From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Wed Jun 27 2007 - 10:32:49 CDT
Hi Ilya,
it seems to me that any velocity reassignment ought to be done strictly
by the thermostat, to make sure that your simulation is controlled and
consistent with the methods you think you're using. Even if you are
using a velocity reassignment thermostat, reinitializing the velocities
prior to each run will (at the very least) add additional reassignments
at more or less random points. If you're using (say) a langevin
thermostat, then you'll be adding velocity reassignments at random
timesteps to a method that is more specific about temperature control.
The problem with a random velocity reassignment is that it breaks up any
concerted motions that might be occurring at the point when you reassign
velocities. Most thermostats do this to some extent, but adding a
complete randomization to a previously existing thermostat may cause
problems and should probably be avoided (just don't to reinitvels in
your restarts and everything should be fine).
Best,
Peter
Ilya Chorny wrote:
>
> Peter,
>
> Does it really matter if I assign new velocities when I restart a job?
> Some thermostats work by reassigning the velocities every so often. I
> do use bin coordinates to restart the job.
>
> Thanks,
>
> Ilya
>
>
>
> On 6/27/07, *Peter Freddolino* <petefred_at_ks.uiuc.edu
> <mailto:petefred_at_ks.uiuc.edu>> wrote:
>
> When you have to restart a run, you should do so by using
> binarycoordinates and binaryvelocities (set to the restart files from
> your previous runs) and *not* assign new initial velocities. This
> may or
> may not be responsible for what you saw (the COM motion may also
> be an
> issue), but is certainly the only way to actually continue an existing
> simulation.
> Peter
>
> Ilya Chorny wrote:
> > Hello All,
> >
> > After analyzing my trajectories of a large protein in a membrane, I
> > noticed the my protein shifts about 3 A after about 10
> ns. Would one
> > expect diffusion on this time scale? Taking a diffusion constant of
> > about 10^-8 cm^2/s the conversion into MD units is .1A^2/ns (if
> I did
> > the math right), thus 3 A seems a bit much. Any thoughts?
> >
> > I am also working with a manic cluster which kills my jobs every so
> > often, at which point I restart the trajectories with new initial
> > veloocites. Could restarting with new initial velocities cause a
> > shift/diffusion.
> >
> > Finally, I do not zero the CM motion of the system. Does that mean I
> > have a non-zero CM velovity, and thus I should expect the system to
> > drift as a whole?
> >
> > Thanks,
> >
> > Ilya
> >
> >
> > --
> > Ilya Chorny Ph.D.
>
>
>
>
> --
> Ilya Chorny Ph.D.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:44:54 CST