[Wannier] Can I submit job parallely?

[Jonathan Yates]

On Tue, 22 Jan 2008, r95222066 at ntu.edu.tw wrote:

> All Dear Wannier90 user,
>           I counter a problem about when I calculate PWscf for scf
> and nscf calculation,I used parallel calculation.It generated as the
> same number wavefuction as the number of nodes .But when I want to
> calculated about "pw2wannier90.x "step ,the ouput data show
> --------------------------------------------------------

>  My system has 22 atoms.If I only use one node to calculate all step
> , I think it may be slowly.Can I has other method to deal with this
> problem ?

Bing-Hong,

  pw2wannier90 can be run correctly in parallel. I suggest using pwscf 
3.2.3 and the version of pw2wannier90.f90 available on the wannier.org 
website.

  One important constraint is that pw2wannier90 will not work when kpoints 
are distributed (ie using npools). You can only use R+G vector 
distribution.

  If you want to use npools for the scf and nscf runs, or you want to run 
pw2wannier90 on a different number of nodes, you can use the wf_collect 
parameter in the control namelist (see the pwscf documentation). This will 
cause a single wavefunction file to be written.

  There wasn't enough information in your email to exactly diagnose your 
problem - it is related to reading in the wavefunctions, so it could be 
that you had used npool>1? If you still encounter problems you need to 
show us your pwscf input files, and the commands you execute (eg if you 
run in a batch queue, your job submission script).

  Yours
   Jonathan


-- 
Dr Jonathan Yates         |    Theory of Condensed Matter Group
Corpus Christi College    |    Cavendish Laboratory
Cambridge, CB2 1RH, UK    |    Cambridge, CB3 OHE, UK
email jry20 at cam.ac.uk     |    Tel +44 (0)1223 337461