The Influence of Initial Vector Selection on Tridiagonal Matrix Enhanced Multivariance Products Representation

Abstract

Enhanced Multivariance Products Representation (EMPR) is a function decomposition method formed by generalization of High Dimensional Model Representation (HDMR). EMPR may be utilized as a matrix decomposer also. The method here builds upon recursive EMPR and it decomposes a matrix into a product of three matrices: an orthonormal matrix, a rectangular tridiagonal matrix and another orthonormal matrix. The initial vectors of the recursion of the formulation are two normalized support vectors. This work focuses on implementation of the method and the choice of these support vectors.