what is the appropriate convergence criterium for CI-NEB calculations?

Message

tracy · #1 Post by **tracy** » Mon Apr 20, 2009 10:34 pm

Dear all,
I used CI-NEB calcualtions to study diffusion, I applied EDIFF=1e-05 and EDIFFG=-0.01 as the convergence criterium. First Using IBRION=3 and a small POTIM=0.01 to run 20 cycles; Then I use a more larger time step POTIM=0.3 and IBRION=1 to run 200 cycles. While after so many cycles, I couldn't get it converged. I'm wondering if I used the right convergence criterium???? If so, then what is the proper value for EDIFF and EDIFFG? will it still be reliable calculation if I choose EDIFF=1e-04 and EDIFFG=-0.1? Thanks! Any comments are welcomed! If you can give me a strategy to get CI-NEB calcualtions converge quickly, I'll be very appreciated! ^_^
The detailed INCAR is as follows:
INCAR:
SYSTEM= transition state
PREC=Med
EDIFF=1e-05
EDIFFG=-0.01
ISMEAR=1
SIGMA=0.1
VOSKOWN=1
IBRION=3
#IBRION=1
LCLIMB=.TRUE
ICHAIN=0
SPRING=-5
LREAL=Auto
IMAGES=11
ISIF=2
POTIM=0.01
#POTIM=0.3
NSW=100
ISPIN=2
MAGMOM=32*4 0
ENCUT=350.0

graeme · #2 Post by **graeme** » Tue Apr 21, 2009 1:06 am

Always use a force criteria (EDIFFG<0). Your criteria of 0.01 eV/Ang is quite reasonable. You will probably not see much change in barrier heights if you go up to ediffg=-0.05. EDIFF just needs to be low enough so that the forces are accurate enough for the specified optimizer to work properly. Again, 1e-4 is fairly safe.

IBRION=3 is a conservative optimizer, but it is not particularly efficient. Try IBRION=1, or the methods described at http://theory.cm.utexas.edu/vtsttools/optimizers/ which are designed to work efficiently with the NEB. You may have to lower ediff for curvature-based (second order) optimizers.

Franky · #3 Post by **Franky** » Thu Apr 23, 2009 10:42 am

From my experience (CI-)NEB calc. converge much slower than
regular relaxation for a (local) energy minimum unless the PES
is rather "simple" in its structure.

Make sure your calc. is actually converging, i.e. the forces are decreasing. Sometimes a PES might just be flat in certain
regions. This can look like convergence!
If the forces or positions oscillate or behave noisy,
one needs to be careful. Using IBRION=3 might help.
For my part, I am having difficulties with optimizers
other than IBRION=3 even if I use EDIFF=1e-08.
Hope this was helpful.

tracy · #4 Post by **tracy** » Fri Apr 24, 2009 4:16 pm

Many thanks to graeme and Franky!!!

To Franky, what's the meaning of PES? Thanks!

Franky · #5 Post by **Franky** » Tue Apr 28, 2009 3:02 pm

PES = potential energy surface

tracy · #6 Post by **tracy** » Fri May 01, 2009 3:28 pm

Thanks for all your replies!
one more question: what's the proper output file we should look at to see the convergence of EDIFF and EDIFFG? Thanks!