Figure 5.

The EM algorithm results are consistent. (a) The program was run ten times with randomly generated starting models, and the final output models have nearly the same correlations compared with the theoretical models. (b) Pairwise comparison of the models from ten runs, the results from the program ambigator in CrystFEL and the twinned data. The lower triangular matrix shows the correlations when compared in CIMs and the upper triangular matrix shows the correlations in AIMs. Note that the correlations are very similar for twinned data in these two modes (dark-grey colour, last row and last column) while the other models are very consistent, giving correlations larger than 0.90 when indexed in the same modes.