[MMTK] symeig change to eigh

Konrad Hinsen research at khinsen.fastmail.net
Mon Apr 2 06:44:38 UTC 2012


ra.biehl writes:

 > I was trying to reduce a bit the memory usage of the normal mode
 > calculation of a large system.

That's not easy if you want to remain general. The only approaches I
know that can reduce memory use significantly are

- reduce the system size (coarse-grained models)

- reduce the number of modes (e.g. by using MMTK.Subspace)

- use iterative methods with a sparse representation of the force constant
  matrix to calculate selected modes

 > This project or better the source is not available anymore, but the wrapper for the
 > underlying lapack functions has moved to scipy and is included as scipy.linalg.eig
 > Changing the small pieces attached at the end allow the usage of scipy.linalg.eigh
 > instead of symeig in NormalModes/Core.py

I'll apply those modifications to MMTK, thanks!

 > Testing with AmberFF and insulin.pdb in mode='all' the Normal mode
 > calculation is a bit faster (232 s instead of 291s ) but
 > unfortunately the memory usage is not changed.

That's what I would expect.

If you want to try iterative approaches, I should still have some old
interface to ARPACK that does the job, assuming that ARPACK didn't
change too much over the last 10 years.

Konrad.
-- 
---------------------------------------------------------------------
Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: research AT khinsen DOT fastmail DOT net
http://dirac.cnrs-orleans.fr/~hinsen/
---------------------------------------------------------------------



More information about the mmtk mailing list