[gmx-users] Re: domain decomposition trick for highly parallel, non-equilibrium runs

[Kathleen Kirchner]

Sorry for double post, first message had the wrong subject.

Dear Vitaly,

I had to struggle with the equilibration of an ionic liquid at charged
electrodes - it means a spatial inhomogeneous system and high forces due
to the electrostatics). I found the following tricks working for me:

* Pre-equilibration with a reduction of the step size (e.g. from 2 fs to
0.2 fs): After a several thousands steps (depends on the system of
course) the jobs can be restarted with the required time step simply by
using the .gro output file.

* Reducing the temperature and slowly heating the system by annealing.

* Don't use "generate initial velocities"!

* Assigning zero charges to some particles for the first equilibration
(E.g. in my case I had to charge the electrodes, which nearly always
crashed if I didn't to an equilibration before hand with neutral
electrodes. - Might be not useful in your case).

* Mixture of all steps above :)

I hope you will find some idea that helps you with the equilibration.

Regards
Kathleen


On 10/06/2012 07:27 PM, gmx-users-request at gromacs.org wrote:
> Message: 3
> Date: Sat, 6 Oct 2012 18:03:54 +0200
> From: "Dr. Vitaly Chaban"<vvchaban at gmail.com>
> Subject: [gmx-users] domain decomposition trick for highly parallel
> 	non-equilibrium runs
> To:gmx-users at gromacs.org
> Message-ID:
> 	<CAPXdD+Z+_2WvVR4nV+nG84SPgn362czmxJip1SGfezX5LK8+5A at mail.gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Dear all -
>
> Would anyone perhaps suggest some tricks to increase the stability of
> the highly parallel jobs (NP > 100), which use domain decomposition
> among nodes. Especially, for the case, where beginning state of the
> system is not equilibrium state (although it contains no bad contacts,
> etc), i.e. the potential energy of the system decreases during MD.
>
> Thank you for insights,
> Vitaly Chaban
>
> -- Dr. Vitaly V. Chaban MEMPHYS - Center for Biomembrane Physics
> Department of Physics, Chemistry and Pharmacy University of Southern
> Denmark Campusvej 55, 5230 Odense M, Denmark

-- 
Kathleen Kirchner
PhD student
Max Planck Institute for Mathematics in the Sciences
(MPI f. Mathematik in den Naturwissenschaften)
Inselstr. 22-26, D04103 Leipzig
e-mail: kirchner at mis.mpg.de
web: http://www.mis.mpg.de/scicomp/CompPhysChem/
Tel +49 341 9959 725
Fax +49 341 9959 999