Loss of precision in Parrinello-Rahman pressure coupling at high pressure

It seems there is an issue with Parrinello-Rahman pressure coupling when coupling to high pressure. The box velocity may become so small that the update to the box is less than the floating point precision. This leads to a density that changes in discrete steps (see figure dens.xvg). In double precision the fluctuations are continuous and the box follows the same "trajectory".

In addition to this there maybe be an issue with the coupling in that it makes the box grow indefinitely (but this may be a force field issue as well).

The example is a CGenFF model for carbon dioxide.co2.zip