[CP2K-user] Very Slow NEB Calculations

[Travis]

Hi,

I think you may have set NPROC_REP incorrectly. If you use a total of 1800 
processors (with openmp threads = 1; mpiexec -np 1800 cp2k.psmp in > out), 
for example, NPROC_REP for 9 images should be 200. Get rid of the KPOINTS 
section, drop the smearing and mixing parts, and turn on OT. Change 
EXTRAPOLATION USE_GUESS to EXTRAPOLATION ASPC or PS with OT. Should run 
much quicker. I can't remember if CP2K treats the user supplied endpoints 
as frames in the number of replicas or if it's 2 + NUMBER_OF_REPLICA. 
Someone else can clarify or you can run a couple test calculations assuming 
you divide your total processor pool by 9 or 11 in setting NPROC_REP. If 
you kept the output files from your first attempt, you'll know based on how 
many bands were created.

-T

On Thursday, December 31, 2020 at 4:36:17 PM UTC-5 ti... at hotmail.com 
wrote:

> Dear Sir or Madam, 
>
>
> I'm currently learning how to run CI-NEB simulations, but I found that my 
> NEB calculations were really slow. There's only one iteration finished in 
> 23 hours. I tried to change some parameters including MAX_SCF and EPS_SCF, 
> and remove the DFT-D3, but they didn't help. I wonder if you know how to 
> solve this kind of problem? My input file is attached below.
>
> Thanks in advance!
>
>
> Best,
> Xiao Wang
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20201231/14d84427/attachment.htm>