Page 1 of 1

NEB:convergence issues.

Posted: Fri Apr 16, 2010 8:16 pm
by pgopal
Dear all,
I have been running a few NEB calculations for finding a MEP for vacancy diffusion in a simple fcc system. I have created 5 images using nebmake.pl. Every time I run the caculation, I face convergence problems with IMAGE 3 which is likely to be the saddle point. The error message I end up getting is Error EDDDAV: Call to ZHEGV failed. Returncode = 76 4 2. I know this error is associated with the IALGO and have also tried using IALGO=FAST and VERY FAST but there is no difference. I have also checked my position file to see if there is a warning associated with the positions of atoms being too close but there is no such error in the OUTCAR file.
I would appreciate if anyone could help me with this.

Regards,
PG

p.s. The INCAR file I use is the following.

PREC = Normal
ENCUT = 400
IBRION = 1
POTIM = 0.1
NSW = 25
ISIF = 2
ALGO = NORMAL (blocked Davidson) # have also changed this to IALGO=FAST and VERY FAST.
NELM = 60
NELMIN = 2
EDIFF = 1.0e-05
EDIFFG = -0.02
VOSKOWN = 1
NBLOCK = 1
ISPIN = 1
INIWAV = 1
# parameters for Nudged band
LCLIMB = .T.
ICHAIN = 0
IMAGES = 5
SPRING = -5.0
LWAVE = .FALSE.
LCHARG = .FALSE.
ADDGRID = .T.
ISMEAR = 0
SIGMA = 0.05
LREAL = .TRUE.
ADDGRID = .T.
RWIGS = 1.15 1.18

Re: NEB:convergence issues.

Posted: Fri Apr 16, 2010 9:52 pm
by graeme
The problem might be different if the program is crashing on the initial energy calculation, or if it was able to calculate the energy of image 3 at least once.

First though, I would start by making a movie of the initial band (e.g. with nebmovie.pl) so that you can see that the path makes sense.

Then, if the energy of image 3 converges at least once, I would suggest switching to IBRION=3 and POTIM=0.1 until the forces are on the order of 0.1 eV/Ang. Then you can switch to IBRION=1.

If the electronic structure of image 3 does not converge, even once, then something is fishy with the electronic calculation. The most conservative setting is IAGL0=58 with a small value of TIME. You could at least make sure that you can converge the electronic structure (even separately from the NEB calculation).

Re: NEB:convergence issues.

Posted: Sat Apr 17, 2010 6:17 am
by pgopal
Hello Graeme,
Thanks for the suggestions.

Regards,
PG

Re: NEB:convergence issues.

Posted: Sun Apr 20, 2014 11:27 am
by mishralu
Hi,
I have similar issues. In CI-NEB calculations, Middle image do not converge even one ionic iteration, it stops after 1-3 electronic steps Error EDDDAV: Call to ZHEGV failed. Returncode = 32 2 18. I have tried NEB calculations with 3, 5, 8, 12 images but all having same problem.
I have tried ALGO normal, fast, veryfast, also played with AMIX and BMIX parameters. The problem is always middle image do not converge after 1-3 electronIc convergence and error is same as I mentioned.
There is no sense of playing with potim value because even one ionic iterations is not achieved.
My system is hydrogination of CO2 on surface.

Any help will be highly appricetaed.
Many thanks,
Dr. Abhishek Mishra
UCL, London

Re: NEB:convergence issues.

Posted: Sun Apr 20, 2014 1:36 pm
by graeme
It does not sounds like this is specifically related to the NEB or climbing image because the problem is with the electronic optimization.

The first thing that I would check is the geometry of the climbing image. You can pull that POSCAR out of the calculation and try to calculate the force and energy in a regular vasp calculation. I expect you will see the same problem. But first, make sure that no atoms are on top of each other due to the construction of the band. If they are, move them away from each other, either by hand or with our nebavoid.pl script.

If the geometry looks reasonable, then I would suggest some of the things that you tried related to the electronic structure optimization. Reduce the mixing parameters, and if really stuck, used the damped dynamics routine. If there are any issues of the initial spin, start with a net spin of zero.

If you are really stuck and you think it does have something to do with the NEB, you can post it so we can take a look.

Re: NEB:convergence issues.

Posted: Fri Apr 25, 2014 9:48 am
by mishralu
Dear Graeme,
Thank you very much. There was overlap of 2 atoms in my structure, though I have corrected this by manually moving atoms and at present calculations are running, Could you please let me know the right way to run nebavoid.pl, I tried to run this code but I was only getting POSCAR_orig file only not the POSCAR file(modified file). I first created 5 images (00-06) folders in a parent folder using nebmake.pl and then running nebavoid.pl.
Thanks and regards,
Abhishek