vasp neb optimization

[jpg]

Hi,

A user on one of our clusters at SDSC ( https://www.sdsc.edu/systems/expanse/user_guide.html ) has a question about optimizing an NEB job with 6 images. He has tried assigning each image to a node (6 nodes). Each node has 128 cpu's ( two 64-core AMD EPYC 7742 processors) . He has tried 21 cpu/s per node and I have added more but it does not seem to make any difference.

The job runs to completion in ~ 14 hrs. Any guidance on optimization would be appreciated.

Attached is the INCAR, POTCAR , KPOINTS file and one of the POSCAR files from image #1 (Please forgive the ".tgz" suffix, its the only way I was able to get the browser to upload it)

Please let me know if any additional files would be useful

[andreas.singraber]

Hello!

I apologize for this very late reply! I am afraid something went wrong with the attached files, I can only see the POTCAR file (with tgz ending). Anyway, even with all files you mentioned we cannot help you without a proper problem description (is the job not fast enough, not converging?) and a detailed list of things you or the user have already tried. In particular NEB calculations are tricky because not just the ab initio calculation must be set up correctly but also the starting points for the images often require tuning. If the issue is convergence I would strongly advise to use less images for starting, 6 is definitely hard to set up correctly right away. Please carefully study the hints on our Wiki page and our tutorial on NEB calculations. If issues persist please upload a complete NEB setup with all relevant input and output files according to our forum posting guidelines. Thank you!

All the best,
Andreas Singraber