Weighted matrix means and symmetrization procedures

TL;DR

This paper proves the convergence of certain matrix mean procedures and introduces weighted matrix means, showing their properties and cubic convergence for all such means.

Contribution

It establishes convergence results for the Ando-Li-Mathias and Bini-Meini-Poloni procedures and defines weighted matrix means for symmetric matrix means.

Findings

Proves convergence of matrix mean procedures.

Defines weighted two-variable matrix means.

Shows cubic convergence of the Bini-Meini-Poloni procedure.

Abstract

Here we prove the convergence of the Ando-Li-Mathias and Bini-Meini-Poloni procedures for matrix means. Actually it is proved here that for a two-variable function which maps pairs of positive definite matrices to a positive definite matrix and is not greater than the square mean of two positive definite matrices, the Ando-Li-Mathias and Bini-Meini-Poloni procedure converges. In order to be able to set up the Bini-Meini-Poloni procedure, a weighted two-variable matrix mean is also needed. Therefore a definition of a two-variable weighted matrix mean corresponding to every symmetric matrix mean is also given. It is also shown here that most of the properties considered by Ando, Li and Mathias for the $n$-variable geometric mean hold for all of these $n$-variable maps that we obtain by this two limiting process for all two-variable matrix means. As a consequence it also follows that the Bini-Meini-Poloni procedure converges cubically for every matrix mean.