On Thu, 2015-04-30 at 07:51 -0300, Gustavo Seabra wrote:
> > Em 29/04/2015, à(s) 16:55, David A Case <case.biomaps.rutgers.edu> escreveu:
> >
> > On Wed, Apr 29, 2015, Robert Molt wrote:
> >
> >> I apologize for this very elementary question, but I am having
> >> difficulty following parts of this conversation (and I would very much
> >> like to understand all of the wisdom imparted). It was written, below:
> >>
> >> "5ns windows is also fine, you might want to extend this to longer if
> >> that is easier for you to manage - I tend to try to shoot for 1 hour or
> >> so run time per simulation -"
> >
> > I don't know why Ross does it this way, but it's just a matter of taste and
> > convenience. I generally target about 1 day per individual run: if a machine
> > crashes, I don't loose more than one day's calculation. But as long as your
> > script is automatically running job "n+1" as soon as job "n" is completed,
> > it's up to you how long each job lasts. There is no *scientific* reason to
> > prefer 1 hr vs 1 day vs 1 week.
> >
>
> Risking being wrong (only Ross can give the exact reason here), I’d
> guess it may be an (overzealous?) attempt to avoid the Langevin
> synchronization effect, described above by Jason, since you get a
> chance to change the random seed about every hour.
Indeed, if this is the reason (as Ross indicated it wasn't), then it
would be very misguided :). Synchronization only occurs if you have
separate simulations whose PRNGs are seeded with the same value. Amber
never "re-seeds" its generators after the simulation starts (I'm not
aware of any program that would, and it would be inadvisable to do so).
So there is only one stream (ergo, no synchronization is possible).
Note that random generators *do* cycle (i.e., they have a period) so
there is a theoretical rationale behind this reasoning, but the period
on modern generators (including Amber's) is so large that it would be
impossible to use them all up in any modern simulation (I wonder how
long it would take a supercomputer to even *generate* the full period,
let alone use them for something productive).
All the best,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Apr 30 2015 - 11:00:02 PDT
