Problems with charged vacancy

[weiyidan]

dear professor,
when i did the calculation with iop=7 and timestep=0.05, after a long period it still fluctuated at about 0.1-0.5. I do not know how to improve my parameters, can you give me some advice? whether reduce the ediff or timestep will help a lot?[attachment=0]neb.rar[/attachment]

[graeme]

The energy of the saddle is changing monotonically; it is fine.

Make sure that you use the same k-point sampling for the endpoints as the band.

[weiyidan]

dear professor,
i am sure that the kpoints is the same, but it seems the energies differ from each other.....
as for the clneb, i should wait until it converges. But is there a appropriate parameter to improve the performance?
thanks for your attention and response.

[graeme]

The k-point mesh is not the same.

You can continue your calculation and look for steady convergence. Using IOPT=1 or 2 can give more rapid convergence, yet less stable, as I think you found previously.

[weiyidan]

yes, .... you are right, i will correct it in the successive calculation, how about doing a rough process with less accurate kpoints ,and switching it to a more accurate one with the converged contcar from the former calculation?

[graeme]

That's a fine strategy. I think it is best to have a set of minimal settings which lets you rapidly explore the potential surface, finding barriers for example, and then more accurate settings that you use to reconverge the most interesting structures and mechanisms to get converged energies.

[weiyidan]

thanks, i continued my calculation with IOP=7 for a long time, however, it does not converge. can you tell me how to judge whether it can converge or not? ‘(*>﹏<*)′
[attachment=0]neb.rar[/attachment]

[wc5879]

Have you considered that your giving this system 2 negative charge made it difficult to converge somehow? What about make it converge first, then add the negative charge?
You could also try decreasing EDIFF, or different optimizers such as LBFGS

And honestly these 2 negative charge doesn't make sense to me.

[weiyidan]

will it make the system harder to converge by reducing ediff? ?
i have do the calculation with IOP=1, which is also helpless.
as for the vacancy without charge, i have never thought this before. but will not it still hard to converge after adding some charge?

[weiyidan]

still not converge 0.0