Unexpected Termination of VASP Simulation

[burakgurlek]

Hi,

I am runnung a phonon calculation with VASP via phononpy, e.g. same job with different positions of atoms, however, some of those jobs randomly cancelled. I tried to debug this with the people from the computing facility and according to them the result is not enough memory. However, once I rerun the same job it again randomly works. Other simulations with same settings having different atom positions works well. I am puzzelled with that. Would someone experience a similar issue or have an idea?

I attach the files and numstat data for the simulation.

files.7z

[alex]

Hi burakgurlek,

I've checked your POSCAR also wrt. symmetry and first thought: sometimes you'll have a symmetric structure, so it works, because of less memory requirements, mostly not: fail.

Another note: The attached POSCAR shows 96(!) f.u. of (C3H2)., where I found a P-1 structure with just 12 f.u., 8 times less atoms, also far less effort in later (frequency) analysis.
I'd suggest if you need phonopy, go for that 12 f.u. structure, make a P1 with no symmetry out of it and end up with 60 atoms(!) instead of 480. And see if this is sufficient for your purpose.

Cheers,

alex

[burakgurlek]

Dear Alex,

thanks for the response. It could be the symmetry issue, but I have no experience how symmetry effects scf calculations. In addition, I did not get your second point. Yes I will use phonopy but this is for 2x2x2 unit-cell that is why it is 480 atoms. I do not see how you do reduce this number and expect the same accuracy.

Regards,
Burak

[alex]

Hello Burak,

about your questions:
1) VASP exploits symmetry incertain ways, which leeds also to reduced resource requirement, hence less memory.
2) what I was suggesting(!) is to reconsider your approach, because there is the much smaller cell with symmetry P-1 (no memory issues later).

I do not understand why you want to calculate a cell containing 12 times the necessary atoms, when you can get along with the smaller set.

Best regards,

alex

[burakgurlek]

Dear Alex,

thanks for your answer

1- I guess the symmetry is not the issue here as the simulation works in resubmission of the same files. There should be a problem mostly regarding the nodes I use but hpc guys cannot see other than memory error. I do not think something inherently wrong with symmetries
2- In the phonon calculation, the size of the unit-cell has certain effects. Thats why simulation via primitive-cell could be problematic.

Regards,
Burak