Circuit Cutting Improves Quantum Computing Performance

Scientists are actively seeking methods to overcome the limitations of current quantum computers, which are hampered by a small number of qubits and high error rates. Michel Meulen from Fontys University of Applied Sciences, alongside Niels M. P. Neumann and Jasper Verbree from The Netherlands Organisation for Applied Scientific Research et al., present a detailed evaluation of circuit cutting, a technique that decomposes complex quantum circuits into smaller, manageable sub-circuits. Their research benchmarks the performance of various cutting strategies when applied to the Quantum Approximate Optimisation Algorithm (QAOA), revealing that Randomized Clifford measurements significantly outperform Pauli and random unitary alternatives. Importantly, the simulations demonstrate the challenges circuit cutting faces in noisy quantum environments, particularly as the number of required circuits increases, offering crucial insights into the scalability and practical implementation of this promising mitigation technique.

Comparative analysis of wire cutting variants and probability reconstruction methods reveals significant performance differences

Wire cutting, a technique for extending quantum circuit capacity, forms the basis of this research. The study decomposes quantum circuits into smaller sub-circuits, enabling execution on hardware with limited qubit counts. Three variants of wire cutting were compared using both exact and noisy simulations to determine optimal performance.

This involved defining auxiliary states and output states to reconstruct the probability of a basis state, as exemplified by calculating the probability of measuring |010⟩ with a cut second qubit using a formula incorporating probabilities of measurements in Pauli bases. Sub-circuits of two qubits resulted from cutting the middle wire of a three-qubit circuit, and their results were classically recombined as described by Equation 1.

An alternative approach, utilizing two distinct quantum channels to reduce sampling overhead, was also investigated. One channel applied a random Clifford gate prior to measurement and its conjugate for qubit reinitialization, while the other acted as a depolarizing channel with random basis state reinitialization.

The method randomly selected between these channels with probabilities of d+1 over 2d+1 and d over 2d+1, where d represents 2k for k cut qubits, to calculate the probability of a state b using Equation 2. These randomized measurements leverage the properties of unitary 2-designs, specifically Clifford gates, to improve efficiency.

Further exploration included investigating rotational Pauli gates as a generalization of standard Pauli gates, assessing whether they could further reduce sampling overhead despite not forming unitary 2-designs. The performance of these different wire cutting strategies was benchmarked and then applied to the quantum approximate optimization algorithm (QAOA) addressing the MaxCut problem, evaluating the potential and overhead of circuit cutting for a practical application in both ideal and noisy simulation environments. This work aimed to bridge the gap between theoretical predictions and the realities of noisy quantum computation.

Randomized Clifford measurements optimise quantum circuit decomposition and exhibit noise sensitivity, particularly in the presence of depolarizing channels

Randomized Clifford measurements yielded superior performance to both Pauli and random unitary measurements in wire cutting experiments. Specifically, the research demonstrated that utilizing Randomized Clifford measurements resulted in a demonstrable advantage when decomposing quantum circuits into smaller sub-circuits.

These measurements were then compared against Pauli and random unitary approaches, establishing a clear preference for the Randomized Clifford technique. The study then investigated the challenges of circuit cutting in noisy environments, particularly as the number of circuits increased. Results indicated that providing accurate answers became increasingly difficult with rising circuit counts, highlighting a limitation of the technique under realistic conditions.

This was observed through simulations designed to mimic the effects of noise on quantum computations. Further analysis focused on a five-qubit GHZ state, chosen for its relevance to distributed quantum computing and suitability for wire cutting due to its cascade-like structure. The addition of rotation gates before and after CNOT gates complicated the state, ensuring contributions from all measurement bases and initial states.

This setup allowed for a detailed comparison of exact wire cutting methods based on Pauli measurements with the randomized measurement approach and a modified version employing random Pauli rotation gates. Applying the best performing method to a QAOA algorithm addressing the MaxCut problem on three layered graphs revealed further insights.

The cost operator was implemented using parametrized ZZ-gates, while the mixing operator utilized parametrized RX-gates. Parameters were optimized classically using Simultaneous Perturbation Stochastic Approximation, a method known for potentially faster convergence with fewer quantum circuit executions.

The layered structure of the graphs simplified the determination of optimal solutions, facilitating verification of the obtained results. The research demonstrated that increasing the number of circuit cuts introduces statistical inaccuracies, particularly when the total number of computational steps remains constant.

This unexpected behaviour arises because each sub-circuit receives a smaller share of the available computational resources, making the results more susceptible to statistical fluctuations. Experiments on QAOA algorithms for solving MaxCut problems on different graphs showed a noticeable decline in performance with increased circuit cuts, a trend exacerbated in noisy simulations.

The authors acknowledge limitations stemming from the fixed computational budget used throughout their experiments, noting that an exponential increase in computational steps is theoretically required with the number of circuit cuts. Future research should explore the use of skewed computational budgets, allocating more resources to impactful terms within the quantum circuit. Additionally, larger sample sizes, achieved through increased computational steps and repeated trials, could help reduce statistical variance, and crucially, testing these methods on actual quantum hardware is necessary to validate their practical performance.