Document MonteCarloBarostat behavior with periodic ML potentials?

[jchodera]

While MonteCarloBarostat is a molecular-scaling barostat, my understanding is that findMolecules() will put every atom into a separate molecule if only a periodic ML potential is defined (such as the MACE-OFF23 as periodic potential), effectively performing as an atomic-scaling barostat. This would be the desired behavior, since waters may exchange protons and continuing to use the wrong molecule definitions would degrade barostat efficiency significantly.

I'm just wondering if we want to document this somewhere (here in the README? in the User Guide?) since this was non-trivial to figure out by looking through the implementations.

cc @saratk1

[peastman]

Note that this isn't specific to ML. For example, GFN-FF is a conventional force field that doesn't define explicit bonds, so it also scales each atom separately. Also, if you use createMixedSystem() then the force field may cause atoms to be treated as molecules even if the ML potential doesn't.

[jchodera]

Right. It would be useful to clarify this behavior in a note somewhere relevant so it isn't unexpected.