ABINIT Discussion Forums

Dear all,

Usually, there is no need to setup nband parameter, however, in rf calculation we are required to specify the nband number.
here is the question:
How does DFT decide the nband?
For example, for MgO cell and Si fcc cell, the nband is the 5 and 4 bands, respectively, which is understandable; but for anatase primitive cell and rutile unit cell, both having 2 Ti and 4 O atoms, and ZnO unit cell, the bands are 25 and 10, respectively. I have noe idea how these numbers add up. Even from the calcualtion, 16 is the valence band for 2(TiO2).
Any insight is greatly appreciated in advance.
Also, is there any role to specify occopt parameter for metal oxides? Is it necessary?

Kindly regards,
Rundong

The bands come from counting the valence electrons. In TiO2, suppose you are using 1s2 2s2 2p6 core, and 3s2 3p6 4s2 3d2 valence for titantium and 1s2 core 2s2 2p4 valence on oxygen. You have 2 Ti atoms and 4 O atoms in the unit cell so you have 48 electrons altogether. If you then treat it strictly as an insulator with no empty bands, you use occopt 1 and nband 24. The default nband number is probably 25, because the convergence is sometimes easier with an empty state or two. If you have a metal you will have to allow for some conduction bands.

However, if you tried to make a Ti pseudopotential based just on 4s2 3d2 valence, you'd have only 32 total valence electrons and hence 16 bands. This would be a faster calculation but would likely fail to give a reasonable answer because usually transition metals aren't well described using so few levels.

Some of the tutorials in the respfn sections give nice examples of how to study metals like Al and Li using the respfn formalism. With insulators it's easier and also usually best to set occopt 1 and nband explicitly to the number of filled bands.

Dear Prof. Zwanziger,

Thank you again for the detailed explanation and all the helps.
I will set the RF calculation following your expertise's opinion and keep you posted.

Best regards,
Rundong