CI-NEB don't coverge to MEP...

whzhang81 · Post by **whzhang81** » Mon Jul 17, 2006 2:13 am

I want to calculate a surface reaction between two small molecules. Below it my input file:

SYSTEM = surface reaction

LREAL = .TRUE.
PREC = Norm
EDIFF = 1E-04
EDIFFG = -0.03
NELMIN = 6
ISPIN = 2
VOSKOWN = 1

ISMEAR = 1
SIGMA = 0.01

ICHAIN = 0
LCLIM = .TRUE.
SPRING = -5
IMAGES = 8

ISIF = 0
POTIM = 0.1

IALGO = 38
NSW = 3000

IBRION = 1

#ICHARG = 11

LWAVE = .FALSE.
LCHARG = .FALSE.

After about 400 steps, it doesn't coverge yet. The 2nd image has the higgest energy. The forces of image1 , image5, image6, imge7, image8 are under 0.03ev/A. But those of image2, image3, image4 are still large ; some of them are bigger than 1.0 .

So what can I do to accelerate the covergence?

Thanks a lot.

Post by **graeme** » Mon Jul 17, 2006 4:14 am

We would need more information to see what's going on. Can you post a link to a tar file of the run, so we can see what's happening.

Also, I see you have LCLIM instead of LCLIMB in your INCAR file, so you are not using the climbing image.

andri · Post by **andri** » Mon Jul 17, 2006 9:17 am

Also, it is usually better to start with IBRION = 3. What is the length of the path? When estimating the local tangent to the path then it is best not to have the image too far from each other. Personally I try not to have images further than about 0.5 A from their neighbors.

whzhang81 · Post by **whzhang81** » Thu Jul 20, 2006 2:30 pm

I have changed the input file according to your suggestion, but it seems that the calculation still coverge slowly...

You can get my input files and the output file through below address . It is about 37M.

Thanks a lot.

[admin note] the original link didn't work. The files for the run can be found at: http://theory.cm.utexas.edu/neb_run.tar.gz

Post by **graeme** » Fri Jul 21, 2006 7:29 am

Well, it's an interesting run.

All the dynamics are happening in the first few images, so there is fairly poor resolution of the (complex) reaction coordinate.

The small surface also appears to be shifting a lot during the reaction. I wonder how much of this was due to an overly aggressive optimizer when you started the NEB. I would be interested in seeing how the following run goes:

Start again from a linear band. You can probably use image 6 as your final state. Still use 8 images for this shorter pathway -- this will give you better resolution during the reaction. Start by using a very conservative IBRION=3, POTIM=0.01 until the forces drop between 1.0 and 0.5 eV/Ang. Then you can raise POTIM to 0.1. Make sure that the band is converging smoothly, even if it is doing so slowly. Check to see if the motion of the top metal layer is reduced. You can use the climbing image throughout both optimizations.

If this converges properly, you can try using the more aggressive IBRION=1 (once you have a semi-converged band. You might also try continuing the band you posted with IBRION=3, but to be safe, start from a clean, linear band first.

Other things to try:
IALGO=48 (better for parallel runs)
SIGMA=0.1 (you have a very small sigma)
LREAL=.FALSE. (you have a small cell, and this will improve your forces)

Good luck, and let us know how it goes.