Hi Marek,
> My actual interpretation of the implementation of general formula
>
> V(L) = (1-L) V0 + L V1
>
> is that during the MD run with the given L value in the given time step ti
>
> a) the atom interactions (potentials, forces) are calculated according to
> potential V0, using the actual coordinates
> b) the atom interactions (potentials, forces) are calculated according to
> potential V1, using the actual coordinates
but a and b are done on different processors, so its not slower than a
non-TI run.
> c) the final atom interactions (potential energy, final forces necessary
> to propagate the system in time) are calculated
> as the L-linear combination of those calculated in a) and b).
>
> Then also dV/dL is calculated (here simple as V1-V0).
that is exactly what thermo_inf.f does, plus mix the forces as well to
propagate with the mixed potential. The only additional complication is
that if softcore atoms exist, parts of the potential are no longer linear
in lambda. All the additional derivative and energy components for that
case are collected at many different points in the code, so this is a bit
more tricky to go through (module softcore.f provides a good start).
> I thing that for all technical details the best way is really to check
> relevant source files.
That is technically true, but I know that sander code (and any other
program written by someone else) is hard to read, especially the optimized
and parallel parts. Nevertheless, sometimes only the source can tell you
whats really going on :-)
Kind Regards,
Thomas
Dr. Thomas Steinbrecher
formerly at the
BioMaps Institute
Rutgers University
610 Taylor Rd.
Piscataway, NJ 08854
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 09 2011 - 12:30:05 PST
