[gpaw-users] calculation with UnitCellFilter

[Jussi Enkovaara]

On 2012-12-11 10:59, Marcin Dulak wrote:
> Your example runs OK (i run 10 steps) on 12 and 24 cores on our system.

The example crashes on our system with 12 and 16 cores, and
I traced the problem to the calculation of stress tensor in GPAW. At
the moment I am suspecting that memory runs out due to creation of
several temporary arrays within the function
"_stress_tensor_contribution" in wavefunctions/pw.py, but I'll try
investigate that in more detail.

Best regards,
Jussi


> I want to add that it may be worth trying to fix the number of
> planewaves during optimization of the cell
> (see
> https://listserv.fysik.dtu.dk/pipermail/gpaw-users/2012-November/001821.html).
>
> I just run you example with PW(400), and PW(400, cell=emimbf4.cell) and
> while
> both are far from convergence at 40 optimizer steps, the latter one
> seems to be closer to converging
> (at least the total energy seems decreasing with iterations).
> Could other users of UnitCellFilter/StrainFilter experiment with this
> cell=atoms.cell setting?
>
> Marcin
>
> On 12/07/12 14:36, Vladislav Ivanistsev wrote:
>> Interesting, I can run my crystal script on 16 cores without a problem.
>>
>>
>>     this should not be the case. Can you run one of your failing jobs
>>     on 8 cores?
>>
>>
>> If I find a light example i would start a new topic.
>>
>> Vladislav
>
>
>
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>