Issue with electronic convergence of Yb3+ cation (4f valence electrons)

[mateusz_domanski]

Hello all,
As mentioned in the topic, recently I have encountered convergence issues of electronic steps during single-point calculations of compounds with formally Yb3+ cation, using a Yb pseudopotential (ZVAL = 24.0) in VASP 5.4.4. Such run starts normally but then the program is unable to decrease dE and after 10-20 electronic steps messages in std_out come up like these below:

DAV: 21 -0.837352644035E+06 -0.77948E+06 -0.35731E+05 22526 0.170E+02BRMIX: very serious problems
the old and the new charge density differ
old charge density: 106.00400 new 105.98939
0.143E+02
WARNING: Sub-Space-Matrix is not hermitian in DAV 33

As a result, I can not calculate converged energy, nor an optimized geometry of course.
I have tried different starting geometries (e.g. experimental Yb2KF7, Yb2Ti2O7, or a molecule preoptimized with Yb_3 pseudopotential) but everywhere the same problem shows up. On the other hand, I have managed to calculate and optimize similar compounds with Gd (also 4f valence).
Is there some method that I have not used (see INCAR) that can help with this convergence issue? Or there may be some issue with the installation of the VASP version I am using?
I have searched the forum for the problem of convergence among lanthanoids with valence 4f electrons, but have not found the same problem or there are some inactive links in explanations (e.g. in topic forum/viewtopic.php?f=3&t=14808&p=14808 ... ons#p14808).
I attach the files to check the calculations.

With kind regards,
Mateusz

[ErnekMB]

I have had same exact issues with my system when including 4f valence treatment of my Yb dopant:
See here:
https://mattermodeling.stackexchange.co ... stems-in-v

[andreas.singraber]

Hello Mateusz and ErnekMB!

I will try to reproduce the problem and discuss this with my colleagues (f-orbitals are often difficult to converge) to find some solution to this.

In the meantime, here are the two missing links you mentioned:

forum/viewtopic.php?f=4&t=96
forum/viewtopic.php?f=3&t=1192

but I have not checked yet whether they contain any useful hints for your problem.

Best,

Andreas Singraber

[mateusz_domanski]

Thank you ErnekMB and Andreas for your messages.
It seems that I have got a similar problem as ErnekMB with Yb case described in the link.
The Delta-SCF approach might be a workaround solution (probably using FERWE and FERDO tags), however, before doing it I have calculated non spin-polarized calculations. In my case (exp. structure of Yb2Ti2O7) I have managed to converge SCF without spin-polarization (ISPIN=1) and then to load it to a spin-polarized calculation. The latter has not converged yet but looks much better than previously, maybe setting the occupancies of states will help further.
But still, some other structures that I have got resist converging in even not spin-polarized SCF calculations, while they should be reasonable since they were pre-optimized with Yb_3 pseudopotential. This procedure worked for Gd/Gd_3 case.
I have tried also many different ALGO as was mentioned in one of the links, but this did not work at all.

[ErnekMB]

For my systems , I cannot get the non spin-polarized calculations either. Every time I use Yb POTCAR with 4f in valence space the calculations are not good. Did you manage to get some systems with Yb to work Mateusz_domanski.
Thank you