Minimization error with CG

Luca bellu... at unisi.it
Fri Oct 5 17:18:43 CEST 2007


On Oct 5, 4:38 pm, Teodoro Laino <teodor... at gmail.com> wrote:
> Luca,
> The error you get with CG means that gradients are too large..
> Rescaling the forces may help.. have a look at:
>
> http://cp2k.berlios.de/input/
> InputReference~__ROOT__~FORCE_EVAL~RESCALE_FORCES.html

Ok  it work but don't stop.
                Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036
................................
 ENERGY| Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036
 ENERGY| Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036
 ENERGY| Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036
 ENERGY| Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036
 ENERGY| Total FORCE_EVAL ( FIST ) energy (a.u.):
-8.639417626820036

but there isn't a problem.
(SAVE_MEM has same problem...)

> or just starting from a geometry that has been annihilated with MD..
>
> Anyway the very large forces may be due to your input file.
> There are quite several things wrong in it:
>
> 1) Important: CONSTRAINTS at the moment DO NOT work with the geometry
> optimizer..
> 2)
 Ok.. Can I use fixed atom for fixed backbone? It seems to work .

> You're using just 1 spline.. so the spline generation cannot be slow
> (as you said).. If you use the full set of spline than.. yes.. it
> could be slow...

It is TRUE...

> 3) You're not using any kind of electrostatic (EWALD_TYPE NONE)!! is
> this really what you want??????????
>       Periodic NONE is not supported for MM run..  all MM run can be
> performed only in full periodicity (at the moment)...

A this moment I want only real part...
EWALD NONE=standard real-space coulomb potential is computed together
with the non-bonded contributions.
am I right?

There are a lot of knobs to be adjusted!!!!!

Thanks




More information about the CP2K-user mailing list