Energetically lower transition state leads to nonphysical KRA

[tanmoy1989]

Dear Prof. Henkelman,

I am doing around 50 CI-NEB calculations on oxide perovskites, where we randomly choose a vacancy site and see its migration barrier to a nearby oxygen site. Out of 50 CINEB calculations, some of the calculations show that the transition state (TS) is actually lower in energy than the end-points leading to a non-physical kinetically resolved activation (KRA) energy value (eq. 14 in PHYS. REV. B, VOLUME 64, 184307, 2001).

I am wondering am I doing a technical mistake here as I don't expect TS to be lower in energy than the end-points. Intuitively, there has to be an energy barrier for an oxygen atom (or vacancy) to migrate from one place to the other.

Any suggestion is highly appreciated, Thanks in advance.

Tanmoy

P.S I have attached a particular CINEB calculation folder where I see that the TS is lower in energy than the end-points leading to a weird migration path and nonphysical KRA.

[graeme]

I think it has to do with the magnetic moment. You have a completely different magnetic state in our transition state structure.

By default, vasp starts with a very high magnetic moment and then attempts to relax it. If/when it gets stuck in a local minimum, (a problem which is exacerbated with DFT+U) you can find very different energies.

You can check if this is the problem by turning off spin and/or DFT+U and see if the barriers are reasonable. Then, to include spin, try setting reasonable initial magnetic moments on the metal atoms, using the magmom tag. You may have to do some handwork to get the lowest energy electronic state.

[tanmoy1989]

Thanks, that was an excellent point which I completely missed out. May I ask you a couple of queries here?

1. When you say that TS has a completely different magnetic state, do you do 'grep magnetization OUTCAR' and look at the 'number of electron' part to conclude that it has a different magnetic state than the IS and FS? I am a little confused here as I don't see the 'magnetization (x)' block in the OUTCAR to see the magnetic moment on each atom.

2. When I try redoing the DFT calculations, shall I turn off spin and/or DFT+U on both the IS and FS first and do the NEB calculation with the same setting (basically re-doing everything with non spin-polarized setting first and then include spin) OR I just take the spin-polarized IS and FS (what I already have) and then do a non-spin polarized NEB calculation first followed by spin-polarized calculation?

Thanks,
Tanmoy

[graeme]

1. Yes, you can grep the OUTCAR(s) to determine the magnetization.

2. You need to determine if DFT+U and/or spin polarization are important for your calculation. My point is just that your NEB calculation needs to be consistent with your endpoints.