Pressure for atom_style ellipsoid (probably a bug)

I have been working with ellipsoids for a while, and one issue still bugs me. The asphere variant of time integrators deals with finite-size particles, each with 3 translational and 3 rotational degrees of freedom. The manual clearly explains how these fixes (npt/asphere, nph/asphere, and nvt/asphere) work: they create their own computes, as if these commands had been issued.

compute fix-ID_temp all temp/asphere
compute fix-ID_press all pressure fix-ID_temp

The temperature accounts for the translational and rotational degrees of freedom, and the pressure uses the same temperature. The problem is that the rotational component of the kinetic energy should not contribute to the pressure tensor, which depends only on the x,y,z components of the kinetic energy tensor.

In a discussion in 2013, @sjplimp affirmed that npt/asphere computes the temperature and pressure correctly, but that the compute temperature needed to be modified for uniaxial ellipsoids. The source of confusion there was that the thermo output shows a temperature and pressure computed for point particles, but that’s now well understood.

My point is that the pressure should only include the translational components of the kinetic energy. This issue has already been discussed here. In this other thread, @Syd_Lin pointed out that the pressure in compute_pressure.cpp is computed correctly when the translational temperature is used, as it comes with the correct dof per particle (3).

To settle the question, I have performed MD simulations on coarse-grained models of water, which are (almost) identical to the corresponding atomistic models (reference), thus providing a way to test the effect of barostating. The results clearly show that the properties of the CG model closely match those of the reference atomistic calculations, except for the larger inertia tensor of CG water. Unfortunately, in the paper, we do not discuss how the pressure was computed, but we provided the input files to reproduce the calculations.

Since the CG models of water are based on a forked version of LAMMPS, I will discuss this example in the relevant thread in the #talk forum.