A restrained-parameter structure-factor least-squares refinement procedure for large asymmetric units

A rapidly converging method for refining approximate atomic models is presented. It combines the conditional structure-factor least-squares procedure described by Waser [Acta Cryst. (1963). 16, 1091-1094] with the conjugate gradient method for solving linear systems [Hestenes & Stiefel, J. Res. Natl. Bur. Stand. (1952). 49, 409-436]. The method allows simultaneous variation of all of the structural parameters, although less than 1% of the derivative matrix need be calculated and stored for large systems and less than th of the diffraction data accessible with Cu radiation need be used. Applications involving a 240 atom mineral and an 812 atom protein are mentioned.