Fitting with Differential Evolution¶
A review of using the Differential Evolution algorithm to fit reflectivity data has been published in J ournal of Applied Crystallography [Bjorck11]. If you do not have access to the journal just send me an e-mail artur.glavic@psi.ch.
In summary, one can say that choosing a minimum population size of 30-50 and k_r = k_m = 0.5-0.9 provides a rather robust algorithm for reflectivity refinements. See the summary figure below for good parameter ranges. If you need more stability the population size can be increased and/or the k_r be lowered. For more details see the paper.
Good parameter ranges for the relevant DE methods and their influence on convergence (Figure 9 from [Bjorck11])¶
