On the gap between deterministic and probabilistic joint spectral radii for discrete-time linear systems (Q2229449)

A discrete-time switched linear system \(\Sigma({\mathcal A})\), associated with a finite set \(\mathcal A\) of matrices, is considered. Its asymptotic behavior can be measured by the deterministic joint spectral radius \(\rho_ d({\mathcal A})\) and by its probabilistic joint spectral radius \(\rho_ p(\mu, {\mathcal A})\) with respect to some probability measure \(\mu\). In the general case we have that \(\rho_ d({\mathcal A}) \le \rho_ p(\mu, {\mathcal A})\). So, a natural problem is to find conditions on \(\mathcal A\) and \(\mu\) under which this inequality becomes an equality. The authors study the family of probability measures obtained from discrete-time shift-invariant Markov chains and obtain conditions for the sets \({\mathcal A}\) under which the deterministic joint spectral radius is equal to a probabilistic joint spectral radius (Theorem 3.1) or is equal to the supremum of the probabilistic joint spectral radii over the considered family of probability measures (Theorem 3.13).