[Pw_forum] hybrid functionals: dexx<0

[Matteo Gerosa]

Dear all,

I've been trying to calculate band gaps for some semiconductor oxides 
using hybrid functionals (pbe0, hse, b3lyp).
I've learned the choice of the q-grid for computing the exchange energy 
can be critical. Indeed in some cases I cannot achieve self-consistency 
and an error message appears in the output file warning that the dexx 
correction term becomes negative.
I guess this is related to the small-q divergence issue for some 
particular q-grid patterns. In fact, if the k-grid is shifted away from 
gamma (keeping the q-grid fixed) convergence is achieved. Unfortunately, 
this doesn't allow me to compute the gap at the gamma point, which I'm 
interested in. On the other hand, I also obtain convergence if I set 
"x_gamma_extrapolation=.false.", but this induces large errors for a 
small number of points in the q-grid.
How is it possible to solve the problem and, more importantly, predict 
in which configurations of the q-grid could the problem arise? Is there 
any way to shift the q-grid away from gamma?

Thanks for any help you can provide.

Best regards,

Matteo Gerosa

-- 
Matteo Gerosa
NanoLab - Micro and Nanostructured Materials Lab
Department of Energy, Politecnico di Milano
Via Ponzio 34/3 I-20133 Milano, Italy

tel. +39 02 2399 6364
web: www.nanolab.polimi.it