oh-my-librpa-fhi-aims-qsgw

name	oh-my-librpa-fhi-aims-qsgw
description	Stack-layer workflow for FHI-aims -> LibRPA QSGW/G0W0 cases. Use when users ask to mirror existing FHI-aims + LibRPA cases, prepare staged k-point or basis campaigns, run aims before LibRPA, submit case-local Slurm scripts, or debug QSGW band workflows such as qsgw_band and qsgw_band0. Keep this layer separate from ABACUS INPUT/KPT/STRU workflows.

oh-my-librpa-fhi-aims-qsgw

Treat existing case directories as the source of truth when the user says follow, mirror, same settings, or same path as Si/MgO/....

Treat this skill as the FHI-aims-side router below the top-level oh-my-librpa entrypoint.

If the request is about building or repairing the FHI-aims executable itself rather than preparing or running a case, stop and hand the task to skills/fhi-aims-build/.

Core Behavior

• Determine the requested task early:
  • g0w0_band
  • qsgw_band
  • qsgw_band0
  • qsgw
  • qsgwa
• Distinguish two execution modes:
  • FHI-aims + LibRPA fresh run
  • LibRPA-only reuse
• Explain major decisions with why + risk + verification.

Intake Markers

Treat these as strong FHI-aims markers:

• control.in
• run_librpa_gw_aims_iophr.sh
• explicit task names such as qsgw_band, qsgw_band0, qsgw, and qsgwa

Treat these only as supporting markers:

• geometry.in
• librpa.d/
• self_energy/

Do not use geometry.in, librpa.d/, or self_energy/ alone to claim FHI-aims ownership. They can also appear in mixed, reused, or postprocessed ABACUS-side bundles.

If the bundle instead centers on INPUT_scf, INPUT_nscf, KPT_*, or STRU, stop and hand the task to skills/oh-my-librpa-abacus-librpa/.

Stage-Only Workflow

Use this branch when the user wants directories and scripts prepared first, then confirmation before submission.

1. Create the new campaign root and case directories.
2. Copy the reference inputs.
3. Keep basis and species settings from the reference control.in.
4. Change only the requested axes:
   • k_grid
   • task
   • job name
   • node count
   • root or mode label
5. Keep executable paths aligned with the chosen reference case unless the user asks to switch builds.
6. Create expected runtime folders such as librpa.d/ and self_energy/.
7. Verify:
   • k_grid
   • frequency_points
   • task
   • aims path
   • librpaexe path
   • #SBATCH --nodes
   • line endings
8. Stop before sbatch until the user confirms.

Fresh FHI-aims + LibRPA Workflow

• Derive nfreq from control.in.
• Keep the aims stage active before LibRPA unless the user explicitly wants to reuse existing generated inputs.
• In shared cluster scripts, use OMP_NUM_THREADS=1 for the aims stage.
• For band workflows, the common librpa.in baseline is:
  • option_dielect_func = 0
  • replace_w_head = t
  • use_scalapack_gw_wc = t
  • parallel_routing = libri
  • binary_input = t

Submission Rules

• Submit only through Slurm:
  • cd <case> && sbatch run_librpa_gw_aims_iophr.sh
• Do not run production mpirun from the login node.
• If the user wants a helper submit script, it may only call sbatch.
• If one case already has trusted results, skip it explicitly instead of mutating the old directory.

Monitoring and Debug

• Use squeue and case-local outputs to classify the state as:
  • queued
  • running aims
  • running librpa
  • failed
  • finished
• For OMP inconsistency, deterministic reduction, or evidence-led reports, reuse the relevant local LibRPA debugging and validation workflows when available.