[Pw_forum] phonon elph espresso 3.2

Andrea Floris afloris at physik.fu-berlin.de
Thu Feb 22 11:50:07 CET 2007


Dear all,

I am encountering a problem in the phonon-e-ph calculation in the version
3.2. I previous versions I was using the q list given in the script "by
hand". In espresso 3.2 this is still possible, using ldisp=.false.
For some reason (see below) I wanto to give the q list by hand but I
want to  profit of the possibility to obtain (via the interpolation
made in lint.f90) more converged phonon linewithts (gammas) and via the
interpolation of the latter, through q2r and so on, a more converged
Eliashberg function a2F.

For Al I use ldisp=.false. and  a single q qpoint coming from a
grid q_phonon=  2 2 2;  k_fit= 8 8 8 and k_phonon = 4 4 4
(where k_fit and  k_phonon are the meshes of the dense and coarse grids of
the scf calculations with  la2F = .true. and  la2F = .false.
respectively). In this case the requirements mentioned in previous mails
of the forum are satisfied, namely:

1) k_fit/q_phonon = 4 = integer  (compatible k_fit and q grids otherwise
   e.g. the
   \epsilon_(k_fit+q) (for the elph calculation) are not found in the
   dense grid, right?)


2) k_fit/ k_phonon = 2 = integer (is this condition necessary? Does  the
phonon code really perform
a band calculation in to get the \epsilon_(k_phonon+q)? Or it simply reads
all the \epsilon_(k_phonon+q) from the \epsilon_(k_fit)?


3) another condition mentioned
k_fit/(k_phonon + q_phonon)  it means really 8/(4+2) ?? (it would not be
satisfied in this case)

In any case with this meshes I get the error
          1           2           4

  0.500000000000000      -0.500000000000000       0.500000000000000
 -0.250000000000000       0.250000000000000      -0.250000000000000
  0.000000000000000E+000  0.000000000000000E+000  0.000000000000000E+000


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     from phq_init : error #         1
     wrong order of k points

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


If I use, instead, ldisp=.true., everything is fine even if k_fit/k_phonon
is not an integer. (but of course k_fit/q_phonon must be integer).

Is it a bug of espresso 3.2 or there is something that I don't understand?
Sometime when one wants to perform phonon-e-ph calculation in the
Brillouin zone BZ, it is
faster to divide by hand the q's (i.e. paralelizing by hand:), and then
compute in parallel each q. That's why the option ldisp=.false. can be
useful also in the contest of an a2F calculation, in which by definition
one needs the q's in BZ.
Many thanks for your answer.
Andrea





--
 Andrea Floris
 Institut f"ur Theoretische Physik
 Freie Universit"at Berlin
 Arnimallee 14
 D-14195 Berlin
 Germany
 Tel ++49-30-838-53029
 Fax ++49-30-838-55258
 e-mail: afloris at physik.fu-berlin.de



More information about the users mailing list