[Pw_forum] Two problems with ph.x and nimage option

[Florian Altvater]

Hi,
I am using QE 5.0.0 and am trying to do phonon calculations on the
naphthalene crystal with the nimage option. I encounter two different
problems:

1) When I use a k-point grid of 2x3x2 for the scf calculation I run the
following commands:

    aprun -n 192 pw.x -nimage 1 -npool 8 < scf.in <http://scf.in> > scf.out
    aprun -n 1728 ph.x -nimage 9 -npool 8 < phG.in > phG.out
    aprun -n 192 ph.x -nimage 1 -npool 8 < phGc.in > phGc.out

phGc.in is the same file as phG.in with the additional parameter
recover=.true.
Looking at the output files (phG.out and out.[1-8]_0), everything seems
fine. When I run the second phonon calculation to finish up, it ony
recognizes that the first 11 modes have been done (they are in phG.out),
but states for the other irreps "to be done". It then starts calculating
the modes starting with representation 12 but then crashes with a davcio
error. Apparently the davcio error could be related to a lot of things.
But why doesn't ph.x recognize that all the irreps have already been
calculated?

2) Using a k-point grid of 3x4x3 for the scf calculation I run the same
commands but use -npool 19 and 456 and 4104 cores respectively. This
time the calculation already crashes during the first phonon
calculation, just a few seconds after the first image is done
(representations 1-11). It spits out lots of errors like:

PGFIO-F-204/CLOSE/unit=20/illegal use of a read-only file.
 File name = ./naph_crys.wfc6    unformatted, direct access   record = 1
 In source file close_phq.f90, at line number 38

I couldn't find anything related to this error. Any ideas? Which input
files or other information would you need in order to debug?

On another machine, running the 2x3x2 calculation on just one image
works just fine. I just would like to speed up the calculation using the
image method.

Thanks for your help,
Florian

--
PhD candidate
UC Berkeley/LBNL
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20141007/90ac7d4d/attachment.html>