Insights into the determination of molecular structure from diffraction data using a Bayesian algorithm

The determination of the molecular ordering in a liquid is still a controversial subject. There is no general consensus either on the methods to obtain reliable liquid structures or on the way to analyze them. Regardless of the method, it is very important to have a realistic molecular structure available that allows simulations to faithfully reproduce the sample features, and that minimizes the computing time in structure refinements. However, attention is not always paid to this point and molecular models coming from general force-fields are frequently used to undertake many of the analyses. We propose in this work to use a Bayesian scheme to fit the experimental data and produce reliable molecular models that can be used as the starting point of any simulation or refinement. The algorithm behind the proposed method is based on a Markov chain Monte Carlo procedure, as many other refinement programs such as reverse Monte Carlo or empirical potential structure refinement. (paper)