Skip to content
This repository was archived by the owner on Jun 19, 2026. It is now read-only.

ligo_h1l1_mini_pipeline_test_plan.md

Latest commit

 

History

History
310 lines (214 loc) · 7.37 KB

ligo_h1l1_mini_pipeline_test_plan.md

File metadata and controls

310 lines (214 loc) · 7.37 KB

LIGO H1/L1 Mini-Pipeline: Test Plan

Version: 1.0 | Date: 2026-05-19 Authors: Carmen N. Wrede, Lino P. Casu Status: ACTIVE — Phase 0-2 complete, Phase 3-10 pending

Objective

Determine whether the L1 excess in the 20-210 Hz band is a local detector/noise artifact, or whether a physically interesting residual remains after all diagnostic checks.

Not the objective:

"Prove SSZ."

The objective is:

"Cleanly exclude artifacts, noise, phase errors, and local L1 disturbances before any physical interpretation is considered."

Phase 0 — Dataset Documentation

Record for every run:

EVENT_NAME:
GPS_TRIGGER_TIME:
H1_FILE:
L1_FILE:
SAMPLE_RATE:
WINDOW_LENGTH:
FREQUENCY_BAND:       20-210 Hz
WHITENING_METHOD:
PSD_WINDOW:
BANDPASS_METHOD:
CODE_VERSION:

Gate 0: No claim without these fields fully populated.

Phase 1 — Baseline: Identical H1/L1 Processing

Verify:

same sample rate?
same window length?
same bandpass?
same whitening?
same PSD method?
same normalization?
H1/L1 not swapped?
strain / ASD / PSD / whitened not confused?

PASS: H1 and L1 are processed identically. FAIL: Do not interpret result.

Phase 2 — Bandpower Test

Compute for windows Trigger, Off -100s, Off -300s, Off -500s, Off +100s, Off +300s:

Bandpower H1  20-210 Hz
Bandpower L1  20-210 Hz
Ratio L1/H1

Decision:

  • L1 elevated only at trigger → potentially interesting
  • L1 elevated off-source too → local L1 noise likely

Phase 3 — Gaussianity Test

For H1 and L1, trigger and off-source:

mean / std / skewness / kurtosis / excess kurtosis
count |z| > 3 / > 4 / > 5
QQ-plot / histogram

Gate 3:

If L1_trigger approximates L1_off-source:
    L1 is stationary non-Gaussian.
    No event-related claim permitted.

Current GW240925 result: L1 excess kurtosis +44.9 trigger vs +1.1 off-source. Status: L1 DIAGNOSTIC-ONLY — gate blocked.

Phase 4 — Frequency Resolution

Resolve the spectrum. Do not treat 20-210 Hz as one block.

Produce: PSD H1 trigger / PSD L1 trigger / PSD H1 off-source / PSD L1 off-source / peak list.

Check for peaks at 20-40, 50, 60, 100, 120, 150, 180, 200 Hz.

Also check: Are peaks harmonic? f_n approximately n * f0?

Decision:

  • Peaks at 50/100/150/200 Hz → European mains suspect
  • Peaks at 60/120/180 Hz → US mains suspect
  • Broad hump → general low-frequency detector noise
  • Decaying peak → resonance/oscillator

Phase 5 — Harmonic Oscillator Test

If peaks found, test: f_n approximately n * f0

Time domain check: x(t) = A * exp(-gammat) * cos(omegat + phi)

Gate 5:

If L1 shows a stable harmonic pattern:
    Detector/resonance artifact is more likely.
    Do not interpret as astrophysical signal.

Phase 6 — H1/L1 Time-Delay Check

Scan tau = -10 ms to +10 ms.

CRITICAL: Use peak_tau = argmax(|C(tau)|), not argmax(C(tau)). H1/L1 may be negatively correlated due to opposite arm orientations — a negative peak at the correct delay is a valid coherence result.

Cross-correlation: C(tau) = sum_t H1(t) * L1(t + tau)

PASS: Clear |xcorr| peak at physically allowed |tau| <= ~10 ms. FAIL: No stable |xcorr| peak, or peak explained by L1 spectral excess alone.

Current GW240925 result (updated): Full-band trigger is COHERENT_WITH_SIGN_FLIP. |xcorr| ≈ 0.991–0.9998 at Δt ≈ 0 ms. Negative sign is physically expected.

Phase 6 full-band: PASS (sign flip accounted for).

Remaining question for Phase 6b — Subband xcorr:

For each subband (20-40, 40-80, 80-120, 120-160, 160-210 Hz):
    peak_tau  = argmax(|C(tau)|)
    peak_corr = C(peak_tau)

Key question: Is the 20-40 Hz L1 excess part of the coherent signal, or a detector-local component not coherent with H1?

Phase 7 — Phase and Coherence

Compute: coherence gamma_sq(f) / phase difference Delta_phi(f) / cross-spectrum H1* L1

Decision:

  • L1 high power, no H1 coherence → L1-local artifact
  • H1/L1 coherent with correct time delay → potentially interesting

Phase 8 — Sonification

Produce:

L1_trigger_30s_20-210Hz.wav
H1_trigger_30s_20-210Hz.wav
stereo_H1_L1_trigger_30s.wav
L1_off_minus300s_30s.wav
stereo_L1_trigger_vs_offsource.wav
optional: x4 / x8 speedup

If L1 sounds the same at trigger and off-source: the excess is not an event signal.

Phase 9 — Matched Filtering / Standard-Model Comparison

Only after Phases 1-8 do not block. Check:

Does H1/L1 match the official GW waveform?
Time delay / phase / amplitude ratio / antenna pattern match?

If excess is explained by standard GR waveform: no new physics claim. Otherwise: document residual.

Phase 10 — SSZ Test: Last Step Only

Only now ask whether SSZ makes a concrete prediction for:

- why 20-210 Hz?
- why L1 stronger?
- which phase?
- which time delay?
- which amplitude scale?
- why not H1 equally strong?

Without a concrete prior prediction:

SSZ_CLAIM = NO

With a concrete prior prediction AND all gates passed:

SSZ_CANDIDATE = POSSIBLE

Still not "proven".

Hard PASS/FAIL Gate

READY_FOR_REAL_LIGO_SSZ_CLAIM = YES
only if ALL of:

[PASS] identical processing verified
[PASS] off-source does not explain L1 excess
[PASS] Gaussianity check does not block
[PASS] no harmonic detector pattern found
[PASS] H1/L1 correlate with physical time delay
[PASS] phase / coherence consistent
[PASS] standard GR does not fully explain residual
[PASS] SSZ delivered concrete prior prediction

If any check fails:

READY_FOR_REAL_LIGO_SSZ_CLAIM = NO
DATA_QUALITY_FINDING = YES   (if anomaly is documented and reproducible)

Current Status — GW240925 (2026-05-19)

Phase Status Note
0 — Dataset documentation PASS Complete
1 — Identical processing PASS Verified
2 — Bandpower test PASS L1 excess found, persistent off-source
3 — Gaussianity BLOCKED L1 kurtosis +44.9 trigger = off-source pattern
4 — Frequency resolution PENDING —
5 — Harmonic oscillator PENDING —
6 — Time delay (full-band) PASS COHERENT_WITH_SIGN_FLIP, abs(xcorr)=0.991, Dt~0 ms
6b — Subband xcorr PENDING 20-40 Hz: is it coherent or L1-local?
7 — Phase / coherence PENDING —
8 — Sonification PENDING —
9 — Matched filtering PENDING —
10 — SSZ test PENDING Gate 3 not passed; Gate 6 full-band passed

Overall:

READY_FOR_REAL_LIGO_SSZ_CLAIM: NO
DATA_QUALITY_FINDING:          YES (L1 20-40 Hz excess, kurtosis +44.9, persistent)
NEXT STEP:                     Phase 6b — Subband xcorr with |corr| to determine
                               whether 20-40 Hz L1 excess is H1-coherent or L1-local

Result Statement for Lino

We have not released an SSZ claim. The full-band trigger is H1/L1 coherent under sign flip at ~0 ms delay (|xcorr| ≈ 0.991). The previous PERSISTENT_SYSTEMATIC classification was an xcorr implementation artifact from using argmax(xcorr) instead of argmax(|xcorr|). The full-band coherence result does not resolve the L1 20-40 Hz non-Gaussianity: that band requires subband coherence analysis to determine whether the excess is part of the coherent event signal or a detector-local component. No SSZ claim.

Authors: Carmen N. Wrede, Lino P. Casu License: Anticapitalist License 1.4