[Pw_forum] modification of phonon programs

[Paolo Giannozzi]

Eyvaz Isaev wrote:

> I introduced some changes to phonon programs so that
> now one can specify iq_start and iq_end keywords in
> ph.in file to calculate phonons for a given set of
> q-points from dyn0-file. 
> 
> If this sounds interesting I can send these modified files.

it is interesting, but it is just part of the story.

The phonon code presently uses only pools of processors,
for k-point and plane-wave parallelization. There are
however two more possible "cheap" parallelizations:
over irreps, and over q-vectors. Both involve loosely
coupled calculations that can be performed almost
independently and thus are ideally suited for a "Grid
computing" environment. Rather than via MPI calls,
parallelization could be achieved by a script
distributing the various calculations across many jobs,
collecting the results again with a script or with an
auxiliary code reading the outputd data file. This will
require some changes to both the input data and the data
files, but nothing dramatic.

There are already two methods to supply a single q-point:
the one described by Kostantin Kudin, and directly as
qx, qy, qz. Both may require in the future to be modified,
in order to better fit into the above picture.

Paolo
-- 
Paolo Giannozzi, Democritos and University of Udine, Italy