[ase-users] Script won't continue when convergences uses too many ionic steps (Vasp2-calculator)

[Rasmus Vester Thøgersen]

Hi Michael,


Thanks for the reply. The cluster is using SLURM. Were you able to solve this issue? Would the issue lie in settings out of my control, or is there some tag I could use in the jobfile to circumvent this?


Best regards,

Rasmus

?

________________________________
From: Michael Joseph Waters <michael.j.waters at northwestern.edu>
Sent: 08 April 2020 23:58
To: ase-users at listserv.fysik.dtu.dk; Rasmus Vester Thøgersen
Subject: Re: [ase-users] Script won't continue when convergences uses too many ionic steps (Vasp2-calculator)

Hi,

Are you using running this on a computing cluster with SLURM or PBS? I've had issues with doing something similar due to SLURM settings.

Best,
-Mike
________________________________
From: ase-users-bounces at listserv.fysik.dtu.dk <ase-users-bounces at listserv.fysik.dtu.dk> on behalf of Rasmus Vester Thøgersen via ase-users <ase-users at listserv.fysik.dtu.dk>
Sent: Wednesday, April 8, 2020 4:59 AM
To: ase-users at listserv.fysik.dtu.dk <ase-users at listserv.fysik.dtu.dk>
Subject: [ase-users] Script won't continue when convergences uses too many ionic steps (Vasp2-calculator)


Hi,



I'm using ASE with the Vasp2-calculator, and I'm having an issue with my script not continuing if a particular run struggles to converge and uses too many ionic relaxation steps. The code is something like this:



from ase.calculators.vasp import Vasp2

from ase import io



at = io.read('POSCAR')

calc = Vasp2(isif=2, ....) # Excluded detailed initial parameters for brevity

at.set_calculator(calc)

at.get_potential_energy()



calc.set(isif=3)

at.set_calculator(calc)

at.get_potential_energy()





Between changing the ISIF-parameter from 2 to 3, the script stops running while the job on the hpc-resource is still running on all allocated cores, just not doing anything. This only happens for runs where VASP struggles to converge (typically automatically generated defect structures where the initial guess might be poor), typically at least 100 ionic steps. It does not happen every time this occurs, and I've never experienced it when running on structures that converge more quickly.



I'm not sure where this problem lies - it could perhaps also be a problem with the hpc-cluster itself where something just times out while waiting for VASP to finish its job? I struggle to find out where to begin with the troubleshooting.



Have you encountered this issue before?



Best regards,

Rasmus Vester Thøgersen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/ase-users/attachments/20200409/09ebd8eb/attachment.html>