CI-NEB: Hubbard U term influences the result a lot.

[yangxx]

Hello professor Henkelman and dear community,

Now I'm using VTST (version 3.2) interfaced with VASP 5.4.4 to calculate ion/vacancy diffusion energy barriers using CI-NEB method in three quite similar structures, and intend to cross compare the results. But I met some issues to obtain converged results with consistent parameter setting. Specifically, if I have parameter setting working well for two of the structures, it doesn't work for the third structure. The difference
in parameter setting mainly lies in whether to include U term (for transition metal) during CI-NEB calculation. I have attached one calculation without U term with negative energy barrier. The parameter setting works well in another two structures (which are not attached in this post).

What I did for CI-NEB calculation is that
1. Optimize the pristine structure with U term;
2. Insert the ion/vacancy in the optimized structure from step 1(CONTCAR), optmize the structure with the fixed cell volume. In this step, U term is either turned on or off.
3. Perform CI-NEB calculation as instructed. Whether to turn on U term corresponds to what has been done in step 2. If step 2 includes U, CI-NEB includes U as well. The remaining parameter setting keeps the same and CI-NEB calculation parameter is set.

What I have checked and the ways I tried to solve the issue before this post are as follows.
1. Checked literature to make sure that the initial and final state are indeed the starting and the ending point along the diffusion path.
2. The U value is in the range of those setting in the literature for the element.
3. The atom sequence in the initial and final state is the same to make sure that only moving species is different.
4. Turn off the symmetry with ISYM = 0 (this has been done in both geometry optimization of initial and final states and CI-NEB, and also has been done just in CI-NEB calculation);
5. Tried different optimizers. Previously, I have used IOPT=1, I also tried IOPT=7 and IOPT=3.

What's more interesting as you can see in the attached file, the intermediate image is very close to the final state but the energy is even more negative. I don't think it is becasue that the final state provided here hasn't been fully optimized. So this makes feel quite confusing.

As mentioned before, if I have U term included with other parameters the same during both step 2(geometry optimization) and step 3(CI-NEB calculation), the negative energy issue can be solved in this attached structure but didn't work for another quite similar structure, making it hard to cross compare the energy barrier result. The reason why I care about U term is becasue I have seen in some paper mentioning that U doesn't help on energy barrier but sometimes make convergence hard. I'm not sure if there is any cases like this before and I wonder if you could please go through my case here and give me some advice.

Thanks,
Xixian