Large long-range coulombic energy using PPPM compared to PME on Gromacs

Dear Lammps users,

I am simulating coarse grained model of a polysaccharide network. I am using the MARTINI 2.2 force field, and using MARTINI polarizable water model. I did the bulk of my work on Gromacs, and shifted to Lammps to perform some mechanical simulations.

The system was very unstable on Lammps and while investigating the reason, I measured the long-range pairwise electrostatic energies between the 1-3 excluded dummy-charge particles on polarizable water (-0.46 and +0.46 electron charge units). Lammps calculated an energy of the order of -3e-01 kcal/mol. The large magnitude of the energy was surprising/

I simulated one single polarizable water particle in a ~160Angstrom box on Lammps and gromacs to compare the energies directly. This was done to overcome the issue of the gromacs PME solver being unsuited for decomposition of kspace energy into pairwise interactions. Lammps, again, calculated an energy of the order of -3e-01 kcal/mol, and, on the other hand, Gromacs calculates the energy to be of the order -2e-04 kcal/mol (converted to kcal/mol from gromacs energy units). The 1000x difference was concerning to me, and so I am writing to the mailing list.

I checked the Lammps simulations on version 30Jul16 (to compare with old simulations) as well as the latest build released on 7Aug19.

I am using PPPM kspace solver on Lammps with 7e-5 kcal/mol-Ang tolerance, 1 Angstrom grid size, and interpolation order of 4.

On Gromacs (version 4.5.4) I using the PME kspace solver with 1.2 Angstrom grid size, interpolation order of 4, which, according to gromacs documentation, equates to a tolerance of ~3e-4kJ/mol-m = 7e-5kcal/mol-Ang.

Short-range cutoff of 16Ang and dielectric constant of 2.5 is used for both simulations as per MARTINI guidelines. All other details are in the attached files.

Has anyone else noticed similar issues? I am attaching the lammps and gromacs input files. Any help is appreciated.

gromacs_files.tar.xz (48.6 KB)

water.test.nvt.log (43 KB) (1.51 KB) (469 Bytes)

it looks to me that you are comparing apples with oranges here.

from looking at the output of your gromacs run, it seems that this version of gromacs handles consideration of excluded pairwise interactions differently than LAMMPS. so what you need to compare is the sum of short range and long range contributions. and those don’t look that different.

also, you are including results from a very, VERY, VERY old gromacs version. have you checked with a more recent version?

apart from that, absolute energies may include arbitrary offsets, what matters are forces.