Questions: vasp-interactive cannot run a scf step but no error return when submitting a script to SLURM system

[phys-chem]

I am using VaspInteractive calculator to optimize cell.
If I submits it to a job scheduling system, no error return though no scf calculations are performed. However, if I run it on a node directly, scf calculations have be performed, and optimization is running with the ASE interface.

[alchem0x2A]

hi @phys-chem sorry for the late reply. could you provide the actual setup of you slurm job and what are the output from the slurm job? One possibility may be related to the pausing of job on a slurm system, since VaspInteractive's slurm handler may have some issue on your specific system.

[phys-chem]

hi @phys-chem sorry for the late reply. could you provide the actual setup of you slurm job and what are the output from the slurm job? One possibility may be related to the pausing of job on a slurm system, since VaspInteractive's slurm handler may have some issue on your specific system.

Thanks for your reoply! I am using the genetic algorithm of ASE with the VASP-interactive calculator. I found my job stopped in
"e = atoms.get_potential_energy()" where the VASP-interative has been set as the calculator of "atoms".
There ouputs are normal before running this code. However, my task stops in the line and no ouputs. If I run vasp_std directly without using ASE interface, single-point calculation finish quicklys. Here is my slurm script:
#!/bin/bash #SBATCH -n 64 # #SBATCH -N 1 # num of node #SBATCH -t 2-5:00 # max time: D-HH:MM #SBATCH -p defaultp # list #SBATCH --mem=10000 # memeory MB python main.py

[phys-chem]

Although I only run a single-point calculation, the OUTCAR is updated slowly. More than one ionic steps have been calculated and my program keeps running VASP-interactive without running other codes. Do you know some possible reasons? Thank you very much!

[alchem0x2A]

interesting, regarding the single-point calculation, the OUTCAR file may indeed be updated slower than your ionic step (the interactive code in vasp may not flush the buffer timely), you will normally see the OUTCAR file updated around 4-5 ionic steps as I remember. one way to avoid this is to apply our custom patch https://github.com/ulissigroup/vasp-interactive#enhanced-interactive-mode-by-patching-vasp-source-codes, but may be advanced topic.

on the other hand, could you specify whether your vasp code

1. only performs 1 scf loop or so, never updated
2. finishes 1 ionic step (i.e. one single point calculation related with ASE), thus you can get the energy but not going beyond