[gmx-developers] introduce effective potential

[Nguyen Hoang Phuong]

> >I would like to replace the non-bonded potential energy functions
> >(Coulomb and vdW) by the effective potential
> >
> >U_eff = KT*q/(q-1)*log(1 + (q-1)*[U + epsilon]/KT)
> >
> >where q and epsilon are the parameters, U is the original
> >nonbonded potential.
> >
> >This is equivalent to introduce the scale factor:
> >
> >q/(1 + (q-1)*[U + epsilon]/KT)
> >
> >into the forces which acting on the atoms.
>
> Aside from practical problems with the code there are other problems.
> For most bonded interactions it is possible.
> But for non-bonded interactions you will run into conceptual problems.
>
> Coulomb potentials can be both postive and negative, depending on
> the signs of the two charges. Maybe you could circumvent this by
> using abs(U) instead of U in your formula.
> But Lennard-Jones interactions can be positive and negative
> depending on the distance. One option could be splitting the
> repulsion and the dispersion, but I don't know if this is what you want.
>
> For the normal potentials one does not need to worry about how
> the terms are split up, since the potentials are additive.
> But your conversion is non-linear.

you are right but in my study, I have to chose the parameter epsilon such
that (U + epsilon) > 0, q > 1. Note that q and epsilon are parameters. In
this case I don't have any problem.

Phuong