A memory-efficient fast Fourier transformation algorithm for crystallographic refinement on supercomputers

Abstract

A vectorizable algorithm for fast computation of structure factors and derivatives during refinement of macromolecular structures is presented. It is based on fast Fourier transformations on subgrids that cover the unit cell of the crystal. The use of subgrids allows reduction of the total memory requirements for the computations without producing large overheads. The algorithm is applicable to all space groups. The performance of the algorithm on a conventional scalar computer as well as on supercomputers is discussed.