[gmx-users] Overflow problem with test-particle insertion

[Kevin Daly]

Hello Gromacs users,

I've been doing some simple NVT simulations of Lennard-Jones particles using
the built-in 12-6 potential and a tabulated version (vdwtype=user) of this
same potential. Both give practically identical results for the density and
pressure, but differ in the chemical potential computed using the "tpi" mode
in mdrun. The chemical potentials end up having the same order of magnitude,
but the chemical potential for the built-in function seems to be
systematically lower, e.g. 3.48 kJ/mol (built-in) vs. 6.26 kJ/mol
(tabulated). The difference is smaller for the double precision version of
mdrun (e.g. 7.19 kJ/mol vs. 7.55 kJ/mol).

I ran some simulations in debug mode, and looked at the energies and
coordinates of each insertion. It seems like 99.9% percent of the them are
identical, but when the insertion energy is especially high, the simulation
using the built-in function just outputs some random number, whereas the one
with the table computes  something around 5e18, the number to which the
table plateaus as r goes to 0.

Here is an example using mdrun_d (I've removed extraneous lines from the
logging file):

Built-in function:

          Insertion #     Energy           x           y          z
TPI    6205  2.23227e+06      0.00961      0.91659      1.56009
TPI    6206  2.29990e+06      0.01536      0.84120      1.57042
TPI    6207  6.44107e+07     -0.01544      0.89706      1.55119
TPI    6208  2.57252e+07     -0.00345      0.85349      1.55140
TPI    6209  1.65644e+07     -0.02364      0.87913      1.57432
TPI    6210  6.26137e+03      0.83206      0.68462      0.46606
TPI    6211  9.31542e+09      0.78445      0.71232      0.44283
TPI    6212  9.63046e+11      0.80033      0.67289      0.44387
TPI    6213  7.00802e+10      0.78840      0.69046      0.44094

Tabulated function:

          Insertion #     Energy           x           y          z
TPI    6205  2.23227e+06      0.00961      0.91659      1.56009
TPI    6206  2.29990e+06      0.01536      0.84120      1.57042
TPI    6207  6.44107e+07     -0.01544      0.89706      1.55119
TPI    6208  2.57252e+07     -0.00345      0.85349      1.55140
TPI    6209  1.65644e+07     -0.02364      0.87913      1.57432
TPI    6210  6.01461e+18      0.83206      0.68462      0.46606
TPI    6211  9.31542e+09      0.78445      0.71232      0.44283
TPI    6212  9.63046e+11      0.80033      0.67289      0.44387
TPI    6213  7.00802e+10      0.78840      0.69046      0.44094

Notice that for insertion #6210, the built-in function outputs 6.26137e+03.
I checked the coordinates of the other particles in this frame, and there is
a particle at (0.829, 0.686, 0.466), so I would think the insertion energy
would be really high; hence the value of 6e18 for the tabulated function.

I am using Gromacs 4.0.7 straight from the Ubuntu lucid repository (
http://packages.ubuntu.com/lucid/gromacs). I am running it on the i386
architecture.

Thanks,

Kevin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20100630/042a867f/attachment.html>