[gmx-users] charge group radii

[Mark Abraham]

On 30/11/2010 2:25 AM, Gavin Melaugh wrote:
> Cheers Mark for the reply
>
> I was more concerned about the Van der Waals interactions. If the
> distance between the centre of geometry of two charge groups is larger
> than rlist then these interactions are not included. I was therefore
> concerned if I had a situation in which certain atoms of a charge group
> (6 atoms) where close enough to interact with atoms of another charge
> group, but the COGs of the charge groups were larger than rlist, then
> these interactions would not be accounted for. Is this true?

Something along those lines is true - I forget the details. Charge 
groups moving in and out of neighbour lists is a source of algorithmic 
noise that affects non-PME models. Making them small in radius, small in 
numbers of atoms, of integral charge, and with larger neighbour-list 
radius reduces these effects. However the VDW interaction strength has 
decayed to zero faster than the Coulomb, so the latter is a more 
important effect.

Mark

> Gavin
>
> Mark Abraham wrote:
>> On 26/11/2010 3:01 AM, Gavin Melaugh wrote:
>>> Hi all,
>>>
>>> I have recently been testing out the new version of Gromacs. To do so I
>>> have used files from previous simulations in Gromacs-4.0.7. When feeding
>>> the three files (mdp,gro, and top) into grompp, the following note is
>>> displayed:
>>>
>>> NOTE 3 [file pbc.mdp]:
>>>     The sum of the two largest charge group radii (0.602793) is larger
>>> than
>>>     rlist (1.500000) - rvdw (1.400000)
>>>
>>>
>>>
>>> This note did not occur in the previous version and it has led to me to
>>> ask some questions.
>>>
>>> 1)  I had previously hydrocarbon chains (6-7 carbons long) assigned to
>>> individual charge groups, and never any note. Does this mean that my
>>> previous simulations have artefacts? (all charge groups have zero net
>>> charge)
>> Maybe. Some electrostatics models are sensitive to charge group size
>> for dealing with cut-offs. PME isn't.
>>
>>> 2) If I am happy with the values of rlist and rvdw, will it not be
>>> pretty difficult to assign a charge group with radius<   0.1nm?
>> Not really. A charge group is normally chosen to be a heavy atom, plus
>> perhaps some associated hydrogen atoms, such as a methylene group. Its
>> radius is smaller than a C-H bond length, so much less than 0.1nm.
>> Normally the total charge of each charge group should be an integer if
>> your electrostatics model cares.
>>
>> Mark