[gmx-users] TIP5P water and Test Particle Insertion

Joerg Sauter Joerg.Sauter at mpikg.mpg.de
Wed May 28 15:55:59 CEST 2014

Dear Gromacs users,

I did some further investigations on this. It seems to be a different 
problem then a (fixed) issue described previously ( 
https://www.mail-archive.com/gmx-users@gromacs.org/msg33349.html ). To 
be sure, I ran it using double precision and the error still occurs.

The likelihood of the occurrence is dependent on the number of 
insertions performed. For few insertions the problem does not occur (for 
reasonably long trajectories), but I do not get converged results. 
However, for 200.000 insertions per frame the problem occurs within the 
first few frames.

When I modify tpi.c to exclude the -inf case there remain some (rare, 
but significant) large negative values of the excess chemical potential 
for some frames. The effect of this is that the total result differs by 
about a factor of 10 from the experimental value (and the value of TIP3P 
and SPC/E water). However, I do not expect the TIP5P value to be that 
off, and -inf should not occur at all.

Judging by the Widom insertion formula, there must be large negative 
insertion energy differences for this to occur. Since the contributions 
from the Lennard-Jones potential are bounded below, LJ interactions 
should not be the cause. On the other hand: The charges in the TIP5P 
models are smaller than in the TIP3P model so electrostatics should not 
be an issue either.

I would appreciate any thoughts on this problem.


On 05/14/2014 11:44 AM, Joerg Sauter wrote:
> Dear Gromacs users,
> I am using Test Particle Insertion with Gromacs 4.6.5. The method 
> works well when I use TIP3P and SPC/E water to obtain the excess 
> chemical potential of water and I get a reasonable agreement with 
> experimental data.
> However, when using TIP5P water the method diverges:
> [...]
> mu -1.159e+01 <mu> -1.791e+01
> mu       -inf <mu>       -inf
> mu -1.387e+01 <mu>       -inf
> [...]
> I have tried different trajectories and different ld-seed values, but 
> the problem remains.
> I wonder if maybe the lone-pairs might cause the problem.
> Best,
> Joerg

More information about the gromacs.org_gmx-users mailing list