SVD row or column symmetric matrix

Abstract

A new architecture for row or column symmetric matrix called extended matrix is defined, and a precise correspondence of the singular values and singular vectors between the extended matrix and its original (namely, the mother matrix) is derived. As an illustration of potential, we show that, for a class of extended matrices, the singular value decomposition using the mother matrix rather than the extended matrixper se can save the CPU time and memory without loss of numerical precision.