Skip to content

SPDX-License-Identifier: AGPL-3.0-or-later

Commercial license available

© Concepts 1996–2026 Miroslav Šotek. All rights reserved.

© Code 2020–2026 Miroslav Šotek. All rights reserved.

ORCID: 0009-0009-3560-0851

Contact: www.anulum.li | protoscience@anulum.li

scpn-quantum-control — Experiment Roadmap: March-June 2026

Experiment Roadmap: March-June 2026

Backend: ibm_fez (Heron r2, 156 qubits) Budget: 10 min QPU / month (free tier) Repository: scpn-quantum-control v0.9.0+

Update (March 17, 2026): 9 v1.0 modules implemented (PEC, trapped-ion, ITER disruption, quantum advantage, SNN/SSGF adapters, identity binding, QSNN training, fault-tolerant UPDE). All 20 experiments validated on AerSimulator. Hardware runs pending IBM Quantum API token.

Update (March 28, 2026): All 20 experiments completed on ibm_fez hardware. 22 jobs, 176,000+ shots. Results in results/ibm_hardware_2026-03-28/. March and April roadmap experiments are now COMPLETED, not planned.

Experiment Inventory (20 total)

# Experiment Qubits Est. QPU Month Science Question
1 noise_baseline 4 10s Mar Calibration drift Feb→Mar
2 kuramoto_4osc_zne [1,3,5] 4 90s Mar Raw vs mitigated (linear ZNE)
3 kuramoto_8osc_zne [1,3,5] 8 120s Mar Extend mitigation to depth-233
4 vqe_8q_hardware 8 60s Mar Scale VQE from 4→8 qubits
5 upde_16_dd 16 60s Mar DD on full 16-layer system
6 kuramoto_4osc_trotter2 4 30s Mar Order-2 vs order-1 Trotter
7 sync_threshold 4 50s Apr Kuramoto phase transition
8 ansatz_comparison_hw 4 90s Apr Prove Knm ansatz wins on hardware
9 zne_higher_order [1..9] 4 120s Apr Optimal extrapolation order
10 decoherence_scaling 2-12 120s Apr Extract per-gate decay rate γ
11 vqe_landscape 4 0s Apr Barren plateau detection (sim only)
12 kuramoto_4osc 4 30s Baseline (Feb, done)
13 kuramoto_8osc 8 60s 8-osc trajectory (Feb, done)
14 vqe_4q 4 30s VQE ground state (Feb, done)
15 vqe_8q 8 0s Statevector only (Feb, done)
16 qaoa_mpc_4 4 20s Binary MPC (Feb, done)
17 upde_16_snapshot 16 180s 16-layer snapshot (Feb, done)
18 bell_test_4q 4 ~20s Mar CHSH violation certifies K_nm entanglement
19 correlator_4q 4 ~25s Mar ZZ cross-correlation validates K_ij topology
20 qkd_qber_4q 4 ~15s Mar QBER from hardware vs BB84 threshold

March 2026 Plan (~370s QPU)

Priority experiments

1. noise_baseline (10s) — Repeat Feb depth-5 circuit. Detects backend drift. Compare hw_R to Feb value (0.8054). If drift > 2%, flag all subsequent results for calibration correction.

2. kuramoto_4osc_zne [1,3,5] (90s) — Gate-fold at 3 noise scales, Richardson linear extrapolation. Feb baseline: 7.3% error at depth 85. Expected: ZNE reduces to ~3-4%. Produces the "raw vs mitigated" figure for the paper.

3. kuramoto_8osc_zne [1,3,5] (120s) — Same ZNE protocol at 8 qubits. Feb: 20% error at depth 233. Expected: ZNE reduces to ~10-12%. Key question: does ZNE maintain effectiveness at higher depth?

4. vqe_8q_hardware (60s) — Statevector VQE optimization followed by single-shot hardware energy evaluation. Feb 4q result: 0.05% error. Expected 8q: 0.1-0.5% error (deeper ansatz, more CZ gates). Establishes scaling trend for VQE accuracy.

5. upde_16_dd (60s) — XY4 dynamical decoupling on all 16-qubit UPDE basis circuits. Compares R(DD) vs R(no-DD) vs classical. Tests whether idle-qubit DD helps at depth ~770 where decoherence dominates.

6. kuramoto_4osc_trotter2 (30s) — Suzuki-Trotter order 2 at identical dt/steps as order 1. Direct comparison: does higher Trotter order gain accuracy faster than decoherence penalty from deeper circuits?

Total: ~370s. Buffer: ~230s for reruns or calibration.

April 2026 Plan (~380s QPU)

Priority experiments

7. sync_threshold (50s) — Sweep K_base ∈ {0.05, 0.15, 0.30, 0.45, 0.60, 0.80} at 4 qubits. Each value = 3 circuits (Z/X/Y) at dt=0.1. Maps the Kuramoto synchronization bifurcation on quantum hardware: below critical coupling K_c, R stays low; above K_c, R jumps to order ~0.5+.

Science value: first quantum measurement of the Kuramoto phase transition. Validates quantum XY ↔ classical Kuramoto correspondence at the critical point. Publication-quality standalone result.

8. ansatz_comparison_hw (90s) — Three VQE ansatze (Knm-informed, TwoLocal, EfficientSU2) optimized on Statevector, then evaluated on hardware via Estimator. Compares hw_energy for each.

Science value: Feb sim-only benchmark showed Knm-informed converges fastest. This proves the advantage survives real hardware noise. Physics-informed circuit design is a hot topic — demonstrating it on real hardware is publishable in PRX Quantum or similar.

9. zne_higher_order [1,3,5,7,9] (120s) — 5-point ZNE with polynomial orders 1 (linear) and 2 (quadratic). Tests whether higher-order extrapolation recovers more signal or overfits.

Science value: systematic ZNE study on the same circuit family. Determines optimal extrapolation strategy for XY evolution on Heron r2. Contributes to the error mitigation literature.

10. decoherence_scaling (120s) — Run 1-Trotter-step evolution at 2, 4, 6, 8, 10, 12 qubits. Records depth and R for each. Fits R_hw = R_exact * exp(-γ * depth) to extract per-gate depolarization rate γ.

Science value: γ is a single number that characterizes the backend for our circuit family. Enables predictive modeling: "at depth D, expect error E." Compares to IBM published T1/T2 and gate errors.

11. vqe_landscape (0s QPU, sim only) — Sample 50 random parameter vectors for each ansatz, compute energy variance. Low variance = barren plateau. Tests whether Knm-informed ansatz avoids barren plateaus.

Science value: barren plateaus are the #1 obstacle to VQE scaling. Showing Knm-informed ansatz has higher variance (trainable landscape) is publishable. Reference: McClean et al., Nature Comm. 9, 4812 (2018).

Total: ~380s. Buffer: ~220s.

May 2026 Plan (Candidates)

Experiment ideas (not yet implemented)

A. Layer-selective qubit assignment — On Heron r2, assign strongly-coupled SCPN layers (L3, L4, L10) to the lowest-error physical qubits. Compare UPDE-16 R vs default Qiskit layout.

Budget: ~120s QPU. Requires reading backend calibration data (backend.properties()) to rank qubits by error rate.

B. Readout error mitigation (M3) — Build measurement calibration matrix from all-0 and all-1 circuits, apply inverse to 4-osc and 8-osc counts. Separates readout error from gate error.

Budget: ~30s QPU (calibration circuits) + reprocessing of existing data.

C. Entanglement entropy measurement — Compute von Neumann entropy S(ρ_q) of per-qubit reduced density matrices during Kuramoto evolution. Requires randomized measurements (shadow tomography) or direct state tomography on ≤4 qubits.

Budget: ~180s QPU for 4-qubit partial tomography at 3 time steps.

D. QAOA with ZZ coupling terms — Extend binary MPC cost Hamiltonian to include inter-timestep correlations (ZZ terms). Tests whether richer cost structure improves QAOA control quality.

Budget: ~60s QPU. Requires extending QAOA_MPC.build_cost_hamiltonian().

E. Depth-optimal circuit decomposition — Rewrite Kuramoto evolution using native Heron r2 CZ+RZ+SX gate set directly (skip Trotter synthesis). Hand-optimized 4-qubit circuit should be 40-60% shallower.

Budget: ~30s QPU. High implementation effort but large depth savings.

June 2026 Plan (Candidates)

F. Quantum phase estimation — QPE on the XY Hamiltonian to extract eigenvalues. 4 qubits + 3 ancilla. Tests whether QPE is practical on NISQ for our Hamiltonian.

G. Variational quantum simulation — VQS (Li-Benjamin algorithm) as alternative to Trotter. Uses ansatz evolution instead of product formulas. Should be shallower for same accuracy.

H. Multi-circuit quantum error correction — Run toric code d=3 with MWPM decoder using Knm-weighted distances. Tests whether physics-aware decoding reduces logical error rate.

Paper Claims Strengthened by This Roadmap

Claim Feb Data March+ Extension
VQE 0.05% 4q 8q (scaling proof)
Decoherence curve 12 pts, qualitative Fit γ, R² (quantitative)
16-layer UPDE 46% error DD + ZNE (reduced error)
Trotter tradeoff Order 1 only Order 1 vs 2 comparison
Error mitigation None ZNE linear + quadratic
Ansatz design Sim only Hardware proof
Phase transition Not measured Bifurcation curve
Barren plateaus Not measured Landscape variance

Hardware Notes

  • Backend: ibm_fez, Heron r2, 156 qubits
  • Native gates: CZ, ID, RZ, SX, X
  • Median CZ error: ~0.5%
  • T1/T2: ~300/200 μs
  • Coherence wall: depth 250-400 (Feb calibration)
  • Budget: 10 min QPU/month (free tier)
  • Scheduling: first-come-first-served, typical queue 30s-5min