[gmx-developers] Excluded charges and PME correction

Berk Hess hess at cbr.su.se
Mon May 3 11:39:27 CEST 2010

David van der Spoel wrote:
> On 5/3/10 11:15 AM, Berk Hess wrote:
>> If you have exclusions in your topology, the ewald_LRcorrection will
>> completely remove
>> the mesh electrostatic forces, up to the mesh precision.
>> But a single dipole in a box will feel the effect of the periodic
>> images.
>> Depending on the size of your box, this can be a large force.
> OK, but isn't it strange that this leads to forces perpendicular to
> the dipole as well (the box is cubic)? And shouldn't the dipole
> correction (epsilon_surface != 0) take care of this force?
It all depends on your box matrix and PME accuracy settings.
The dipole correction never fully takes care of this, it simply reduces
it by 1/epsilon.
The 3dc geometry will remove most of it, but not everything.

I guess the perpendicular forces should be (close to) zero, unless your
box is triclinic.


>> Berk
>> David van der Spoel wrote:
>>> If I have a diatomic molecule with opposite charges, e.g. C=O, which
>>> are excluded in the short range, there will be a small force between
>>> the atoms due to PME. Is it correct that this force is completely
>>> taken away by the routine ewald_LRcorrection?
>>> For a simple test system (I2 oriented along the Y axis, with charges
>>> +/- 10 and bond length at 0.4 nm) we find at time step zero for bond
>>> completely in equilibrium:
>>> With cut-off:
>>>     f (2x3):
>>>        f[    0]={ 0.00000e+00, -2.24787e-03,  0.00000e+00}
>>>        f[    1]={ 0.00000e+00,  2.24787e-03,  0.00000e+00}
>>> With PME (epsilon_surface = 1)
>>>     f (2x3):
>>>        f[    0]={ 3.41797e+01, -1.11967e+02,  1.44752e+01}
>>>        f[    1]={ 3.41948e+01,  7.41194e+01,  1.44833e+01}
>>> With PME (epsilon_surface = 0)
>>>     f (2x3):
>>>        f[    0]={ 3.41796e+01, -1.73351e+03,  1.44758e+01}
>>>        f[    1]={ 3.41922e+01,  1.69567e+03,  1.44829e+01}
>>> Surely this can not be correct?
>>> This is using the git head sources.

More information about the gromacs.org_gmx-developers mailing list