About GWA

Summary:

How useful is it?

About GWA

Fabio Fumi 04 Sep 2012 12:59

Hi,

let me be a bit "destructive" here, but just to stimulate (excite?) convincing arguments against mine… How "useful" is the GWA approximation, in electronic structure calculations? I perfectly understand the importance of the underlying theoretical background, in particular to recover some meaning to DFT DOS calculations with respect to experimental spectra, but from a practical standpoint… I followed the tutorial and applied it to a simple 3C-SiC simulation and what I've noticed is that the result on the band structure curves is "only" an almost linear stretching of the energy scale, about 10%. See for example this plot:

That should also be demostrable, to some extent, I think… I miss the literature here. As explained in the tutorial, careful choice of computational parameters is required to properly fit the GW results with experimental values (typically the gap energy). This is not a true ab-initio, in my understanding, against an increased cost of tenfold, or more, computational resources. Why then not using just the energy "stretching" as the fitting parameter and forget the GW at all? Any practical use case where GW is unavoidable (in calculations, I mean)?

Thanks all for the great job!
Fabio

Last edited on 04 Sep 2012 14:25 by Fabio Fumi Show more

Re: About GWA

DmitriiNabok 10 Sep 2012 12:31

Hi,

To answer your first question, I'd recommend you to have a look at existing review articles, e.g.,
"Electronic excitations: density-functional versus many-body Green's-function approaches" by Giovanni Onida, Lucia Reining, and Angel Rubio, Rev. Mod. Phys. 74, 601 (2002).

Second, it's nowhere in tutorial mentioned that one has to tune any of parameter to reproduce experimental data. As it's correctly said, this concept completely contradicts to the idea of ab-initio.
Convergence tests must be performed to assure the numerical accuracy of results, as well as to help others to reproduce these results.

Next, the mentioned almost rigid shift between GW and KS energies of unoccupied states is very well known for s/p semiconductors.Therefore so-called "scissor" approximation is widely used to calculate optical properties of these materials. But if to have a look closer, one can convince yourself that the amplitude of this shift is band dependent.
This dependence become more and more pronounced for d/f elements. Moreover, this rigid shift does not work for valence states (the deeper state the larger is difference between GW and KS values).

Next point, where to take the values for the "scissor" operator? Normal practice is to take experimental values (if they are or simply tune them). What about unknown materials? Other simple example - a physisorbed molecule on a surface, where the respective order of bands would play a crucial role and where DFT is known to have problems. Well, one can think of other even more spectacular examples if to scan the literature.

Cheers,
Dmitrii

Re: About GWA

Fabio Fumi 11 Sep 2012 08:08

Thanks Dmitri for the useful references and clarifications.
Enough to convince me that for more complex system, were actual prediction is requested, GW is necessary :)
Indeed, my idea was towards the "scissor" approximation, but of course this looks like limited mostly to s/p semiconductors.