On the joint spectral radius of nonnegative matrices (Q2080245)

The author studies the joint spectral radius of nonnegative matrices and gives an effective bound of the joint spectral radius of a finite set of nonnegative matrices. Given a finite set \(\Sigma\) of square matrices in \(\mathbb{C}^{d\times d}\), the joint spectral radius \(\rho(\Sigma)\) of \(\Sigma\) is defined to be the limit \[ \rho(\Sigma)=\lim_{n\to \infty} \sqrt[n]{\|\Sigma^n\|}, \] where \(\|\Sigma^n\|=\max_{A_1,\dots,A_n\in \Sigma} \|A_1\dots A_n\|\). The existence of the limit is shown in Proposition 1. Though \(\rho(\Sigma)\) is also defined for infinite bounded sets \(\Sigma\), only finite sets are considered in this paper. For a component \(C\) and every \(m\), denote \[ \|\Sigma^n\|_C=\max_{A_1,\dots,A_n\in \Sigma} \ \max_{i,j\in C} \ |(A_1\dots A_n)_{i,j}| \] and \[ P_m(\Sigma)=\max_{A_1,\dots,A_m\in \Sigma}\ \rho(A_1\dots A_m). \] Given a finite set \(\Sigma\) of nonnegative matrices and \(n\), the author gives the bound of \(\rho(\Sigma)\) in Theorem 2 as follows: \[ \sqrt[n]{\left(\frac{V}{UD}\right)^D\max_C\|\Sigma^n\|_C}\leq \rho(\Sigma)\leq \sqrt[n]{D\max_C\|\Sigma^n\|_C}, \] where \(D\times D\) is the dimension of the matrices, \(U\), \(V\) are respectively the largest entry and the smallest entry over all the positive entries of the matrices in \(\Sigma\), and \(C\) is taken over all components in the dependency graph (see Definition 1). The method for estimating the joint spectral radius in Theorem 2 is better than the one for the popular bound \( \sqrt[m]{P_m(\Sigma)}\leq \rho(\Sigma)\leq \sqrt[m]{D\|\Sigma^m\|}\) in [\textit{R. M. Jungers}, The joint spectral radius. Theory and applications. Berlin: Springer (2009; \url{doi:10.1007/978-3-540-95980-9})] by a root of a polynomial of degree \(r\). The author also gives the following bound on \(\|\Sigma^n\|\): Theorem. If \(\rho(\Sigma)>0\), then there exist a non-negative integer \(r\) and two positive numbers \(\alpha\), \(\beta\) so that for every \(n\) there holds \[ \alpha n^r \rho(\Sigma)^n \leq \|\Sigma\|^n \leq \beta n^r\rho(\Sigma)^n. \]