Volume 8, pp. 21-25, 1999.

An optimum iteration for the matrix polar decomposition

A. A. Dubrulle

Abstract

It is shown that an acceleration parameter derived from the Frobenius norm makes Newton's iteration for the computation of the polar decomposition optimal and monotonic in norm. A simple machine-independent stopping criterion ensues. These features are extended to Gander's formulas for full-rank rectangular matrices.

Full Text (PDF) [64 KB]

Key words

Matrix polar decomposition, Newton iteration.

AMS subject classifications

65F30, 65F35.

< Back