[gmx-users] tabulated potential - why table length is rc+1nm

[Erik Lindahl]

Hi,

On Oct 12, 2007, at 7:54 PM, Nickle Fan wrote:

> Dear gmx-users:
>
> I am working on incorporating a customized non-bonded interaction  
> into my simulation. I am trying to implementing it through the  
> tabulated interaction potential.
>
> The manual says that the potential should be tabulated up to rc+1.0  
> nm (rc is r_vdw or r_columb). Why is this necessary? My impression  
> is that for distance larger than r_vdw/r_columb, the interaction is  
> truncated and thus set to zero.

Two reasons:

1. Neighborsearching is based on "charge groups" for performance  
reasons. Thus, even if two charge groups are just inside the cutoff  
distance some of the atomic pair interactions between the two groups  
can be beyond it. This depends entirely on your force field - if you  
don't like it you can simply assign a separate charge group to each  
atom.

2. The neighborlist is only updated every nstlist (10 or so) steps.  
Since the charge groups move during these 10 steps you might have two  
charge groups just inside the cutoff that diffuse out of the cutoff  
before the next neighborsearching.


In particular item (2) gets worse when you have a system very far  
from equilibrium, and then it is very handy to have a buffer zone so  
the simulation doesn't crash and die just because some charge groups  
moved slightly more than they should.

The actual interaction isn't truncated (unless you use switch/shift),  
it is just that we will try not to calculate it outside the cutoff.

Cheers,

Erik