[CP2K-user] [CP2K:13100] SCAN XC density functional: Do we have to use specific PSEUDOPOTENTIALS for SCAN? Are they available?
Matt W
mattwa... at gmail.com
Tue Apr 14 21:41:12 UTC 2020
There could be two factors to the reported huge cutoff for the MgCl2 salt:
(i) Mg is nearly pathological, you need a huge cutoff with unsmoothed GPW
scheme to get the 2s semi-core states reasonalbly (so GAPW is much better)
(ii) SCAN is known to be 'difficult' in some sense, needed much tighter
settings (meta GGA). I am not sure if it is compatible with GAPW method?
For other sytems the cutoff should be less extreme.
Matt
On Tuesday, April 14, 2020 at 8:18:46 AM UTC+1, Marcella Iannuzzi wrote:
>
> Dear Slava,
>
> Have you tried GAPW. A much lower cutoff should be sufficient and it
> possibly improves the convergence.
> Kind regards,
> Marcella
>
>
> On Monday, April 13, 2020 at 7:55:08 PM UTC+2, Vyacheslav Bryantsev wrote:
>>
>> Dear Juerg and All,
>>
>> Here I provide an update for a large band gap system, consisting of Mg2+
>> and Cl- ions (molten salt).
>> Very high CUTOFF 2300 and REL_CUTOFF 100 were necessary to get converged
>> SCF results with SCAN-D3 XC, suitable for MD.
>> Using lower cutoffs, such as CUTOFF 2000 and REL_CUTOFF 100 were not
>> sufficient, because it required more > 100 SCF steps for some fraction of
>> MD steps, which is not acceptable. Because of that, the average time for
>> AIMD was higher than for the case with CUTOFF 2300.
>>
>> The portion of used input file is shown below. The average time for each
>> SCF step is 5-6 times slower compared to PBE-D3.
>> Is this the expected behavior? Are there any ways I can explore to speed
>> up SCAN-D3 calculations, used in the context of AIMD.
>>
>> Thank you,
>> Slava
>>
>> &MGRID
>>
>> CUTOFF 2300
>>
>> REL_CUTOFF 100
>>
>> ! NGRIDS 4
>>
>> &END MGRID
>>
>> &SCF
>>
>> MAX_SCF 150
>>
>> EPS_SCF 3.5E-6
>>
>> SCF_GUESS RESTART
>>
>> &OUTER_SCF
>>
>> EPS_SCF 3.5E-6
>>
>> MAX_SCF 40
>>
>> &END
>>
>> &OT T
>>
>> MINIMIZER CG
>>
>> PRECONDITIONER FULL_ALL
>>
>> &END OT
>>
>> &END SCF
>>
>> &XC
>>
>> &XC_FUNCTIONAL
>>
>> &LIBXC
>>
>> FUNCTIONAL MGGA_X_SCAN
>>
>> &END LIBXC
>>
>> &LIBXC
>>
>> FUNCTIONAL MGGA_C_SCAN
>>
>> &END LIBXC
>>
>> &END XC_FUNCTIONAL
>>
>> &vdW_POTENTIAL
>>
>> DISPERSION_FUNCTIONAL PAIR_POTENTIAL
>>
>> &PAIR_POTENTIAL
>>
>> R_CUTOFF 40.0
>>
>> TYPE DFTD3
>>
>> D3_SCALING 1.0 1.324 0.0
>>
>> PARAMETER_FILE_NAME dftd3.dat
>>
>> ! REFERENCE_FUNCTIONAL SCAN
>>
>> &END PAIR_POTENTIAL
>>
>> &END vdW_POTENTIAL
>>
>> &END XC
>>
>> &END DFT
>>
>>
>> &SUBSYS
>>
>> &CELL
>>
>> ABC 24.450 24.450 24.450
>>
>> PERIODIC XYZ
>>
>> &END CELL
>>
>> &TOPOLOGY
>>
>> COORD_FILE_FORMAT XYZ
>>
>> COORD_FILE_NAME last_frame_wrapped.xyz
>>
>> &END TOPOLOGY
>>
>> &KIND Mg
>>
>> BASIS_SET DZVP-MOLOPT-SR-GTH
>>
>> POTENTIAL GTH-SCAN-q10
>>
>> &END KIND
>>
>> &KIND Cl
>>
>> BASIS_SET DZVP-MOLOPT-SR-GTH
>>
>> POTENTIAL GTH-SCAN-q7
>>
>> &END KIND
>>
>> &END SUBSYS
>>
>>
>>
>> On Monday, April 13, 2020 at 10:09:19 AM UTC-4, Vyacheslav Bryantsev
>> wrote:
>>>
>>> Dear Juerg and All,
>>>
>>> Thank you for recommendations.
>>> With regards to very high cutoffs, what should I try?
>>>
>>> I startred with CUTOFF 600 and REL_CUTOFF 100, and this does not seem
>>> to work.
>>> Which setting would you recommends for SCAN XC?
>>>
>>> Thank you,
>>> Slava
>>>
>>>
>>> &MGRID
>>>
>>> CUTOFF 600
>>>
>>> REL_CUTOFF 100
>>>
>>> &END MGRID
>>>
>>> &SCF
>>>
>>> MAX_SCF 150
>>>
>>> EPS_SCF 1.0E-6
>>>
>>> SCF_GUESS RESTART
>>>
>>> &OUTER_SCF
>>>
>>> EPS_SCF 1.0E-6
>>>
>>> MAX_SCF 40
>>>
>>> &END
>>>
>>> &OT T
>>>
>>> MINIMIZER CG
>>>
>>> PRECONDITIONER FULL_ALL
>>>
>>> &END OT
>>>
>>>
>>>
>>> On Friday, April 10, 2020 at 6:31:42 AM UTC-4, jgh wrote:
>>>>
>>>> Hi
>>>>
>>>> 1) Many people have used PBE pp previously.
>>>> SCAN optimized pp can be found at
>>>> https://github.com/juerghutter/GTH/blob/master/SCAN/POTENTIAL
>>>> 2) Newly published SCAN parameters for D3 are available in the current
>>>> version of CP2K from Github. For older version you need to
>>>> add a line in the input with the parameters.
>>>> 3) Use very high cutoffs, depending on your system. If the cutoff is
>>>> not
>>>> high enough SCF will not converge smoothly to a low value (10^-7 in
>>>> OT).
>>>> I don't have experience if the smoothing methods work to reduce the
>>>> cutoff.
>>>>
>>>> regards
>>>>
>>>> Juerg Hutter
>>>> --------------------------------------------------------------
>>>> Juerg Hutter Phone : ++41 44 635 4491
>>>> Institut für Chemie C FAX : ++41 44 635 6838
>>>> Universität Zürich E-mail: h... at chem.uzh.ch
>>>> Winterthurerstrasse 190
>>>> CH-8057 Zürich, Switzerland
>>>> ---------------------------------------------------------------
>>>>
>>>> -----c... at googlegroups.com wrote: -----
>>>> To: "cp2k" <c... at googlegroups.com>
>>>> From: "Vyacheslav Bryantsev"
>>>> Sent by: c... at googlegroups.com
>>>> Date: 04/09/2020 04:50PM
>>>> Subject: [CP2K:13100] SCAN XC density functional: Do we have to use
>>>> specific PSEUDOPOTENTIALS for SCAN? Are they available?
>>>>
>>>> Dear CP2K Community,
>>>>
>>>> Is there a set or recommended setting for using SCAN in cp2k?
>>>>
>>>> More specifically,
>>>> 1. Can we use PBE pseudopotentials for SCAN or this is a bad idea? If
>>>> not, where one can find specific pseudopotentials reparametrized for SCAN?
>>>> 2. Does the D3 correction work automatically now with SCAN. If not, how
>>>> to specify it?
>>>> 3. Recommendations for grid when using SCAN
>>>>
>>>> Thank you,
>>>> Slava
>>>>
>>>>
>>>>
>>>> --
>>>> 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 c... at googlegroups.com.
>>>> To view this discussion on the web visit
>>>> https://groups.google.com/d/msgid/cp2k/43781524-ffdc-45d0-8899-258ace32638e%40googlegroups.com.
>>>>
>>>>
>>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200414/c8482e8a/attachment.htm>
More information about the CP2K-user
mailing list