Convergence problem

[wvno2]

Dear Prof. Henkelman and users,

I'm simulating a dissociative chemisorption of molecule (XeF_6) on surface (h-BN), and trying to make the reaction pathway using NEB method to find barrier energy. The 2 F atoms are dissociated and chemisorbed on the surface, and I'd like to draw the reaction pathway for this process.

I'm expecting the middle image at highest barrier point and other images at some intermediate points, but the problem is that all images converge to the first or end point. In the folder "Figures for first calculation" in the attached zip file, I saved the atomic position figure and all converged images are gone to the almost near end points. In this calculation, I used 3 images, IBRION = 3, EDIFFG = -0.05, and POTIM = 0.12 which is gradually increased from 0.01 during calculation.


After this, I calculated again, with 5 images and IBRION = 3. Also, during second calculation, first I used EDIFFG = -1.0 with POTIM = 0.01, and after rough convergence I changed the setting to EDIFFG = -0.5 with POTIM = 0.02, and again after rough convergence I changed the setting to EDIFFG = -0.3 with POTIM = 0.03. It seems that it was going well and I obtained the result with EDIFFG = -0.3 (rough, though), and I drew the graph with this result (the graph is in the zip file, and the calculation result is between II and III in the graph).

So, I aimed to more reliable result with lower EDIFFG = -0.1. I read Prof. Henkelman's reply that a force convergence criterion of 0.1-0.01 eV/Ang is generally suitable for saddle point calculations. Until convergence with EDIFFG = -0.3, I continuously checked the change of forces and these had been decreasing gradually. However, during further calculation with EDIFFG = -0.1, at some point forces suddenly increased to 0.7~0.8 eV/A with image 3 and it didn't decrease. I worried that I might obtain the same result with the first calculation, and it was right.. the tendency of atomic movement along the time steps was same as the first calculation.

So, because I thought that it may be due to the increasing POTIM from 0.01 to 0.03, I reduced POTIM = 0.01 and calculate again from some point which didn't converged with EDIFFG = -0.3 yet, but the result was same.

(Note that, I attached the calculation files in the folder "Calculation files for second calculation" in the zip file. in the folder, I numbered the result files (CONTCAR, OSZICAR, OUTCAR, and additionally POSCAR also) as calculation order. Convergence was obtained at 4th, 11th, and 21th with EDIFFG = -1.0, -0.5, and -0.3, respectively, and POTIM was reduced from 0.03 to 0.01 since 17th calculation. Folders named ""Previous17~21" contains the results before reducing POTIM.)


Is there any possible solution for my problem? Please any comments are very thankful for me.

If you need more information for analysis, please let me know. I'll write or upload in reply.

Note that,
1. I didn't turn on LCLIMB = .TRUE. for finding the highest barrier because I read one reply from the forum that it is recommended to turn on LCLIMB after some convergence.
2. I'm using VASP version 5.3.5


Thank you very much.

[graeme]

If the forces start increasing it can be indicating that the reaction mechanism is more complicated than what you assumed. You might have to use a small time step again and continue convergence to see where the path converges. If you see any intermediate minima develop along the path, I suggest minimizing them separately to understand the reaction mechanism and perhaps break up your pathway into two calculations.

Looking at your path, I would expect the F atoms to dissociate and adsorb to the surface separately, rather than in the concerted way that you assumed. It may be that your increase in forces corresponds to the path switch to that sequential F-adsorption mechanisms. You might also check an intermediate state with one F adsorbed and check if the barrier for that mechanism is simple and lower in energy.

Note that you do not have our vtstcode linked into your vasp binary and so tags such as LCLIMB will not work.

[wvno2]

Dear Prof. Henkelman
Thank you very much for your reply. I'll check and try your suggestion.
I thought that VASP has all of your codes. It seems that only NEB method is available in the original VASP 5.3.5.
Thank you very much.

[wvno2]

Dear Prof. Henkelman

Thank you and I really appreciate your help.
I used vtst module to use CI-NEB, and tried calculating it by dividing into two reactions (folders "II-III" & "III-IV" in the zip file attached) with smaller time step (POTIM 0.002). I made convergences up to -0.1 eV/A criterion for each calculation. I'd like to ask you if CI-NEB calculation was well-activated, and the results are well-converged into the creteria.

Also, I attached the ppt slide which contains the summary of the calculation results (Reaction pathway graph and the snapshots for endpoints and images. The collective movie for snapshots in the right-middle side of the slide can be seen by slide show). I obtained the saddle point for the 1st reaction (II-III), but there is no saddle point in the 2nd reaction (III-IV).
Actually, I checked by AIMD calculation that 2nd reaction occurred right after the 1st reaction. Also, I know that there exists XeF6 or XeF4 molecule, but XeF5 doesn't exist because Xe cannot form five stable Xe-F bonds at once. In that sense, is it possible that 2nd F atom will be dissociated spontaneously once the 1st reaction of F atom occurs, and thus there would be no saddle point for the 2nd reaction?

Thank you very much.

[graeme]

This all looks fine to me.

[wvno2]

Dear Prof. Henkelman

I really appreciate your helpful reply.
Thank you very much.