Page 1 of 1

NEB vs CI-NEB differences

Posted: Tue Sep 09, 2014 4:29 pm
by chrisH
Dear all,

I have been trying to reproduce the pathway for a simple hydrogenation step for ethylene over a metal surface, which was calculated previously with NEB by a previous group member.
When repeating the calculation with NEB (in VASP-5.2.12-vTST) I can reproduce the pathway and maximum image energy, but when I apply the climbing image modification to the calculation, strange things occur.


In one system (C2H4 + H over Au (111)), using the output of the NEB calculation as input for the CI-NEB, I find the NEB pathway to be notably lower in energy than the final CI-NEB one, but in this case the CI-NEB had found a true saddle point, and the NEB had not. This would be a pleasing result for CI-NEB, except when I look at the geometries in the images, they are virtually identical, and so I can't reconcile this with the significant difference in energy between the two (0.32 eV).

In another, similar system (C2H4 + H over Os(111)), I again used the output of the previous NEB calculation as input for the CI-NEB. In this case, both the NEB and CI-NEB highest energy images are found to have one imaginary frequency after vibrational analysis (with CI giving a 0.25eV higher saddle point energy). This seems to imply the CI-NEB finds a higher energy path, and misses a lower energy TS, despite starting from the previously converged NEB pathway.

I'm not certain what the conclusions I should draw from these two calculations should be, or indeed what information to add to this post. I find over a whole range of metals that CI gives significantly higher energy pathways without changing the path noticably. If someone does have some advice, I can add images, energies, input, etc as necessary.

The INCAR files are like so:

===============================
ENCUT = 420.000000
SIGMA = 0.07
EDIFF = 1.00e-5
PREC = Normal
ISPIN = 1
ISMEAR = 0
NSW = 700
NELMIN = 6
NELM = 200
AMIX = 0.08
MAXMIX = 100
LCHARG = .FALSE.
LASPH = .FALSE.
LWAVE = .FALSE.
LREAL = Auto
ALGO = Fast

IMAGES = 8
SPRING = -5
NPAR = 3
ICHAIN = 0
LCLIMB = .TRUE.
EDIFFG = -0.05
IBRION = 3
POTIM = 0.10
=========================

I really am stuck on this.
Thanks for your time!

Chris

Re: NEB vs CI-NEB differences

Posted: Wed Sep 10, 2014 1:17 am
by graeme
I would start by checking that the settings in the INCAR files for the endpoint calculations, NEB calculations, and CI-NEB calculations are identical. Also, turn off symmetry using ISYM=0. Check that the KPOINT settings are the same as well as any fixed atoms in the geometry files. If you used a different version of vasp and/or different pseudopotential files, recalculate them so that everything is consistent.

If you still have a problem and believe that it is related to the NEB code, please post the calculation so that we can see what's going wrong.

Re: NEB vs CI-NEB differences

Posted: Wed Sep 10, 2014 9:38 am
by chrisH
Thanks for getting back to me.

The endpoints, and both NEB and CI-NEB have identical KPOINTS, POTCARS and fixed atoms.

The CI-NEB and NEB calculations have identical INCAR files to eachother. The endpoints have slightly different INCARS, with the same energy and force cutoffs as the band calculations, but with MP smearing of order 1, rather than the Gaussian smearing of the NEB and CI-NEB. This was changed to help the NEB pathways converge. However, both NEB and CI-NEB use the same endpoints, so I wouldn't have thought the difference would manifest itself between the two string pathways (perhaps this is not true?).

The NEB and CI-NEB calculations both use the same version of VASP (5.12-vTST), and really should be identical in every way except for the switching on of the climbing image.


I will try rerunning with ISYM=0, and see if this brings NEB and CI-NEB in line with eachother. If not, I'll post the calculations as you suggest.


Thanks,
Chris