Convergence issues with CI-NEB/NEB calculation for alloy system in VASP

[hl94vul3h6]

Hello everyone,

I am working on NEB calculations for an alloy system consisting of 128 atoms. I have defined and relaxed both the initial and final structures. Using the VTST scripts, I generated the intermediate images. However, despite multiple attempts with the following settings, the calculations fail to converge.

Due to the time limit on my university's cluster (7 days), I copy the CONTCAR file to POSCAR after each run and restart the calculation. It has now been running for over one month without success.

During this process, I visualized the structures in VESTA and noticed that the atoms seem to move significantly (and energy value is oscillating in OSZICAR), but I am not sure why this is happening. I suspect it might be related to the inability to find a stable intermediate state, even though both the initial and final states were relaxed before starting the NEB calculation.

I have also tried increasing the number of intermediate images from 5 to 9, but the results remain the same.

To force the calculations to complete, I temporarily adjusted the settings to NSW = 10 and EDIFF = 1E-5, but the results image were unphysical and inconsistent.
For more information, I run the calculation in the cluster and VASP version is 6.3.2. And below is the INCAR file that I used.

I read the previous post related to this topic, and I'm trying increase the intermediate images, set CL-NEB to FALSE, SPRING = -5, and the calculation is still running.
If my explanation is unclear, please let me know what additional files or information I can provide. Any advice or suggestions would be greatly appreciated.

Thank you!

Code: Select all

## Output writing ##
LWAVE = .T.    

## electronic settings ##
ISMEAR = 0      
SIGMA = 0.01    
ENCUT = 450   
EDIFF = 1E-7

## Ionic / structure settings ##
NSW = 500
IBRION = 3
ISIF = 2       
EDIFFG = -2E-2
POTIM = 0

IMAGES = 5
ICHAIN = 0
IOPT = 3
LCLIMB = .TRUE.

[ahampel]

Dear hl94vul3h6,

thank you for reaching out on the official VASP forum.

The larger the unit cell for a NEB calculation the fewer images you should try to start with, e.g. here I would maybe recommend 4 images or maybe even only 1 image if necessary. May I ask if you used selective dynamics for the 128 atoms or all atoms free to move? In that case converging the NEB calculation will be very hard.

If your atoms start moving significantly this is most likely a sign that the generated intermediate structures are not very good. Please also check this post and discussion: https://vasp.at/forum/viewtopic.php?t=19769#p29382 for more details. It might be also a good idea to use one of the optimizers provided by VTST instead of VASP: https://theory.cm.utexas.edu/vtsttools/ ... optimizers as suggested in the discussion.

Could you please share your POSCAR files as well do understand better the system you are working on?

Best regards,
Alex

[hl94vul3h6]

Hi Alex,

Thank you for the quick response and advice!
I already have tried with 3 intermediate images and 1 intermediates. Now the calculation is running, will let you know if I get any update!

For the selectrive dynamics, no I think I didn't use it, I just provided two POSCAR (initial and final) to the VTST to generate the interimediate images. So I am doing the wurtzite structure, between the initial and final state, only the anion is moving.

I already used the optimizers provided by VTST such as IOPT=3, IBRION=3 and POTIM=0 in my previous setting. Should I try other setting?

I attached the POSCAR below. Let me know if you need anything else, thanks!

Best,
Watson

[ahampel]

Hi Watson,

I had a look at the structure and I fear these are too complicated changes for a NEB calculation. If I look at the initial and final structure all of the anions (many of them) are moving. Maybe it would help if you could flag the other atoms as fixed for the relaxation, i.e. flag them in the POSCAR via Selective dynamics. Is it possible to mimic the transition first for an alloy that allows for a smaller cell? I talked to a colleague and he things if a few atoms are moving it is fine, but dozens of atoms will be very hard to relax during the NEB. You can try the selective dynamics.

Alternatively, my colleague suggested to perform a phonon calculation on the initial structure and see if on of the imaginary modes bring you to the final state. Follow that mode and plot the energy vs mode amplitude to mimic the transition path.

Otherwise I do not have many ideas how to solve your problem.

Best,
Alex