[gmx-users] PME and particle insertion

[Jakob Wohlert]

Dear all,

I see that issues concerning PME and particle insertions have been discussed before, and some issues were apparently resolved for the latest (5.1.2) version. However, I have noticed a strange thing when using TPI for inserting (tip3p) water in liquid water at 300 K and 1 atm; the distribution of DU using PME becomes very different from when using a plain cutoff. Using reaction field electrostatics on the other hand gives a distribution which looks essentially the same as for cutoffs. You can have a look at the distributions at: http://www.wohlert.se/insrun.png

The reported chemical potentials also differ; cutoff and RF give reasonable <mu>, but with PME they differ by an order of magnitude. I also tried using water at lower density where convergence shouldn't be an issue, and basically got the same effect: http://www.wohlert.se/insrun2.png

I would of course expect some difference between PME and plain cutoffs, but this seems unreasonable to me. Or isn’t it?

Any suggestion what’s going on would be appreciated!

Regards,
Jakob

PS
I use gmx 5.1.2, with 7000 tip3p water in cubic box, and the mdp file looks as below, with the only difference between the runs is that coulomb type is either cut-off, PME, or Reaction-field.

integrator               = tpi
dt                       = 0.002
nsteps                   = 100000
comm_mode                = Linear
comm_grps                = System
nstcomm                  = 1
pbc                      = xyz

nstlist                  = 1
ns_type                  = grid
rlist                    = 1.2
cutoff-scheme = group

coulombtype              = Cut-off
rcoulomb                 = 1.2
vdw-type                 = Cut-off
rvdw                     = 1.2
epsilon-rf              = 80

Tcoupl                   = berendsen
tc-grps                  = System
tau_t                    = 2.0
ref_t                    = 300
ld_seed                  = -1

Pcoupl                   = no