openmx-data-processing

name	openmx-data-processing
description	Prepare split-local OpenMX preprocessing inputs and submit recognized OpenMX processing jobs through `inspire-openmx-submit`.

OpenMX Data Processing

1. Prepare one split-local OpenMX workdir per dataset split. Do not run train and test processing against one shared in-place interfaces/openmx directory.
2. Seed each split-local workdir from <skill_root>/templates/openmx/:

• poscar2openmx.yaml
• run_openmx.sh
• graph_data_gen.yaml
• dat_gen.sh
• graphgen.sh

3. Render the copied templates into the split-local workdir. Do not edit the skill-owned templates in place.
4. In run_openmx.sh, only render split-specific values such as SYSTEM_NAME, INPUT_ROOT, and DATASET_ROOT. Preserve the runtime environment contract.
5. Use inspire-openmx-submit for the actual OpenMX submission. inspire-openmx-submit is the authority for workspace, compute group, spec, image, and HPC submission shape.
6. Default split-local execution policy:

• default OpenMX task budget is tasks=16, cpus-per-task=2
• omp_threads defaults to 2
• default material parallelism is therefore 16
• jobs defaults to auto-calculation from the task's total CPUs
• total CPUs must come from the submitted task budget
• detect total CPUs from SLURM first, then from explicit task environment variables such as OPENMX_TOTAL_CPUS or OPENMX_TASKS * OPENMX_CPUS_PER_TASK
• do not infer total CPUs from the local interactive session CPU set
• explicit jobs / omp_threads overrides are still allowed

7. Submit the split-local entry command that runs <skill_root>/scripts/data_processing.sh --workdir <openmx_split_root> [--jobs <jobs>] [--omp-threads <omp_threads>].
8. Use as many CPUs as practical for the split, but do not exceed the number of data points.
9. Return the generated .npz path for the processed split so downstream HamGNN stages can consume it.