The matrix rotation technique is applied to three-dimensional (3-D) multile
vel fast multipole algorithm (MLFMA) at very low frequencies (LF) to save t
he storage without increasing the order of the floating-point operations. S
ome symmetrical properties of the translation matrices along the z-directio
n for the 3-D LF-MLFMA and the static MLFMA are also derived to further red
uce the memory requirement Numerical results shown that the order of errors
introduced by the matrix rotation is much smaller than the order of errors
from the plain LF-MLFMA or plain static MLFMA. The implementation of the m
atrix technique does not change the precision order of the LF-MLFMA and the
static MLFMA for fixed truncation of multipole expansions. (C) 2000 John W
iley & Sons, Inc.