ciNEB barrier plot not smooth!

[mingyang644]

I am calculating the ciNEB using 7 intermediate images.
It is very slow(3000 steps) to get to convergence.
My criteria is EDIFF=1E-4, EDIFFG=-0.04.
I attached the result, but notice the POSCAR in the file is the one after convergence.
The main issue now is I can not get a smooth plot for the barrier.
[attachment=1]mep image.jpg[/attachment]

Could you please have a look at my plot and give me some suggestion?
Thanks.

[graeme]

I would check to see if there is any net translation or rotation in your system. Freezing at least one atom can help to prevent this (if that is indeed the problem).

Also, I can't tell how the neb.dat file was generated. Was it using the nebresults.pl script?

As far as the convergence rate, I would have to see a .tar.gz file of the calculation to investigate.

[mingyang644]

[quote="graeme"]I would check to see if there is any net translation or rotation in your system. Freezing at least one atom can help to prevent this (if that is indeed the problem).

Also, I can't tell how the neb.dat file was generated. Was it using the nebresults.pl script?

As far as the convergence rate, I would have to see a .tar.gz file of the calculation to investigate.[/quote]

There are translation in the system, such as an organic linker moving away from a metal cluster due to the bondbreaking.
The neb.dat is generated using nebresults.pl.
I attached the tar.gz file of the whole calculation from starting image.

The POSCAR1 in each folder is the initial image.

[mingyang644]

Hi graeme,
I attached the neb.tar.gz file for my calculation.

I still can not get smooth barrier plot.

Thank you very much.
Best

[graeme]

Ok, I took a look and I see that there are a lot of soft modes in the system. The distance along the path is also very long, because of the accumulated oscillations along these soft modes. So the problem with the mep.eps plot is that the energy difference between the points divided by the (large) distance less than the force perpendicular to the path, which determines the slope at each point. Unless you really care about the path, I suggest continuing with the climbing image until you have a converged saddle. At that point, you can also minimize along each direction from the saddle to trace out the MEP.

[mingyang644]

Hi graeme,

Thank you for your suggestion.
The result I showed here is already converged at EDIFF=1E-4, EDIFFG=-0.04.

You suggest me to continue the calculation, do you mean I need to use more strict criteria?
1.What kind of criteria do you think will be good?

2.Do you think adding more intermediate images will be helpful to deal with the large distance?

Thank you very much.
Best

[graeme]

The force on the climbing image is 0.034 eV/Ang, so it is not really at a saddle. I suggest dropping ediffg to -0.01.

There is no real problem with your barrier calculation; the improved convergence will give you a converged activation energy. At that point, adding intermediate images will not improve the (converged) barrier, or the smoothness, since this is related to the soft mode, and not the resolution along the band.

Try minimizing image 04 to see what kind of intermediate minimum exists along the path.

[mingyang644]

Hi graeme,

Thanks for the suggestion, I will try that.

Here is my INCAR file, do you have any suggestion about the parameters?

SYSTEM = MOF5_H2O
ISTART = 0
ICHARG = 2
ISMEAR = 0
SIGMA = 0.1
PREC = Med

ENCUT = 500
IALGO = 38

LPLANE = .True.

LREAL = Auto

ISIF = 0
IBRION = 3
POTIM = 0.05
SMASS = 0.4
NSW = 200
EDIFFG = -0.01
EDIFF = 1E-4
LWAVE = .False.
LCHARG = .False.

IMAGES=7
SPRING=-5
ADDGRID= .True.
LCLIMB = .True.


GGA=ML
Zab_vdW = -1.8867
LUSE_VDW=.TRUE.
AGGAC = 0.0000