[CP2K-user] RPA/MP2 in CP2K

[Vladimir Rybkin]

Dear Katarina,

I believe you are doing a good job in how you treat the exchange and number 
of quadrature points. A few points:
1) I would not use D3 together with RPA, as it is double counting: RPA is 
responsible for intermolecular interactions. I see that you don't actually 
do it, which is good.
2) You may actually reduce CUTOFF      900  in the WFC_CORRELATION section 
to speed-up you calculation and without much effect on the accuracy.
3) If 40 quadrature points the largest value? With 50-60 it should converge.
4) SCREEN_ON_INITIAL_P affects efficiency, not the result.
5) EPS_DEFAULT are sensitive parameters EPS_SCHWARZ (but you already 
checked them). They should be sufficiently small.
6) What you have done is actually RPA with exchange *E*totEXX/RPA (thus HF 
section in WFC section). This is an accurate method, but I am not sure 
whether you computed the same in VASP. 
7) You .xyz file contains only water molecule, so I can't say much about 
the system. You may try to increase the z dimension of the cell to make 
more empty space between you periodic images.
8) Did VASP employ k-point sampling? CP2K is Gamma-point only.
9) Yes, it may still be basis set incompleteness responsible for the 
difference. But there are options to check (see above).

Yours,

Vladimir

пятница, 9 августа 2019 г., 12:03:19 UTC+2 пользователь katarína 
stančiaková написал:
>
> Dear Vladimir,
> thank you for your reply. I applied BSSE correction and for my system I 
> got BSSE error around 4kJ/mol.  The Eads I got from CP2K is around -88 
> kJ/mol while the VASP data that I have available predicts Eads for a water 
> = -74 kJ/mol. As I do not have much experience with RPA, I use the VASP 
> data as a reference to "tune" my CP2K setup. Please, do you have some 
> suggestion what can influence the accuracy (or to be the source of the 
> discrepancy)? As an input I use PBE+d3 (I tried also PBE only) wavefunction 
> optimized with the same basis set and I've already tested the following 
> criteria:
> EPS_DEFAULT;
> EPS_SCHWARZ;
> RPA_NUM_QUAD_POINTS;
> SCREEN_ON_INITIAL_P;
> CUTOFF RADIUS for the Coulomb potential (I use max. value, i.e. 1/2 of the 
> unit cell)
>
> Attached are my input files for both PBE-D3 pre-optimization as well for 
> RPA.
>
> Any help is really apreciated!
> Thank you
> Kind regards,
> Katarina
>
>
> Dňa utorok, 6. augusta 2019 14:51:39 UTC+2 Vladimir Rybkin napísal(-a):
>>
>> Dear Katarina,
>>
>> cc-TZ are correlation-consistent triple-zera quality basis sets. 
>> Normally, they are OK, however, sometimes one needs to go for the complete 
>> basis set extrapolation, e.g. for cohesion energies. In data/BASIS_RI_cc-TZ 
>> there are cc-TZ basis sets for O, H, Si and Al. So, I would try these 
>> first. You may also (and actually should) do BSSE correction. Bear in mind 
>> that 4- and 5-zeta basis sets will make the computation much more 
>> expensive, so it may happen that you will not be able to run those 
>> calculations. In any case, if you really need the 4- and 5-zeta basis, 
>> please write again.
>>
>> Yours,
>>
>> Vladimir 
>>
>> понедельник, 29 июля 2019 г., 17:54:59 UTC+2 пользователь katarína 
>> stančiaková написал:
>>>
>>> Dear all,
>>> I would like to perform RPA and MP2 calculations on a periodic system of 
>>> cca 40 atoms to get reference adsorption energy for small molecules in 
>>> alomunosillicates. In my system I have H, O, Al and Si atoms. Which basis 
>>> set would you recommend to use? I tried cc-TZ in combination with RI-TZ 
>>> (with GTH_POTENTIALS), however, I was wondering whether there is something 
>>> "better". I found a tutorial how to make better RI basis set, however, I 
>>> could not find bigger primary basis set. I don't have much experience with 
>>> these techniques, so any help would be very much appreciated!
>>>
>>> Thank you
>>> Kind regards,
>>> Katarina
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190809/454455dd/attachment.htm>