NEB calc won't converge with VASP minimizers, fails with VTST

[olderandcolder]

[attachment=0]FORUM.tar.bz2[/attachment]Hi there,

I am studying a strained system with NEB, using the VTST implementation.

When I use the vasp minimizer (IBRION =1 ) I can't get the forces down below my convergence criterion of 0.03 eV/A, so I thought I would see if the VTST minimizers would manage it. When I try them the forces in the system initially reduce, but then begin to increase, eventually ending up in a very unrealistic geometry.

(Actually the LBFGS run failed with the following error: internal ERROR: RSPHER: NONLR_S%LMMAX is wrong )

I was wondering if anyone had any thoughts on how to reduce the forces further (and even if the system will allow for it. My feeling is that it probably won't due to the strain, and in fact the max force is always on the atom that is in transition).

I have included the set-up and output and should anyone have the time and inclination to have a look I would be grateful for any bright ideas!!

Thanks!

[attachment=0]FORUM.tar.bz2[/attachment]

[graeme]

I don't think this has anything to do with the optimizer.

First, figure out why there is a change of 15 eV across your band. Check your settings, and specifically your initial spin.

In the attached, I reconverged your initial and final states using what I would consider more reasonable settings, including PAW potentials, normal precision, a cutoff closer to 300 eV, and orthonormal cell, and a reduced kpoint mesh. The resulting endpoints are closer to iso-energetic and the band converged rapidly to a reasonable path.

[olderandcolder]

Hi Graeme,

Firstly thanks very much for taking the time to have a look.

My start and end points were more or less iso-energetic - maybe you have looked at the final energy resulting from my attempts to optimise using the VTST optimisers ( though the same happens with the vasp IBRION = 3 optimiser), which led to a very unrealistic geometry for some reason. While the IBRION = 1 vasp optimiser didn't, but wouldn't get the forces down below the convergence criterion. For these settings the barriers are very reasonable, and don't fall apart in the same way.

In any case, I will switch to PAWs to test and see if that helps the situation, as it certainly looks like it helped when you used them and seems to be the only significant difference. (the reason I was using ultrasofts was for a direct comparison to other another work examining the unstrained case which used them).

Thanks again - much appreciated!