Ensure MP VASP sets don't use auto_ismear, few other fixes

[esoteric-ephemera]

In response to issue #672, all MP VASP input sets now set auto_ismear = False by default. This change leads to the following behavior:

• GGA workflow always uses tetrahedron smearing (ISMEAR = -5, SIGMA = 0.05). This is consistent with the atomate GGA workflow and left for consistency
• meta-GGA workflow uses Gaussian smearing (ISMEAR = 0, SIGMA = 0.05) for relaxations and tetrahedron smearing (ISMEAR = -5, SIGMA = 0.05) for the final static. This is not quite the behavior from the atomate workflow, but is a minor change given that the atomate wf didn't include a static or CHG/WAVECAR initialization like the atomate2 one does
• The meta-GGA workflow behavior is consistent with the wf structure moving forwards with MP
• Updated test files to reflect these changes

Few other fixes/improvements:

• Removed dependence on pkg_resources.resource_filename as this has been deprecated in favor of importlib.resources.files
• Froze the MACE model used in the elastic forcefield workflow test - I saw that the test could fail (link) when the model couldn't be downloaded on the fly

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (4cd1e19) 76.26% compared to head (3d00f3e) 76.13%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #673      +/-   ##
==========================================
- Coverage   76.26%   76.13%   -0.13%     
==========================================
  Files          87       87              
  Lines        7132     7140       +8     
  Branches     1055     1055              
==========================================
- Hits         5439     5436       -3     
- Misses       1369     1381      +12     
+ Partials      324      323       -1     

Files	Coverage Δ
src/atomate2/cp2k/sets/base.py	64.15% <100.00%> (ø)
src/atomate2/vasp/sets/base.py	73.98% <100.00%> (ø)
src/atomate2/vasp/sets/matpes.py	87.50% <100.00%> (ø)
src/atomate2/vasp/sets/mp.py	93.02% <100.00%> (+1.59%)	⬆️

... and 1 file with indirect coverage changes

[mkhorton]

Asking to clarify: is this issue in relation to the MP GGA (PBE) workflow, or the MP GGA (PBEsol) workflow used to bootstrap the R2SCAN calculation?

If the former, I thought the historical behaviour was to use Gaussian smearing (ISMEAR=0) as a sensible default for relaxations, and tetrahedron only for statics?

[esoteric-ephemera]

It ends up being both workflows (not just the PBEsol pre-relax) because the value of ISMEAR changes according to bandgap. We don't necessarily want Methfessel-Paxton smearing in case a system is incorrectly predicted to be a metal in the GGA pre-relax.

If the former, I thought the historical behaviour was to use Gaussian smearing (ISMEAR=0) as a sensible default for relaxations, and tetrahedron only for statics?

I recall looking into the atomate workflows and finding that the OptimizeFW defaults to MPRelaxSet with no further changes to ISMEAR. That ends up giving ISMEAR = -5 in relaxations.

I've also done some sanity checks against the MP task docs, and it looks like ISMEAR = -5 was always used in GGA relaxations.

[utf]

Thanks @esoteric-ephemera. Can I just ask about the motivation for disabling auto_ismear - is that for consistency with atomate1/pymatgen input sets or is there another reason? (Sorry if I missed it in another issue somewhere)

[esoteric-ephemera]

Hey @utf, for the GGA workflow, it's consistency with atomate. To be consistent with atomate, the meta-GGA workflow would have to start with Methfessel-Paxton smearing (ISMEAR = 2, SIGMA = 0.2) and then switch to tetrahedron (ISMEAR = -5, SIGMA = 0.05) if the gap in the PBEsol step is nonzero

I will say I'm in the process of running the "new" (updated pseudopotentials, parameters and hopefully +U's) workflows for both meta-/GGA workflows. So if we'd rather have these exist just as their atomate 1 equivalents, I'm happy to make that change

[utf]

Understood! I think this is ok for now. Thanks very much.

[mkhorton]

Thanks @esoteric-ephemera, not the first time my memory has been incorrect :)

[Andrew-S-Rosen]

FYI, the MACE model downloading issue is fixed in main of the MACE repo. It's a security flag raised by Google Drive.

[Andrew-S-Rosen]

@esoteric-ephemera --- with this change, is ISMEAR always equal to 0 and SIGMA always equal to 0.05 in the MP meta-GGA relax and pre-relax steps (unless changed by Custodian)? I'm not opposed -- I just want to make sure I'm understanding this correctly.

[esoteric-ephemera]

That's right @Andrew-S-Rosen. To clarify and document these changes:

For the pre-relaxation , ISMEAR = 0, SIGMA = 0.05, KSPACING = 0.22. The r2SCAN relax continues with 0.22 (metal) <= KSPACING <= 0.44 (wide-gap insulator) and ISMEAR = 0. The final r2SCAN static uses ISMEAR = -5. SIGMA = 0.05 is unchanged through this.

This is mostly to avoid situations where Methfessel-Paxton (ISMEAR > 0) smearing is used to relax insulators, or tetrahedron (ISMEAR = -5) method is used to relax metals. The atomate 1 r2SCAN flow had the former problem, the PBE flow has the latter problem

[Andrew-S-Rosen]

Thanks!! For clarification: presumably you meant KSPACING instead of SIGMA for the 0.22 to 0.44 comment.

[esoteric-ephemera]

Whoops yes thanks! Corrected above