[gmx-users] neighbor list

[Berk Hess]

>its not so much about the noise, the point with charge
>groups is that you end up with electrostatic interactions between only 
>dipoles (which go with 1/r^3) and no
>more monopole-monopole interactions ( 1/r ).
>Thus the electrost. interaction is (seeemingly) shorter ranged (though 
>still not converging for r-cut -> infinity) giving results that are less 
>system-size dependent; whether they
>are correct is a different question ...

That is exactly what I was referring to, but maybe not stated very clearly.
Interactions get much stronger as suddenly a charge iso a dipole is felt.
This appears as 'noise' on the potential and leads to incorrect sampling.

>but there still is the question: is in Gromacs only the generation the 
>neighbour-list group-based or also the calculation of
>the electrostatic interactions (which is nor necessarily the same
>thing)

Everything is group based. That is, atoms pairs are put in the neighborlist
based on distances between charge groups. All these atom pairs are used
for the electrostatics and LJ calculation.

>I am not sure whether group-based cut-off in combination
>with PME is a good idea, and If you don't use PME there are still reasons 
>to avoud group-based cut-offs, see, e.g., Biophys J, September 2000, p. 
>1537-1553, Vol. 79, No. 3;
>there is also a more rigorous investigation of the subject by Hummer G., 
>the exact reference eludes me for know ...

In principle it is better not to have charge groups with PME.
But the difference is quite small, especially with the relatively
small charge groups in the available force-fields.
Neighbor searching is expensive, so charge groups save a lot of time.
At the same cost you can probably use a larger cut-off distance
that almost compensates for the charge group size.

It might be nice though to have the option to turn charge groups off,
even if it is only used to check the accuracy.
I think it has been in the development code at some point, but it 
disappeared.

Berk.

_________________________________________________________________
MSN Search, for accurate results! http://search.msn.nl