Can doped parse defect complex calculations?

[jumpingfermions]

It seems the examples are always about single site defect, e.g. a vacancy or an interstitial. If I prepare VASP calculations about defect complex such as two neighboring vacancies, can doped recognizes the results, processes further, and plots transition level diagrams?
Thank you!

Before taking developer time, I have searched the following curated resources for answers:

• doped documentation site – tip: use the search bar on the left.
• doped tutorials
• doped Python API docs
• Previous doped issues where the same question may have been asked – tip: use the search bar.
• All other doped docs pages; such as Tips, Troubleshooting and Installation

If your question is about general defect calculation methodology (rather than specific to doped), please refer to the following papers, and other relevant literature:

• Quick-Start Guide on Defect Calculations – Kim et al.
• Large Review on Defect Calculations – Freysoldt et al.
• Defect Structure Searching – Mosquera-Lois et al.
• Free Energies of Defects – Mosquera-Lois et al.

[kavanase]

Hi @jumpingfermions, good question.
It is possible to simulate complex defects using doped, but they are not automatically handled at present (though this is on the development plan!) and so it requires some code customisation on the part of the user.

Firstly, for generating complex defects, the recommended approach is to generate one of the constituent point defects in the complex using the usual DefectsGenerator approach, then feeding in the defective supercell to DefectsGenerator again, for it to generate 2-defect supercells. You can then ensure that this only includes symmetry-inequivalent combinations by taking only those with unique defect-defect distances. This is discussed in #91, and this approach was used in https://arxiv.org/abs/2412.19330 (for which the notebooks used will be shared soon, but some example code is given in #91), as well as e.g. https://pubs.rsc.org/en/content/articlelanding/2021/sc/d1sc03775g & https://pubs.rsc.org/en/content/articlehtml/2023/ta/d3ta00532a.

For parsing, a similar approach can be used, where you use a neutral charge state of one of the constituent point defects as the "bulk" supercell reference, to then parse the complex defect calculation outputs in the normal way (which will get the appropriate charge correction etc), before then accounting for the formation energy of the neutral 'bulk' reference in this case. Alternatively, you can directly use the different doped sub-functions (e.g. for charge corrections; see https://doped.readthedocs.io/en/latest/doped.corrections.html; site determination etc) and calculated chemical potentials to determine the formation energies etc. This approach was used in: https://pubs.rsc.org/en/content/articlelanding/2021/sc/d1sc03775g

I will try to update this method when further updates to doped are added for automatically handling complex defects!

[jumpingfermions]

Thanks @kavanase for the detailed illustrations.

I'm trying to follow the procedures above. In the two-step approach for the charge correction, supposing that A and B are two defects, we get the A&B charged defect complex supercell, and A neutral defect supercell as the bulk. Now the model charge distribution is centered at B site.

Another way for the charge correction is sxdefectalign by FNV author. I tried that before. https://sxrepo.mpie.de/attachments/73 In the manual page 9, Multiple charges, they set two Gaussian-like distributions at two defect sites as the model charge distribution. So I wonder if these two approaches produce consistent results.

[kavanase]

Yep, what you say is correct. Nice to see you are thinking critically about this. As you say, the approach I mention assumes the charge is centred at the B site defect. In practice, for reasonably large supercells and moderate charge states I think this approximation has a relatively small effect, but of course this is system dependent.

The manual approach with sxdefectalign accounting for multiple charges is presumably the most accurate (though requires assignment of the individual charge states and locations). You can use this and then manually edit the DefectEntry.corrections dict to apply it to your parsed defect calculations in doped.

[jumpingfermions]

Thanks for more methods to solve the problem.

It seems that if we can combine FNV multiple charges correction with atomic site potential treatment, and make this robust and automatic, that would be a nice approach. For now, I will try treating this within doped or combine sxdefectalign manually. Doped is a convenient tool allowing for getting data from other sources.