Silicon interstitial NEBs

[zwei-beiner]

Pre-review checklist for PR author

• I've confirmed the contribution guidelines

Summary

Adds a new NEB benchmark for silicon interstitials (MP0 A.1).

Linked issue

Resolves #317

Progress

• Calculations
• Analysis
• Application
• Documentation

Testing

Tested on CPU with:

• mace-mp-0a
• mace-mp-0b3
• mace-mpa-0
• mace-omat-0
• mace-matpes-r2scan
• orb-v3-consv-inf-omat
• pet-mad

New decorators/callbacks

None.

[ElliottKasoar]

Thanks for this, @zwei-beiner! From a first pass, it looks really nice!

I've just left a few minor suggestions/requests.

The only other thing is I wonder if for convenience when choosing a test, we may want a slightly more concise name e.g. si_defects (since it's already in the nebs category), or do you think that's too vague?

Not a problem, but I'm also curious about some of the NEBs, in particular the 216 di-to-single, seem to show little-to-no change until very near the end of the NEB. Is this expected?

[zwei-beiner]

The only other thing is I wonder if for convenience when choosing a test, we may want a slightly more concise name e.g. si_defects (since it's already in the nebs category), or do you think that's too vague?

The name si_defects is good, I've renamed everything.

Not a problem, but I'm also curious about some of the NEBs, in particular the 216 di-to-single, seem to show little-to-no change until very near the end of the NEB. Is this expected?

Yes that was the original NEB, the images are just clustered there. In principle I could remove very similar images as this gives a 5x speedup of the test, and currently we are just testing against energies and forces anyway. Only issue is that this would technically not be a NEB anymore which might become an issue if we want to do actual NEB relaxations with models later on (not just single point comparisons on a fixed NEB).

Let me know if you think the NEB should be de-duplicated.