Preprint on artifacts with neighbor lists in GROMACS

I noticed the preprint “Neighbor List Artifacts in Molecular Dynamics Simulations”(Neighbor List Artifacts in Molecular Dynamics Simulations | Theoretical and Computational Chemistry | ChemRxiv | Cambridge Open Engage) from the Hummer group discussing how there are issues in GROMACS defaults resulting in short outer cutoffs and infrequent neighbor list updates that lead to missing long-range attractive Lennard-Jones interactions, as well as small but systematic imbalances in the apparent pressure tensor then induce unphysical asymmetric box deformations that crumple the membrane. Have people seen this yet? I didn’t see any other post in the developer forum. Does it appear to indeed be an issue, or did they get something wrong?

1 Like

This is an actual issue. But, as far as I have seen, it is only leads to measurable effects in systems without electrostatics. Thus in practice this is only an issue for coarse-grained systems and very hydrophobic polymers. When there are some electrostatic interactions present, these will lead to a sufficient Verlet buffer and/or sufficiently short pair list lifetime such that the missing LJ interactions are negligible. So nearly all atomistic simulations are unaffected by this.

I have a change ready that prints the average pressure error to the log file. We should likely add a warning when using semi-isotropic or anisotropic pressure coupling. For the 2024 release we need to find the best way to always set a sufficient buffer.