Quantitative Universal Approximation for Noisy Quantum Neural Networks

Battle Count: 59

The paper is directly motivated by quantitative finance (option pricing, risk measures, sensitivities) and provides explicit error bounds for Black-Scholes Put pricing on noisy quantum hardware. The theoretical framework for approximating expectation functions E[Φ(x+L)] is directly applicable to derivative pricing. However, practical relevance is currently limited by NISQ-era noise levels (α=0.365), the small number of qubits (5), and the fact that classical methods (Monte Carlo, PDE) remain vastly superior for current problem sizes. The paper is more foundational/theoretical than immediately deployable for trading systems, but provides important error guarantees for future quantum advantage claims in finance.

High complexity: requires quantum circuit design (Hadamard gates, uniformly controlled rotations, multi-qubit entangling gates), density-operator formalism for noise modelling, CPTP channel theory, Kraus operator decomposition, hardware calibration data integration, Qiskit Runtime execution with SamplerV2, and sophisticated optimisation (L-BFGS-B, Adam). The theoretical framework involves advanced quantum information theory and stochastic analysis. Practical deployment requires access to quantum hardware and expertise in both quantum computing and quantitative finance.

The paper provides detailed circuit architecture, hardware parameters for three platforms (IBM, Quantinuum, Rigetti), explicit noise calibration formulas, and uses standard Qiskit Runtime primitives (SamplerV2). However, execution on specific IBM hardware (ibm_fez) introduces non-determinism. The theoretical framework is fully specified with explicit constants. No public code repository is mentioned.