[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ESPResSo] p3m tune
From: |
Axel Arnold |
Subject: |
Re: [ESPResSo] p3m tune |
Date: |
Thu, 25 Feb 2010 15:46:20 +0100 |
User-agent: |
KMail/1.10.3 (Linux/2.6.27.42-0.1-default; KDE/4.1.3; x86_64; ; ) |
On Wednesday 24 February 2010 05:50:50 Yang Ye wrote:
> Hi,
>
> I have a script running fine in single process but stalls at p3m tune
> (or tunev2) when running with more than one process with MPI (lam).
> The MPI-Espresso binary has been tested with other scripts running in
> parallel (inside sample directory). How shall I approach to resolve
> this problem?
The problem is presumably, that the tuning takes too long, e.g. because it
tries either too small or too large numbers of cells, or too many different
mesh sizes. You can drastically speed up things by trying to manually
restricting the mesh size and the number of cells. The first is a parameter of
the p3m tune command; typically, the mesh size should be a power of two for
the FFT to be efficient, and the resulting mesh spacing should be
approximately the typical size of your charges. With particles of diameter 1
and a 10x10x10 box, that is a mesh of 32, for example. If fixing the mesh size
doesn't speed up the tuning, you can limit the number of cells by "setmd
min_num_cells" and "setmd max_num_cells". However, it is difficult to give
likely parameters here, typically, you want the cells to be of a size that
there are between 4 and maybe 100 particles in on average. Basically, you can
simply try to fix rigid bounds first, in case P3M doesn't find appropriate
parameters, you can still relax them.
Axel
--
Dr. Axel Arnold Tel: +49 711 685 63593
ICP, Universität Stuttgart Email: address@hidden
Pfaffenwaldring 27
70569 Stuttgart, Germany