Why does the force keep falling below EDIFFG when running NEB?

[tjgus6142]

Hello,

I'm calculating NEB of charged molecules. I set the EDIFFG to -0.1 and the force fell well below 0.1 in all images except one (image 05).
The curious part is that when I checked the forces of every images using "nebef.pl", some images have much lower force than the criteria like 0.03 of images 01 and 04.
I expected that optimization would be completed for images whose force was lowered than 0.1, and optimization would continue until only images that do not satisfied the criteria fell below 0.1.

Does the calculation also continue for images that already meet the EDIFFG until all images are converged?

I attached the INCAR file and the output of "nebef.pl" is shown below.

$ nebef.pl
>>>
0 0.049068 -153.119900 0.000000
1 0.039224 -153.398000 -0.278100
2 0.097189 -153.896200 -0.776300
3 0.067681 -152.957900 0.162000
4 0.035775 -154.167600 -1.047700
5 0.123284 -153.075400 0.044500
6 0.047936 -152.962800 0.157100

[michael_wolloch]

Hi tjgus6142,

could you please attach a minimal working example with all the necessary input files to run? It would be great if your minimal example is something I can run rather quickly, so maybe a smaller version of your problem, or with reduced k-point density, but still something where you observe the issue. Otherwise, I cannot try and reproduce the issue, and I cannot be sure what is happening. Your OUTCAR and stdout would also help.

I see that you have set LCLIMB = .TRUE. in your INCAR file. Are you using VTST tools? If yes, are you sure that you installed them correctly?

Generally, if EDIFFG is negative, not only the forces are checked, but also the stress tensor has to be below a certain threshold. This is not well documented on the wiki, something we will change. But for your calculations, the cell shape and volume are fixed, so I doubt that this is the problem here.

Thanks a lot, Michael Wolloch

[tjgus6142]

Hi michael_wolloch,

Because of the filesize issue, I compressed only 00(initial), 04, 06(final) directories containing CONTCAR, and OUTCAR. Furthermore the INCAR, KPOINTS, POTCAR, run files for NEB calculation are also compressed. I hope this gives you sufficient information. If you need more files or information please let me know.

I optimized the initial and final images with the tag "EDDIFG = -0.05". So it makes sense for the force to come out below 0.05, but when I turned NEB, I designated EDDIFG as -0.1, so I thought the calculation would end when the image from 01 to 05 came out about 0.09, but I wonder that the image from 04 fell to 0.03 in force.

Yes, I use VTSTtools and considering that there were no problematic issues when using vtstscripts, I think it is installed well.

In addition to the force, you said that the stress tensor should fall below a certain threshold, is that specific threshold also a value that can be specified by INCAR? I wonder what does the stress sensor mean and how can I check it in OUTCAR.

Thanks in advance!

seohyeon

[michael_wolloch]

Dear Seohyeon,

I think I was too focused on technical things initially. I think the answer is quite simple: As long as one image is moving, spring forces will change on the neighboring images, causing them to move. This translates until the fixed ends of the band. Thus no calculation can stop independently because they are all coupled together and all calculations can only finish if every calculation is converged.

So what you see is completely normal if I am not mistaken.

The stress tensor components for each ionic step can be found by searching for

Code: Select all

FORCE on cell =-STRESS in cart. coord.  units (eV)

in the OUTCAR file.

[tjgus6142]

I totally understood.
Thank you for the explanation!