Frobenius Iteration for the Matrix Polar Decomposition
Augustin A. Dubrulle · 1994
Higham's iterative computation of the matrix polar decomposition uses an acceleration parameter derived from economical approximations of the £2 norm and nearly optimum for that norm. It is shown here that the iteration based on a parameter optimum for the Frobenius norm converges as fast as the £2 iteration and lends itself to more efficient implementation and easier iteration control. The description of a practical algorithm is included for illustration.