Convergence problem of climb NEB calculation

[can_liu]

Dear administrator,

I'm running a calculation of methoxy diffusion on rutile(110) with climb NEB method. There are five middle states. It has run for more than 5000 cycles, but not converged yet. The energies are changing quite slowly. I'm not sure if there is any problem. Would you please have a check on my calculation? Here are the current files

report 0201.rar

including POTCAR, INCAR, KPOINTS, CONTCAR(states 01-05), OSZICAR(states 01-05), and OUTCAR(states 01-05). Thank you very much.

Sincerely yours,
Liu

[alexey.tal]

Hi,

Could you please provide POSCARs for all images?

[can_liu]

Hello,
Thank you for your reply. Here are the POSCAR files (

report-POSCARs.rar

) of the current calculation. They have been iterated after 5000 cycles. If you need the very beginning POSCARs, please let me know.

[alexey.tal]

As EDIFFG is not specified in your INCAR, it is set to the default value, which is 1E-4 eV in your case. If EDIFFG is positive, VASP has an additional convergence criterion for the forces g(Force) < EDIFFG and this condition is not satisfied in your calculation. Also, considering that g(Force) is calculated as a sum over the images, it can be rather difficult to converge. So it might be easier to achieve the convergence by setting a criterion for the forces explicitly EDIFFG < 0.

You can also try to increase ENCUT and the number of k-points to improve the accuracy of the calculated forces and get smaller residual forces. Another thing that can improve the convergence is fixing more layers in the slab.