EFS in 'save-selected' file not correct.
This may or may not be a problem, but I have noticed in my LAMMPS runs that the energies, forces and stresses in the config files generated during AL (to the file out/preselected.cfg in my case) are not correct. I checked this a number of ways, but the easiest was to include 'write-cfgs all_configs.cfg' in the .ini, and comparing the preselected.cfg and all_configs.cfg files for the configurations in the former. I wanted to check that this is expected behaviour for MLIP-2. It occurred to me that the EFS may be 'wrong' in the preselected file because they are not actually used (after all, we just need to calculate the grade to determine these configs, and that only requires the configuration and potential parameters; not the EFS). In that case the supplied EFS in preselected.cfg would just be stubs. (In fact they actually correspond to the previous step in the MD, which seems to support this hypothesis.)