Failure-Guided Fuzzing Boosts Hybrid Quantum-Classical Program Testing

Hybrid quantum-classical (HQC) algorithms, such as VQE and QAOA, are vital for near-term quantum computing but are notoriously difficult to test due to their joint optimization space of classical hyperparameters and quantum circuit parameters. Standard random fuzzing often misses failure-prone regions under realistic budgets. Lei Zhang's paper introduces failure-guided fuzzing: a two-phase strategy that first identifies non-convergent seed configurations, then performs local fuzzing of circuit parameters around those seeds. The approach reuses failure information to systematically explore risky areas.

The study implements five budgeted testing strategies on VQE and QAOA MaxCut instances in Qiskit. Results show that failure-guided local fuzzing is the primary driver of improvement over random testing, catching more faulty configurations in fewer iterations. Concolic seed discovery (combining concrete execution with symbolic analysis) provides extra gains for VQE but is less stable for QAOA, indicating that the value of symbolic reasoning depends on the algorithm. These findings offer a practical roadmap for testing HQC programs by leveraging past failures.