Add fixed PseudoDojo 0.5 potentials and change rcut to 5.99

[mfherbst]

This PR adds a version of the PseudoDojo 0.5 UPF potentials after a bug in the UPF exporter discovered by @azadoks.

@Technici4n Do we also want to change the rcut values ? I just kept the rcut = 10 which is also used for the other PseudoDojo potentials.

[Technici4n]

I would set rcut to 6.0 as well since it matches what the people verifying the pseudos use (and otherwise some pseudos like Cu are just bad :/). It may be nice to document that we set rcut to 6.0 and that we regenerated some pseudos somewhere (maybe this is a better fit for PseudoPotentialData).

[mfherbst]

We can make the same change to all PseudoDojo potentials (i.e. rcut=6.0) and just bump a breaking release across the software stack, including PseudoPotentialData. I find it weird to change the rcut for one PseudoDojo potential, but not the rest.

[Technici4n]

Yeah true. Changing it to 6 for all existing pseudos sounds good to me.

[azadoks]

I'd tend to agree with @Technici4n; the Dojo potentials were basically universally tested in Abinit with the psp8-format files, where the cutoff would have been 6.0 (exclusive) / 5.99 (inclusive) Bohr. In some cases, they seem to have been essentially overfit assuming that cutoff.

[mfherbst]

Ok, so let's do that then. Only question is rcut = 6 or rcut = 5.99 ?

[azadoks]

Looks like 5.99 for DFTK

[Technici4n]

Maybe make it 6 and change DFTK to be exclusive (C-style)? I really don't like this 5.99 😅

[mfherbst]

For me a cutoff should be inclusive (err on the side of caution), so 5.99 it is.

[Technici4n]

ONCV likely used an exclusive cutoff of 6. 5.99 just doesn't look right, anyone finding that will think we did something wrong, no?

[azadoks]

ONCV likely used an exclusive cutoff of 6. 5.99 just doesn't look right, anyone finding that will think we did something wrong, no?

This is true, but I wasn't a huge fan of that behavior either; it's even inconsistent with how core radii are interpreted internally (first mesh point greater than or equal to the input value). 🤷🏻‍♂️

[mfherbst]

I don't think that it was exclusive was on purpose. It just happened to be that way because someone used < instead of <=. Just because someone did that does not mean we have to do the same.

Also I don't think most users question the pseudo parameters at all. Most will not notice whether it's 5.99 or 6 ...

[Technici4n]

True, it was likely not a deliberate choice.