Possible bug with Gaussian smearing

[jmmshn]

The code might have an issue.

https://github.com/materialsproject/pymatgen/blob/4f906e3578cc20372ff4f12fcda9db3aaebec975/pymatgen/analysis/path_finder.py#L388

I used a Si CHGCAR and ran

from pymatgen.analysis.path_finder import StaticPotential
sp = StaticPotential(chgcar.structure, pot=chgcar.data['total'])
sp.gaussian_smear(r=2)
sp.get_v().shape

This should return an sp where potential has the same dimensions as the original but instead give a smaller array. Also just looking at the code I think something is wrong with how the number of required images are computed for skewed cells. (There's also no tests for these classes) I have a working solution for this kind of smearing in PyRho but might take a while until things are migrated here.

I just wanna create this issue for future reference. But maybe since testing is lacking we should just add a deprecation warning for now?

[mkhorton]

But maybe since testing is lacking we should just add a deprecation warning for now?

Deprecation warning is only for if we're going to remove the function. A general warning that it is untested would be useful; we should either add tests asap or remove though, untested code shouldn't be in pymatgen.

[mkhorton]

Thanks for reporting/noticing this.

[shyuep]

I think the code has been migrated to pymatgen-diffusion. Can you check? If it is already there, I will just delete this code.

[jmmshn]

The pathfinder in pymatgen diffusion is different from this pathfinder and the code that I added to pymatgen-diffusion uses both pathfinders but I cannot find anywhere in our codebase that uses this StaticPotential that contains this incorrect Gaussian smearing implementation.