Re: [ESPResSo-users] Energy minimization at C level

[Axel Arnold]

On 01/17/2013 04:17 PM, Rudolf Weeber wrote:
> Hi,
> On Thu, Jan 17, 2013 at 04:07:00PM +0100, Lei Liu wrote:
> > Dear all,
> >
> > I need to do lots of energy minimization with Espresso at the moment.
> >
> > Dose anyone has experience in implementing energy minimization functionality,
> > with whatever method (e.g. deepest descent), in the core of Espresso ?
> > It is best not at the TCL interface level, because I hope it is faster.
> >
> > Any advice would be much appreciated.
> Simulated annealing, i.e., lowering the temperature over the course of the 
> simulation, can easily be done from the tcl level.
> Side remark: Don't forget to adjust the Bjerum length of electrostatic or 
> magnetic interactions after changing the temperature.
Hi!

That is not quite the same. Energy minimization by steepest descent or 
CG is mostly for local energy minimization, while simulated annealing 
and co try to get to the global minimum, at a much higher computational 
cost, of course. Therefore, implementing some energy minimization would 
be useful. One could implement it similar to the integrator, as for 
steepest descent, only forces are necessary.

> An other alternative is to set the temperature to zero after a warmup and just 
> keep a very small friction force and then let the system follow the forces.

This could actually come close to a steepest descent, with sufficient 
dissipation, i.e. the friction should not be too small.

Axel

--
JP Dr. Axel Arnold
ICP, Universität Stuttgart
Pfaffenwaldring 27
70569 Stuttgart, Germany
Email: address@hidden
Tel: +49 711 685 67609