NEB get a larger diffusion barrier

[Tiger-paw]

I am calculating the diffusion barrier of a solute in BCC exchanging with a nearest-neighbor vacancy site. The diffusion path is well defined as [111], but I still used two methods: One is simply by total-energy sampling along the solute-vacancy direct path [111] - during relaxation, only the solute was fixed at each sampling image; The other is through NEB - for convergence, both force and energy criteria result in the same barrier curve.

I am surprised to see that NEB yields a much larger barrier, although NEB still reported a same path as [111]. Shouldn't I expect a minimized barrier path after NEB?

[graeme]

One possibility is that the non-solute ions are participating in the reaction. If so, it is very dangerous to assume that the reaction coordinate involves only the solute atom. In your drag approach, you could be relaxing on either side of the saddle, and never really constraining the system to the saddle point.

To check this, you can see if any of your constrained relaxation configuration have a small total force (including on the solute atom). If these forces are large, you are not near the saddle. A good second test is to evaluate the normal modes and see if you have one negative mode.

[Tiger-paw]

Thanks for your reply, Graeme.

So you believe in the NEB results more than the drag results. I fixed the solute atom along a certain path during relaxation, so that I could correlate the obtained energies with specific displacements along that path. Otherwise the solute may move around, and one can hardly get the energies vs. displacements curve afterwards. Moreover, some images might possibly simply relax back to a same local-minimum position. You think this constrainted drag-approach resulted in the wrong barrier, but if so, shouldn't this approach yield a higher barrier (than NEB), supposed that the resulted saddle point were not the true saddle?

BTW, I noticed several forces are tabulated in OUTCAR after NEB: TANGENT-FORCE, CHAIN-FORCE, TOTAL-FORCE, CHAIN+TOTAL-FORCE. There is another force reported: "tangential force". Which force or all forces should be literally zero by a right NEB for those images and the true saddle point?

Thanks in advance for any incoming inputs

[graeme]

An interpolated path can result in a barrier that is higher or lower than along the minimum energy path. It can be lower if your interpolated points either skip the saddle region, or, in the case of a predefined (drag) coordinate if the system is not well constrained in saddle region.

In your calculation, for example, fixing the solute and allow the rest of the system to relax can result in structures that always look like minima. They will be precisely minima if there are no other frozen atoms. This is an example of a case where the system is not being constrained to the saddle region.

The total NEB force goes to zero at convergence. But I was suggesting that you take the highest energy image along your drag coordinate. Check if the forces are zero - including those on the solute atom - and if you have a single unstable mode. This will determine if you have a first-order saddle. I suspect that you do not in your drag calculation and this is why the barrier appears lower than the NEB calculation.