[CP2K:1246] Re: hybrid DFT for bulk

Juerg Hutter hut... at pci.uzh.ch
Mon Jul 28 14:09:06 CEST 2008


Hi
>
> Your are absolutely right. The correct result should be something
> slightly below -4 a.u. because the binding energy is small.
> Actually, I just did a g03 calculation using
> 2916 k points, and the result becomes -4.564a.u. which is very
> close to the cp2k result, -4.426a.u., with \Gamma only! Is this
> surprising?

No. Currently CP2K calculates integrals using 1/r but truncates
the infinite sum of integrals. As mentioned in an earlier post this
can lead to instabilities for small unit cells. However, at what
size of unit cell this approach is problematic, cannot be said
beforehand. That this approach works has been shown by
Challacomb et al. and also in our recent paper in JCP.
Still we think that one can do better, and we are working on that.


> Is the more reasonable cp2k result due to the error cancellation
> between the lack of long-range exact exchange and the \Gamma-only
> k-sampling? or some special techniques are applied in cp2k?

No. I wouldn't call this error cancelation. We are making several
well defined approximations. The HFX term is rather small and
the difference between your converged (?) k-point sampling and
the cp2k gamma point value is still large.

Another comment. I assume that you know how to correctly use
the HFX in cp2k. You have certainly checked that you correctly
understand all the keywords by for example testing on small
isolated molecules.

regards

Juerg


>
> regards,
> molecule
>
>
> On Jul 24, 6:08 pm, Juerg Hutter <hut... at pci.uzh.ch> wrote:
>> Hi
>>
>> just a few additional remarks.
>> You are aware that your system is pathological? It most likely
>> has a degenerate ground state (for non spin polarized) or
>> no gap for spin polarized calculations.
>> A quick sanity check for the total energy gives
>> 8 * -0.5 = -4 a.u.
>> Looks like cp2k is less wrong than G03, if something like that
>> exists.
>>
>> Try an infinite chain of He atoms or H2 molecules. There is a
>> small chance that you will get reasonable numbers.
>>
>> regards
>>
>> Juerg
>>
>> ----------------------------------------------------------
>> Juerg Hutter                   Phone : ++41 44 635 4491
>> Physical Chemistry Institute   FAX   : ++41 44 635 6838
>> University of Zurich           E-mail: hut... at pci.uzh.ch
>> Winterthurerstrasse 190
>> CH-8057 Zurich, Switzerland
>> ----------------------------------------------------------
>>
>> On Thu, 24 Jul 2008, molecule wrote:
>>
>>> Hi, Teodoro
>>
>>> I don't want to argue about the wording.
>>> I just did comparative calculations
>>> using cp2k and g03 for an infinite H chain (1A of separation).
>>> The settings are the same for both calculations:
>>> The unit cell contains 8 H atoms
>>> and its dimension is 8A x 8A x 8A.
>>> The basis set is 6-31G* and 4 shells are used for the periodicity.
>>> xc=b3lyp, k-sampling=\Gamma
>>
>>> E(g03_PBC)= -27.6442625829 A.U.
>>> E(cp2k_PBC)= -4.426086805926252 A.U.
>>
>>> Any comment on this?
>>
>>> cheers,
>>> molecule
>>
>>> On Jul 18, 9:28 pm, Teodoro Laino <teodor... at gmail.com> wrote:
>>>> Dear molecule,
>>
>>>> If you consider the gamma point sampling a numerical error.. well..
>>>> guess there's not too much one can comment.. but be cautious to do
>>>> such claims in a more aggressive audience!
>>
>>>> Moreover, keep in mind that Juerg also mentioned the 1/r problem for
>>>> the exact exchange.. and since your cell is small.. I wonder.. do you
>>>> consider
>>>> that a numerical problem as well?
>>>> Maybe you think that the whole story of long-range interactions is..
>>>> a numerical error? Eager to know your point of view..
>>
>>>> cheers,
>>>> Lost_somewhere_with_more_than_6_hours_delay.
>>
>>>> On 18 Jul 2008, at 23:09, molecule wrote:
>>
>>>>> However, the behavior looks more like a bug rather than the
>>>>> numerical errors caused by \Gamma-only sampling and the
>>>>> minimum images.
> >
>



More information about the CP2K-user mailing list