Dear Admin, users
I am trying to run O diffusion from fcc to hcp position on pt slab. I am having problem in convergence of my calculations even with low convergence criteria ( EDIFF = 1E-04, EDIFFG = -0.1, IBRION = 3, POTIOM = 0.1). The run goes through NSW steps and writes the wavefunctions but doesnt reach the required accuracy. please provide me some information on how can I solve this problem.
thank you
best
Hom
Convergence problem with CI-NEB
Moderator: moderators
Re: Convergence problem with CI-NEB
Here is the input files for the calculation.
thanks
Hom
thanks
Hom
- Attachments
-
- O-diffusion-Pt.zip
- Input files
- (108.49 KiB) Downloaded 1057 times
Re: Convergence problem with CI-NEB
I need the OUTCARs to see what happened in the calculation (the geometries look fine).
Re: Convergence problem with CI-NEB
Thank you very much for your reply Proff. Graeme. I have attached the OUTCARs (due to the size, I had to attach zipped OUTCAR separately for each image).
thank you
regards
Hom
thank you
regards
Hom
- Attachments
-
- OUTCAR-2.7z
- OUTCAR-2
- (978.41 KiB) Downloaded 940 times
-
- OUTCAR-1.7z
- OUTCAR-1
- (957.56 KiB) Downloaded 929 times
Re: Convergence problem with CI-NEB
More OUTCARS
- Attachments
-
- OUTCAR-4.7z
- OUTCAR-4
- (955.31 KiB) Downloaded 926 times
-
- OUTCAR-3.7z
- OUTCAR-3
- (960.11 KiB) Downloaded 926 times
Re: Convergence problem with CI-NEB
one more
- Attachments
-
- OUTCAR-5.7z
- OUTCAR-5
- (951.56 KiB) Downloaded 916 times
Re: Convergence problem with CI-NEB
There is certainly a major problem with the dipole correction. Notice that the energy of the images can jump by thousands of eV. Start by turning off the dipole correction and the symmetry (by setting ISYM=0). Then run IBRION=3, POTIM=0.01 for about 50 iterations. The forces should drop systematically. Then you can switch to a more efficient optimizer or a larger time step of POTIM=0.1.
There is no need for such a high k-point mesh for the initial convergence of a band. You can do pretty well with a 2x2x1 grid for getting the band converged. Then you can switch to a higher mesh and quickly refine it. Debugging a calculation like with with such expensive settings doesn't make sense in terms of computational efficiency. You can also using ISPIN=1 since there is no magnetization.
There is no need for such a high k-point mesh for the initial convergence of a band. You can do pretty well with a 2x2x1 grid for getting the band converged. Then you can switch to a higher mesh and quickly refine it. Debugging a calculation like with with such expensive settings doesn't make sense in terms of computational efficiency. You can also using ISPIN=1 since there is no magnetization.
Re: Convergence problem with CI-NEB
Thank you very much Proff. Graeme,
I followed your suggestion and did calculations. after turning of the dipole and w/o spin, Now I see that the force decreased and no jump in the energy. After initial run with IBRION 3 and POTIM 0.01, what would be the best optimizer (IOPT) in this case ? or should I use IBRION 1 and POTIOM 0.1?
Since, the slab is Asymmetric, IDIPOLE 3 tag need to be put back for later run?
thank you very much for your help.
best
Hom
........................................
ENCUT = 400
LCHARG = .TRUE.
LWAVE = .TRUE.
ALGO = Fast
ISPIN = 1
LREAL = Auto
#IDIPOL = 3
IMAGES = 5
SPRING = -5
LCLIMB=.TRUE.
Electronic Relaxation 1
EDIFF = 1E-05 :stopping-criterion for ELM
EDIFFG = -0.1
Ionic Relaxation
NSW = 100
IBRION = 3 :ionic relax: 0-MD 1-quasi-New 2-CG
ISIF = 2 :stress and relaxation
POTIM = 0.01
# NFREE = 2
ISYM=0
I followed your suggestion and did calculations. after turning of the dipole and w/o spin, Now I see that the force decreased and no jump in the energy. After initial run with IBRION 3 and POTIM 0.01, what would be the best optimizer (IOPT) in this case ? or should I use IBRION 1 and POTIOM 0.1?
Since, the slab is Asymmetric, IDIPOLE 3 tag need to be put back for later run?
thank you very much for your help.
best
Hom
........................................
ENCUT = 400
LCHARG = .TRUE.
LWAVE = .TRUE.
ALGO = Fast
ISPIN = 1
LREAL = Auto
#IDIPOL = 3
IMAGES = 5
SPRING = -5
LCLIMB=.TRUE.
Electronic Relaxation 1
EDIFF = 1E-05 :stopping-criterion for ELM
EDIFFG = -0.1
Ionic Relaxation
NSW = 100
IBRION = 3 :ionic relax: 0-MD 1-quasi-New 2-CG
ISIF = 2 :stress and relaxation
POTIM = 0.01
# NFREE = 2
ISYM=0
Re: Convergence problem with CI-NEB
As you get closer to the converged band, you can use more aggressive optimizers. For high forces above 1 eV/Ang, the IBRION=3, POTIM=0.01 is very safe. Below that, IBRION=3, POTIM=0.1 should also be fine but slow. IBRION=2, POTIM=0.1 is also pretty safe and generally faster. IBRION=1, POTIM=0.1 should be the fastest of those, and also the least stable.
You can also check out our own optimizers, for which we have a good deal of experience with the NEB:
http://theory.cm.utexas.edu/vtsttools/optimizers/
You can also check out our own optimizers, for which we have a good deal of experience with the NEB:
http://theory.cm.utexas.edu/vtsttools/optimizers/