A variable relaxation parameter for the parallel one-sided JRS SVD algorithm

The computation of the singular value decomposition (SVD) of an m×n matrix A is important in many fields. Many sequential and parallel algorithms such as Jacobi, QR based methods have been proposed. In this paper, we study the one-sided JRS algorithm which is based on traditional cyclic one-sided Jacobi algorithm by using the relaxation technique and give a new method in which the relaxation parameter λ is variable in contrast to the original JRS algorithm. The experiments show that when the size of A and the number of processors used in the cluster are different, the variable λ can decrease the sweeps and accelerate the whole SVD process.