[MMTK] Re: Minimization

khinsen at cea.fr khinsen at cea.fr
Wed Sep 7 09:08:58 CEST 2005

On 07.09.2005, at 00:14, Matias Saavedra wrote:

> Hi Konrad. i'm trying to perform a minimization with a
> small protein (50 residues) in order to go faster...
> but each minimization step takes around 6 seconds, and
> after one hour and a half of calculation and 1300
> steps, i still have a gradient norm of 3.2.

How do you do the minimization?

When you are far away from a minimum, it is best to start using  
steepest descent, adjusting the step size for best performance. Once it  
starts to slow down, switch to conjugent gradient for precision.

An alternative to steepest descent is MD at very low temperatures, e.g.  
starting at 10K and going down to zero. This can be more efficient in  
some cases.

> I would like to now the right moment to stop a
> minimization to perform a normal modes analysis.

It depends on how precise you need the first six modes to have zero  
frequency, but I'd say you should go for a gradient norm of no more  
than 10^-2.

> If a small 50 residues protein takes more than 2 hours
> to get minimized, a 300 residues protein takes several
> days?

Not necessarily. The initial configuration has a significant impact on  
the duration of a minimization. A big protein close to a minimum is  
easier to handle than a small one in an unfavourable starting position.

Konrad Hinsen
Laboratoire Leon Brillouin, CEA Saclay,
91191 Gif-sur-Yvette Cedex, France
Tel.: +33-1 69 08 79 25
Fax: +33-1 69 08 82 61
E-Mail: khinsen at cea.fr

More information about the mmtk mailing list