NEB convergence problems

[Aiden83]

Dear Prof. Henkelman,

I am trying to do a cNEB calculation of CO oxidation on Pt single atom anchored on perovskite. I use IOPT = 3, TIMESTEP = 0.01, EDIFF = 1E-5 at first, to make the forces drop to around 0.5 eV/Ang systematically, and this set up work effectively to drop the high forces. However, when I use a more strict convergence set up(IOPT = 3,TIMESTEP = 0.05, EDIFF=1E-7, EDIFFG= 0.1), the force only fluctuate between 0.15~0.5 ev/ Ang. Is there any suggestion to make the structure converge better, or is it just unnecessary to converge any more?(because the energy barrier doesn't be changed radically, and there is reasonable imaginary frequency, as you can see in my given example)

I also tried to use IOPT = 2 or (IOPT=3, TIMESTEP = 0.1) in the fine convergence(EDIFF = 1E-7), but it seems that the force will eventually rise up to above 1 eV/ Ang, especially in the middle images, and there would even be two energy highest point in the reaction path in some canses. Can you explain why would this happen?

I have attached a .tar.gz file with all the data used. It will be of great help if you could take some time to have a comment.

Yours sincerely,

[Aiden83]

Dear Ingle:

Thanks for your replying! I will try to run the set up with more accurate EDIFF. I am looking forward to your later reply!

Best regards,

[graeme]

I have a somewhat different take on the issue. I'm going to attach 3 tests that I did.

In SrAgO3, I used the conservative settings of IOPT=3 TIMESTEP=0.05. After 132 iterations (it is still running) the force on the climbing image has increased from 0.15 up to 0.54 eV/Ang and the barrier remains over 1 eV.

In SrAgO3_1, I increased TIMESTEP to 0.1 eV, and again the force on the climbing image increased, now to over 6 eV/Ang. You might think this indicates a problem, but the barrier is steadily decreasing, now to 0.76 eV.

Finally, in SrAgO3_2, I used a the second order optimizer (IOPT=1 INVCURV=0.005) and the force on the climbing image was as high as 50 eV/Ang. Again, you would have reason to be worried, but it resulted in the fastest optimization and converged with a barrier of 0.71 eV.

So I think the take home message here is that your initial band was in a low force region, but actually on a ridge so that it needed significant relaxation to converge to a saddle. The second point is that just trying to keep the forces low is not nessecarily a good strategy if the system needs to undergo significant relaxation from your initial guess.

[Aiden83]

Dear professor,
Last time you suggested me about using IOPT =1, INVCURV = 0.005 to converge the structure that need a significant relaxation, and this setting works effectively in some of my structures. However, when I used this setting (even a smallar INVCUR = 0.001) in some other structures, a more general situation is that some additional movements occurred in CO, which was not moving toward the transition state, especially after many steps of iteration, and the forces remain very high. While in structures that converge successfully (like the SrAgO3 that converged in previous discussion), the CO only rotate in one direction to the final state. Is there any way to solve this issue?

Best regards,