ABINIT Discussion Forums

Hello,

I would like to compute the GW band gap for some semiconducting oxides. I started doing some convergence studies on nband, ecutsigx, ecuteps using a single k-point in Gamma for computing the KSS file and, subsequently, epsilon and sigma, as is done in the first lesson on GW in the Abinit tutorial. However I got a negative GW correction to the GGA band gap, i.e. the band gap gets smaller than in DFT. This looks very strange to me given that also using a denser k-point mesh does not always solve the problem. For example, while for MgO using a 2*2*2 k-point mesh is enough for getting a larger gap than in GGA, for ZnO even a 6*6*6 mesh is not enough. So, I would like to ask for your support concerning the following issues:
- Are convergence studies on nband, ecutsigx, ecuteps, ets. reliable even when the k-point mesh is obviously too coarse?
- Is there any good criteria for choosing a converged k-point mesh for the GW calculation?

Thanks for any help you can provide

Regards,
Matteo

In general, for semiconductors and insulators, there should be very little inter-dependence between the convergence of ngkpt and other parameters (see the supp. materials of J. Phys.: Condens. Matter 35,105503 or ibid. 26,125503). It therefore probably suffices to do the convergece studies with a coarse brillouin zone sampling (BZS) and test only a few points with a denser BZS. How well converged are your other parameters with ZnO? Perhaps it helps to study the already published results [e.g. PRB 84, 241201]. I'm not aware of any general rules for choosing BZS or any other convergence parameter. The better you guess the starting parameters the fewer rounds of convergence studies you need to do.