From: Alexandre Suman de Araujo (asaraujo_at_fcfrp.usp.br)
Date: Thu Jan 14 2010 - 06:25:33 CST
On Wed, 13 Jan 2010 14:22:13 -0500, Axel Kohlmeyer wrote
> On Wed, Jan 13, 2010 at 1:25 PM, Alexandre Suman de Araujo
> <asaraujo_at_fcfrp.usp.br> wrote:
> > Hi all.
> >
> > I'm performing a simple simulation of a POPC membrane in water. After
1000
> > steps of minimization I started the MD using NPT ensemble, time step of 2
> > fs, pbc and PME. After 1000 steps a had the follow error:
> >
> > FATAL ERROR: Periodic cell has become too small for original patch grid!
> > Possible solutions are to restart from a recent checkpoint,
> > increase margin, or disable useFlexibleCell for liquid simulation.
> >
> > I tried a lot of solutions (changing simulation parameters) and I got the
> > same error. The simulations stopted to crash just after I change the
> > /pairlistdist/ parameter from /13.0 /to/ 14.0. /So I remembered another
> > simulation with this same problem and when I looked at /pairdistlist/ it
was
> > set as 17.0. Another membrane simulations that NOT show this kind of
errors
> > had /pairdistlist/ = 15.0.
> >
> > Is this just a coincidence that the crashed simulations had pairdistlist
as
> > prime numbers or this can be a bug or some limitation of any algorithm?
>
> yes. this is neither a bug nor a real limitation of sorts. it just is
> what it is.
> your system has shrunken a lot, and thus the patch grid cell lists
> would have to be recomputed. NAMD does not do this automatically,
> since it is effectively the same as restarting the calculation. so
> you get a message like you saw and then NAMD aborts. just make sure
> that in scenarios like this (i.e. at the beginning of an
> equilibration), you write restarts often or run only small chunks of
> trajectory and then restart repeatedly, until the system cell is
> closer to equilibrium size. then this message won't show up.
>
> > Changing /pairdistlist/ from 13 to 14 or from 17 to 16 is not a problem
to
> > me, but if this is a (known or unknown) problem/bug I think it is
important
> > to be discussed.
>
> this message has been discussed on this mailing list many, many times.
>
> the message NAMD prints out is the proper summary of what you should
> do.
>
> cheers,
> axel.
>
> >
> > Cheers
> >
> > --
> > **************************************************************
> > Alexandre Suman de Araujo *
> > Faculdade de Ciências Farmacêuticas de Ribeirão Preto *
> > Universidade de São Paulo *
> > Dep. de Física e Química *
> > Grupo de Física Biológica * e-mail: asaraujo_at_fcfrp.usp.br*
> > Av. do Café, s/n° * e-mail: ale.suman_at_gmail.com *
> > CEP: 14040-903 * Phone: +55 (16) 3602-4172 *
> > Ribeirão Preto, SP, Brasil * Phone: +55 (16) 3602-4222 *
> > **************************************************************
> >
>
> --
> Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> Institute for Computational Molecular Science
> College of Science and Technology
> Temple University, Philadelphia PA, USA.
I think I was not as clear as I desired.
My question is not about the error which, I know, was discussed many times
on this list.
My question is about a possible bug/limitation/problem in any part of NAMD
code that can be triggered by the choice of prime numbers as pairdistlist
(and possible as other cut-off parameters), since the impossibility to split
this kind of number in small ones can bring problems in the parallelization
process.
As I said, this happened in my simulations and it stopped to crash when I
exchanged the prime number by any other non-prime number (lower or higher
than the used prime number).
As I also said, this can be just a coincidence and I'd like to know if
anybody else also observed this behavior.
Cheers
-- Alexandre Suman de Araujo Faculdade de Ciências Farmacêuticas de Ribeirão Preto Universidade de São Paulo Dep. de Física e Química Grupo de Física Biológica e-mail: asaraujo_at_fcfrp.usp.br Phone: +55 (16) 3602-4172
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:55:20 CST