NEB converge problem with different number of images

[musanna]

Dear Professor Graeme Henkelman,

I am trying to do a cNEB calculation of the Zn atom. I used 2 different setup ( 1st case: images =5, EDIFFG=-0.1, NSW=500 and 2nd case: images = 10 EDIFFG = -0.01,NSW=2000). The migration energy barrier plots are totally different and don't make any sense to me. Also, as nebbarrier.pl uses cubic spline to plot migration energy barrier, how many images are optimum to use?

Also, I am confused about how to optimize the endpoint. Is it cell relaxation with ISIF=3?

I attached a .tar file with all the data used. Can you please comment on this?

Sincerely,
Musanna

[graeme]

I don't see an 11 directory in your 10-image calculation.

[musanna]

Dear Professor Graeme Henkelman,

Thanks for the reply. I am adding the updated file with folder 11. However, I didn't add the OUTCAR in that as I am confused about how to optimize the endpoint. Is it cell relaxation with ISIF=3?

I am also adding the migration energy barrier plots for the two cases.

Sincerely,
Musanna

[graeme]

The endpoint in the second calculation is crazy, but I am suspecting that it is because you didn't provide a final structure (in directory 11).

For now, just start with a fixed cell - I don't see any reason to relax the cell for the transition that you are interested in.

[musanna]

Dear Professor Graeme Henkelman,

Thanks for the reply. I am working on adding the endpoint outcar using the following INCAR
General:
SYSTEM = O deposition on Zn surface
ISTART = 0 !begin from scratch
ICHARG = 2 !Default: ICHARG = 2 , if ISTART=0, the initial charge distribution are the atomic charges.
ENCUT = 400 !Default: largest ENMAX on the POTCAR file
ALGO = Normal ! IALGO=38
PREC= Normal
ISPIN=1 ! non spin polarized calculation
LWAVE = T (Write WAVECAR)
LCHARG = T (Write CHGCAR)

Electronic Relaxation:
EDIFF = 1E-4 ! recommeded value for high precision

Ionic Relaxation:
NSW = 2000 !Relaxation steps
NELM = 100 !the maximum number of electronic SC (selfconsistency) steps which may be performed
POTIM = 0.5 !scaling constant for the step widths
IBRION = 2 !ionic relaxation (conjugate gradient algorithm)
ISMEAR = 1 !for metal ISMEAR > 0 , Fermi smearing
SIGMA = 0.026 !0.026 ev = 300K
EDIFFG= -0.01 !Stopping criteria for ionic relaxation loop
ISIF = 2 !To optimize ionic positions within the cell only






However, I think the 10th image gives a wired value because suddenly the total energy without entropy suddenly changes to -251.94295477 ev whereas
for image 0: -265.08708763
image 1: -264.75564109
image 2: -264.76238955
image 3: -264.36111199
image 4: -264.11838228
image 5: -264.18269062
image 6: -263.93421908
image 7: -264.36963351
image 8: -264.62854703
image 9: -263.54035317

Can you kindly take a look if I am doing any mistakes?

Sincerely,
Musanna

[graeme]

I would have to look at the entire calculation - I'm almost sure there will be a different setting for the calculation of image 10 as compared to the rest of the images.

[musanna]

Dear Professor Graeme Henkelman,

Thanks for the reply. I am adding the updated simulation with the endpoint here. Still, the NEB plot seems wrong to me. Can you kindly give me some suggestions on it?

Regards,
Musanna

[graeme]

It seems like a combination of small mistakes.

It seems that you used the POSCAR in the product state (11) to generate your band. But then look at the relaxation for image 11 - the Zn atom doesn't want to be in the POSCAR geometry; rather, it moves well above the surface into the CONTCAR geometry.

Relax both the initial and final states and then make a new band between those converged structures.

[musanna]

Dear Professor Graeme Henkelman,

I completed another set of simulations but couldn't interpret the migration barrier plot and why there are multiple saddle points along the path. Can you kindly take a look? I am adding the simulation files and the plot here.

Also using nebefs.pl, the images (1,5,8,9,10) do not converge even after 2000 ionic steps! Can you please suggest how to improve this?
Image Force Stress Volume Magnet Rel Energy
0 0.00000000 0.00000000 1656.93 0.00000000 0.00000000
1 0.15917200 1.55049600 1668.46 0.00000000 0.67610000
2 0.07793200 3.85019500 1668.46 0.00000000 1.63420000
3 0.03457900 3.55009700 1668.46 0.00000000 0.79850000
4 0.04085600 9.85564400 1668.46 0.00000000 1.59940000
5 0.11284600 7.41968300 1668.46 0.00000000 1.19400000
6 0.04062600 1.94853500 1668.46 0.00000000 1.62790000
7 0.09243500 4.47679100 1668.46 0.00000000 1.15590000
8 0.29593300 5.84993500 1668.46 0.00000000 2.19710000
9 0.17488600 0.63649500 1668.46 0.00000000 1.58080000
10 0.10262000 3.83230800 1668.46 0.00000000 1.64640000
11 0.00000000 0.00000000 1668.46 0.00000000 0.87760000

Sincerely,
Musanna

[graeme]

Again, there is a problem with your initial band. If you look at the initial movie (attached) you can see that atom 22 moves vertically as it should from images 0 to 10, but then in image 11, it jumps back down to the 11.7 Ang plane in place of where atom 23 was in image 10. Then the ordering of the atoms in that plane are scrambled between image 10 and 11. Given that crazy initial path, the NEB path is over 80 Ang trying to move atoms around between their positions in image 10 and 11.

Again, minimize your initial and final states, which are images 0 and 11, and then make sure that you have a sensible initial path between them before starting the NEB calculation.

[musanna]

Thanks for the reply. The idea is the atom will be diffused away from the surface. That's why I removed the atom from the final image (atom 22 is deleted in image 11) and relaxed it then. Therefore, can you kindly suggest what should I do next?

Also, the intermediate images are not converging even after 2000 ionic steps.

Sincerely,
Musanna

[graeme]

You have not set up a reasonable initial band, and so you have no sense of convergence. You can remove and atom. Again, find reasonable minimized initial and final structures and then generate a band between them. You can post the initial path before you run the NEB if you want me to check it before you waste more computer time.

[musanna]

Hello Professor,

Thanks for the reply. I tried to optimize my initial and final image with a few trials. Though all atoms are not converging below 0.01 ev/A, the free atoms are converging around 0.01ev/A. I attached the plots and files here. Can I use these as reasonable initial bands? Also, for the final image, I kept the moving atom in the middle of the vacuum space rather than deleting it. Because nebmake.pl cannot work with different numbers of atoms in the initial and final files. I added the initial path using these structures and added it here. Can you please comment on these?

Sincerely,
Musanna

[graeme]

Unfortunately, it seems that once again your initial band is not reasonable. Do you have a way of visualizing it? You can use our nebmovie.pl and xyz in the vtsttools, or the ase-gui; whatever - you need a way to see your reaction path.

In what you posted, one Zn atom is moving up in the vacuum, which is what you want, but the slab is being compressed at the same time. This doesn't make sense.

[musanna]

Hello Professor,

Thanks for the reply. I checked the movie and the cell was shrinking. I optimized the structure again and added the initial structure here. Though all atoms are not converging below 0.01 ev/A, the free atoms are converging below 0.01ev/A. Should I consider the attached path reasonable to start neb simulation?

Sincerely,
Musanna