On Mon, Oct 7, 2013 at 2:27 PM, Niel Henriksen <shireham.gmail.com> wrote:
> Hi Eugene,
>
> Try doing a simple R-REMD run with ntpr=1, ntwx=1, and nstlim=1.
> Everything should match nicely. I think the combination of the
> EPtot/mdcrd mismatch and the order of operations creates a confusing
> situation when you only write/print every 500 steps or so. I don't have
> time to carefully go through the code to determine what is happening, but I
> believe the exchange attempt is also a half-step off of the printed
> energies and so things don't match nicely unless you print every
> integration step.
>
> So my reply to your questions:
> 1) the potential energies in the saveene file should be taken from sander
> post-processing of the frames in the reservoir
> 2) Unless you print energies every step during MD (not usually a good idea
> for performance), only post-processing will give you the correct energy of
> your mdcrd frames
>
Further to Niel's comments here, even if you _were_ able to write out the
trajectory with the 'correct' snapshots (I'm not sure if the mdcrd should
correspond to the REMD energies or not), it is likely that post-processing
with sander will _still_ not give you identical energies for this
reason: Coordinates are stored in full double precision internally, yet are
written either in laughably low precision (ASCII trajectories) or single
precision (NetCDF trajectories), meaning that the exact values of the
coordinates themselves will never be exactly identical. [1] This is
another argument for following Niel's suggested steps.
HTH,
Jason
[1] NetCDF restart files are written in double precision, and if enough
care is taken you _can_ exactly reproduce sander energies using these
files, as long as care is taken to evaluate energies deriving from the
exact same time step. I have done it myself.
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 08 2013 - 05:30:04 PDT
