GROMACS version: 2021.3
GROMACS modification: Yes
(version modified, but same question applies to 2021.3 without modification. Our modification should not alter the pressure calculation.)
I have a hopefully quick question about interpretation of the pressure tensor as printed on the .log file output by mdrun. Our particular system is a liquid crystal with a custom force field --our gromacs is modified to accept this alteration with significant direction from Hess (thank you!). The molecules of the LC are oriented along the z-axis of the simulation box and the simulation is run as pcoupltype = semiisotropic. Most of the simulation has been with the z-axis compressibility set to zero, but I’ve lifted this and run continuations where pcoupltype is set to isotropic and where it’s still semiisotropic but with the compressibility for z shifted from zero to 100-fold and 1000-fold smaller than x/y.
The oddity that I’m trying to understand is that when the pcoupltype is set to semiisotropic, the xx and yy elements of the pressure tensor in the log appear to have enforced opposite signs… after much equilibration, to where the volume of the simulation box is nearly unchanging, the magnitudes are close, but the signs are always opposite. And, I’ve seen this running a semiisotropic simulation of the same general sort in an unaltered version of gromacs as well. With the current simulation if I lift the forced semiisotropic type and make it fully isotropic, x and y become able to have the same sign. I thought that this might be a side effect of zeroing the compressibility along z, so I set z to be merely a small non-zero value and I still see the same effect.
This is troubling to me because the referencing value has no hope of being achieved in the x or y directions simultaneously and not really ever in z because the lack of compressibility there doesn’t allow it to change since x and y are canceling each other out. This is on the assumption that the system is trying to come to Pref supposing that P = (1/3)Trace(P-tensor).
Is there something here that I’m missing?
Quick Addendum: I’ve tried this with Berendsen and P-R barostats and seen the same in both cases. I am trying an experiment with anisotropic pcoupltype for completeness.