TURBOMOLE Users Forum

TURBOMOLE Modules => Escf and Egrad => Topic started by: martijn on April 21, 2021, 11:11:30 AM

Title: qsGW vs evGW convergence
Post by: martijn on April 21, 2021, 11:11:30 AM

Hi,

I have observed that qsGW needs typically many more (10+) iterations than qsGW (3-4) in order to converge. Is that to be expected?

Thanks,

Martijn

Title: Re: qsGW vs evGW convergence
Post by: chris.hol on April 22, 2021, 11:55:07 PM

Hi,

indeed this is to be expected; qsGW will in general need many more iterations than evGW and sometimes even more than HF/DFT calculations. qsGW updates all the Fock-Matrix elements, making a re-diagonalization of the Fock-Matrix necessary in every iteration. Especially in cases were the self-energy has peaks close to the Fock-Matrix elements a steep dependence of the latter to small changes can be expected. This leads to a rather slow convergence, and DIIS is not helping too much in these cases. In evGW only the eigenvalues/quasiparticle energies but not the eigenvectors of the Fock-Matrix are updated, a re-diagonalization is not needed. Therefore, it converges pretty fast especially with DIIS.

All the best,
CHristof

Title: Re: qsGW vs evGW convergence
Post by: martijn on April 23, 2021, 10:45:55 AM

Thanks Christof