Re: [AMBER-Developers] vdW radii in pmemd.cuda

From: Jason Swails <>
Date: Wed, 1 Aug 2012 19:21:49 -0400

On Wed, Aug 1, 2012 at 10:16 AM, David A Case <>wrote:

> On Mon, Jul 30, 2012, Jason Swails wrote:
> >
> > Actually, pmemd.cuda automatically applies a value of 0.5 angstroms as
> > the sigma value (or epsilon? I always get those confused) for atoms
> > whose vdw parameters are 0 and we've seen no ill effects so far (and the
> > code gets tested pretty thoroughly).
> (Expresses surprise!) Is this documented anywhere? What value of epsilon
> is
> used? Since Amber inputs never use sigma, is it Rmin that is set to 0.5
> Ang?)

Hmm... looking into this more, this may be a slightly false alarm. Rmin
(or sigma, as it's called in the CUDA code) is set to 0.5, but epsilon is
still set to 0. Given the functional form of the LJ equation, this would
not actually have any effect, since the geometric mean of epsilons will
still be 0.

I'm not sure why sigma was set to 0.5, but it doesn't seem to have any
effect, unless I'm butchering the obvious. I had originally thought that
setting sigma > 0 and epsilon = 0 would add a hard repulsive radius with no
attractive term, but this was before doing the maths. When this is sorted
out, I'll clarify my response on the Amber list.

For the interested, the relevant lines in gpu.cpp are 1499-1518 in the
master branch (1942-1962 in spfp-dev). nbtype is -1 for atoms with 0 LJ

All the best,

Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
AMBER-Developers mailing list
Received on Wed Aug 01 2012 - 16:30:03 PDT
Custom Search