[gmx-developers] Problems with artificial minima when using soft-core potentials

[Berk Hess]

Hi,

I introduced identical scaling of for LJ and electrostatics in Gromacs 
more than a decade ago
exactly to avoid issues of double minima which arose with the Gromos 
setup where the electrostatics
were soft-cored with r^2 with a second alpha parameter.
When creating (unphysical) paths between states it's difficult to avoid 
additional minima at intermediate states.
A few years ago we noticed such issues with force fields that have H's 
with small sigma's for which
For that I introduced a simple fix. In general it is advisable to 
decouple electrostatics and LJ separately, which
avoids the issue. With the above mentioned fix, we haven't had much 
problems and I haven't heard about
issues recently from other groups.

Can you provide more details about such minima?

Cheers,

Berk

On 08/30/2012 03:58 AM, Alan E. Mark wrote:
> A number of people have noted potential problems with the way the 
> solft-core potential is implemented in GROMACS.
>
> Specifically if the same parameters are used for the soft-core for 
> both LJ and electrostatic interactions this gives rise to temporary 
> minima at intermediate lambda values.
>
> That using a single value for both the LJ and Coulombic interactions 
> would give rise to this problem was recognized in the original paper 
> of Beutler et al. http://dx.doi.org/10.1016/0009-2614(94)00397-1
>
> The simple and logical solution to this problem is simply to introduce 
> separate alpha values for the LJ and Coulombic interactions. That is 
> to use the form of the soft-core that was proposed originally. If the 
> alpha value for the charge is set larger than that of the LJ these 
> artificial minima no longer occur.
>
> Alan Mark and Alpesh Malde
>