Quantum Transformer BSDE Solver via Multi-Layer Fully-Connected Variational Quantum Circuits

Battle Count: 72

The paper directly addresses the Black-Scholes PDE/BSDE framework, which is foundational to options pricing and derivatives valuation in quantitative finance. The model-based RL interpretation of BSDE solvers connects to optimal execution and stochastic control problems in trading. However, the work is primarily a methodological contribution to quantum ML for PDE solving rather than a direct trading strategy. The d=36 dimensionality is relevant for multi-asset portfolio problems. The quantum advantage demonstrated is modest and at compact widths, limiting immediate practical trading applications. The approach could eventually enable faster high-dimensional PDE solutions for real-time risk management and derivative pricing.

The architecture combines multiple complex components: FC-VQC modules with shifted-ring mixing, causal multi-head self-attention, time-coordinate tokenization, Euler-Maruyama BSDE simulation, and hybrid quantum-classical training. Understanding requires knowledge of quantum circuits, variational quantum algorithms, Transformer architectures, BSDE theory, and stochastic calculus. The FC-VQC construction itself references a separate paper. Training involves 10,000 epochs with Monte Carlo path simulation. No open-source code is provided, making reproduction non-trivial.

The paper provides detailed experimental setup including hyperparameters (T=1, N=10, 1000 MC paths, batch size 128, 10000 epochs, Adam lr=5e-3, seeds {42,123,456}), model architectures, and benchmark PDE formulations with analytical solutions. However, no GitHub repository or code link is provided. The FC-VQC construction references a separate paper [10]. Results are based on ideal quantum simulations with only 3 random seeds, limiting statistical robustness.