Fault-Finding Technique Boosts Reliability of Emerging Quantum Computers

Lei Zhang, University of Maryland, and colleagues investigate a new approach to testing hybrid quantum-classical algorithms, which are key for near-term quantum computing but notoriously difficult to verify. Failure-guided fuzzing identifies problematic configurations by first locating non-convergent starting points and then refining quantum circuit parameters around them. Implementation on Variational Quantum Eigensolver and Quantum Approximate Optimisation Algorithm instances within Qiskit reveals that using failure information sharply enhances testing effectiveness compared to random approaches, with concolic seed discovery offering further advantages for specific workloads. These findings highlight a promising pathway towards more strong and reliable hybrid quantum-classical program testing.

Failure-guided fuzzing substantially improves quantum circuit error detection

A five-fold increase in detected crashes resulted from using failure-guided local fuzzing compared to random hybrid testing, overcoming a key barrier to robust verification. Hybrid quantum-classical (HQC) algorithms, such as the Variational Quantum Eigensolver (VQE) and the Quantum Approximate Optimisation Algorithm (QAOA), delegate computationally intensive tasks between a quantum processor and a classical computer. This division of labour, while promising for near-term quantum devices with limited qubit counts, introduces a complex interplay of classical optimisation and quantum evaluation, creating numerous potential failure points. The expected number of crashes drops below one for even moderately sized quantum circuits, previously hindering error identification; this is because the search space for problematic configurations grows exponentially with the number of qubits and classical optimiser settings. Reusing information from previous failures significantly enhances the efficiency of hybrid quantum-classical program testing, especially when combined with targeted fuzzing around problematic configurations. This is important because the probability of finding errors diminishes exponentially with the number of qubits, making exhaustive testing impractical. The core principle behind failure-guided fuzzing is to leverage the knowledge gained from previous test runs to intelligently guide the search for new errors, rather than relying on purely random exploration.

Qiskit, a popular open-source quantum computing framework developed by IBM, was used for the implementation, providing a flexible platform for defining and executing HQC algorithms. The framework allows for the manipulation of both the quantum circuit parameters and the classical optimisation routines. Concolic seed discovery, employing symbolic execution to find diverse classical inputs, boosted crash detection on the Variational Quantum Eigensolver (VQE) instance. Symbolic execution involves representing classical inputs as symbolic variables, allowing the testing framework to explore a wider range of input values and identify edge cases that might trigger errors. However, this benefit proved less consistent when applied to the Quantum Approximate Optimisation Algorithm (QAOA) MaxCut instance, suggesting workload-specific effectiveness. The MaxCut problem, a classic combinatorial optimisation challenge, presents different characteristics than the VQE’s ground state energy estimation, potentially requiring different fuzzing strategies. Classical enumeration, without subsequent fuzzing, identified fewer failures than even random hybrid testing, highlighting the importance of actively mutating circuit parameters. Simply trying a predefined set of parameter values is insufficient to uncover the subtle errors that can arise in complex HQC algorithms.

The findings suggest a one-size-fits-all approach to HQC testing may be unrealistic, with the optimal strategy appearing heavily dependent on the specific algorithm and problem being tackled. Different algorithms and problem instances exhibit varying sensitivities to specific parameter ranges and optimisation techniques. Current work focuses on small instances with only two or four qubits, but scaling these techniques to larger, more complex quantum programs remains a significant challenge. The expected number of crashes diminishes exponentially with qubit count, represented as B · 2−Nq, where B is the budget and Nq is the number of qubits. This exponential decay underscores the need for more efficient testing strategies as quantum computers scale up. Further investigation will focus on tailoring testing strategies to specific algorithms, paving the way for a more durable quantum ecosystem. This includes exploring adaptive fuzzing techniques that dynamically adjust the mutation strategies based on the observed failure rates and the characteristics of the algorithm being tested.

Algorithm-specific performance of concolic testing in hybrid quantum-classical verification

Progress is being made in verifying hybrid quantum-classical programs, essential for unlocking the potential of near-term quantum computers, but a fundamental tension remains between the need for thorough testing and the limited resources available. Failure-guided fuzzing consistently improves testing, even without perfect seed generation, offering a practical pathway to enhance the reliability of near-term quantum programs. The ability to identify errors despite imperfect seed generation is particularly valuable, as generating comprehensive and diverse seed inputs can be computationally expensive. Hybrid quantum-classical algorithms, such as the Variational Quantum Eigensolver and the Quantum Approximate Optimisation Algorithm, are prone to errors arising from both the quantum and classical components, making strong testing vital. These errors can manifest as non-convergence of the classical optimiser, inaccurate quantum state preparation, or incorrect measurement results.

The technique consistently outperformed random approaches, revealing vulnerabilities in these complex programs. Random testing, while simple to implement, suffers from low efficiency due to its inability to focus on potentially problematic areas of the parameter space. This represents a major step forward in verifying hybrid quantum-classical programs, moving beyond the limitations of random testing methods. Traditional software testing techniques are often inadequate for HQC algorithms due to the inherent complexity of quantum computations and the challenges of simulating quantum systems on classical computers. By concentrating testing on previously identified failure points, it efficiently uncovers vulnerabilities within these complex systems. The efficiency gains are particularly significant for larger quantum circuits, where the search space for errors is vast. Intelligently targeted testing improves the reliability of hybrid quantum-classical computations, as this approach is crucial given the difficulty of finding errors in quantum computations. Verifying the correctness of quantum computations is inherently challenging due to the no-cloning theorem and the probabilistic nature of quantum measurements.

The research demonstrated that failure-guided fuzzing consistently improves the testing of hybrid quantum-classical programs, such as the Variational Quantum Eigensolver and the Quantum Approximate Optimisation Algorithm. This matters because these algorithms are susceptible to errors from both quantum and classical components, and robust testing is essential for reliable results. By focusing testing on areas where failures have already been found, the technique proved more efficient than random testing methods. The authors suggest that reusing failure information is a promising avenue for improving the reliability of near-term quantum computations.