CINEB system shift

[xi007]

Dear All,

Recently, when I used the CINEB to study the migration barrier of defect, such as interstitial and vacancy, I found some strange problems. It is difficult to converge for some configurations. And then when I checked the configuration for each image, I found that the system has been shift, that is, the mass point of the system has been shift. I know this may cause the difficulty convergence. But I don't know what's the reason to result in the shift. I tried to increase and decrease the number of images, but it didn't work. I hope you can give me some suggestions. Thanks very much!

Here is my INCAR[attachment=0]INCAR.zip[/attachment]

Jianqi

[graeme]

If you can freeze any atoms, that would constrain that center of mass.

Second, try an initial choice of IOPT=3 with TIMESTEP=0.1 to make sure that convergence is steady (and it will be slow). IOPT=1 is a faster optimizer but with high forces the atoms can move around, leading to an elongation of the path.

For more specific help I would need to see the other files in the calculation.

[xi007]

Thanks very much for your quick reply. I will try to do that. And here is the other files

[graeme]

Something is clearly inefficient with the optimizer. So yes, do check with IOPT=3 or 7 (first-order optimizers) until the forces drop below 1 eV/Ang and then if you go back to IOPT=1, try reducing the INVCURV parameter (note that this variable is not correct in your INCAR file).

And a general thing: I would recommend converging your band using less expensive settings (e.g. prec=normal, a 2x2x1 kpoint mesh, ENCUT=400, or 270 if you use the C_s potential). You should be able to get a nicely converged pathway at a fraction of the cost. Then, it is trivial to reconverge with more accurate settings to check for sensitivity of your diffusion barrier.

[xi007]

[quote="graeme"]Something is clearly inefficient with the optimizer. So yes, do check with IOPT=3 or 7 (first-order optimizers) until the forces drop below 1 eV/Ang and then if you go back to IOPT=1, try reducing the INVCURV parameter (note that this variable is not correct in your INCAR file).

And a general thing: I would recommend converging your band using less expensive settings (e.g. prec=normal, a 2x2x1 kpoint mesh, ENCUT=400, or 270 if you use the C_s potential). You should be able to get a nicely converged pathway at a fraction of the cost. Then, it is trivial to reconverge with more accurate settings to check for sensitivity of your diffusion barrier.[/quote]


thanks very much for your suggestions.