Incorrect band gaps?

I’m just playing around with the API and data, and noticed many incorrect band gaps being reported.

For example, I get band gaps of 0.189 eV and 0.194 eV for GaAs, whereas wikipedia reports a band gap of 1.424 eV.

I know the MP Calculations Manual states that the DFT calculations underestimate band gap, as much as 50% from literature. But, in the above example, this error is a whole order of magnitude off.

Am I misinterpreting what’s being reported by MP and the API?

Thank you for your interest in MP. In our calculations guide, under the Accuracy of Band Structures section, we plot experimental versus computed band gaps for 237 compounds. You can see there is quite a bit of variation. Our “50%” statement is further down, in the first subsection:

In general, band gaps computed with common exchange-correlation functionals such as the LDA and GGA are severely underestimated. Typically the disagreement is reported to be around 50% in the literature. Some internal testing by the Materials Project supports these statements; typically, we find that band gaps are underestimated by about 40% (Figure 2). We additionally find that several known insulators are predicted to be metallic.

We have more information in the guide about the origin of the error, methods to improve accuracy, and a way to interpret DFT band structures (because the structure does provide useful information) by shifting conduction bands.