Analyzing Yambo Bse Optical Convergence

• Search local maxima in a physical window (for Si-like tasks, ~3.0–3.5 eV; broader fallback 2.5–4.5 eV).
• Exclude endpoints and tiny pre-edge noise using a significance threshold (e.g., fraction of local/global max).
• Apply one consistent method across all runs.

Common Pitfalls

• Using global “first local max” without physics constraints.
  In all runs, this produced false peaks (e.g., ~0.1 eV) and boundary artifacts (~10 eV), corrupting convergence logic.
• Searching filesystem too broadly (find /) and drowning in /proc//sys noise.
  Use targeted roots (/data, project dirs).
• Declaring “no converged candidate” due to inconsistent peak definition.
  Changing detector rules across attempts flipped outcomes (e.g., selecting highest-parameter fallback). Keep one stable peak rule.
• Assuming test failure always means wrong science result.
  Here, 7/8 tests passed consistently; failing test checked a different data path prefix than runtime data location. Treat as harness-path mismatch, not convergence-method failure.

Verification Strategy

1. Numerical sanity checks

   • Print per-run table: params, peak energy, peak intensity.
   • Confirm selected peak is physically plausible (e.g., 3.0–3.5 eV for silicon tasks).

2. Convergence checks

   • Print max ΔE and max %ΔI for selected candidate versus all dominating higher-parameter runs.
   • Confirm thresholds are satisfied with the same detector used for all runs.

3. Output contract checks

   • Ensure output file exists and has exactly 4 lines in required key=value format.
   • Re-read and parse values as the verifier would.

4. Directory existence checks (task + harness compatibility)

   • Verify selected parameter directory exists in the actual dataset root.
   • If verifier expects a different mirror path, check that path too (or create a compatibility symlink/copy when allowed).
   • Do not change scientific selection solely to satisfy a path-construction bug.

References to Load On Demand

• Reusable Python snippet for:
  • regex parameter parsing from directory names
  • constrained first-peak detection with fallback windows
  • dominance-based convergence comparison (>= in all params, one strict >)