Add FlashMD

[johannes-spies]

Add a new fix that uses a model to predict future positions and momenta.

[Luthaf]

Suggested change

	if (igroup == atom->firstgroup) nlocal = atom->nfirst;
	if (igroup == atom->firstgroup) {
	nlocal = atom->nfirst;
	}

[frostedoyster]

The same line is used by all other LAMMPS fixes with the same formatting

[Luthaf]

what's the difference between rmass and mass?

[frostedoyster]

LAMMPS allows per-type masses as well as per-atom masses (to support e.g. different isotopes of the same element within the same simulation). rmass contains one mass per atom in the latter case, otherwise it's undefined and you have to use mass[type[i]]

if (atom->rmass)
  m = atom->rmass[i];       // per-atom mass array exists
else
  m = atom->mass[type[i]];  // fall back to per-type mass

[frostedoyster]

@copilot This code integrates a new machine-learning-driven integrator into LAMMPS, through an interface called metatomic. The prototype you see here works well.

I have three tasks for you:

• Cleaning up the code. This includes adding comments, including, but by no means limited to, explaining the role of the force copy (isolating the Langevin forces to be applied to FlashMD), isolating the cryptic unit conversions and adding comments about them, erroring out nicely when the units are not metal, and any improvement that DOES NOT MODIFY THE FUNCTIONALITY OF THE CODE
• Writing a kokkos version of this code. You should take inspiration from the relationship between pair_metatomic.cpp and pair_metatomic_kokkos.cpp and re-use code where it is possible
• Erroring out nicely in case of multiple processes (they are not supported yet)

[Copilot]

@frostedoyster I've opened a new pull request, #30, to work on those changes. Once the pull request is ready, I'll request review from you.