[CP2K-user] [CP2K:15589] Stuck when using BFGS

[ma...@gmail.com]

Hi Fabian,

Thanks for the information! I will give it a try.

Cheers,
Hongyang

在2021年6月21日星期一 UTC+10 下午8:42:02<fa... at gmail.com> 写道：

> Hi
>
> If you compile cp2k yourself you can modify the code to force BFGS to 
> select SCALAPACK and still use ELPA for everything else. In 
> src/motion/bfgs_optimizer.F replace every occurence of "choose_eigv_solver" 
> with "cp_fm_syevd".
> If you are in the cp2k root directory simply run
>
> sed -i.bak src/motion/bfgs_optimizer.F 's/choose_eigv_solver/cp_fm_syevd/g'
>
> and recompile the code. This works for me but I have not extensively 
> tested it. 
>
> Cheers
> Fabian
>
> On Friday, 18 June 2021 at 12:45:44 UTC+2 ma... at gmail.com wrote:
>
>>
>> Hi Matt,
>>
>> Got it. Although I do like using ELPA because it is much faster. Thank 
>> you very much. Wish ELPA would be more compitable with cp2k in the future 
>> (figure crossed).
>>
>> Regards,
>> Hongyang
>> 在2021年6月18日星期五 UTC+10 下午6:58:54<Matthias Krack> 写道：
>>
>>> Hi
>>>
>>>  
>>>
>>> It seems that is an issue rather related to ELPA than to BFGS. 
>>> specifically, because BFGS involves a diagonalization of the Hessian 
>>> whereas LBFGS does not. When BFGS works with ScaLAPACK, the problem stems 
>>> most likely from ELPA. There have been issues observed with CP2K/ELPA, 
>>> recently. So, just use ScaLAPACK, since the advantage of using ELPA instead 
>>> of ScaLAPACK is most likely minor for your case.
>>>
>>>  
>>>
>>> M.
>>>
>>>  
>>>
>>> *Von:* c... at googlegroups.com <c... at googlegroups.com> *Im Auftrag von *
>>> ma... at gmail.com
>>> *Gesendet:* Freitag, 18. Juni 2021 02:44
>>> *An:* cp2k <c... at googlegroups.com>
>>> *Betreff:* [CP2K:15589] Stuck when using BFGS
>>>
>>>  
>>>
>>> Hi,
>>>
>>>  
>>>
>>> I'm a new user of cp2k v8.2. When I do GEO_OPT using BFGS, it has been 
>>> stuck many times in the step "OPTIMIZATION STEP: 1" as shown below without 
>>> showing any errors. If I don't kill the job manually, the job will just 
>>> keep running without printing any new results. This does not occurs when 
>>> using SCALAPACK instead of ELPA or using LBFGS/CG instead of BFGS. I'm 
>>> wondering is there any way to solve this BFGS issue? Thanks.
>>>
>>>  
>>>
>>>
>>> ---------------------------------------------------------------------------------------------
>>>
>>> [image: 微信截图_20210618103738.png]
>>>
>>>
>>> -----------------------------------------------------------------------------------------------
>>>
>>>  
>>>
>>> Regards,
>>>
>>> Hongyang
>>>
>>> -- 
>>> You received this message because you are subscribed to the Google 
>>> Groups "cp2k" group.
>>> To unsubscribe from this group and stop receiving emails from it, send 
>>> an email to cp... at googlegroups.com.
>>> To view this discussion on the web visit 
>>> https://groups.google.com/d/msgid/cp2k/1e3675c8-622e-4f70-b320-ae8951a1f233n%40googlegroups.com 
>>> <https://groups.google.com/d/msgid/cp2k/1e3675c8-622e-4f70-b320-ae8951a1f233n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210621/ff4693b8/attachment.htm>