A four-mean theorem and its application to pseudospectra

TL;DR

This paper proves a new inequality relating means of positive numbers and applies it to matrices with similar pseudospectra, establishing a sharp bound on polynomial norms of such matrices.

Contribution

Introduces a four-mean theorem and applies it to derive a sharp bound on polynomial norms for matrices with super-identical pseudospectra.

Findings

Established a new inequality for positive numbers with equal means.

Showed that matrices with super-identical pseudospectra satisfy a specific polynomial norm inequality.

Proved the inequality is sharp for the case N=4.

Abstract

Let $N\ge 4$. We show that, if $x_1,\dots,x_N$ and $y_1,\dots,y_N$ are $N$-tuples of strictly positive numbers whose arithmetic, geometric and harmonic means agree, then \[ \max_j x_j <(N-2)\max_j y_j \quad\text{and}\quad \min_j x_j <(N-2)\min_j y_j. \] This is used to show that, if $N\ge4$ and $A,B$ are $N\times N$ matrices with super-identical pseudospectra, then, for every polynomial $p$, we have \[ \|p(A)\|< \sqrt{N-2}\|p(B)\|, \] unless $p(A)=p(B)=0$. This improves a previously known inequality to the point of being sharp, at least for $N=4$.