[CP2K:1885] Re: Grimme correction and supercell

Teodoro Laino teodor... at gmail.com
Tue Mar 17 19:47:32 UTC 2009


Andrzej,

I asked my nephew (named Giovanni, almost 7) to multiply the following 
two numbers:

0.16665*5

after some time, extremely proud, he came back with the answer !! On a 
couple of sheets of paper he wrote

0.16665*5 = 0.83325
(of course with all details and the proper algebra..)

I had a quick look to your input files and well.. you wrote 0.8333..

You can't expect to get the same energy per atom if the coordinates are 
different.. (when you build
the NNN systems you have to handle EXACTLY the proper operations if you 
wanna get the right
numerics!!!)

Anyhow.. after correcting this issue the result is just PERFECT.
Thanks for making feel my nephew proud and useful to this world! :-)

Teo

Andrzej wrote:
> Dear Theo,
> thanks for answering my questions.
>
> I have uploaded gra_grim.tar file with inputs and outputs.
>
> The last one (442) is calculated with revpbe, but I think it doesn't
> matter, as it uses the same parametrization for Grimmie correction as
> pbe.
>
> Andrzej
>
> On Mar 16, 5:13 pm, Teodoro Laino <teodor... at gmail.com> wrote:
>   
>> The neighbor list is always built across all possible periodic images
>> included within the CUTOFF specified
>> for the dispersion (as long as you don't have PERIODIC NONE in the &CELL
>> section! this case would
>> explain the problem then..)
>> .
>> The problem must be somewhere else..
>> Could you please post a simple input file for both cases (111 and 422) ?
>> Thanks
>> Teo
>>
>> Axel wrote:
>>
>>     
>>> On Mar 16, 12:54 pm, Andrzej <kwa... at o2.pl> wrote:
>>>       
>>>> Dear users,
>>>>         
>>>> I performed some calculation for graphite, just for testing some
>>>> features.
>>>> Here are some results obtained from single point calculations made for
>>>> experimenrtal crystal structure under periodic conditions.
>>>> Calculations were performed for single unit cell (111), and extended
>>>> unit cells: 222, 332 and 442.
>>>>         
>>>> Here are values of dispersion (Grimme) energy per C atom.
>>>>         
>>>> 111: -0.004719
>>>> 222: -0.0041627
>>>> 332: -0.00405803
>>>> 442: -0.00405803
>>>>         
>>>> As the calculations were performed under periodic conditions I would
>>>> expected that all these values should be the same. Grimme correction
>>>> is purely classical expression independent of electron density,
>>>> cutoff,  lack of k-point sampling and so on.
>>>>         
>>> but it is not independent of minimum image convention and
>>> the classical cutoff. i haven't checked the cp2k implementation,
>>> but the equivalent code in cpmd would provide an option to
>>> compute the real space interactions across multiple copies of
>>> the principle unit cell, so that for small unit cells, you
>>> would still be able to compute those interactions up to a
>>> significant distance beyond the restrictions of using only
>>> the closest image.
>>>       
>>> cheers,
>>>    axel.
>>>       
>>>> I will be grateful for any comment.
>>>>         
>>>> Andrzej
>>>>         
> >
>   




More information about the CP2K-user mailing list