The ABINIT Users Mailing List ( CLOSED )

[Fabien Bruneval <fabien.bruneval@polytechnique.fr>]

Dear Manolo,

As you said, the spectral function is indeed the imaginary part of the 
Green's function. When the exchange-correlation is accounted for by a 
non-dynamic potential (like the KS potential), this imaginary part 
reduces to a usual DoS.
However, in the GW approximation, the exchange-correlation is described 
by a dynamical self-energy and the spectral function contains more than 
the usual DoS.
There is a part of the spectral function, which is DoS-like (named the 
quasiparticle part) and a remainder, which does not resemble to any 
defined one-particle excitation (named incoherent part or satellites).

The GW subroutine of Abinit (version > 5.1) is able to output the matrix 
elements of the spectral function:

Im ( < i | G(\omega) | i > )

To obtain in practice the spectral functions from the code, you need to 
use the GW part of Abinit without the plasmon-pole model:

gwcalctyp   2

and you need to define a mesh of real frequencies where the spectral is 
calculated, for instance:

nfreqsp      1000
freqspmax       50. eV

The spectral function will be output in the last column of file "out.sig".

Be aware that the satellites of the spectral function within the GW 
approximation are believed to be not so reliable... (see F. Aryasetiawan 
PRL's on nickel and on Aluminum)

Please let me know if you have problems in calculating this.

Fabien




mperez@mpi-halle.mpg.de wrote:
> Dear *,
>
>  I believe the spectral function is computed for the GW band structure, ie, 
> after a GW computation (either standard one-shot or self-consistent).
>
>  Since spectral function and density of states is sometimes deemed the same 
> thing, i would like to have this clarified: how do you define the spectral 
> function, is it the fully integrated energy-dependent DOS or, else, it is the 
> k,k'-dependent imaginary part of the Green's Function.
>  
> I would appreciate if you could let me know how you calculate it, and where in the code.
>
> thanks
>
> Manolo