On 4/7/24 6:47 AM, Mario Bianchet via modeller_usage wrote: > The script below exceeds molpdf limits immediately when run this > automodel job on parallel, but run starts with a low molpdf and ends > well when commenting the line "a.use_parallel_job(j)" Any hint of how > to solve will be appreciate?
The model building procedure is the same for both serial and parallel jobs; however, parallel jobs run in a slightly different environment (a separate process, with the random number generator in a different state). So you should not expect to get exactly the same models.
There is always a possibility when building models that the randomized starting coordinates are such that the optimizer cannot recover (e.g. a knot in the backbone, or atoms placed on top of each other) and in some cases this will trigger the max_molpdf cutoff. The solution is the same though for both serial and parallel jobs - build multiple models and discard the bad ones.
Ben Webb, Modeller Caretaker