GROMACS version: Any version
GROMACS modification: No
Hi everyone,
I am running into an error which I encounter only in NPT equilibration phase with both LINCS and frozen solute (a small molecule in a solvent box). If one of these is not present, everything proceeds smoothly (of course, if I get off LINCS, I put dt = 0.001 ps). Here’s the issue:
- Fatal error:
23 particles communicated to PME rank 9 are more than 2/3 times the cut-off out of the domain decomposition cell of their charge group in dimension y. This usually means that your system is not well equilibrated.
I tried to go round this by raising both tau-p and tau-t, also putting dt = 0.005 in NPT and precedent NVT and lengthening the NVT to 4 ns (instead of 2), but this just postpones the error, which in the ends happens in NPT at a certain step. The suggestion of using a single rank via -ntmpi 1 is actually unfeasible (I tell this since one could come up with that by looking at the reported error), I really need to run too many trajectories for such an option in a reasonable time. Even with 4 I have the issue, and I would like to have at least the freedom to go for 10-12.
Temporarily, I am just getting off the constraints and using a 1 fs timestep. Any clue? Do you consider it a mistake?
Seems to me its more of a clash between the triad I mentioned, as if a scaling of the coordinates from the NPT calculation clashes with those two when both are adopted. Moreover, should the NPT + frozen solute features be avoided together, as the scaling of the coordinates of the latter would not be carried out, whereas of all the rest of the environment would? If so, may this be an ideal to which one should stick to, but practically, if the box does not changes dimension that much, still the run would be ok? E.g. : from 6 x 6 x 6 A^3 to 5.92 x 5.92 x 5.92 A^3, which to me seems tiny.
Hope I stimulated some thoughts and at least your interest, any answer to these three different questions is way more than welcomed.
P.S. Given that I am writing about this frozen solute issue, would the ‘sd’ integrator make sense in that case? The tau-t - related to the friction coefficient value if adopting that integrator - on the one hand would have to deal with moving particles (the solvent), on the other with frozen ones (the solute), contemporarily.
Best Regards,
Jacopo