[Pw_forum] i/o question

[Luke Shulenburger]

Hello,

I'm using pwscf to generate wavefunctions for quantum Monte Carlo
calculations and have run into an I/O oddity I wanted to solicit advice
about.  My calculations proceed as follows:  scf calculation -> nscf
calculation with specific k-points -> pw2casino for output.  Thinking that
the nscf calculation would need significantly fewer resources than the
scf, I used 64 processors for the scf, 16 for the nscf and than 1 for
pw2casino (I need everything written to a single file at the end).  In
order for this to work, I specified wf_collect=.true. in both the scf and
nscf input files.  Both of those tasks completed without incident but
pw2casino died with an I/O error (unable to read file).  However, I rerun
everything using the same number of processors for scf and nscf, and only
having wf_collect=.true. in the second step, everything runs perfectly.

Looking into the output directory I notice two things.  When I have
wf_collect in both files, I still end up with 64 .wfc* files in the
directory as well as having directories for all of the k-points from the
scf calculation in the outdir/prefix.save/ directory.  Is it possible this
is confusing pw2casino?  Is there some way this problem could be remedied
or am I making some logical mistake with the way I'm attempting to use the
code?

Thanks,

Luke Shulenburger
Geophysical Laboratory
Carnegie Institution of Washington
(lshulenburger at ciw.edu)