Optimizing the input parameters for powder charge flipping

Over the past few years, the powder charge-flipping algorithm has proved to be a useful one for structure solution from powder diffraction data, so a semi-systematic study of the effect of the different input parameters on its success has been performed. Two data sets were studied in these tests: a zirconium phosphate framework material and D-ribose. The Superflip input parameters tested were the reflection overlap factor, the intensity repartitioning frequency, the isotropic displacement parameter, the threshold for charge flipping and the number of cycles/runs. By varying the values of these parameters within sensible ranges, an optimized set could be found for the zirconium phosphate case, but no combination of parameters allowed the D-ribose structure to be solved. Reasoning that starting with nonrandom phases might help, an approximate (but incorrect) structure was generated using the direct-space global-optimization method implemented in the program FOX. This structure was then used to calculate initial phase sets for Superflip by allowing the calculated phases to vary in a random fashion by a user-defined percentage. With such phases and reoptimized input parameters, some fully interpretable solutions with the correct symmetry could be produced, even with fairly low resolution data. Unfortunately, it was not possible to recognize these solutions using the Superflip R values, so other criteria were sought. Both cluster analyses and maximum entropy calculations of the solutions were performed, and the latter, in particular, look very promising. A set of guidelines derived from these two structures could be applied successfully to a further two inorganic and seven organic structures.