solid state CI-NEB, not converge

[graeme]

The fact that you have a number of images at the same energy as the reactants and products indicates that the system may be translating. This will lead to a poor spline interpolation (since the distances between points are too large). Try freezing an atom, or some equivalent constraint so that translation (and rotation) are prevented.

[rlzhou]

Yes, when I watched the movie, I found the first three images really just translate. I'll try to fix one atom. The question is whether I sould fix that atom in all images or just tat in the first and last images?

[graeme]

All images, in the same location.

[rlzhou]

In my another job, there is no translation between images. However, I still get a mep curve where the highest energy image does not sit on the saddle point. I thought may be I select two few images (5), so I increase the number of images to 8. However, I get almost the same curve. The same cases are found in other jobs. I wonder whether the climb NEB are implanted correctly into the ssNEB. Do you have any suggestion to resolve this problem?

The mep curves are attached.

[rlzhou]

Another question. Must I use IOPT=3 and IBRION=3 in solid state NEB calculation. Can I use other optimization method?
The normal fixed-cell NEB calculation using IBRION=2 can get perfect mep curve.

[graeme]

You should be able to use any of our optimizers with the SS-NEB (by setting IOPT) as described here: http://theory.cm.utexas.edu/vtsttools/optimizers/

The one setting that you describe, using the built-in conjugate gradients IBRION=2, is not appropriate to use with the NEB. While it might have converged in your case, it is not generally safe to use because the line optimizer uses the energy as well as the force. In the NEB, the forces are not conservative and the energy is not consistent with the force. Thus, the bracketing method in the line search can fail.

[rlzhou]

I'm studying the phase transition using ss-NEB from graphite to a carbon three-dimensional structure. I did it under different pressure. I found that under 10GPa, the mep curve is excellent where the line is smooth and the highest-energy image sit on the saddle point. However, under 30GPa, the highest-energy image doesn't sit on the saddle point. And the curve seems to be not very smooth. I checked the OUTCAR. In both case the stresses converge well. So, what's the possible reason for the problem under a higher pressure?

[xph]

Could you share the neb.dat file?
If the force converged but the path is still not smooth, adding more images usually helps.
Also try to start with your the path converged at 10 GPa. Don't forget to minimize the end points again at 30 GPa.

[rlzhou]

10GPa, nebss.dat
0 0.000000 0.000000 0.000000 0
1 0.102707 0.004305 -0.014135 1
2 0.207701 0.020212 -0.109129 2
3 0.314128 0.063382 -0.349505 3
4 0.416795 0.127964 -0.517634 4
5 0.518140 0.174005 0.000045 5
6 0.612178 0.138514 0.926411 6
7 0.706679 0.056008 0.751208 7
8 0.801659 0.042941 -0.001375 8
9 0.897025 0.043444 0.000000 9

30GPa
0 0.000000 0.000000 0.000000 0
1 0.086610 0.008906 -0.041874 1
2 0.173283 0.052320 -0.226542 2
3 0.257584 0.095037 -0.000443 3
4 0.324234 0.094551 0.619449 4
5 0.393998 0.045058 1.293884 5
6 0.472430 -0.030205 1.026625 6
7 0.558317 -0.032268 -0.019924 7
8 0.647030 -0.032868 0.000017 8
9 0.736243 -0.032599 0.000000 9

[xph]

Seems like another minimum shows up on the path. I would suggest minimize the 6 image at 30 GPa and use it as the new final state.
What happens for the last 3 images? Is this because of translating or the tangent along some ultra soft mode?

[rlzhou]

No matter what I do, I can not get a smooth mep curve where the highest image sit on the saddle point under 30GPa. Under 10GPa, I always get curves with images sit on the saddle points. It seems that under high pressure, the climb neb method not work well. I'll calculate the frequency to check wether the highest energy image only have one negative frequency.

[xph]

Apparently, your climbing images converged to the saddle point, which can be seen from the force in neb.dat (the second last term, climbing image is close to zero). The problem you saw is from spline. That means the path near the saddle needs to be zoomed in. It is mainly because there is a level part at the end of the path, which wastes your resolution. So I asked you why it is like that, translation or soft mode? Since your 10 GPa path is not very smooth either (force is small in the beginning, seems like the path is stretched), I suspect you still have some problem there.
I will also check our code with another example.

[rlzhou]

There are image translating in my previous calculation. I recalculated them again by fix one atom in each image. Now, there are no tranlating. However, there is still no image sitting on the saddle point. I calculated the phonon of the highest energy image and found there are two negtive frequency. So, it is not the saddle point. The force on the highest image is also very small. Where the forece is read from? I can not find the force listed in nebss.dat in the OUTCAR file. The nebss.dat is as below:

0 0.000000 0.000000 0.000000 0
1 0.111224 0.039074 -0.162204 1
2 0.221582 0.090756 -0.002338 2
3 0.316085 0.082286 0.762257 3
4 0.415262 0.016162 1.037590 4
5 0.519767 -0.036253 0.490129 5
6 0.627296 -0.043628 0.000000 6

[xph]

That second last terms are the magnitude of the real forces on images. You can find details of how to grep and calculate them in nebbarrier.pl
For the frequency calculation, make sure that you excluded the three translation modes, whose frequencies are close to zero and can be imaginary, when you count the number of negative mode. If there are two real negative mode, then you find a monkey saddle point in a solid state system. That would be very interesting.

[rlzhou]

I used DFPT in VASP to calculate the phonons at gama point. I got two imaginary frequency. One is 190cm-1, the other is 301 cm-1. I don't think there are problems in my calculation of phonons. I have done the neb calculation on some different systems. Under 10GPa, for all the cases, I can get good mep curves where an image sitting on the saddle point. However, under 30GPa, for all the cases, I can not get curves with image sitting on the saddle point. I'm trying to increase the number of images and set the EDIFFG= -0.01 to check whether the problem can be solved. Did you encounter similar problems when you calculating expample systems?