Suppose we have two real, symmetric and positive semidefinite matrices $A$ and $B$, and we know that they approximate each other well in the sense that

$$\| A - B \|_2 \le \epsilon,$$

where $\epsilon$ is positive and $\|\cdot\|_2$ is the $\ell_2$ operator norm. Is it possible to bound the difference between their leading eigenvalues, $|\lambda_1(A) - \lambda_1(B)|$, in terms of $\epsilon$?

I know it's possible to bound the angle between the leading eigenvectors of $A$ and $B$, but I have no idea how to prove a similar bound for eigenvalues.

  • 1
    $\begingroup$ The difference is bound by $\epsilon$ as long as both matrices are normal, see the Bauer-Fike theorem. $\endgroup$ – Conifold Jan 2 at 4:00

We may assume that $\lambda_\max(A)\ge\lambda_\max(B)$. Then \begin{aligned} |\lambda_\max(A)-\lambda_\max(B)| &=\lambda_\max(A)-\lambda_\max(B)\\ &=\|A\|_2-\|B\|_2\\ &\le\|A-B\|_2 \le\epsilon. \end{aligned}


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.