Segmentation Fault in BSE Calculation Post-G0W0

[marco_vitek]

I am encountering a persistent issue when attempting to run a Bethe-Salpeter Equation (BSE) calculation following a G0W0 calculation. Despite sufficient memory resources, the job consistently crashes with the following error messages:

Code: Select all

allocating two-electron 4 orbital integral table    4    8    4    8
BSE setting up matrix
srun: error: b035: tasks 0-7: Segmentation fault (core dumped)
slurmstepd: error: mpi/pmix_v3: _errhandler: b035 [0]: pmixp_client_v2.c:211: Error handler invoked: status = -25, source = [slurm.pmix.8105789.4:1]

The calculations are being run on a node with approximately 1TB of RAM, and monitoring shows that the job crashes while around 700GB of memory is still available, suggesting that this is not a memory-related issue.

I have attached a tar file containing all input and output files, along with the run script, for your reference. Any insights or suggestions to resolve this issue would be greatly appreciated.

[pedro_melo]

Dear marco_vitek,

Your system is quite big for us to use it in debugging. Do you have a smaller one where the same issue occurs?

Kind regards,
Pedro

[marco_vitek]

Dear Pedro,

Thank you for your message. Unfortunately, I do not have a smaller system available. However, I can confirm that the exact same error occurs even if I rerun everything with just a gamma point instead of a 2x2x2 k-point grid.

Kind regards,
Marco Vitek

[pedro_melo]

Dear Marco Vitek,

Thank you for your reply. After having a look at your POSCAR, it seems to me that this is a system where two Li atoms are in a solution of NH3. Is this correct? If so, I think that you not use more that the Gamma point in your KPOINTS file. Otherwise you will be enforcing boundary conditions.

While I look into this I have a question: did you try running this job with other flavours of VASP, namely the Gamma-only version, vasp_gam?

Kind regards,
Pedro

[marco_vitek]

Dear Pedro,

Thank you for your assistance.

The system is actually more complex, involving two lithium cations and a solvated dielectron in ammonia. The k-points are used because the GW DoS fits well with experimental PES measurements (https://www.science.org/doi/10.1126/science.aaz7607), though I'm not entirely sure why. I am now interested in optical properties.

I also tried using vasp_gam, but encountered the same segmentation fault.
Kind regards,
Marco

[pedro_melo]

Dear Marco,

Thank you for your reply. I am looking into your files and trying to understand where the issue is and if your workflow is appropriate for the system in question. Reading your INCAR.BSE I found that you are using ANTIRES=2. Any reason as to why you want to go beyond the Tamm-Dancoff approximation? Usually this is not needed, so I am wondering if you have a reason to think that the coupling between the resonant and anti-resonant blocks is important.

Kind regards,
Pedro

[marco_vitek]

Dear Pedro,

Thank you. I chose ANTIRES=2 based on suggestions that it might give more accurate results, but since I'm still learning BSE and TDDFT, it was mainly exploratory.

Kind regards,
Marco Vitek