[gpaw-users] trajectory and pickle files do not print

[Ask Hjorth Larsen]

Hi

On Fri, 11 Mar 2011, Ferguson, Glen A. wrote:

> Hi,
>
> I'm running calculations on the Bluegene/P cluster at Argonne national 
> Laboratory. The calculations run many jobs in parallel over a large 
> number of nodes. So 64 jobs can be run over 512 nodes simultaneously.
>
> The jobs run but there is a problem with printing the trajectory and 
> pickle files. The output files are printed fine but only the trajectory 
> and pickle files for the first job are printed.
>
> The pertinent part of the script is below.
>
> structure = io.read('disp_%d.xyz'%(j),format='xyz')
> calc = GPAW(...)
> calc.set(txt='disp_%d.out'%j)
> structure.set_calculator(calc)
> job_name = 'disp_%d'%(j)
> relax = BFGS(structure,trajectory=job_name + '.traj', restart=job_name + '.pckl')
> relax.run(fmax=0.05)
>
> With this script I get the disp_#.out files for all jobs but only get 
> the disp_1.traj and disp_1.pckl files.
>
> Any idea how to get these files to print?
>
> Glen Ferguson,
> Postdoctoral Fellow
> Material Science Division
> Argonne National Laboratory
>
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>

Normally it is assumed that only the "master" writes, where the master has 
rank 0.  If you have multiple "submasters" that should write, you can 
instantiate a PickleTrajectory manually and supply the "master" boolean as 
appropriate on the different ranks.

Unfortunately the optimize module is not coded to allow a similar feature, 
but it may be possible to "restart" by using the replay_trajectory method 
(this I think is documented somewhere on the web page).

Regards
Ask