From: Blake Charlebois (bdc_at_mie.utoronto.ca)
Date: Tue Dec 07 2004 - 20:12:18 CST
This old thread may be of interest:
overlaps the "Variation of results in iterations" thread but contains a
little additional information.
From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf
Of Edward Patrick Obrien
Sent: December 6, 2004 11:13 PM
To: Pijush Ghosh
Subject: Re: namd-l: Variation of results in iterations
I run MD simulations in the NVE ensemble, so I believe the only time a
random number is used (if not speciefied in the configuration file) is to
randomly assign velocities from the Maxwell Boltzmann distribution.
Other ensembles though, such as NVT, use randomness which may or
may not be seeded with the random number supplied, maybe someone else
knows about use of random numbers in these ensembles.
On Mon, 6 Dec 2004, Pijush Ghosh wrote:
> I really appreciate the suggestions i am getting regarding my
> problem. In response to what Obrien said i would like to mention that
> I am using same seed for all
> my runs and identical in the sense i am using running same configuration
> file with different name may be with different or same number of
> processors. I am running most of the runs for 400,000 thousands and this
> variation in the results i am observing is terms of the final
> coordinates of atoms.
> Robert suggestd this could be because of some arithmatic operation
> affecting round-off error.
> Does NAMD for any of its calculation picks up some default random
> number everytime it runs resulting some differernce in numerical
> results, which should be theoretically zero though.
> Any further discussion on this will be highly appreciated.
> Edward Patrick Obrien wrote:
> >Deaj Pijush,
> > What do you mean by identical conditions? Do you start these 3
> >simulations from the same exact configuration? Do you use the same random
> >number, or vary it?
> >This info is needed to answer your questions.
> >On Mon, 6 Dec 2004, Pijush Ghosh wrote:
> >>NAMD Friends,
> >> I have a question.
> >> When we run simulations(optimisation or MD) for any
> >>system under identical condition three times, we end up with results
> >>which are not exactly same but
> >>vary within a range, which we denote by error bar in the plot. My
> >>question is inspite of everything remaining same, what could be the
> >>possible reasons giving rise to
> >>these differences in results? Is it because of computation error ? or
> >>anywhere in calculation in NAMD any other parameters involved which
> >>causes so.
> >> Any suggestions and references highly appreciated.
> >>Pijush Ghosh.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:03 CST