[gmx-developers] energy re-normalization

[Berk Hess]

Hi,

If you want to keep forces between atoms, that has to be done in the 
non-bonded kernels. This is a useful feature in general, but then I 
would suggest making the potential linear and thus the force constant. 
There are currently no plans to implement this. Also for it to work 
well, the cut-off should likely be based on the LJ parameters, as for 
the soft-core interactions use in free-energy calculations.

Cheers,

Berk

On 2018-01-18 12:19, Christian Krekeler wrote:
> Dear all,
>
> I have a short question. I'm working on replacing the old GC-AdResS
> implementation. I would like to avoid to enter into the force kernel and
> adjust it there.
>
> In AdResS the interface between hybrid and coarse grained part of the
> simulation is the problem. The introduction of the atomistic details can
> lead to atoms which are way too close and thus the forces, energies and
> velocities are way too high and the simulation crashes. In the old
> nb_generic_adress force calculation a force capping was introduced.
>
> Another approach seems to be to re-normalization of specific atom-atom
> interaction at that interface whenever an interaction between an atom i
> and an atom j is so high that will make the system explode, i.e. if
> U_{ij}>U_{cutoff}, then U_{ij}=U_{cutoff}
>
> I'm now looking for some advice on where to best put it.
>
> Thanks!
>
> Christian
>
>
>
>