NpT - grid GPW

marc marcvanh... at gmail.com
Sun Jun 24 18:21:00 CEST 2012


Hi Matt,
thanks a lot for the reply, your information is worth a million.  I read 
some papers already that discussed the dependence of the simulation on the 
cut off or the basis set, but they were all about vapor-liquid coexistence 
curves. I will look a bit more into this.

For now I'm going with the 800Ry cut-off, and with the constant gird 
density method.  You are right that proper testing should be done, but 
since the MD is ab initio the computations are very expensive (my systems 
are between 645 and 1161 atoms) and I want to waste a minimal amount of 
computer resources (and time).

cheers!
marc

On Saturday, June 23, 2012 8:10:39 AM UTC+2, Matt McGrath wrote:
>
> Hi Marc.
>
> The default method is to use a constant grid density, using grid sizes 
> that correspond to the "magic" numbers for FFTs.  You can change this by 
> putting in the CELL_REF flag, e.g
>
>     &CELL
>       ABC 17.28  17.28  17.28
>       &CELL_REF
>         ABC 18.0 18.0 18.0
>       &END CELL_REF
>     &END CELL
>
> Now the cell will always have the same number of grid points as a cell of 
> size 18x18x18 using the default grid density, regardless of the actual size 
> of the cell.
>
> You probably don't have to use a cutoff of 1200 Ry.  In later studies, we 
> use a cutoff of 600 Ry (you can look up the papers by McGrath, Kuo, and 
> Siepmann), and it seems to work.  Of course, you should also do some tests 
> for your system.  We reported I think in at least one paper that increasing 
> the cutoff (or, at least, using a more converged basis set) can actually 
> push the result away from the experimental properties.  This is a 
> well-known probably that arises from the lack of a variational principle in 
> DFT once a form for the exchange-correlation functional has been decided 
> on, though I caution you that I'm not an expert in this.  I wasn't an 
> author of the paper you cite, so I'm not sure why they chose 700.
>
> Good luck!
>
>                                                  Cheers, Matt
>
>
> On Friday, June 22, 2012 5:31:29 PM UTC+9, marc wrote:
>>
>> Dear all,
>>
>> I have a question concerning the GPW method, more specific about how the 
>> regular grid used in the computation of the electronic density, is 
>> generated.  In the following reference:
>> ChemPhysChem, 2005, 6, 1894
>> it is found that when using an (almost) constant density for the grid 
>> points, one should take the cut-off high enough (i.e. a 1200 Ry) to avoid 
>> discontinuities in the potential energy vs. volume curve.  However, when 
>> employing a constant number of grid points, by making use of the reference 
>> cell method, these jumps can be avoided (but in the paper the authors use 
>> also the very high 1200 Ry-cutoff in this method). 
>> Knowing this, I have two questions about this:
>> i) what is the 'default'-method the generate the grid in CP2K: constant 
>> density of points or constant number of points?
>> ii) if the default-method is to generate a constant number of points, is 
>> it then also necessary to take such a big cut-off of 1200 Ry? Or is it then 
>> good enough to go a bit smaller? In this paper: J. Phys. Chem. B, 2009, 
>> 113, 11959, the authors take 700 Ry for the reference cell method, but why 
>> exactly this value? It is 500 Ry less then the proposed 1200 Ry as it was 
>> proposed in the first reference...
>>
>> many thanks
>> marc
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20120624/f87f4796/attachment.html>


More information about the CP2K-user mailing list