The ABINIT Users Mailing List ( CLOSED )

["Matteo Giantomassi" <Matteo.Giantomassi@uclouvain.be>]

> Dear all
>
> will you please let me know whether the following warning is
> important, ie, something needs to be changed, or not:
>
> "screening: WARNING - enforcing OLD GW mesh for density"

  Hi Manuel,

  Despite its name, it's not a so serious warning. It just means that
  the density and other related quantities (v_xc and, if needed, v_hartree)
  are calculated using the "coarse" FFT mesh defined according to
  ecutwfn and (ecuteps|ecutsigx).

  This mesh is usually coarser than the standard augmented FFT mesh used
in the
  GS part of the code and it introduces some numerical errors as
  it doesn't correctly describe the convolution defining rho in
  reciprocal space. According to my experience the matrix elements
  of vxc, at least for LDA, are not so sensitive to the kind of FFT mesh
  used. On the other hand the Hartree term needed in case of SCGW
  can undergo changes also of the order 0.1 eV thus affecting the _absolute_
  position of the QP energies.
  Fortunately, in most of the cases I've tested, the QP energies are
changed by the same
  amount and the change in relative positions and band gaps is much smaller.

  The new version of abinit uses by default an augmented FFT mesh
  to calculate the density while a second coarser grid is used for
  the oscillator strengths i.e <k-q|e^{-i(q+G).r|k>.
  Therefore that annoying warning has been removed.

>
> and the other thing is, in my current configuration, i get the following
> FFT
> info:
>
> "FFT parallelization level .......  0"
>
> is there any possibility of using distribution/parallelization wrt FFT, in
> GW
> calculations? i guess this is possible for ground state (lda or gga)
> calculations,
> through Goedecker's routines, but what about gw?

  No, unfortunately it's not yet possible to use the FFT parallelism in
the GW part.
  It's tricky since the FFT parallelism should be implemented
  in conjunction with a Scalapack distribution of the inverse dielectric
matrix.
  Some parts have to be rewritten, first of all the Scalapack treatment of
epsilon^-1

  On the other hand it's possible to run a GS calculation taking
  advantage of the band+FFT parallelism however the output wavefunctions
  cannot be directly re-used to generate a KSS file. One has to
  collect the Fourier components spread across the different processors
  in order to have a complete WFK file. MPI-IO would greatly facilitate
such task.

  There are developers testing and using this part but I don't know if
  this part can be considered in production.
  Maybe the MPI-IO developers or Fabien Bruneval have additional comments
on this.

  Best Regards
  Matteo Giantomassi

>
> regards
>
> Manuel Perez Jigato
> KU Leuven
>
> Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm
>