Skip to content

SCPN/FIM IBM pilot analysis

Date: 2026-05-05

Backend: ibm_kingston

Public run label: ibm-run-4c0bd60c3fc2c532

Artefacts

Raw counts:

  • data/scpn_fim_hamiltonian/fim_ibm_pilot_raw_counts_2026-05-05_ibm-run-4c0bd60c3fc2c532.json
  • SHA256: be284b9b2f71dfecd978703d979a8893e79b35dcc4537d7a372b83ba48305790

Analysis:

  • scripts/analyse_fim_ibm_pilot.py
  • data/scpn_fim_hamiltonian/fim_ibm_pilot_analysis_2026-05-05_ibm-run-4c0bd60c3fc2c532.json
  • SHA256: 90ca87916a0ae92bbf2d7212ee7def265f76b59e73925963eaedca3955c8d98b

Row metrics:

  • data/scpn_fim_hamiltonian/fim_ibm_pilot_row_metrics_2026-05-05_ibm-run-4c0bd60c3fc2c532.csv
  • SHA256: 84ed542516dd69161ece84cbb936058f78fbf1a443e474d19f25bdacead50c6f

Lambda trends:

  • data/scpn_fim_hamiltonian/fim_ibm_pilot_lambda_trends_2026-05-05_ibm-run-4c0bd60c3fc2c532.csv
  • SHA256: 967bcd8d51373e514347351dc52ff35d59bde1eabe7dad8fba9b503460b8171b

Reproduction command:

env PYTHONPATH=src /home/anulum/.local/bin/python scripts/analyse_fim_ibm_pilot.py --verify-integrity

Run size

  • Circuits: 61
  • Main FIM rows: 45
  • Readout baseline rows: 16
  • Shots: 249,856
  • Maximum live transpiled depth: 540
  • Maximum live two-qubit gates: 157

Readout baseline

The readout-only basis-state baseline over all 16 computational basis states gives:

  • Mean exact-state retention: 0.9847564697265625
  • Mean magnetisation leakage: 0.015228271484375
  • Mean parity leakage: 0.0150909423828125

This is sufficient for readout sanity checks and state-specific parity-flip correction. It is not a full 2^n x 2^n confusion-matrix inversion.

Descriptive hardware outcome

This pilot does not support a hardware-protection claim for the SCPN/FIM term.

For lambda = 1 relative to lambda = 0, across 15 matched state/depth comparisons:

  • Mean state-retention delta: -0.08836263020833333
  • Mean magnetisation-leakage delta: 0.11277669270833333
  • Mean parity-leakage delta: 0.080615234375
  • Lower magnetisation leakage: 0 / 15
  • Higher magnetisation leakage: 15 / 15

For lambda = 4 relative to lambda = 0, across 15 matched state/depth comparisons:

  • Mean state-retention delta: -0.09288736979166666
  • Mean magnetisation-leakage delta: 0.11559244791666666
  • Mean parity-leakage delta: 0.08427734375
  • Mean parity-leakage delta after state-specific readout-flip correction: 0.08717736719154022
  • Lower magnetisation leakage: 0 / 15
  • Higher magnetisation leakage: 15 / 15

Claim boundary

Supported:

  • The submitted circuits executed and returned complete count dictionaries.
  • The data are sufficient for sanity checks, readout-baseline comparison, and planning repeated randomized campaigns.

Not supported:

  • Formal hardware p-values, because there is one hardware sample per lambda/depth/state condition.
  • A claim that the FIM term improves hardware coherence.
  • A hardware many-body-localisation claim.
  • Full confusion-matrix readout mitigation, because only basis-state readout baselines were collected.

Scientific interpretation

The pilot is valuable as a falsification and design step. Under this circuit family, transpilation, backend calibration, and initial-state set, increasing lambda did not reduce measured leakage. The next hardware experiment should therefore be smaller but replicated and randomized, with layout/state randomization and repeated samples per condition before any manuscript-level hardware claim is made.