[gmx-users] User tabulated potential and VdW Modifier

[Kiesel, Matthias]

Dear everyone,

My question concerns the effect of the mdp option "vdw-modifier" when vdwtype is set to user. I have a tabulated 8 - 6 Mie potential and I'm simulating a coarse grained water, NaCl system, which all interact with the same potential form but different dispersion coefficients. I wanted to check the energy conservation of that system and therefore set up a simulation as follows:

dt = 0.008

cutoff-scheme = group
rlist = 2.3

vdwtype = user
vdw-modifier = potential-shift
rvdw = 2
rvdw-switch = 2

coulomtype = PME-switch
rcoulomb = 2
rcoulomb-switch = 1.95

I will limit my discussion to the VdW energies, since I trust the PME algorithm and I did not change any of the settings in the following discussion. As far as I can judge, this system should be energy conserving, because the potential is shifted, and there is enough space between rvdw and rlist to allow for diffusion in between updates of the neighbourlist. However, this system exhibits an energy drift (total energy) of over 1000 kJ/mol over the course of
30.000 integration steps. On a side note, I also calculated the LJ long range corrections, which is -1888 kJ/mol.
After this simulation I thought, that the vdw-modifier does not affect the tabulated potential. Therefore, I calculated a shifted version of my potential and fed it to gromacs with the same parameters except:

vdw-modifier = none

which effectively should enforce a cutoff equal to that in my tabulated potential (2nm). This setup conserves the energy (Drift ~ 0.4 kJ/mol) and gives me a slightly higher LJ energy, which I expected. However, the long range correction was different: -946 kJ/mol (appr. half). This seemed odd to me, so I played around with some settings and found that the LR correction took the value of -946 for  vdw-modifier = none and -1888 for vdw-modifier = potential-shift. This was also true for a setup with rlist = rvdw = rvdw-switch = 2 (and nstlist = 1), so it wasn't the difference between rlist and rvdw-switch.

According to the manual and grompp notes, the LR correction is calculated as -C6/r_c^6, which should be the same for all simulations, but apparently the vdw-modifier changes something somewhere. Does anyone have an explanation on what's the reason for the behaviour of the LR correction? And also, more generally, are vdw modifier supposed to work for user lists (and if so, why did my initial setup not work)?

Thank you very much in advance,

Matthias