
  • $\begingroup$ @jmsac $\|V^\top x\| = \sqrt{x^\top VV^\top x} \le \lambda_{\max}(VV^\top) = \lambda_{\max}(V^\top V)$ $\endgroup$ – angryavian Nov 23 '18 at 3:21
  • $\begingroup$ Ahh ok thank you! $\endgroup$ – jmsac Nov 23 '18 at 3:23

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.