other questions about the calculation error

Axel akoh... at gmail.com
Sun Oct 7 23:45:18 UTC 2007

dear chun-min,

before continuing trying to understand cp2k, i recommend
taking a piece of paper and write on it a hundred times:

cp2k is not gaussian.

as teo already mentioned, you first need to get a grasp
on the implications of the quickstep method wrt. to accuracy.
this is not always that obvious though. for examples,
whenever plane waves are used, you always compute
in periodic boundary conditions. 'turning off' PBC with a
poisson solver is actually done after the fact, i.e. you
compute everything in PBC and then use the poisson solver
to remove the impact of the periodic images. for any affordable
calculation and due to numerical limitations, there will always
be small changes when changing box sizes and grids.


On Oct 7, 6:11 pm, Teodoro Laino <teodor... at gmail.com> wrote:
> Chun-Min,
> please have a look at
> http://www.theochem.ruhr-uni-bochum.de/research/marx/marx.pdf
> (this manuscript is about CPMD but other similar publications about
> other PW codes are fine as well)
> to have an idea how g-vectors are treated/stored in PW calculations.
> Regarding the other question about what of your two results is correct:
> looking at your input file none of them is correct.
> The keyword PERIODIC:http://cp2k.berlios.de/input/
> controls the PBC of the atoms and  therefore also the construction of
> the neighbor lists used in QS/FIST or other part
> of the package, but not the periodicity of the electrostatics.
> The presence of this keyword alone does NOT decouple the coupling
> between periodic replica due to the
> treatment of the electrostatic with PW (have a look always at the
> manuscript of Hutter/Marx for details).
> The correct input looks like (showing only modified sections):
>       [..]
>      &MGRID
>        CUTOFF 250
>        # Use a larger cutoff
>      &END MGRID
>      [..]
>      &POISSON
>          PERIODIC NONE
>          # no periodicity for the electrostatic. Decoupling achieved
> through the Martyna-Tuckerman Scheme..
>      &END
>      # (POISSON inhttp://cp2k.berlios.de/input/
> InputReference~__ROOT__~FORCE_EVAL~DFT~POISSON.html)
>    &END DFT
>    [..]
> To be correct, the MT algorithm requires a computational box at least
> double the size of the dimensions of the molecule you
> want to simulate. If the box is smaller than this value the result
> will not be correct.
> Teo

More information about the CP2K-user mailing list