Number of bands in GW calculation

[maria_cg]

Hi, I am interested in knowing something about the number of bands needed in VASP to get a converged GW calculation with the number of bands. I have noticed that in VASP, compared to other GW codes, you need less number of bands, with similar other convergence parameters. I would like to know if, in the calculation of the polarizability, there are any special mathematical or numerical procedures that ensure the convergence of the calculations.

[michael_wolloch]

Dear Maria,

Thanks for your question.

There are two ways in VASP in which GW calculations can be performed:

• Single-step procedure: GW in one go

• Two-step procedure: DFT first, GW on top of it.

Both of these methods, and their advantages and drawbacks, are detailed on the wiki in the Practical guide to GW calculations page.
The single-step procedure uses all available bands for the selected energy cutoff ENCUT, so no convergence study with respect to bands must be performed.
For the two-step procedure, the number of bands needs to be increased until the property of interest (e.g., the bandgap) is converged. But a large number of empty bands is necessary to get meaningful results. How "large" this large number is depends on the system and the quantity of interest.

Since you are talking about polarizability, our wiki page about GW and dielectric matrix might also be of interest to you, as well as the GW category page and our tutorial (and lecture) on GW.

Let me know if this post and the resources linked answer your question.
All the best, Michael