Dear all,

I have two questions concerning convergence vs NOMEGA in GW0 calculations:

1) I'd like to reach convergence vrt NOMEGA. E.g. I set NBANDS to maximum allowed for given Ecut. Then, if I understand right, if I will increase NOMEGA at fixed k-sampling, than from certain value I shoud see not the convergence but noise increase. So the only straitforward way is to increase k-sampling and repeat NOMEGA variation and compare results. Am I right?

Taking into account GW0 computing time, it should be quite long calculations. So, another option is to use recommended values. But there is another question:

2) It is recommended to set NOMEGA<=100 for 4x4x4 k-points/atom and <=200-300 for 6x6x6 points/atoms. I expected that increasing number of atoms means increasing of unit cell, then I can decrease number of points in k-sampling. But the number of atoms in recepy is in denominator (k-points/atom). I am not sure, why? E.g. if I have 1 atom per cell then I can use NOMEGA=100 for 4x4x4 grid, what will be recommended NOMEGA for 2 atoms per cell and same 4x4x4 grid?

Best regards,

Dmitry