Random matrix-improved estimation of covariance matrix distances

Given two sets and (or ) of random vectors with zero mean and positive definite covariance matrices and (or ), respectively, this article provides novel estimators for a wide range of distances between and (along with divergences between some zero mean and covariance or probability measures) of the form (with the eigenvalues of matrix ). These estimators are derived using recent advances in the field of random matrix theory and are asymptotically consistent as with non trivial ratios and (the case is also discussed). A first "generic" estimator, valid for a large set of functions, is provided under the form of a complex integral. Then, for a selected set of 's of practical interest (namely, , , and ), a closed-form expression is provided. Beside theoretical findings, simulation results suggest an outstanding performance advantage for the proposed estimators when compared to the classical "plug-in" estimator (with ), and this even for very small values of .
View on arXiv