Improving spectral-variation bounds with Chebyshev polynomials (Q919436)

The main result is a positive answer to a conjecture of \textit{S. Friedland} [Linear Multilinear Algebra 12, 81-98 (1982; Zbl 0491.15002)]. Let A,B be complex n-by-n matrices with eigenvalues \(\lambda_ i,\mu_ i\) \((i=1,...,n)\) and \(\| \|\) some operator norm and \(\nu (A,B)=\min \max_{i} | \lambda_ i-\mu_{\pi (i)}|\) where min runs over all permutations \(\pi\) of \(\{\) 1,...,n\(\}\), then \(\nu (A,B)\leq c_ n\| A-B\|^{1/n}(\| A\| +\| B\|)^{1-1/n}.\) Friedland conjectured that there is a global bound for \(c_ n\). Here it is shown that \(c_ n\leq 8\) for all n. It should be added that in the meantime Bhatia et al and Krause have shown by using similar tools that \(c_ n\leq 3.08\).