[Pw_forum] Is it possible to separate the phonon calculation from the el-ph calculation and use the GRID technique for phonons?

[Lorenzo Paulatto]

Hello Thomas,
If I remember correctly, in the grid parallelisation different nodes compute 
different irreps independently, right?

In that case, and also if I remember correctly, the dvscf file created by each 
node is incomplete, as it only contains the dvscf computed at that node, it is 
prettu starightforward to recombine the files; it take a 20-lines fortran 
program, which I had written some time ago and then lost. Or you can do a tiny 
modification to the ph code and then attach the files with "cat".

However, I'm not sure if you'ld be able to do el-ph calculation afterwards.

hth!

On Wednesday, 7 September 2016 19:28:38 CEST Thomas Brumme wrote:
> Dear all,
> 
> my question is already given in the subject, but I want to
> elaborate on it a little bit more. I have big system of 14 atoms
> and thus 42 representations. Unfortunately, it is not possible
> to use the image parallelization when doing an el-ph coupling
> calculation. However, as 42 representations need quite a lot of
> time, I run into the time limit of our HPC (just 24h) and always
> need to restart calculations for the different q points.
> 
> I would like to know if it is possible to use the GRID parallelization
> with el-ph and how to setup it properly.
> I now tried a couple of hours and also searched the mail archive
> for help, but maybe I used the wrong search terms!?
> 
> Calculating the phonons on a grid is in fact pretty easy and I
> have no problems with that. I thought that I could maybe calculate
> the phonons on a grid, then (or at the same time?) also calculate
> the dvscf and then start the el-ph calculation with trans=.false.
> but it is always complaining about something, mostly about
> errors while reading the dvscf file...
> 
> Following the 1. grid example I calculated the phonons (I try at
> the moment with simple Al) for different q points and
> representations. Then - I thought - it would be enough to set
> the fildvscf to something once I make the finale calculation to
> collect the phonon results. But then I get problems. It seems
> the code does not use/recalculate the density. The charge density
> files in the iq subfolders of _ph0 were empty. So I copied them
> from the grid subdirectories to the iq subfolders into the _ph0 dir.
> Now, it seems, that the code is correctly calculating also the
> dvscf files. Yet, if I now start the el-ph calculation the code
> complains about errors in the dvscf file...
> 
> I could also provide the changed bash script I used to do all
> these steps but I think it won't be very helpful.
> 
> If anyone has an idea how to use the grid parallelization for
> a phonon calculation with a subsequent el-ph calculation, I
> would be very grateful for help.
> 
> Thomas


-- 
Dr. Lorenzo Paulatto
IdR @ IMPMC -- CNRS & Université Paris 6
+33 (0)1 44 275 084 / skype: paulatz
http://www.impmc.upmc.fr/~paulatto/
23-24/4é16 Boîte courrier 115, 
4 place Jussieu 75252 Paris Cédex 05