[CP2K:143] Re: GAPW and GTH

[Matthias Krack]

On the basis of my experience an EPS_FIT value of 0.01 is by far too large 
to obtain reasonable forces even if energies might look ok. Basically, 
this is a (dirty) workaround for a long present bug in the GAPW 
implementation which pops up whenever no hard basis is present, ie. the 
full atomic basis is equivalent to the soft basis as already outlined by 
Juerg. The O basis is much harder than the Cl basis set and thus you could 
try smaller EPS_FIT values for water (even smaller than the default 
EPS_FIT value). Moreover, for GAPW calculation the atomic KIND parameters
RADIAL_GRID, LEBEDEV_GRID, MAX_RAD_LOCAL might have an impact on the
accuracy (rotational invariance) of your results.

Matthias

> Sorry, it took me a while to test things, but EPSFIT at 10^-2 seems to
> work really well. Thanks a lot. :)
> 
> Though while testing I noticed something 'odd':
> If I use H2O instead of HCl (and essentially the same input as above),
> the results also look good, the grid effect is quite a bit smaller
> than the
> default EPS_FIT, MD seems stable, but if I just do a energy_force
> calculation of the following configuration:
> 
> O 0.0 0.0 0.0
> H 0.0 0.59124 -0.77291
> H 0.0 0.59124 0.77291
> 
> I get a strange force on the oxygen in z direction:
> 
> ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):
> -17.211597494294843
> FORCES| (a.u.)
>   O         0.0000000127       -0.0024393452        0.0001621868
>   H        -0.0000000043        0.0012914592        0.0010373482
>   H         0.0000000041        0.0011673908       -0.0011995375
> 
> Now, the force isn't that large, but if I simply switch axis (from z
> to x) to:
> 
> O 0.0 0.0 0.0
> H -0.77291 0.59124 0.0
> H 0.77291 0.59124  0.0
> 
> I get basically the same energy, and the force on oxygen looks much
> better as
> one would expect from symmetry:
> 
> ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):
> -17.211597494294828
> FORCES| (a.u.)
>   O        -0.0000000021       -0.0024393452        0.0000000127
>   H         0.0011184426        0.0012294258       -0.0000000043
>   H        -0.0011184430        0.0012294243        0.0000000041
> 
> How come the different axis give such different results? Is it really
> just the ordering
> of the loops over axis in the code? I wouldn't have thought the impact
> quite that
> large from -2.1e9 to 1.6e-4....
> Or is this an unlucky extreme example (due to symmetry)?
> 
> BTW the only force term that is really different in both calculations
> is the vhxc_atom part and
> for default EPS_FIT the difference is far less pronounced.
> 
> -H
> 
> 
> > 
> 

Dr. Matthias Krack
Computational Science
Department of Chemistry and Applied Biosciences
ETH Zurich
USI-Campus, via G. Buffi 13
6900 Lugano
Switzerland

Phone: +41 (0)58 666 48 05 (direct)
       +41 (0)58 666 48 00 (secretary)
Fax:   +41 (0)58 666 48 17

Email: kr... at phys.chem.ethz.ch
URL:   http://www.phys.chem.ethz.ch