Entry sets use wrong Yb pseudo-potential

[janosh]

The docs are incorrect on what pseudo-potential the EntrySets use for Yb. Docs say Yb_3, actual is Yb_2. Just to list a few:

pymatgen/pymatgen/io/vasp/MPRelaxSet.yaml

Line 171 in a7ee5ad

Yb: Yb_2

pymatgen/pymatgen/io/vasp/MPSCANRelaxSet.yaml

Line 123 in a7ee5ad

Yb: Yb_2

pymatgen/pymatgen/io/vasp/MVLRelax52Set.yaml

Line 177 in a7ee5ad

Yb: Yb_2

pymatgen/pymatgen/io/vasp/MPAbsorptionSet.yaml

Line 113 in a7ee5ad

Yb: Yb_2

The A-lab revealed that as a result of using Yb_2 the energy on Yb compounds is off by a lot, resulting in supposedly stable things being unsynthesizable. While an unfortunate mistake, it's also great to see how experiment can help surface simulation errors. (@rekumar) (Although if tales are correct, @arosen93 originally brought this up with @mkhorton in Oct 2021?)

Planned solution is to update all input sets to actually use Yb_3. Why not Yb so as not to force an oxidation state? Apparently Yb has stability issues. Additional context in

• Add MP Base Sets atomate2#27 (comment)
• https://www.vasp.at/forum/viewtopic.php?t=18367

@shyuep Any concerns on your end? We plan to highlight this change in the pymatgen release notes and in the next MP database release. Any additional venues?

Unfortunately, this comes too late for the large data set Google generated which used the current MPRelaxSet and MPSCANRelaxSet but better fix this now before another large effort gets this wrong.

[shyuep]

I don't really have any concerns. I think whether Yb_2 or Yb_3 depends on what you are calculating. VASP "recommends" Yb_2 as the PSP. See https://www.vasp.at/wiki/index.php/Available_PAW_potentials. This was probably the reason why we used that in the first place. For HT efforts, we have to select one. I am fine with switching to Yb_3 given that most Yb exists in 3+ oxidation state.

[janosh]

This docs page needs updating. It incorrectly states we use Yb. It even lists "thermo data off with Yb_2" as the reason not to use Yb_2 even though VASP recommends. 🤦

[shyuep]

The docs page need to be updated for sure. But the docs seem to be based on the old HT framework at MIT. When we did MP, there was a deliberate change in PSP used. Anyway, like I said, no issue with switching to Yb3+.

[janosh]

@munrojm Change request for docs page submitted on gitbooks.

[munrojm]

@janosh merging now

[mkhorton]

Although if tales are correct, @arosen93 originally brought this up with @mkhorton in Oct 2021?

I don’t recall, but I believe it. The consequence of changing the MP Input Set is that current data on MP will the be inconsistent with the MP Input Set; since a lot of people generate data intentionally to be compatible with MP, this can cause a lot of downstream problems (even in this case, where the change is sensible and necessary).

Around that time we were also talking about reviewing other pseudopotential choices in MP, eg some pseudopotentials we’re not used because of ghost states (perhaps Na? I forget), but these issues were later fixed in the newer 54 pseudopotentials that the MP SCAN Input Set uses. The SCAN Input Set presented an opportunity to cleanly make this change. However this was not pursued, I think more out of conservativeness than anything.

All this to say, I’m glad this change was made, but if I could offer a viewpoint to the MP team, it’s that a new data release to replace the Yb data should also be done promptly (although I realise this may not be possible!), as well as a louder mention of this change in the relevant places (eg docs changelog, or a banner on the linked docs page) might be a good idea too, to let people know this change is coming.

[munrojm]

Just an FYI for everyone here that I am moving to re-calculate all the Yb compounds right now. The plan is to stop working on the current MP build to get these done and incorporated.

[janosh]

as well as a louder mention of this change in the relevant places (eg docs changelog, or a banner on the linked docs page) might be a good idea too, to let people know this change is coming.

Yes this is definitely planned! 👍

new data release to replace the Yb data should also be done promptly (although I realise this may not be possible!)

This is also planned but the timeline is less firm.