NEB:convergence

[2008zdd]

Dear Graeme,
Somtimes my system just does not converge under the force criteria ediffg=-0.01, in this situation, can I use a larger force criteria eg. ediffg=-0.015,
and would this setting have a large influence on the results? Thanks!

[graeme]

Probably not, but it's a little worrisome if you can not achieve good convergence. Perhaps try a conservative optimizer such as IBRION=3 and POTIM=0.1 and make sure that the force drops systematically.

You can also use just the force on the climbing image if you only care about the energy of the saddle point.

[pgopal]

Hello,
I have a problem with convergence of saddle points. I am trying to find the migration energy of a vacancy diffusing across an interface. My system contains 200 atoms. The force criteria I use is -0.01 eV/Ang. Is this criteria too strict for a system with vacuum? The problem with my NEB calculation is the forces just keep oscillating between 0.05 and 0.08 for almost 50 ionic iterations. If I stop the calculation and look at the energies, the barrier energy seems too large. I used IBRION = 3 and POTIM = 0.1 for the first few iterations until the forces came to 0.5 and then used IOPT = 2 for the next few iterations.

Thanks,
PG

[graeme]

That doesn't sound good. If you upload it, we'll take a look.

[pgopal]

Dear Graeme,
Should I send you the files by email or upload all the files here in this forum?

Regards,
PG

[graeme]

Either is fine. Uploading is a bit better.

[pgopal]

Dear Graeme,
Here are the files of my run..

Regards,
PG

[graeme]

The initial movie of POSCAR files look very strange. The path does not correspond to a single atom moving; instead there is a rearrangement of the atoms. To accommodate this, the NEB is creating a vacuum gap in the middle of the material, which is leading to the high energy barrier. I think there is some problem with how your initial path was generated. Make sure that the atoms are in exactly the same order in your initial and final structures.

[pgopal]

Dear Graeme,
I didnt quite understand how there is a vacuum created in one my images. When there is a vacancy created how does one do a one-one correspondence between atoms to track the path? For all my images I used nebmake.pl to create the path.

Regards,
PG

[graeme]

Perhaps this is not the first NEB calculation then, because the POSCAR files in the 00-04 directories do not look like a linear interpolation. Let me see the initial calculation.

Wait, it looks like you have frozen the x and y positions of the moving atom! This will certainly cause problems.

[pgopal]

I wanted to fix the x-y position of the atoms since it ian interface calculation. Should I let all the atoms move?

PG

[graeme]

In your path, an atom is moving in the x-y plane; the atoms must be unconstrained near the vacancy.

[pgopal]

Dear Graeme,
Will It work if I let all the atoms near the vacancy relax (upto 2-3 layers) and constrain the x-y movement of all the remaining atoms? I am not sure if this is physical though? The other option is to treat the system as a substrate system with the top and bottom layers fixed.

Regards,
Priya

[graeme]

That should be fine, and it should also solve the convergence problems. But just make sure that an frozen degrees of freedom have exactly the same positions in the initial and final states (and hence for each intermediate image as well).

[pgopal]

Dear Graeme,
I started the calculations again by letting the atoms relax in a few layers below and above the defect. The initial run was ok with the energy and the forces of the saddle point converging well. Below is the nebef.dat for the first run.
0 0.01768400 -1083.01825100 0.00000000
1 0.02076900 -1082.74840600 0.26984500
2 0.01903300 -1082.18059500 0.83765600
3 0.02190500 -1082.69066100 0.32759000
4 0.01856600 -1082.99058000 0.02767100

The force threshold was -0.01 and the max force on the atom in the saddle point was 0.019033. So I restarted the job again by changing to IOPT = 2 to further relax the forces to -0.01. Is this ok to do? The convergence got worse once this was done. I had to stop the calculation and the results of the second run are as follows.
0 0.01768400 -1083.01825100 0.00000000
1 0.03767000 -1083.19332500 -0.17507400
2 0.25789700 -1082.78931600 0.22893500
3 0.09270700 -1081.50143000 1.51682100
4 0.01824800 -1083.01053400 0.00771700

I am not sure what the problem is..

Regards,
PG.