[CP2K-user] [CP2K:17929] CP2K Hangs for CellOpt+kpoints in certain systems.

[Nicholas Winner]

Hey I have identified the problem area, but not exact cause. I was relaxing 
argon for this example.

I found that the issue is something in the parallelization of kpoints. When 
I run without mpirun, it always runs okay. I was running with 128 MPI 
processes (one per core), and it failed. When I run with Monkhorst-pack 
grid, the calculation runs fine with 128 MPI processes, but the automatic 
grid has many more kpoints. So I rerun with the explicit grid but with 4 
MPI processes instead, and the calculation proceeds.

There is some kind of "over"-parallelization problem that's happening here. 
Has this reported in the community at all?

On Saturday, October 22, 2022 at 10:53:40 AM UTC-7 Matthias Krack wrote:

> On Argon?
>
> Matthias
>
> Am 22.10.2022 um 18:30 schrieb Nicholas Winner <nwi... at berkeley.edu>:
>
> Hi Matthias,  
>
>
> That's very strange. I tried a calculation again on Argon, but this does 
> not fix the problem. 
>
> -N
>
> On Friday, October 21, 2022 at 9:08:29 PM UTC-7 Matthias Krack wrote:
>
>> Hello Nick
>>
>>  
>>
>> The failures are possibly caused by the keyword WAVEFUNCTION COMPLEX. I 
>> observe that the regression test QS/regtest-kp-2/cc2.inp often fails 
>> because of this issue <https://github.com/cp2k/cp2k/issues/2117> related 
>> to zgemv of OpenBLAS.
>>
>>  
>>
>> Best
>>
>>  
>>
>> Matthias
>>
>>  
>>
>> *From: *cp... at googlegroups.com <cp... at googlegroups.com> on behalf of 
>> Nicholas Winner <nwi... at berkeley.edu>
>> *Date: *Saturday, 22 October 2022 at 01:17
>> *To: *cp2k <cp... at googlegroups.com>
>> *Subject: *[CP2K:17924] CP2K Hangs for CellOpt+kpoints in certain 
>> systems.
>>
>> Hello all,
>>
>>  
>>
>> I'm doing some cell optimizations using kpoints with CP2K. I have never 
>> had a problem with the geo_opt module, but cell_opt is hanging very often 
>> for a number of systems. 
>>
>>  
>>
>> Usually, the calculation will proceed for 1-3 optimization steps, and 
>> then it will hang at the start of a new SCF loop. I have found that 
>> sometimes the behavior is fixed by using direct_p_mixing instead of 
>> broyden_mixing, but this is not a consistent fix.
>>
>>  
>>
>> I've noticed the problem using both v9.1 and v2022.1, I've also tried a 
>> build on 3 different clusters and found the same behavior. 
>>
>>  
>>
>> Does anyone have experience with this and can offer advice? I've attached 
>> an example for Argon, which has this problem very consistently. The output 
>> goes until the calculation gets stuck.
>>
>>  
>>
>> Thanks,
>>
>> Nick
>>
>> -- 
>> 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 cp2k+uns... at googlegroups.com.
>> To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/cp2k/d26b287a-624d-4ab4-88ee-052529111991n%40googlegroups.com 
>> <https://groups.google.com/d/msgid/cp2k/d26b287a-624d-4ab4-88ee-052529111991n%40googlegroups.com?utm_medium=email&utm_source=footer>
>> .
>>
> -- 
> 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 cp2k+uns... at googlegroups.com.
>
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/cp2k/35a086c1-1515-4b1b-bbe1-53c35bac7bd0n%40googlegroups.com 
> <https://groups.google.com/d/msgid/cp2k/35a086c1-1515-4b1b-bbe1-53c35bac7bd0n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
>

-- 
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 cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/7a4ad49e-ea23-482e-a932-071e270c21e6n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20221022/31da0c4e/attachment.htm>