AMBER Archive (2005)

Subject: AMBER: GB with Langevin dynamics - solution

From: Peter Varnai (pv232_at_cam.ac.uk)
Date: Mon Apr 11 2005 - 12:18:13 CDT

Dear Amber-users,

First, thanks to David Case, Carlos Simmerling and Andreas Svrcek-Seiler
for helping me to figure out what went wrong with the Langevin dynamics
simulations with sander8.

The key is the "ig" random number seed which has a single default value
that we change normally once at the start of a simulation to initiate
multiple trajectories from the same starting geometry. It was not clear to
me from the manual that random forces in Langevin equation use that
parameter, nor seems to have much rationale to use such default in the
case of Langevin dynamics. I am sure that this is useful for debugging
purposes but should be avoided for production runs. It is easy to write a
script that feeds different seeds into the input. It would however be
useful to generate different seeds automatically for Langevin dynamics
restarts (eg by generating iseed from unix date for the Marsaglias
method).

These identical "random" forces at every restart seem to affect the net
molecular motion in Langevin simulations in a deterministic way. Since the
random forces can produce overall translation and rotation effect, these
repeating forces create quasi-linear translations, especially after
equilibration when the random forces dominate (?) the total force over the
Newtonian and frictional forces. This problem is more visible on systems
with < 500 atoms. The translation speed (not its direction) seems to be
fine at around 5 A/ns, considering the fact that I used gamma_ln=5. I
agree that it would be nice not to remove the rot/trans degree of freedom
as it was pointed out to have adverse effects on the kinetic energy of
small systems. However, in long simulations small molecules can translate
over 999 A and reach the format limits of printing xyz coordinates in
rst/crd. Probably a more elegant solution is to reset COM periodically to
(0,0,0). It is a pity that ptraj center command does not print the
(transformed) forces to its restart format.

Best regards,
Peter

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu