Efficient MPI parallel interpolative decomposition

Abstract

The interpolative decomposition (ID) has been employed to reduce the computational cost for low-rank approximation of matrices, such as the method of moments (MoM) system with many right-hand-sides. However, the memory requirement may be the bottleneck of the ID when the scale of the problem becomes very large. A parallel scheme of the ID is proposed to alleviate the difficulty. Numerical experiments have been conducted to validate efficiency of the proposed parallel scheme.