[CP2K-user] Co(111) Geometry optimization with kpoints

[Manuel Corral]

Dear Fabien and Marcella

Thanks to both of you for your replies. 

Yes, my slab is small indeed. At this very moment I am just learning to use 
the code for this kind of systems and benchmarking results against to what 
I am familiar to. In a production run I will use a significantly bigger 
cell. But I was planning to do that next, after I manage to fine tune the 
SCF convergence with smaller/faster systems.

Best regards,
Manuel 

Le mardi 19 janvier 2021 à 15:25:05 UTC+1, Marcella Iannuzzi a écrit :

>
> Dear Manuel, 
>
> You can avoid the Wavelet Poisson solver and use the surface dipole 
> correction instead
>
>      SURFACE_DIPOLE_CORRECTION
>
>       SURF_DIP_DIR Z
>
> When the optimisation method is the diagonalisation, the ASPC is not a 
> good extrapolation method.
>
> For metals and G-space mixing, the best choice is probably USE_PREV_WF
>
> Running MD for such a small cell using k-points sounds a little bit weird 
> to me. 
>
>
> Kind regards
>
> Marcella
>
>
> On Tuesday, January 19, 2021 at 2:14:42 PM UTC+1 Manuel Corral wrote:
>
>> Dear Fabian,
>>
>> Thank you very much for the commented file. It works, which it is already 
>> a huge improvement, though it may take about 75 electronic iterations for 
>> each ionic step, which may render a MD calculation too expensive. I wonder 
>> if there is anything that could be done to speed it up.
>>
>> By the way, why do  you need to rotate the cell for the Wavelet Poisson 
>> solver ? Based on this rotation, I would have thought of changing the 
>> Monkhorst Pack set of k-points from 7x7x1 to 7x1x7, which you didn't and, 
>> when I tried it, it did not work properly:
>>
>>  *** WARNING in pw/ps_wavelet_methods.F:236 :: Density non-zero on the ***
>>  *** edges of the unit cell: wrong results in WAVELET solver.
>>
>> Thanks again.
>>
>> Regards,
>> Manuel
>>
>> Le lundi 18 janvier 2021 à 18:03:26 UTC+1, fa... at gmail.com a écrit :
>>
>>> Dear Manuel,
>>>
>>> I changed a couple of things in the input and it's converging now 
>>> (attached below). The changes include:
>>> To use the WAVELET poisson solver I rotated the slab such that B is the 
>>> surface vector. 
>>> I also decreased RELAX_MULTIPLICITY because large values hinder scf 
>>> onvergence. Maybe you need to test various initial guesses (see here: 
>>> https://manual.cp2k.org/cp2k-8_1-branch/CP2K_INPUT/FORCE_EVAL/SUBSYS/KIND/BS.html) 
>>> to find the minimal energy configuration.
>>> I increased the ELECTRONIC_TEMPERATURE in the &SEMEAR section. 300 K 
>>> corresponds to just 0.02 eV in VASP which is very little.
>>>
>>> Cheers,
>>> Fabian
>>>
>>> On Monday, 18 January 2021 at 14:53:36 UTC+1 Manuel Corral wrote:
>>>
>>>> Dear Colleagues,
>>>>
>>>> I have been trying to run a geometry optimization for with a Co(111) 
>>>> 4-layer slab model without success. 
>>>>
>>>> I set up my initial input file according my previous experience with 
>>>> another software (VASP) and then changed different calculation settings: 
>>>> number of additional MOS (10, 30, 100), electronic temperature of the 
>>>> smearing (300 K, 500 K), decreasing the cell in the vacuum direction, 
>>>> changing the number of kpoints (gamma point with an increased cell), the 
>>>> Poisson solver (analytical vs non analytical), the diagonalization 
>>>> algorithm, settings in the mixing algorithm ... nothing worked. The SCF 
>>>> cycle never converges.
>>>>
>>>> I would very much appreciate if you could give some advice or feedback 
>>>> on my input files (enclosed).
>>>>
>>>> Thank you in advance for your help.
>>>>
>>>> Kind regards,
>>>> Manuel Corral
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210120/e790b79c/attachment.htm>