Pull_f and Pull_X graphs showing sharp peaks

User discussions
1

GROMACS version: Simulation using GROMACS 2022 w/ MPI and CUDA, all else 5.1.4
GROMACS modification: No
I have been running some pulling simulations, but keep getting a strange artifact. Every once in a while (a few times per ns) the force jumps as the distance spikes down, causing an artificial unfolding. From inspecting the trajectory using VMD, this causes large shifts and sudden, subsequent unfolding that isn’t anticipated with a build-up in force as might be expected. The pull_x file shows a sudden change in s2 (1 g 1 y), which while not a definitive marker of a change, is not a pbc shift, since with other pbc shifts changing by around the length of one axis of the box, while this change is only around 1 nm.

My hunch is that this is an issue where GROMACS is identifying a periodic image of a pull-group as the correct one and causing issues. If this is the case, how should I proceed? The easiest solution is likely to increase the dimensions of the box, but this could be rather costly in terms of computation time. Is there a way to get rid of these jumps and have proper group detection, without changing the simulation? Or is increasing the box size the only way?

Thanks for any assistance.

2

Although this sounds like a periodic image issue, I think this is unlikely as mdrun will stop with an error when the distance is larger than half the box size.

Which pull geometry are you using?

3

That’s what I was thinking too. I am using pull-geometry: distance. I think my issue was with pull-group definition, as I accidentally included a distant atom to one of the pull group and haven’t seen the same problem since.