We investigate whether the three normalized components of a thermodynamic stability metric, normalized fidelity I, coherence ratio ρ = T2/T1, and readout error S, can serve as training-free input features for machine learning models that predict the Φ-threshold qubit-quality class across IBM Quantum hardware. Using a strict methodology that excludes Φ from the feature set and evaluates generalization via backend-split evaluation, we find that on 445 qubits from three IBM Quantum backends, (I, ρ, S) predict the Φ-threshold class with 98.4% average balanced accuracy on held-out backends without retraining. Feature ablation reveals that ρ alone achieves 94.6% balanced accuracy and carries 70 to 78% of feature importance, while I alone and S alone are near random at 52.2% and 50.9% respectively. All four tested model architectures (Random Forest, Gradient Boosting, Neural Network, SVM) exceed 80% balanced accuracy under the strict protocol. For the narrower task of predicting T2-based hardware quality, a data-driven threshold of 0.60 outperforms the physics-derived threshold of 0.25 by 16.2 percentage points. These results establish (I, ρ, S) as a model-agnostic, training-free feature representation for predicting the Φ-threshold qubit-quality class across the tested IBM Quantum backends, with coherence ratio ρ carrying most of the predictive signal. The methods described in this paper are the subject of U.S. Provisional Patent Application No. 63/956,800 (filed January 9, 2026). No license to implement or commercialize the described methods is granted by this publication. All rights reserved.
We investigate whether the three normalized components of a thermodynamic stability metric, normalized fidelity I, coherence ratio ρ = T2/T1, and readout error S, can serve as training-free input features for machine learning models that predict the Φ-threshold qubit-quality class across IBM Quantum hardware. Using a strict methodology that excludes Φ from the feature set and evaluates generalization via backend-split evaluation, we find that on 445 qubits from three IBM Quantum backends, (I, ρ, S) predict the Φ-threshold class with 98.4% average balanced accuracy on held-out backends without retraining. Feature ablation reveals that ρ alone achieves 94.6% balanced accuracy and carries 70 to 78% of feature importance, while I alone and S alone are near random at 52.2% and 50.9% respectively. All four tested model architectures (Random Forest, Gradient Boosting, Neural Network, SVM) exceed 80% balanced accuracy under the strict protocol. For the narrower task of predicting T2-based hardware quality, a data-driven threshold of 0.60 outperforms the physics-derived threshold of 0.25 by 16.2 percentage points. These results establish (I, ρ, S) as a model-agnostic, training-free feature representation for predicting the Φ-threshold qubit-quality class across the tested IBM Quantum backends, with coherence ratio ρ carrying most of the predictive signal. The methods described in this paper are the subject of U.S. Provisional Patent Application No. 63/956,800 (filed January 9, 2026). No license to implement or commercialize the described methods is granted by this publication. All rights reserved.
We present empirical evidence that a thermodynamic stability metric Φ = I × ρ − α × S serves as a leading indicator of qubit degradation on IBM Quantum hardware and provides practical discrimination across multiple circuit and hardware validation settings. Analysis of 445 qubits across three backends (ibm_fez, ibm_marrakesh, ibm_torino) over 30 days demonstrates 100% detection of IBM-reported status transitions in the tested window, with average lead time of 163 hours (6.8 days) and maximum lead time of 480 hours (20 days) for warnings occurring at least 24 hours before degradation. The coherence ratio ρ = T2/T1 accounts for 70 to 78% of component contribution in backend-split component analysis, suggesting a substantially larger role for coherence information in stability monitoring than is reflected in prior qubit-selection heuristics. Cross-backend transfer achieves 98.4% balanced accuracy without retraining. Additional validation demonstrates 4.34x higher two-qubit gate error for low-Φ qubit pairs, 8 to 18x error discrimination across circuit depths from 10 to 500 gates, and correct identification of all five failed qubits in the tested dataset via Φ < 0. These results establish Φ as an early warning indicator for quantum processor maintenance and a practical qubit selection metric. The methods described in this paper are the subject of U.S. Provisional Patent Application No. 63/952,883 (filed January 2, 2026). No license to implement or commercialize the described methods is granted by this publication. All rights reserved.
We investigate whether a training-free thermodynamic stability metric Φ = I × ρ - α × S, computed from IBM Quantum calibration data before circuit execution, can serve as a routing signal for allocating computation between quantum hardware and classical simulation in the tested setting. Across 23 tests on real IBM Quantum hardware, we validate a Φ-based routing framework and show in direct crossover tests that LOW-Φ quantum execution can be worse than exact classical simulation for the tested classically tractable circuits: in the primary crossover test, classical simulation achieved 0.00% error while LOW-Φ quantum execution produced 8.23% error on the same circuit. Under a strict methodology with verified physical qubit binding and decisive ground truth, HIGH-Φ versus LOW-Φ error ratios ranged from 2.90x (QPE) to 30.47x (Bernstein-Vazirani) across 7 tested algorithms. The strict seven-algorithm tests constitute the primary evidence layer, while the broader 23-test suite provides supporting boundary conditions on size, depth, and reproducibility. Circuit size and depth impose additional routing boundaries: in the tested conditions, LOW-Φ execution at 15 qubits produced 94.85% error and execution at circuit depth 103 produced 67.29% error, both transitions into regimes where classical fallback was more reliable. These results support a routing decision framework in which minimum Φ across circuit qubits determines whether to execute on quantum hardware or fall back to classical simulation, with hybrid strategies retained as a natural extension. The methods described in this paper are the subject of U.S. Provisional Patent Application No. 63/956,752 (filed January 9, 2026). No license to implement or commercialize the described methods is granted by this publication. All rights reserved.
We investigate whether a training-free thermodynamic stability metric Φ = I × ρ - α × S, computed from IBM Quantum calibration or telemetry data, can drive closed-loop intervention decisions during execution-associated monitoring windows on real hardware. Across 10 paid circuit tests on three IBM backends (ibm_fez, ibm_marrakesh, ibm_torino), together with simulated and telemetry-based harness validation, we evaluate a closed-loop controller that monitors Φ during an execution-associated window and supports five intervention actions: CONTINUE, CONTINUE-DEGRADED, CLASSICAL-FALLBACK, CHECKPOINT-MIGRATE, and ABORT-RESTART. In a same-circuit controlled comparison on ibm_fez, swapping LOW-Φ qubits (Φ = 0.097, 0.030) for HIGH-Φ qubits (Φ = 0.9988, 0.9984) reduced error from 18.36% to 2.73% on an identical Deutsch-Jozsa circuit, an 85.1% relative error reduction. Under matched conditions, composite Φ selection outperformed raw T2 selection by 7.79 percentage points on ibm_fez (68.7% relative reduction) and by 6.37 percentage points aggregate across three backends. Mid-circuit dynamic branching on IBM hardware achieved 98.78% per-shot conditional consistency between probe outcome and branch marker, with an estimated 101,220 logical gates avoided on the ABORT path. A Wilson confidence interval non-overlap rule triggered ABORT-RESTART on 1 of 3 backends in the tested setting. Classical checkpoint representation at a segmentation boundary enabled suffix migration from LOW-Φ to HIGH-Φ qubits with 82.15% final fidelity, without storing an unknown quantum state. Honest counterexamples are preserved: in a three-policy cost comparison a seeded random pair outperformed the Φ-guided pair on ibm_fez, and on ibm_marrakesh raw T2 selection outperformed Φ selection under matched compiled circuit conditions. These results support a closed-loop intervention framework in which Φ can operate as a control variable during execution-associated monitoring rather than only as a pre-execution selection score. The methods described in this paper are the subject of U.S. Provisional Patent Application No. 63/973,723 (filed February 2, 2026). No license to implement or commercialize the described methods is granted by this publication. All rights reserved.
Variational quantum classifiers (VQCs) are increasingly evaluated on real quantum processors, yet their deployment behaviour can depend on post-measurement readout and variation across independently trained models. We study a four-qubit VQC for three-class traffic-state classification on two IBM Heron r2 backends. In a single-seed comparison, a four-channel linear-projection readout achieves higher clean balanced accuracy than a three-channel direct-argmax readout (94.4% versus 78.9%), whereas their hardware recoveries on ibm_kingston are 53.0% and 94.4%, respectively. A matched synthetic-noise control shows that prediction stability does not imply higher absolute noisy accuracy. Across five independently trained direct-argmax models, the clearest backend difference occurs for the moderate class and favours ibm_marrakesh ; the congested-class difference is weaker and shot-dependent, whereas free-flow recall remains stable. Finite-shot disagreement follows the inverse of the class-mean margin ordering and decreases with increasing shots, but the hardware differences do not exhibit the same monotonic behaviour. Register-level assignment-matrix indicators show directional alignment with these differences without establishing causality. Temperature scaling reduces ECE, although the residual near 0.17 is empirical rather than structural. These results demonstrate the complementary roles of readout strategy, class-resolved metrics, decision margins, and seed-level replication in hardware evaluation of VQCs.
We evaluate calibration-based digital twins for IBM Quantum hardware, aiming to reproduce hardware measurement outcomes on classical simulators. We present a workflow that builds twins from downloadable calibration CSV files by mapping coherence times, gate and readout error rates, and operation durations to thermal-relaxation, depolarizing, and readout error channels, while reconstructing a directed coupling map to restore connectivity constraints during transpilation. We compare four twin variants (CSV-built, backend-derived simulator, backend-derived noise model, and fake-backend snapshots) under a common execution and validation protocol. Experiments on two IBM QPUs, ibm_brisbane and ibm_sherbrooke, use randomized five-qubit circuits of depths 10, 20, and 30 across four optimization levels. Weighted Jaccard similarity indicates that twins constructed from downloadable calibration CSV data often achieved the closest agreement with hardware, while backend-derived twins provided competitive and practical baselines. The results further show that agreement depends on both the target device and the transpilation settings, underscoring the need to validate digital twins for the specific execution setup rather than assuming transferability across devices.
This study investigates classification performance and trainability of variational quantum classifiers on near-term noisy quantum hardware across circuit depth and qubit count regimes within the context of quantum computing and machine learning, an area of growing scientific importance given its implications for hybrid quantum-classical algorithm design and practical NISQ hardware benchmarking for near-term quantum advantage assessment. Using gradient-based parameter optimization of parameterized quantum circuits with hardware noise modeling via density matrix simulation and physical execution on IBM Eagle processors, we examine expressibility of parameterized circuits enabling nonlinear decision boundaries in feature Hilbert space, with trainability constrained by barren plateau gradients at large qubit counts in 96 circuit architectures (4 qubit counts x 4 depths x 6 ansatz families) evaluated across 5 classification datasets with 20 random initialization seeds each drawn from IBM Quantum Eagle (127-qubit) and Falcon (27-qubit) processors with noise mitigation via zero-noise extrapolation. Results indicate that 10-qubit, 4-layer VQC achieves 84.2% classification accuracy on the best benchmark, within 3.8 percentage points of classical SVM, but gradient variance decays exponentially with qubit count (variance halving per 2 added qubits), confirming barren plateau scaling (p < 0.001), with 84.2% accuracy, 3.8 pp gap to classical SVM baseline as the primary quantitative benchmark. Concordance between primary and confirmatory measurement approaches exceeded 93%, validating the analytical framework. These findings contribute empirically to quantum computing and machine learning and carry actionable implications for the design of programs and policies targeting hybrid quantum-classical algorithm design and practical NISQ hardware benchmarking for near-term quantum advantage assessment.