[CPMD-list] Energy components in PIMD

[Łukasz Walewski]

Dnia 2008-05-08, czw o godzinie 12:38 -0400, Axel Kohlmeyer pisze:
> On Mon, 5 May 2008, ukasz Walewski wrote:
> 
> dear lukasz,
> 
> LW> Dear Axel,
> LW> 
> LW> I just wanted to give you an update on my progress. I did the NVE runs
> LW> with USPP, CUTOFF=30 Ry and DUAL=8,10 and also with MT, CUTOFF=70 Ry and
> LW> DUAL=4. The EKINC has very similar drift as for previous runs, i.e.
> LW> USPP, CUTOFF=30 Ry and DUAL=4 - see
> LW> http://bioexploratorium.pl/ljw/cpmd/nvt/test1/EKINC.png and
> LW> http://bioexploratorium.pl/ljw/cpmd/nvt/test1/EKINC.agr
> 
> without seeing the corresponding development of the
> ion kinetic energy and the potential energy there is
> very little to say. from running your input briefly it

I did not write this but the rest of the plots (energy components) are
in the same directory, i.e.
http://bioexploratorium.pl/ljw/cpmd/nvt/test1/ 

> looks as if your choice of reference frequency for the
> ionic nose-hoover chains does not couple well to the
> prevalent modes in your system (how did you determine it?).

181 cm^-1 is one of the vibrational eigenmodes calculated with
VIBRATIONAL ANALYSIS option. It was proved to promote the proton
transfer (by making the N-N distance shorter) so I want to have this
mode excited.

> i did some runs of the input by myself (on a very old,
> soon to be decomissioned cluster, so it took a loooooooong 
> time). i reduced the time step and played with the GC-CUTOFF
> parameter and the thermostat coupling and my conclusion is 
> simply that with the ultra-soft pseudopotential support as 
> it is currently implemented in CPMD, you cannot fully conserve
> the fictitious kinetic energy. with your input i always get
> a drift of roughly 0.0005 Ha/ps in EKINC. with norm-conserving
> pseudopotentials, however, this does not happen and once
> i actually run across a large enough number of nodes the 
> NC-run is almost as fast as the USPP run...

I see. Then I wonder what is the origin of that, i.e. is it the
specificity of my system that makes USPPs unusable (and if yes, where
does it come from) ? How is it connected with the implementation in
CPMD ?

Could you please post your input that conserves the fictitious kinetic
energy ? On what number of processors did you run it ?

Regards,
Lukasz