BadELF Exception Raised

[JuliaBaratta]

When trying to model a new system using the BadELF package in Simmate, I got this error:

Exception: At least one atom was not assigned a zero-flux basin. This can result from covalency or from pseudo-potentials with only valance electrons.

After running the command:
simmate workflows run input.yaml

The source code for this error is:

if check_for_covalency: raise Exception( """At least one atom was not assigned a zero-flux basin. This can result from covalency or from pseudo-potentials with only valence electrons (e.g. Al, Si, B in VASP 5.X.X).""" )
Do you have any recommendations for avoiding this error when trying to investigate structures that contain some level of covalency?

@Andrew-S_Rosen

[SWeav02]

if check_for_covalency: raise Exception( """At least one atom was not assigned a zero-flux basin. This can result from covalency or from pseudo-potentials with only valence electrons (e.g. Al, Si, B in VASP 5.X.X).""" )

This error can be caused by one of two issues:

1. If there are covalent or metallic bonds in the structure, they will show up in the ELF as maxima between bonded atoms. BadELF assumes that atoms are divided from each other at minima, so there is currently no method for dividing these maxima. Instead, all of the charge will simply be given to the closest atom. We plan to eventually solve this issue by tracking covalent/metallic bonds in a similar way to how we track electride sites.

2. VASP's ELF calculation only includes contributions from the valence electrons in the pseudopotential (POTCAR) files. If there aren't enough valence electrons in this file, there's a possibility that all of the valence electrons will participate in bonding. There will be no maximum on the atom throwing off the zero-flux surface. Ordinarily this can be solved by using a pseudopotential with more valence electrons. Unfortunately, VASP does not always provide pseudopotentials with additional valence electrons (such as in Al, Si, and B). In this case it's possible to use a software other than VASP (e.g. Quantum Espresso) and then run BadELF by manually loading the charge density and ELF using python (see tutorial).

Do you have any recommendations for avoiding this error when trying to investigate structures that contain some level of covalency?

You can force the program to ignore this error by setting check_for_covalency: false in your input.yaml file. Beware that the oxidation states on atoms probably won't be reasonable. However, if you just want the number of electrons in the electride sites or the dimensionality of these sites you might be ok. The main exception is with problem (2). If one of the problem atoms is adjacent to the bare electride sites it will result in an incorrect zero-flux surface and incorrect electron count.

So in summary, you can do this but be careful and thoughtful about the results.

[Andrew-S-Rosen]

Just chiming in to say thank you @SWeav02 for your thorough and timely reply to @JuliaBaratta who is a PhD student in my group. If I understand correctly, it seems like Issue 1 would potentially make BadELF a challenge to use with organic electrides, and Issue 2 would potentially make something like mayenite (Ca-Al-O) difficult to analyze with the default VASP pseudopotentials. Are those reasonable comments on my part? In any case, this is all very useful to know. We will definitely do some tests. :)

P.S. I think #713 would be a very valuable feature addition!

[scott-materials]

Hi Andrew -- delighted that y'all are finding things of use with Simmate. I had perused your website 6 months or a year ago and was impressed with all your efforts!

Sam may have an alternate suggestion for your first question, but there is probably a manual fix to Issue 1--Sam, correct me if I get something wrong. You can edit the ELFCAR to place a dummy atom within each covalent basin. Then, BadELF should integrate those regions using the same rules as real atoms. Of course, you have to decide whether the BadELF algorithm is a physically sensible approach for these covalent basins. My first guess is that it is a reasonable approach, but I don't believe that we've tested that.

And you are correct that it is not possible to do things like mayenite. If you know of custom VASP pseudopotentials or have other ideas, please let us know!

[SWeav02]

If I understand correctly, it seems like Issue 1 would potentially make BadELF a challenge to use with organic electrides, and Issue 2 would potentially make something like mayenite (Ca-Al-O) difficult to analyze with the default VASP pseudopotentials. Are those reasonable comments on my part? In any case, this is all very useful to know. We will definitely do some tests. :)

Yes, I think this is a great summary!

You can edit the ELFCAR to place a dummy atom within each covalent basin. Then, BadELF should integrate those regions using the same rules as real atoms.

As I see it, this should work pretty well and is basically what I would do to automate the process. BadELF automatically places dummy atoms at electride sites, so this is a similar idea. One thing to note is that the software already uses He atoms as dummy atoms, so if you also use He atoms the covalent bonds will be treated like electride sites. In this case they'd be separated from the nearby atoms using a zero-flux surface. If you use another element, the covalent bond will be treated like an atom and be separated using voronoi-like planes.

I'm not sure which way makes more sense, but I would use the same logic we applied to electrides. If the covalent bonds are fairly spherical (like an atom), the plane separation will probably give better results. If they're not very spherical (like many electride sites) the zero-flux surface will do a better job.

We will definitely do some tests.

If you do end up using the dummy atom strategy, please let us know how it goes! If it works well, I'll try and automate the process so you don't have to do everything manually.

[SWeav02]

P.S. I think #713 would be a very valuable feature addition!

I'm glad there's interest in this! I just posted a script in #713 that I've been using in the meantime.

[Andrew-S-Rosen]

@scott-materials: Thank you for the kind words! And for chiming in! Big fan of everything your group is doing and wonderful to see the emphasis on open-source!

Thank you both for the suggestions. This is great brainstorming material. We will keep you posted once @JuliaBaratta gets some time to try out the dummy atom approach on some tricky cases.

For the pseudopotential problem, one could in principle try using the PAW PBE GW potentials instead, which have options with more electrons treated as valence (e.g. Al_sv_GW, Si_sv_GW). The VASP manual and prior work suggests that the GW potentials can be used for more accurate ground-state DFT calculations (i.e. can be used even if a GW calculation is not run). I suppose your mileage may vary with this one.