Validate global NHC thermostat, make it default for classical MD?

[danielhollas]

Currently, by default we use massive Nosé-Hoover thermostat, where each atom is thermostatted separately (in each cartesian dimension). This setup is crucial to make PIMD work, as it ensures fast thermalization of all degrees of freedom. But for classical MD, it might be suboptimal as it might be too aggressive and hinder sampling of slower motions of the system such as diffusion / barrier crossing etc. The global NHC is implemented as well, but it has been validated much less and I am not sure if many people actually used it for production simulations. We should validate that it works and perhaps make it a default for classical MD.

In particular, we should investigate how global NHC works with respect to sampling translations and rotations. I think it should conserve translational / rotational momentum, in which case if we remove them at the beginning, the system will not sample them. In that case we might need to change the variable f which determines the number of conserved quantities and influences the computation of kinetic temperature.

See also #129.