Incorrect predictions of bandgap

Materials Project Materials Project Data/API
1

Hi,

As it is clear, the plain GGA is unable to correctly predict the band gaps of Strongly correlated materials. This problem has been mentioned before. Why dont you use other methods such as GGA+U or HSE to estimate the gaps? The shift of conduction band to get a more accurate electronic structure may not be useful for all cases (for instance when there are localised electrons).

Is there a way for someone to contribute to your useful site?

Cheers

2

Hi kambiz,

Thanks for sending us your comment and suggestion on the electronic structure calculation. In materials project, we do have GGA+U level of calculation for strongly correlated compounds to assess their properties correctly. For example, the GGA+U bands are available for LiCoO2 ( https://materialsproject.org/materials/mp-24850/), and the calculation summary section at the bottom of the page specifies that the “run type” is GGA+U. The details of how we run GGA+U calculations can be found through this wiki page https://www.materialsproject.org/wiki/index.php/GGA%2BU_calculations, from where we provided info of how we calibrated the U values for each element.

Methods like HSE is under developing and testing stage and will be available in the future. So stay tuned. You are very welcome to join us and contribute to this effort by working on the open access repos of the materials project through GitHub. E.g., we have several HSE-related python classes in pymatgen.io.vasp.sets module to deal with HSE calculations (http://pymatgen.org/_modules/pymatgen/io/vasp/sets.html).

The GGA+U is known to have the underestimated bandgap to some extent, so that we added warning info under the band structure section – “Warning! Semi-local DFT tends to severely underestimate bandgaps. Please see the wiki for more info.” We did not simply scale the GGA+U band gap as there is not a reliable and standard way to correct the band gap to our best knowledge. But, please feel free to let us know if there is a solution. We will be very glad to discuss and work with you.

Best,

Miao

1 Like
3

Hi,

Thanks for your reply. I am not familiar with python; however, if needed I can provide information based on our published results or calculated results (with no restriction). Please let me know If i can be of any assistance.

Regards

4

Just FYI, infrastructure developments for community contributions are ongoing, but I don’t believe they’ve been officially released. Perhaps @tschaume-old might elaborate :slight_smile:?

http://pythonhosted.org/mpcontribs/