Matrix concentration for products (Q2098235)

This paper offers some new tools for studying random matrix products that arise from stochastic optimization algorithms and related problems. The authors develop a collection of nonasymptotic bounds for products of random matrices. These results hold under simple and easily verifiable conditions, and they give accurate predictions about the behaviour of some particular instances (e.g., products of independent identically distributed random perturbations of the identity). The proofs are based on foundational results about the geometry of the Schatten classes, and they can easily be adapted to treat variants of the problems under consideration. These results sharpen and generalize recent work of \textit{A. Henriksen} and \textit{R. Ward} [Linear Algebra Appl. 594, 81--94 (2020; Zbl 1436.65047)], and they are similar in spirit to the results of \textit{R. Ahlswede} and \textit{A. Winter} [IEEE Trans. Inf. Theory 48, No. 3, 569--579 (2002; Zbl 1071.94530)] and of \textit{J. A. Tropp} [Prog. Probab. 71, 173--202 (2016; Zbl 1382.60016)].