[CP2K-user] Extremely slow SCF convergence on O2/metal

[Y Lu]

Dear Fabian,

I see. I will try XZ instead. Many thanks for your help! Very helpful!

Best regards,

You

On Saturday, 14 November 2020 at 07:55:40 UTC fa... at gmail.com wrote:

> Dear You,
>
> The wavelet solver is supported by version 6. But the manual states:
> "[...] *and slab-boundary conditions (but only PERIODIC XZ)*. [...]"
> You might have to rotate your system to accomodate this restriction.
>
> Best,
> Fabian
> On Saturday, 14 November 2020 at 04:50:57 UTC+1 Y Lu wrote:
>
>> Dear Fabian,
>>
>> yes it works like a charm! Amazing!
>>
>> I was running CP2K 6 and it didn't support WAVELET Poisson solver on 2D 
>> but I'll try 7 instead.
>>
>> Thanks a lot!
>>
>> Best regards,
>>
>> You
>>
>>
>>
>> On Friday, 13 November 2020 at 17:13:49 UTC fa... at gmail.com wrote:
>>
>>> Dear You,
>>>
>>> A few pointer that can improve the scf convergence:
>>>  - decrease EPS_DEFAULT to at least 1e-12
>>>  - increase the smearing temperature
>>>  - increase the cutoff, 300 is rather low here
>>>  - use BROYDEN_MIXING
>>>  - I found that ALPHA between 0.15 and 0.2 is often a good choice for 
>>> metallic systems
>>>  - you can use far fewer ADDED_MOS, 20 is plenty here
>>>
>>> Another issue you should address is the periodicity. The settings in 
>>> &POISSON and &CELL have to match unless you really know what you are doing. 
>>> The WAVELET solver allows for slab periodicity, see 
>>> https://manual.cp2k.org/cp2k-7_1-branch/CP2K_INPUT/FORCE_EVAL/DFT/POISSON/WAVELET.html
>>> .
>>>
>>> The attached input (ignoring the problem with the periodicity) converged 
>>> within 120 iterations. This can probably still be improved upon.
>>>
>>> Cheers,
>>> Fabian
>>>
>>> On Friday, 13 November 2020 at 17:15:05 UTC+1 Y Lu wrote:
>>>
>>>> Google didn't allow to post with plain text attachments!
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20201114/9f9667a4/attachment.htm>