Reliability of bad-gap calculations

I’ve seen a little discussion on the forum about reported “incorrect” band gaps. This seems to be due to the intrinsic tendency of DFT to underestimate it.

My question is: how is the situation at the moment? Has any improvement been made on the topic? How reliable would you say the average prediction is, based on the employed methodology?

Please feel free to link to any publication that might prove useful for this!

Hi @acarnevali,

Certainly consult our docs first for a comparison graph: Electronic Structure - Materials Project Documentation

The situation has been improved for some time, there are many techniques that can provide more accurate band gaps in general. The difficulty is using these techniques on a large scale, since these are quite expensive and more difficult to apply at a large scale. Materials Project, at time of writing, still reports “band gaps” predicted using PBE-DFT, which is one of the techniques that systematically underestimates the gap, but which is computationally cheaper.

For the purposes of materials screening, an empirical correction like that shown in this paper could be used to correct the PBE band gap, although this still only gives a very coarse approximation and I would not trust it very much.

For a specific material, better methods should be used. Typically, we use a functional known to give better predictions like HSE06 for this. Methods like GW can improve on this again.

It’s all quite subtle however, it’s a difficult problem!