Evidence-ledgered campaigns on IBM Heron r2: legacy ibm_fez artefacts, promoted ibm_kingston raw-count datasets, follow-up popcount controls, and dated claim amendments.
The XY Hamiltonian preserves parity ideally, and the campaign measured different leakage profiles for selected even- and odd-sector initial states. Follow-up popcount controls found substantial within-sector state variation, so the current evidence supports a backend-sensitive, parity-sector and excitation-number correlated leakage asymmetry — not DLA parity alone as the established mechanism.
Protocol: equal-depth Trotter circuits prepared in known parity eigenstates, measured in the computational basis, with parity leakage defined as the fraction of shots that end in the opposite parity sector. Parity is exactly conserved by the ideal XY evolution, so any leakage is a direct measure of decoherence.
| Trotter Depth | ISA Gate Count | Leakage (even) | Leakage (odd) | Relative Asymmetry | Reps |
|---|---|---|---|---|---|
| 2 | 50 | 8.06% | 8.27% | −2.5% (baseline noise) | 12 |
| 4 | 88 | 9.82% | 8.62% | +13.98% | 21 |
| 6 | 126 | 12.91% | 10.99% | +17.48% | 21 |
| 8 | 164 | 14.43% | 12.84% | +12.41% | 21 |
| 10 | 202 | 16.58% | 14.95% | +10.91% | 21 |
| 14 | 278 | 18.98% | 17.97% | +5.58% | 21 |
| 20 | 392 | 22.95% | 21.14% | +8.55% | 12 |
| 30 | 582 | 27.71% | 25.76% | +7.58% | 12 |
ibm_kingston (Phase 1, April 2026) and ibm_fez (legacy, February 2026): 156 fixed-frequency transmon qubits each. Tunable couplers. Median CZ gate error: ~0.3–0.5%. T1 ≈ 300 µs. T2 ≈ 200 µs. Heavy-hex connectivity. Accessed via Qiskit Runtime with V2 Estimator and Sampler primitives.
| Experiment | Qubits | CZ Depth | Hardware Result | Exact Value | Relative Error |
|---|---|---|---|---|---|
| VQE ground state energy | 4 | 12 | −6.2998 | −6.3030 | 0.05% |
| Kuramoto XY dynamics (1 Trotter rep) | 4 | 85 | R = 0.743 | R = 0.802 | 7.3% |
| Kuramoto XY dynamics (3 Trotter reps) | 4 | 255 | R = 0.581 | R = 0.802 | 27.6% |
| Qubit scaling test | 6 | 147 | R = 0.482 | R = 0.532 | 9.3% |
| UPDE-16 snapshot | 16 | 770 | R = 0.332 | R = 0.615 | 46% |
| Bell inequality (CHSH) | 2 | 3 | S = 2.165 / 2.188 | S = 2√2 | 7.54σ / 8.94σ |
| Legacy QBER artefact | 1 | 2 | Reported 5.5% | Not derivable | Superseded by matched-basis re-run |
| State preparation fidelity | 4 | 8 | F = 0.946 | F = 1.0 | 5.4% |
Gate errors: Median CZ fidelity ~99.7%. Accumulates as 0.3% × NCZ.
Readout errors: ~1% per qubit. Mitigated by readout error correction matrix.
Crosstalk: Heavy-hex topology minimises but does not eliminate simultaneous-gate crosstalk.
T1 relaxation: Circuit execution time < T1 for shallow circuits. Dominates at depth >400.
T2 dephasing: Phase coherence decays exponentially. Primary limit for phase-sensitive Kuramoto measurements.