[CP2K-user] NEB - number of replicas

[Peter Benedek]

Hi Matt,

So I tried what you described below and still get the same "CPASSERT 
failed" error message. Attached the extended .out file and the .inp file. 

I ran the calculations on 12 nodes containing 12 processors each with 10 
replicas. 

Best,
Peter

PS: I also tried setting NPROC_REP 12 to match the amount of processors 
(and not nodes) but I got the same error.


Am Montag, 18. Februar 2019 14:25:45 UTC+1 schrieb Matt W:
>
> Hi,
>
> sorry you didn't get a reply, we normally try to.
>
> I haven't checked this, but I believe that the NUMBER_OF_REPLICA keyword 
> is used for start up and tells CP2K how many _intermediate_ images to 
> create, filling in any not given explicitly using interpolation. So if you 
> set it to 6 and only provide initital and final coordinates there will be 8 
> images running - beginning and end + NUMBER_OF_REPLICA intermediates.
>
> Matt
>
> On Monday, February 18, 2019 at 10:36:15 AM UTC, tobias... at gmail.com 
> wrote:
>>
>> Hi, Sorry I did not get an answer to the issue so far, and I did not work 
>> on it to figure the problem out.
>> But I would be really happy, if someone else could find a suitable answer 
>> to the problem.
>> Let me know, if you are more successful.
>>
>> Best wishes and good luck
>> Tobias
>>
>>
>> Am Donnerstag, 25. Oktober 2018 14:18:04 UTC+2 schrieb 
>> tobias... at gmail.com:
>>>
>>> Dear all,
>>>
>>> I want to perform a nudged elastic band (NEB) calculation using cp2k.
>>> I get the "CPASSERT failed", when I try to calculate too many replica at 
>>> once.
>>> Does anyone know the difference between the "Number of Images" and 
>>> "Number of Replicas"?
>>> Does anyone have experience, how to choose the number of processors per 
>>> replica NPROC_REP  and the number of calculated replica  
>>> NUMBER_OF_REPLICA   
>>> for the best performance? 
>>> Is there a thump rule, how many Open MP threads to use for best 
>>> performance?
>>>
>>> I tried to match the total cpu cores with nproc_rep * number_of_replica 
>>> but the number_of_replica input seems to be ignored, instead cp2k uses the 
>>> number of intermediate structures (replicas) as number of images.
>>>
>>> When I use 6 replicas and 12 nproc_rep the calculation runs on the 72 
>>> core node.
>>> however I set the number of replicas to 6 and provided 6 structures. 
>>> CP2k now gives as output:
>>>
>>> Number of Images :            6 **
>>>  **  ##     ## ######### ##  #### ##     ##   Number of 
>>> Replicas:           1 **
>>>  **  ##     ## ##     ## ##   ### ##     ##   Number of 
>>> Procs/Rep:         12
>>>
>>> Why does it change the number of replicas during the run?
>>>
>>> When I use 9 replicas and 8 nproc_rep the calculation does not run, but 
>>> as far as I understand the input should work this way.
>>>
>>> Best wishes 
>>> Tobias
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190219/971aac85/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NEB_Run.out
Type: application/octet-stream
Size: 58941 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190219/971aac85/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NEB_Run.inp
Type: chemical/x-gamess-input
Size: 2540 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190219/971aac85/attachment.inp>