NEB calculation IBRION=3 vs individual calculation with IBRION=2

[gu301purdue]

Dear colleagues,

I have a question about NEB calculation with VASP. I set IBRION=3 for all the calculations, and get the energies (E1, E2, etc) for each image (transitional structure); however, at the same time, I also tried to get the energy of each transitional structure by running regular structural relaxation (with INCAR, POSCAR, POTCAR, KPOINTS ready in one folder), and IBRION is set to 2. However the energies I got (E'1, E'2, etc) are inconsistent with NEB calculations (E1 does not equal to E'1). Why there is a difference and which method is more reliable?

Thanks!
Tina

[fabien_tran1]

Hi,

If E1 and E1' are not equal it is probably because of the two geometries that are not the same, right? During the NEB calculation there is some constraint on the geometry that is applied, while it is not the case for a regular structural relaxation.

[gu301purdue]

Thanks! That's a good point. The geometry is different slightly, but the input structures are the same, and for both input structures I fixed the positions of one atom that could move around (the rest of the atoms stay in their original positions and relatively stable).
So the question is: which E I should trust? The final results of the free energy with IBRION at 2 and 3 are quite different.

Thanks!
Tina

[fabien_tran1]

Thus, provided that the parameters in INCAR (except those related to NEB, i.e. IMAGES) are the same for the two calculations, then the E1-E1' difference should come only from the different geometries.

What is the value of E1-E1'? Is it large enough to affect significantly the final calculated property and therefore to affect the conclusion of the study? Is there in the literature not such discussions about this "dilemma" in the context of NEB?

[gu301purdue]

Thanks for following up! The difference between the first two images (E1-E'1 and E2-E'2) is relatively large, ~4 eV for each, which I think is quite a lot, but the trend is consistent. I didn't know if there was any previous discussion about this issue, and I thought this might be something in common with different systems, which is why I posted questions here to see if this issue happened before.

Thanks!
Tina

[fabien_tran1]

4 eV seems large, but this seems reassuring that the trend is ok according to you. Besides, this is a good thing to practice NEB to get more insight about the method and to know the details.