[CP2K-user] OT: Bad covergence performance

[Marcella Iannuzzi]

Dear Hongyang, 

It works also with ADMM, just set 

ADMM_PURIFICATION_METHOD NONE

Regards

Marcella

On Saturday, April 24, 2021 at 7:38:05 AM UTC+2 ma... at gmail.com wrote:

> Dear Marcella,
>
> But I remeber that ADMM only works with OT. Is this correct?
>
> Regards,
> Hongyang
>
> 在2021年4月24日星期六 UTC+10 下午3:33:58<Marcella Iannuzzi> 写道：
>
>> Dear Hongyang
>>
>> you can use the diagonalization also with HSE06
>> Best
>> Marcella
>>
>>
>> On Saturday, April 24, 2021 at 5:40:45 AM UTC+2 ma... at gmail.com wrote:
>>
>>> Dear cp2k developers and users,
>>>
>>> I'm a new user of cp2k 8.1.
>>> I'm currently calculating a 1000 atoms system. However, OT does not 
>>> perform very well.
>>> When I use outer-scf method (please have a look of attached 
>>> si-pbe-01-Outer-SCF-out.txt), the first inner-scf cycle performs pretty 
>>> well, but the following outer-scf cycle makes the SCF covergence terrible.
>>> When I do not use outer-scf method (please have a look of attached 
>>> si-pbe-01-No-Outer-SCF-out.txt), the first SCF cycle converges in 151 
>>> steps. However, the following SCF cycle perfoms bad again.
>>> Use DIAGONALIZATION 
>>> <https://manual.cp2k.org/cp2k-8_1-branch/CP2K_INPUT/FORCE_EVAL/DFT/SCF/DIAGONALIZATION.html> method 
>>> can give much better performance. But because I need to do HSE06 
>>> calculation later, OT is necessary. 
>>> Could someone please provide me some suggestions on improving the OT 
>>> performance? Thanks!
>>>
>>> Regards,
>>> Hongyang
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210424/6896c2e0/attachment.htm>