Abstract
This study focuses on the convergence of the iterative methods for solving the complex symmetric systems arising from the electromagnetic field analysis. The COCG method has been widely used to solve such systems, however, it suffers from slow convergence rate. To improve convergence rate, multiple-precision arithmetic techniques are adopted. Especially, the double-double precision operation is investigated in complex number. The double-double precision complex number and its arithmetic operations are implemented using error-free transformation based on Knuth’s and Dekker’s algorithm. Moreover, a mixed-precision COCG method is proposed to save computational cost. Using a developed system and the QD library of commonly used software, some numerical experiments are demonstrated and its performance is evaluated.
