We consider two existing FFT-based fast-convolution iterativesolution techniques for the scalar T-matrix multiple-scattering equation [1]. The use of the FFT operation requires ﬁeld values be expressed on a regular Cartesian grid and the two techniques differ in how to go about achieving this. The ﬁrst technique [6,7] uses the nondiagonal translation operator [1,9] of the spherical multipole ﬁeld, while the second method [11] uses the diagonal translation operator of Rokhlin [10]. Because of its use of the non-diagonal translator, the ﬁrst technique has been thought to require a greater number of spatial convolutions than the second technique. We establish that the ﬁrst method requires only half as many convolution operations as the second method for a comparable numerical accuracy and demonstrate, based on an actual CPU time comparison, thatit canthereforeperformiterations fasterthanthe secondmethod. We then consider the respective symmetry relations of the non-diagonal and diagonal translators and discuss a memory-reduction procedure for both FFT-based methods. In this procedure, we need to store only the minimum sets of near-ﬁeld and far-ﬁeld translation operators and generate missing elements on the ﬂy using the symmetry relations. We show that the relative cost of generating the missing elements becomes smaller as the number of scatterers increases.

}, issn = {1991-7120}, doi = {https://doi.org/}, url = {http://global-sci.org/intro/article_detail/cicp/7726.html} }