Quasi-Block-Cholesky Factorization With Dynamic Matrix Compression for Fast Integral-Equation Simulations of Large-Scale Human Body Models

Abstract

In this paper, a fast direct integral-equation method for simulating human models is presented. Based on the mixed symmetric and skew-symmetric pattern of the impedance matrix, a quasi-block-Cholesky (QBC) algorithm was proposed to reduce both the memory and central processing unit (CPU) time for matrix factorization by half. Dynamic matrix compression via single-level adaptive cross approximation (ACA) was further applied to reduce the computational costs. Validity of the QBC method is provided. Numerical examples further demonstrate the practicality of the proposed method.