ANALYSIS OF DIFFUSE-SCATTERING FROM SINGLE-CRYSTALS VIA THE REVERSE MONTE-CARLO TECHNIQUE - I - COMPARISON WITH DIRECT MONTE-CARLO

The use of the reverse Monte Carlo (RMC) technique for analysing diffu se scattering data from single crystals is compared with the well esta blished direct Monte Carlo (MC) method. Whereas in the MC method a mod el involving only a few interatomic interaction parameters is used, fo r RMC the atom coordinates themselves are the variables and problems r elated to underdeterminacy can arise. Attempts to use the RMC techniqu e to obtain short-range correlation information for a relatively simpl e real physical system, the Tl cation distribution in TlSbOGeO4, are d escribed. It is found that the RMC method has two conflicting requirem ents. If the size of the model system is sufficiently large to give a workably smooth calculated diffraction pattern, then the number of var iables inherent in the structure is so large that it far exceeds the n umber of observed data, and the fit obtained is completely spurious. O n the other hand, if the model system is kept sufficiently small so th at the number of observations greatly exceeds the number of variables, then the calculated diffraction pattern is so noisy that meaningful s hort-range correlation information is difficult to discern. Even for s mall systems, it appears that RMC refinement using the goodness-of-fit parameter, chi(2), results in adjustment of the many longer-range cor relations to obtain the fit rather than the relatively few short-range correlations. Despite the poor performance of the currently implement ed RMC algorithm for extracting short-range correlation information, t here are some grounds for optimism that the method can provide useful information. Although the derived short-range correlation values prese nt in the final refined coordinates were barely significantly differen t from zero, it was nevertheless possible to discern consistent trends as the simulations progressed that could provide useful guidance in e stablishing a better MC model. Ways in which the RMC methodology might be improved have been suggested by the study, although these would re quire even greater computational resources.