Matrix Poincaré inequalities and concentration (Q776819)

The authors show that if a given probability measure \(\mu\) on a Polish space satisfies a matrix Poincaré inequality with respect to a reversible Markov generator, then \(\mu\) also satisfies a concentration inequality, whose bound is expressed in terms of the matrix carré du champ operator associated with this Markov generator. The proof of this result requires the development of new matrix trace inequalities. To demonstrate this general approach for turning matrix Poincaré inequalities into concentration inequalities, the authors also prove matrix Poincaré inequalities (and the corresponding concentration bounds) for Gaussian measures, product measures, and measures \(\mu\) satisfying the stochastic covering property (SCP), a negative dependence assumption closely related to the strong Rayleigh property. Examples of measures satisfying the SCP include determinantal measures and those obtained by running exclusion dynamics from a deterministic state.