[gmx-developers] Reaction Filed crash!

Shirts, Michael (mrs5pt) mrs5pt at eservices.virginia.edu
Sat Dec 17 22:12:50 CET 2011

A couple thoughts:

1. if something was stable before (even if it was not best practices), then
I agree that it should be stable in the current code, unless the previous
algorithm was wrong to begin with.  I haven't looked into the details best
to know what the problem is yet.

2. Massive heating damped with temperature control will drive you away from
an equilibrium to a steady state where some properties become incorrect even
if others are correct.  So you do have to be very careful that the
properties you looked at before are the same ones.  Where exactly the
failure point is is very hard to say very clear information about the
specific system.

> We can also discuss if, as a general rule, using a continuum reaction
> field is better or worse than using a lattice method such as PME, but that
> is a different issue. Like many other people (Wilfred van Gunsterem, Arieh
> Warshel, Alan Mark, Philippe Hünenberger, etc), I'm really not convinced
> that lattice methods offer any real advantage, because I find the
> published evidences for either their benefits or their artifacs to be
> rather weak or contradictory.

For HOMOGENEOUS systems, then I agree; the configurations sampled can be
achieved by cheaper methods than PME -- basically, the configurations are
dominated by short-range effects, and beyond a certain range, all that the
extra interactions do is affect the overall energy, which can be calculated
pretty well by many methods (including reaction field).

When lattice methods are needed is inhomogeneous systems, such as membranes
and interfaces, because the long range order affects the configurations
sampled, and RF and other continuum methods simply can't handle those well;
properties will depend on cutoffs, etc.

Michael Shirts
Assistant Professor
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu

More information about the gromacs.org_gmx-developers mailing list