Why can't cell optimization converge?

Enlong Liu liuen... at gmail.com
Mon Mar 3 11:36:48 UTC 2014


Dear Matthias,

By using LBFGS, the simulation stops in 27 steps and the result is better 
than I got after 1000 steps. But I still can not be sure whether it is 
finally optimized. However, thank you very much! I think it works very well.

Best regards,

在 2014年3月3日星期一UTC+1下午12时14分13秒,Matthias Krack写道:
>
> Dear Enlong,
>
> either you provide a full setup (i.e. including the atomic coordinates) 
> for the actual case, or you provide the input and the corresponding output 
> file (at least part of it) showing the problem.
> Did you try LBFGS?
>
> Best,
>
> Matthias
>
> On Monday, March 3, 2014 11:57:27 AM UTC+1, Enlong Liu wrote:
>>
>> Dear Matthias,
>>
>> Thanks!
>>
>> But I find out that the input I upload is a wrong one. So maybe you can 
>> have look at the right one and give me some suggestion. 
>>
>> Best regards,
>> Enlong
>>
>> 在 2014年3月3日星期一UTC+1上午11时12分52秒,Matthias Krack写道:
>>>
>>> Dear Enlong,,
>>>
>>> I would suggest to use BFGS (or LBFGS for large systems) instead of CG. 
>>> Moreover, you may change the PRINT_LEVEL to medium to get more info. 
>>> However, even if the run will converge somehow, the results won't be 
>>> reasonable given the cell size of only 2.5A in z direction in a gamma point 
>>> calculation.
>>>
>>> Matthias
>>>
>>> On Monday, March 3, 2014 10:41:11 AM UTC+1, Enlong Liu wrote:
>>>>
>>>> Fellows,
>>>>
>>>> I am practicing CP2K by doing cell optimization and geometry 
>>>> optimization to CNT. In the cell_opt, I choose 'DIRECT_CELL_OPT*: 
>>>> Performs a geometry and cell optimization at the same time. The stress 
>>>> tensor is computed at every step* '. And the max SCF iteration is 
>>>> 1000. However, every time I run the program, the program always used all 
>>>> 1000 iteration to get the result, which in my opinion, means the result 
>>>> does not converge. It seems like for every step, the threshold for energy 
>>>> is satisfied but between steps the threshold for geometry can not be 
>>>> satisfied. Could you help me to solve this issue?
>>>>
>>>> PS: the input file is in the attachment.
>>>>
>>>> Thanks a lot for your help!
>>>>
>>>> Best regards,
>>>> Enlong
>>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140303/d6b8d1c3/attachment.htm>


More information about the CP2K-user mailing list