Re: Re: numerical inaccuracy upon restart

Date: Mon Sep 06 2004 - 06:30:22 CDT

> This PDB class, it so happens, stores
> coordinates as single-precision floats, dropping about nine significant
> digits, which nicely explains the observed deviations. :-(

Thanks Jim, this is really enlightening. I hope it can be fixed in
a future release.

> For the record, another factor in non-repeatability is that since NAMD is
> execution-driven, in a parallel simulation forces and energies are added
> in a possibly different order each time a job is run. The small
> differences usually won't show up in the energies on the first step, but
> since MD is a chaotic algorithm, they will accumulate over time and be
> noticable after a couple of hundred steps.

I was aware of that possibility and ran all my test on one node.

Thanks again for tracking this down,


This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:52 CST