The ABINIT Users Mailing List ( CLOSED )

[Valerio Olevano <valerio.olevano@polytechnique.fr>]

> the main GW bottlenecks of the GW are:
> the generation of the KSS file
> the calculation of the screening, of thhe independent particle
> polarizability \chi^0
>
> for the first, from the next coming version, there is an option to 
> avoid
> the full (or partial) diago required to obtain also the empty (many 
> empty)
> states,
> getting them directly from the CG alghorithm. However I guess in some
> cases
> it is more convenient to perform a diago instead of requiring about 20
> times
> the occupied states.
>
> for the second case, also starting from I don't know (I sent it to 
> xavier)
> the real space fft mesh has been reduced. there is an aliasing_factor 
> in
> setmesh.f
> that has been set to 1 (before it was 2) to speed up the calculation of
> chi^0.
> I hope there will not be surprises with the fft aliasing problem.
> the tests seem to show the option is safe.
>
> We can improve, but I don't know how, in the cchi0 routines.
> by running even the first tutorial you can realize that's the real
> bottleneck.
>
> pay attention that there is someone Marc Torrent that is already 
> working
> on
> the parallelization of the code.
> So it's not the case to do twice the same job.
>
> regards
> Valerio Olevano