From: Brian Radak (bradak_at_anl.gov)
Date: Fri Aug 19 2016 - 07:54:54 CDT
Multprocess runs are generally impossible to reproduce due to randomness
in the order of operations. Even a 216 water box on my workstation comes
out slightly differently each time when threading Newtonian dynamics on
4 cores. This is doubly a problem for stochastic methods like Langevin
Your best bet is probably to re-seed trajectories from conditions that
might produce a transition. I believe techniques like this have been
used for constructing Markov state models, but I'm not an expert. Maybe
search the literature?
On 08/18/2016 10:35 PM, sunyeping wrote:
> Dear all,
> I guess I missed some important snapshots which contains important
> atom interaction information for understanding a conformational
> transition in a namd run because I set a low output frequency. It is
> possible to rerun a certain fragment of the trajectory with a higher
> output trajectory so that the new output trajectory contains all the
> original snapshots of that trajectory fragment but complemented with
> the snapshots between them.
> Maybe it is impossible to reproduce exactly the original trajectory.
> My aim is to understand how the molecule transforms from conformation
> A in snapshot A to conformation B in snapshot B. If I cannot rerun the
> trajectory form snapshot A to snapshot B in a normal MD, maybe I can
> use target MD to do it. But I guess the conformational transition is
> caused by the internal atom interaction in the molecule, so would
> target MD -- which adds an extra force to the molecule -- be useful ?
-- Brian Radak Postdoctoral Appointee Leadership Computing Facility Argonne National Laboratory 9700 South Cass Avenue, Bldg. 240 Argonne, IL 60439-4854 (630) 252-8643 brian.radak_at_anl.gov
This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:20:39 CST