MODIFYING THE SINGULAR-VALUE DECOMPOSITION ON THE CONNECTION MACHINE

A fully parallel algorithm for updating and downdating the singular va lue decompositions (SVD's) of an m-by-n (m greater than or equal to n) matrix A is described. The algorithm uses similar chasing techniques for modifying the SVD's described in [3], but requires fewer plane rot ations, and can be implemented almost identically for both updating an d downdating. Both cyclic and consecutive storage schemes are consider ed in parallel implementation. We show that the latter scheme outperfo rms the former on a distributed memory MIMD multiprocessor. We present the experimental results on the 32-node Connection Machine (CM-5).