Handling of NPT trajectories

[gonzalezma]

I have recently learnt (see e.g. this paper https://doi.org/10.1021/acs.jctc.3c00308 and references therein), that the handling of PBC in NPT simulations is far from trivial. Furthermore, different codes will use different wrapping/unwrapping schemes. While the effect should be null or negligible when simulating a solid or performing short simulations, this is no longer true for liquids simulated over very long time scales. The paper presents the effect on the m.s.d. and the self-diffusion coefficient, but any dynamical property would be affected.

I do not think that there is any simple solution, so for the moment I can only think of 3 possible actions:

1. Do nothing and let the user with the full responsibility of how he/she uses MDANSE.
2. Check (when converting or loading a trajectory) if we have an NPT trajectory (i.e. cell dimensions change) and if so, show a warning message preventing of the possible problems and citing this reference, so it is up to the user to decide what to do.
3. Be strict and allow MDANSE to work only with constant-cell trajectories.

A 4th solution could be to agree that the TOR algorithm in the paper is the right way to do the wrapping and implement it, but I am afraid there are at least two difficulties:

1. We need to handle many different types of trajectories (see discussion in the paper about Gromacs, NAMD and Lammps)
2. The algorithm cannot be applied on individual atoms, but on centers of masses, and then the molecules reconstructed with the right dimensions.

I don't think that this will be an easy job, so my preference would go for the 2nd one. But this is open for discussion.