[QE-users] Convergence problem with hydroxide

[RABONE Jeremy]

Hello,

  I have just started a project using QE, and have a question regarding a hydroxide ion in a box. When I first tried to do a relaxation with conv_thr = 1.0d-8, it started off like it was going to converge and then suddenly the convergence went all over the place. Changing all of the other parameters made no difference, but reducing
conv_thr to 1.0d-6 stops the scf step before the problem and the calculation completes. There also seems to be some effect of the randomised starting wavefunctions even with conv_thr = 1.0d-6 (sometimes converges, sometimes goes all triplets). When surrounding the ion with a water cage, the calculation works with no problems. If I switch to a spin polarized calculation, with conv_thr = 1.0d-8 it seems determined to converge to a triplet state with the first antibonding orbital occupied, and does so with no difficulty is the magnetization is set to 2.

I recall there were sometimes issues in DFT when the order of state energies changes during a calculation - is this what is happening here and is there a way to get passed the problem or do I just have to be content with a reduced threshold ?

A S2- ion in a box does something similar - seemingly converging to triplet with a d-state occupied, and even when surrounded by a water cage it fails to converge in a singlet state (although in this case one might question the validity of the calculation).

I'm using SG15 norm-conserving pseudopotentials but the PAW pseudopotential seems to do the same thing.

Regards,

Jeremy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20240124/8575b435/attachment.html>