5
$\begingroup$

In On the variation of the spectrum of a normal matrix, Sun proves the following result (Corollary 1.2):

Let $A$ be an $n$-square normal matrix and $B$ an arbitrary $n$-square matrix. Then $$ \min_{\sigma \in S_n} \max_{1\le i\le n} |\lambda_i(A) - \lambda_{\sigma(i)}(B)| \le C(n) \|A-B\|,\quad C(n) = n, \label{1}\tag{$\star$}$$ where $\|\cdot\|$ is the spectral norm.

This result is a direct consequence of a result for the Frobenius norm (Theorem 1.1), which is shown to be optimal. However, the spectral norm result above is not shown to be optimal: the example provided only shows that the constant $C(n)$ in \eqref{1} must be at least $\sqrt{n}$.

Further work by Li and Sun provides additional hypotheses under which $C(n)$ can be taken to be smaller, but I have not seen any results which improve \eqref{1} under the sole hypothesis of normality of $A$.

I'm interested in the optimality of the constant $C(n)$ in \eqref{1}. Can the constant $C(n)$ in \eqref{1} be improved? Is it possible that $C(n) = o(n)$ works? Is there a lower bound better than $C(n) = \Omega(\sqrt{n})$ as shown by Sun?

$\endgroup$

1 Answer 1

-1
$\begingroup$

Theorem(Sun) Let $A$ be an $n × n$ normal matrix having eigenvalues $λ_1, \dots , λ_n$ and $B$ be any n×n matrix having eigenvalues $μ1, \dots , μn$. Then there is a permutation $p$ on ${1, 2, . . . , n}$ such that $$ (\sum_{i=1}^n |λi − μ_{p(i)}|^2)^{1/2} \leq \sqrt{n}∥A − B∥_F. $$

$\endgroup$
1
  • 2
    $\begingroup$ This response does not answer my question; you’ve just quoted theorem Theorem 1.1 in the linked paper which bounds the optimal matching distance in the 2-norm in terms of the Frobenius norm. I’m interested in bounding the optimal matching distance in the $\infty$-norm in terms of the spectral norm. $\endgroup$
    – eepperly16
    Commented Aug 1, 2021 at 17:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .