[CP2K-user] [CP2K:22398] Re: Problems with cell optimization of graphene

[Frederick Stein]

Dear Riccardo,

Usually, I restart the calculation from a preconverged one, usually PBE. 
and point CP2K to the reference wfn-file using WFN_RESTART_FILE_NAME <name 
of the restart file>. If that one does not converge, I perform a 
PBE0-calculation starting from PBE and restart from PBE0.

A few more notes on your input file:

- EPS_PGF_ORB is too tight (just comment it out to use the default)

- make sure to converge EPS_DEFAULT

- make sure to use a suitable pseudopotential (PBE or GGA for PBE, PBE0 or 
HYB for PBE0 and HF for HF)

- Why are you not applying an exchange correction in ADMM? Just use the 
default (omit EXCH_CORRECTION_FUNC)

- As already mentioned by Augustin, your cutoff for HF might be too low. 
this introduces artifacts. Can you extend your system in the x- and 
y-directions? You need at least 5-6 A for converged results.
- start with a smaller ADMM basis set (AUX_FIT) of dz or dzp-quality. The 
one you are using will not introduce a siginificant acceleration.
- the RI-basis set (RI_AUX) was not optimized for your primary basis set 
and also not tested with it. You may remove them and set AUTO_BASIS RI_AUX 
[SMALL|MEDIUM|LARGE|HUGE] (in the DFT-section) instead.
- Are you sure that you need such a high cutoff in the WF_CORRELATION 
section ? I commonly apply 200 Ry. I have never seen value larger than 300 
Ry in this method.
Best,
Frederick

Riccardo Pezzetta schrieb am Montag, 17. August 2026 um 10:59:10 UTC+2:

> Dear Professor Stein,
>
> Thank you very much for your reply.
>
> I checked the convergence with respect to the smearing temperature. I 
> found that the energy is very similar for electronic temperatures of 100, 
> 200, and 300 K, while a noticeable change starts to appear at 500 K. 
> However, if I completely remove the smearing, the PBE SCF calculation does 
> not converge.
>
> I also checked the convergence with respect to the size of the cell along 
> the z direction.
>
> Following your suggestion, I added the following section to print the 
> orbital occupations:
>
> &PRINT
> &MO
> OCCUPATION_NUMBERS
> &END MO
> &END PRINT
>
> I found that the occupation numbers are either 2 or 0 for the different 
> lattice constants considered, with no partially occupied orbitals.
>
> Finally, I tried to perform a pure HF calculation to investigate whether 
> the problem originates entirely from the Hartree-Fock exchange term. 
> However, the HF calculation does not converge, even when using a relatively 
> high electronic temperature or when mixing HF exchange with PBE exchange or 
> when using PERIODIC XYZ instead of PERIODIC XY.
>
> Do you have any suggestions on what I could try next?
>
> Thank you again for your help.
>
> Best regards,
>
> Riccardo Pezzetta
>
> Il giorno domenica 2 agosto 2026 alle 17:26:34 UTC+2 Frederick Stein ha 
> scritto:
>
>> Dear Riccardo,
>> You are using smearing. Can you turn it off? Are all orbitals fully 
>> occupied? How does the band-gap change with cell volume? Beware that our 
>> RPA-implementation does NOT consider partial occupation. Even worse, 
>> partially occupied orbitals are considered as fully occupied by our 
>> implementation such that the energies may even differ with the available 
>> orbitals.
>> In addition, I am not entirely sure how well the 2D-periodicity is 
>> covered in your executable of CP2K and I would rather use PERIODIC XYZ 
>> instead.
>> Best,
>> Frederick
>>
>> Riccardo Pezzetta schrieb am Mittwoch, 29. Juli 2026 um 14:12:13 UTC+2:
>>
>>> Thank you very much for the answer.
>>> Kind regards,
>>>
>>> Riccardo Pezzetta 
>>>
>>> Il giorno mercoledì 29 luglio 2026 alle 11:15:38 UTC+2 Augustin Bussy ha 
>>> scritto:
>>>
>>>> Dear Riccardo,
>>>>
>>>> your HFX input is not ideal, because a 4 Angstrom cutoff is probably 
>>>> too short. There are 2 things to consider when selecting a value for the 
>>>> cutoff:
>>>> 1) is should not exceed half the cell size (OK in your case)
>>>> 2) it should be long enough. In most cases, a value of ~6 Ang is deemed 
>>>> sufficient, but technically, the ideal cutoff depends on the system band 
>>>> gap (for smaller gap systems, a longer cutoff might be required)
>>>>
>>>> I suggest you take a supercell, and extend the cutoff.
>>>>
>>>> Note that the value of EPS_PGF_ORB you use is way too small (1.0E-16), 
>>>> and it will dramatically impact performance. Please read section 2.2 of 
>>>> https://pubs.acs.org/jpcbfk/article/130/4/1237/5088250/The-CP2K-Program-Package-Made-Simple 
>>>> for an explanation.
>>>>
>>>> If using a supercell makes the RPA part of your calculation too 
>>>> expensive (RPA scales as N^4 with system size), consider using the 
>>>> low-scaling implementation; it is particularly efficient for 2D systems.
>>>>
>>>> Best,
>>>> Augustin
>>>>
>>>> On Wednesday, 22 July 2026 at 14:59:37 UTC+2 Riccardo Pezzetta wrote:
>>>>
>>>>> Dear CP2K community,
>>>>> I am trying to optimize the lattice constant of a graphene cell with 
>>>>> RPA at the Gamma point. I performed energy calculations with different 
>>>>> lattice constants, but I get a very strange behavior close to the minimum. 
>>>>> As shown in the image attached to this mail, the RPA energy shows an 
>>>>> unphysical drop close to the minimum. With PBE the curve is regular, while 
>>>>> the HF at PBE energy (i.e., RPA without the correlation term) shows the same 
>>>>> irregular profile as RPA, so the issue seems to originate from the 
>>>>> Hartree-Fock exchange term. In my input, that you find attached to the 
>>>>> mail, the only parameter I set in the HF section is the cutoff of the 
>>>>> truncated Coulomb operator, I set it equal to 4 Angstrom so that it does 
>>>>> not exceed half the cell size for all values of the lattice constant 
>>>>> considered. Have you ever encountered a similar problem? What can be the 
>>>>> cause of this unphysical behavior? Thank you in advance for your answer.
>>>>> Kind regards,
>>>>>
>>>>> Riccardo Pezzetta
>>>>>
>>>>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/b483482d-3fbf-4b92-93cf-20ce99b94e00n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260817/f74c3d5a/attachment-0001.htm>