[CP2K-user] Having trouble getting SCF convergence for Cell_Opt of bulk Ni
Travis
polla... at gmail.com
Wed Jan 27 19:43:10 UTC 2021
Hi,
I assume you'll add a KPOINTS section in the future, the most obvious
improvement is to use a higher electronic temperature. Probably the most
commonly used figure in literature is 0.2 eV which is ~2300 K, you may of
course be able to use a lower one. You might also see some improvement from
a smaller alpha (0.01 - 0.1), these settings can require a bit of
experimentation to optimize. I prefer Broyden mixing over Pulay - but I
sometimes get jobs that hang when using Broyden for cell optimizations and
switch to Pulay or Kerker instead. My 'preference' is based solely on
Broyden tending to require fewer SCF cycles and being more consistent, it's
not a large improvement but adds up over 1000's of MD steps.
Marcella recommended this recently for a calculation on bulk Au (inside the
&SCF section),
ADDED_MOS 200
&DIAGONALIZATION T
ALGORITHM STANDARD
&END DIAGONALIZATION
&MIXING T
METHOD BROYDEN_MIXING
ALPHA 0.01
BETA 0.5
NBUFFER 8
&END MIXING
&SMEAR
METHOD FERMI_DIRAC
ELECTRONIC_TEMPERATURE 500
&END SMEAR
Another recommendation from Matthias for bulk Cu, includes an example
KPOINT section which you can adapt to your calculation,
&DFT
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME GTH_POTENTIALS
&KPOINTS
SCHEME MONKHORST-PACK 2 2 2
FULL_GRID yes
SYMMETRY yes
VERBOSE yes
PARALLEL_GROUP_SIZE -1
&END KPOINTS
&MGRID
NGRIDS 5
CUTOFF 400.0
REL_CUTOFF 60.0
&END MGRID
&QS
EPS_DEFAULT 1.0E-12
EXTRAPOLATION use_prev_p
&END QS
&SCF
ADDED_MOS 60
EPS_SCF 1.0E-8
MAX_SCF 300
SCF_GUESS restart
&DIAGONALIZATION yes
ALGORITHM STANDARD
&END DIAGONALIZATION
&MIXING yes
ALPHA 0.4
BETA 1.0
METHOD broyden_mixing
NBROYDEN 8
&END MIXING
&SMEAR on
METHOD FERMI_DIRAC
ELECTRONIC_TEMPERATURE [K] 2000.0
&END SMEAR
-T
On Wednesday, January 27, 2021 at 1:21:41 PM UTC-5 GENG YUAN wrote:
> Hello everyone and hope everything goes well!
>
> I started learning CP2K recently and trying to run some DFT simulations. I
> am currently working on performing a cell optimization of bulk Ni, but I'm
> not getting an inner loop SCF convergence. I saw some similar problems in
> this discussion forum and modified my input file based on some suggestions
> from other discussions, such as adding smearing, changing the SCF
> convergence criterion, etc. Nothing has worked so far. Moreover, I thought
> my simulations won't take quite long as each of them only has 32 Ni atoms,
> but none of them finished within 30 minutes (I set up the total wall time
> for 30 minutes, so all simulations will be terminated if they didn't finish
> in 30 mins).
>
> I attached both my input file and xyz file here for your reference, my
> bulk Ni model contains 2X2X2 unit cell (the lattice constant I use for Ni
> in my work is 3.52 Angstrom, the structure of my Ni is FCC), therefore I
> made #ABC 7.04 7.04 7.04 in my input file under &cell subsection. Any
> suggestions on how to get convergence are welcome!
>
> Many thanks in advance,
> Sincerely,
> Geng Yuan
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210127/fb1258e3/attachment.htm>
More information about the CP2K-user
mailing list