3
$\begingroup$

Dr. Tao in his notes on eigenvalue inequalities uses Courant-Fischer min-max theorem to prove the eigenvalue stability inequality. Specifically, I am looking for proof of Eq. (13) where Dr. Tao states as an immediate result of Eq. (6) and (10). But the problem is that the min-max function is not convex. I have read Stewart & Sun's book on Matrix Perturbation Theory, but it seems that they have felt that it is obvious too.

Can someone provide more details on how to derive Eq. (13)?

$\endgroup$
4
$\begingroup$

It is a simple and repeated application of $\min$ and $\max$ operators.

$$v^*(A+B)v=v^*Av+v^*Bv\le v^*Av+\|B\|_{op},\,\forall v\in R^n\wedge |v|=1.$$ Given $V$ where $\dim(V)=i$, $$\min_{u\in V,|u|=1}u^*(A+B)u\le v^*(A+B)v\le v^*Av+\|B\|_{op},\,\forall v\in V\wedge |v|=1,$$ and $$\min_{u\in V,|u|=1}u^*(A+B)u\le \min_{v\in V,|v|=1}v^*Av+\|B\|_{op}.$$ Then $$\min_{u\in V,|u|=1}u^*(A+B)u\le \max_{\dim(V)=i}\min_{v\in V,|v|=1}v^*Av+\|B\|_{op},$$ and $$\max_{\dim(U)=i}\min_{u\in V,|u|=1}u^*(A+B)u\le \max_{\dim(V)=i}\min_{v\in V,|v|=1}v^*Av+\|B\|_{op},$$ In other words $$\lambda_i(A+B)\le\lambda_i(A)+\|B\|_{op}.$$ Similarly, we can prove $\lambda_i(A)-\|B\|_{op}\le\lambda_i(A+B)$ and reach the desired inequality.

$\endgroup$

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.