[CP2K-user] [CP2K:15078] OT/Supercell vs. Kpoints: Abnormal pressure difference

[Lucas Lodeiro]

Hi,

without a full input and output is difficult to say more. But, I insist
with the smearing + UKS... maybe the system is creating an abnormal
magnetic state. Try with RKS + kpoints and without smearing. I do not
believe that PW programs make a huge difference with CP2K. If you want to
try, CP2K has the SIRIUS package, which computes electronic structure as
Quantum ESPRESSO. I tried it later, and gave pretty similar results with QE
and VASP.

Regards

El vie, 2 abr 2021 a las 23:26, Nicholas Winner (<nwi... at berkeley.edu>)
escribió:

> Hi Lucas, good points you made.
>
> The biggest thing that surprised me was not that 250 gamma point was not
> equivalent to 13x13x13 kpoint calculation on a primitive cell. I expected
> some difference. The biggest surprise was that the OT calculation converged
> to something very similar to Materials Project, while the kpoint
> calculation converged to a complete different polymorph. The Materials
> Project structure was calculated with kpoint grid and high plane wave basis
> cutoff, which confirmed Fm-3m is the ground state polymorph in agreement
> with experiments. While Materials Project's VASP calculations should not be
> equivalent to CP2K, with high quality basis/cutoffs/and the same kpoint
> grid, the results should be similar, as they both use good
> pseudopotentials, but it wasn't.
>
> That's why I thought there had to be something off about how I was using
> the kpoints in CP2K.
>
> Also I just forgot to turn UKS off. Probably wasted some core-hours, but
> won't make the results worse.
>
> Any further thoughts?
>
> -Nick
> On Friday, April 2, 2021 at 5:00:16 PM UTC-7 Lucas Lodeiro wrote:
>
>> Hi Nicholas,
>>
>> I am seeing some definitions which are not totally comparable. For
>> example, the primitive cell of Si has 2 atoms, and if you use a 13x13x13
>> kpoint grid, the comparable supercell using just the gamma point is a
>> 13x13x13 supercell containing 4394 atoms. As you mention, for gamma point +
>> supercell calculation, you use a cell with 250 atoms (5x5x5 supercell)
>> which is comparable with a primitive cell calculation with 5x5x5 kpoint
>> grid.
>>
>> Abot the difference using different basis sets, well is not weird, the
>> stress tensor calculation is highly dependent on the scf convergence and
>> the accuracy of the wavefunction.
>>
>> Another thing that caught my attention was the OT vs diagonalization
>> comparison. Both probably give a pretty similar solution, but no equal... I
>> would f¿prefer compare the primitive and supercell calculation using the
>> same wavefunction solver, especially if you use smearing, which is not
>> implemented for OT method, and could be a source of discrepancies (albeit
>> in a semiconductor this is not so important, if you are using a GGA or LDA
>> functional, the band gap is highly underestimated and the smearing affects
>> the calculation).
>>
>> Why are you using UKS instead of RKS?
>>
>> Regards - Lucas Lodeiro
>>
>> El vie, 2 abr 2021 a las 15:14, Nicholas Winner (<n... at berkeley.edu>)
>> escribió:
>>
>>> Hello all,
>>>
>>> I performed two cell optimizations on bulk Si, the first with a
>>> primitive cell and kpoints and the second using OT and a 250 atom
>>> supercell. Both structures started from the Materials Project (mp-149)
>>> lowest energy Fm-3m polymorph.
>>>
>>> I found that while OT only took 3 cell opt steps, the kpoint calculation
>>> took 10 steps, which I found a little strange.
>>>
>>> Upon investigation I found two things:
>>> (a) the kpoint calculation, had a large initial pressure deviation of
>>> -92570 bar, compared to OT, which had an initial deviation of  -708 bar at
>>> DZVP level and -2934 at TZV2PX level. Again these are the same structure,
>>> just kpoints vs supercell. It is a little odd to me that DZVP is closer to
>>> converged pressure, but I don't think its too strange.
>>> (b) the kpoint calculation relaxed from Fm-3m to Im-3m space group,
>>> which is not the ground state polymorph.
>>>
>>> My question is whether or not this issue is due to CP2K's kpoint
>>> implementation, or if this is something more fundamental, and how (if at
>>> all) it can be resolved. Below is the abridged DFT section for the kpoint
>>> calculation, which I assume is where the problem lies, but I can provide
>>> more info if needed.
>>>
>>> Best,
>>> Nick
>>>
>>> &DFT
>>>
>>>                 BASIS_SET_FILE_NAME BASIS_MOLOPT
>>>
>>>                 BASIS_SET_FILE_NAME BASIS_MOLOPT_UCL
>>>
>>>                 POTENTIAL_FILE_NAME GTH_POTENTIALS
>>>
>>>                 UKS True
>>>
>>>                 MULTIPLICITY 0
>>>
>>>                 CHARGE 0
>>>
>>>                 &QS
>>>
>>>                         METHOD GPW
>>>
>>>                         EPS_DEFAULT 1e-12
>>>
>>>                         EXTRAPOLATION PS
>>>
>>>                 &END QS
>>>
>>>                 &SCF
>>>
>>>                         MAX_SCF 100
>>>
>>>                         EPS_SCF 1e-07
>>>
>>>                         SCF_GUESS RESTART
>>>
>>>                         MAX_ITER_LUMO 400
>>>
>>>                         ADDED_MOS 100
>>>
>>>                         &DIAGONALIZATION
>>>
>>>                                 &DAVIDSON
>>>
>>>                                         PRECONDITIONER FULL_ALL
>>>
>>>                                 &END DAVIDSON
>>>
>>>                         &END DIAGONALIZATION
>>>
>>>                         &MIXING
>>>
>>>                                 METHOD BROYDEN_MIXING
>>>
>>>                                 ALPHA 0.2
>>>
>>>                                 NBUFFER 5
>>>
>>>                         &END MIXING
>>>
>>>                         &SMEAR
>>>
>>>                                 ELEC_TEMP 300
>>>
>>>                                 METHOD FERMI_DIRAC
>>>
>>>                                 FIXED_MAGNETIC_MOMENT -100.0
>>>
>>>                         &END SMEAR
>>>
>>>                 &END SCF
>>>
>>>                 &KPOINTS
>>>
>>>                         SCHEME MONKHORST-PACK 13 13 13
>>>
>>>                         EPS_GEO 1e-06
>>>
>>>                         FULL_GRID False
>>>
>>>                         PARALLEL_GROUP_SIZE -1
>>>
>>>                         SYMMETRY False
>>>
>>>                         UNITS B_VECTOR
>>>
>>>                         VERBOSE True
>>>
>>>                         WAVEFUNCTIONS COMPLEX
>>>
>>>                 &END KPOINTS
>>>
>>> &END DFT
>>>
>>> --
>>> 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 cp... at googlegroups.com.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/cp2k/b7c861a2-eb23-4fac-8a42-9846e53f6f24n%40googlegroups.com
>>> <https://groups.google.com/d/msgid/cp2k/b7c861a2-eb23-4fac-8a42-9846e53f6f24n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>> --
> 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 cp... at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/e071d68c-279e-4028-83cb-6d9f73662270n%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/e071d68c-279e-4028-83cb-6d9f73662270n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210403/1bd41ba8/attachment.htm>