[gmx-users] Re: Free energy blowing up

[Michael Shirts]

If you are mutating one atom A->B, there is no need to use soft core,
as there is no problem with singularities.  Actually, it would
probably make things worse.  I'd stick with linear interpolation in
the A and C parameters (i.e., where U(r) = A/r^6-C/r^12 -- this will
likely work better than linear interpolation in epsilon and sigma, as
the "core" that is excluding water will change less rapidly.

Also, with a change like this, I would highly suggest not using a slow
growth type simulation.  As has been pointed out, hysteresis is not
your friend.  Instead, run a number of equilibrium simulations at
intermediate with no change in lambda.  Collect the average <dH/dL>,
and numerically integrate.  Changing A->B, the curve should be fairly
smooth, allowing for higher order integration techniques, and fewer
points.  I bet 5-10 will be sufficient.

Best,
Michael Shirts
Research Fellow
Columbia University