Inconsistent potential energy values computed from Gromacs and openmm

[CongWang98]

I'm planning to simulate some proteins using the a99sb-disp ff. To set up my system, I used Gromacs with the Gromacs ff provided in this repo. I obtained .top files that are compatible with both Gromacs and openmm. However, for some reasons, I need to use openmm. To ensure that openmm runs correctly, I calculated the potential of my initial structure using both openmm and Gromacs. Surprisingly, I discovered that these two produced different potentials. The difference came from non-bonded potential. After delving into the Gromacs ff files and conducting experiments with toy systems, I identified two factors contributing to this inconsistency.

1. In the file Gromacs_FFs/a99SBdisp.ff/a99SBdisp_water.itp, the [bond] and [angle] sections were missing. The absence of the [bond] caused openmm to produce incorrect potential for rigid water. When I compared the Gromacs and openmm potential energies for a pure water system, I observed a difference of 200 kJ/mol (approximately 0.1%). This discrepancy was about ten times larger than what I typically observed with other ff. I added [bond] and [angle] sections based on those found in the tip4p water model. After this adjustment, the difference for the pure water system decreased to about 0.01%, which is more reasonable.

2. In the file Gromacs_FFs/a99SBdisp.ff/ffnonbonded.itp, there is a [nonbond_params] section after the [atomtype] section. This section was overriding the non-bonded parameters for HB and OB pairs that should be calculated based on combination rules. The presence of this section caused openmm to produce potential that were 100% higher than expected for a toy protein system. Removing this section resulted in a perfect potential matching. However, I believe we should keep it. It may be a bug with openmm. But I think it is still necessary to mention it here in case someone has the similar issue .

(BTW, I also compared openmm potential and amber potential using an amber format a99sb-disp ff. I also obeserved different potential. This difference is from dihedrals. I am not familiar with amber so I didn't spend time to fix it)