Skip to content

Add fixed PseudoDojo 0.5 potentials and change rcut to 5.99 - #12

Merged
mfherbst merged 3 commits into
masterfrom
05fix
Feb 23, 2026
Merged

Add fixed PseudoDojo 0.5 potentials and change rcut to 5.99#12
mfherbst merged 3 commits into
masterfrom
05fix

Conversation

@mfherbst

Copy link
Copy Markdown
Member

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

Copy link
Copy Markdown

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

mfherbst commented Feb 23, 2026

Copy link
Copy Markdown
Member Author

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

Copy link
Copy Markdown

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

@azadoks

azadoks commented Feb 23, 2026

Copy link
Copy Markdown
Collaborator

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

Copy link
Copy Markdown
Member Author

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

@azadoks

azadoks commented Feb 23, 2026

Copy link
Copy Markdown
Collaborator

Looks like 5.99 for DFTK

@Technici4n

Copy link
Copy Markdown

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

@mfherbst

mfherbst commented Feb 23, 2026

Copy link
Copy Markdown
Member Author

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

@Technici4n

Copy link
Copy Markdown

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

azadoks commented Feb 23, 2026

Copy link
Copy Markdown
Collaborator

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

Copy link
Copy Markdown
Member Author

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

Copy link
Copy Markdown

True, it was likely not a deliberate choice.

@mfherbst mfherbst changed the title Add fixed PseudoDojo 0.5 potentials Add fixed PseudoDojo 0.5 potentials and change rcut to 5.99 Feb 23, 2026
@mfherbst
mfherbst merged commit c4f291c into master Feb 23, 2026
1 check passed
@mfherbst
mfherbst deleted the 05fix branch February 23, 2026 21:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants