My Community

Let's say you relax a system in VASP and everything seems to go well... you do a "tail -20 OUTCAR" to see "reached required accuracy" ....i.e. the ions have converged to within your, say, force requirements.

But what if one or more of the electron minimisation steps did not converge? Will VASP "complain"? Say you set NELM to 100 and several of the electron minimisations got to 100 and stopped... and then the ions were updated using a non-converged potential.

I could imagine that maybe in a worst case scenario that the ions could accidentally get kicked out of a local minimum (because of "bad" forces coming from the non-converged electronic potential) and then the final answer being quite different from what one would have gotten had they set a larger NELM and always had ionic updates using converged potentials.

So that is a long way of asking this: does someone have a list (short or long) of things they check after a run (maybe you check during the run!) to make sure that the answers they got are sane? I konw I'm sort of sneaking in a big question here... but checks like this can make a big difference not only in saving computer time (stopping a run that's not working) but also but also in preventing bad data.

Thanks...
Jason

no, vasp does not give any warning if a geometry does not converge within the given NELM steps, you have to look at the numbers (dE in OSZICAR) yourself.

The only concern is exactly what you mentioned, if the electrons are not converged well then ions may move in the wrong direction. I'd rather try to make sure that electrons always converge well.

I faced this problem in using Hybrid Functionals, the cure was to preconverge using standard PBE and then get a WAVECAR to start the HYBRID calculations. So, a good strategy is to obtain somehow a good WAVECAR to start with.
Another option is to play with electronic convergence algorithm. (ALGO, IALGO tags).