Band crossing during MD

[heisen]

Dear All,
I am running a NVT MD run on CaCO3, I see a repeated message in the log file of my run: wavefunction orthogonal band
I understand this can occur from band-swopping. I checked and the bands that the message is written about are empty bands far above the Fermi level. I read that using IWAVPR=11 can solve this problem, and indeed when I run with IWAVPR=11 I do not see the error message. My question is, in the manual it says very clearly IWAVPR=12 should be used for MD runs. Which should I use? Can I ignore the band-swopping in the empty bands and use IWAVPR=12?
Thank you for your help,
Dr. Helen Eisenberg

[martin.schlipf]

What you should expect is that IWAVPR=11 is more expensive. When you use IWAVPR=12, VASP will extrapolate the density and the orbitals to the new positions. With IWAVPR=11 only the density is extrapolated. Therefore in the latter case the optimizer needs to do more work to get the orbitals to agree with the positions of the atoms. It may be that these additional steps help to overcome the warnings you observe.

So in general, I would recommend using IWAVPR=12 for pure ab-initio MD simulations because you need the additional speed. However, if you are interested in ab-initio MD, I would also strongly recommend to look at the machine-learning approaches VASP offers. You should not run any production calculation without machine learning unless you have a very good reason not to. For ML-enhanced MD, the recommendation may be not to extrapolate the orbitals because the next position may be very far away due to the intermediate MD steps.

[heisen]

Thank you very much for your reply