Split vacancies benchmark

[ThomasWarford]

Pre-review checklist for PR author

PR author must check the checkboxes below when creating the PR.

• I've confirmed the contribution guidelines.

Summary

Based on this paper: Identifying split vacancy defects with machine-learned foundation models and electrostatics

The metrics:

• Formation energy of split-vacancy defects from fully ionised defects
• Spearman's coefficient for ranking the energies of initial defect structures.
• RMSD of MLIP relaxed structures vs DFT relaxed structures. (Note the initial structure for the MLIP relaxation is currently the DFT relaxed structure.)

Linked issue

#335

Progress

• Calculations
• Analysis
• Application
• Documentation

Testing

New decorators/callbacks

[ThomasWarford]

@joehart2001 @ElliottKasoar do you think this belongs in a "defects" section rather than bulk crystals?

[ThomasWarford]

Here's how the table looks at the moment

[joehart2001]

@joehart2001 @ElliottKasoar do you think this belongs in a "defects" section rather than bulk crystals?

Yes i totally agree. We have some other PRs that could maybe fit in a defects category, like #337 creates an interstitial category, so we could merge you two. but its something we can move around easily anyway

[joehart2001]

Here's how the table looks at the moment

Looking good, potentially don't need rmsd as well as MAE? hopefully we'll have an option soon to switch between different types of errors automatically.

[ThomasWarford]

The RMSD metric is still subject to change, since it is sensitive to the accuracy of the bulk structure as well as the defect structure (which is what we are interested in).