[MMTK] Problems with NMA of big structure

Konrad Hinsen research at khinsen.fastmail.net
Mon Oct 17 09:19:28 UTC 2011


On 13 oct. 11, at 17:54, mmtk-owner at starship.python.net wrote:

> As the consequense I've obtain error -  that array is too big .
...

> Does this error due to big size of my structure ? What restrictions  
> in the atom ammount has classical forse field based NMA?

MMTK imposes no restrictions on the size of a structure, but there are  
other sources of limitations:

1) The available memory on your computer.

2) Limitations imposed by 32-bit architectures: a single process can  
only access 2 GB in total.

If you are using Linux or MacOSX, consider moving to a 64-bit version  
if your computer supports it.

What fails in your script is the allocation of an array for storing  
the Hessian matrix. For an N-particle universe, this matrix requires  
8*(3*N)^2 bytes. For 5000 atoms, that's 1.8 GB.

If you use a force field with a sufficiently small cutoff, you may be  
able to handle big systems by using the sparse matrix storage mode.  
Just add sparse=True to the argument list of the NormalModes  
constructor. However, note that this will slow down your computations.

> Also I wounder to know if it's possible to save results of my NMA  
> ( e.g obtained eigenvalues- freqs ) in the separate text file after  
> simulation?

Everything is possible with a few lines of Python, that's the big  
advantage of MMTK.

> Finally is there a possibility to increase of speed of NM  
> computation by ussage of multiple core of my processor ? (Recenly  
> you've told me abut function "thread='numer_of_cores' '" but I  
> coudnt find a way to use it in my NMA computation or energy  
> minimization)

You can use threads=... for energy  minimization. For the eigenvalue  
analysis, MMTK uses standard library functions (most frequently  
LAPACK, but the details depend on your hardware and your software  
installation). There are implementations of these libraries that  
automatically use multiple cores. On my Mac, a standard installation  
of Python and NumPy uses Apple's accelerated LAPACK version, which  
uses multiple cores.

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
---------------------------------------------------------------------




More information about the mmtk mailing list