engleski

On the Convergence of Block Kogbetliantz Methods for Calculating the SVD

We introduce the block Kogbetliantz method for computing the singular value decomposition of a matrix. The initial matrix is partitioned into blocks, and instead of annihilating a couple of elements by plane (2 by 2) rotations at each step, we consider annihilating a couple of blocks at mutually transposed positions by more general block rotations. The conditions that block rotations have to satisfy in order for a row-cyclic algorithm to converge are discussed. We also prove convergence of left and right singular vectors.

SVD; block methods; convergence

nije evidentirano