Benchmarking near-term quantum computers via random circuit sampling

A major challenge in the development of near-term quantum computers is to characterize the underlying quantum noise with minimal hardware requirements. We show that random circuit sampling (RCS) is a powerful benchmarking primitive that can be used to efficiently extract the total amount of quantum noise of a many qubit system by creating an exponential decay of fidelity. Compared with randomized benchmarking (RB) and its scalable variants, RCS benchmarking has the unique advantage of being flexible with the gate set, as the fidelity decay in RCS benchmarking comes from scrambling properties of random quantum circuits, which hold for generic gate sets and do not rely on any group structure. First, under a first order approximation in the noise rate, we rigorously prove the exponential decay of average fidelity of low-depth noisy random circuits, using a technique that maps random quantum circuits to a classical spin model. Second, we use the unbiased linear cross entropy as a sample efficient estimator of fidelity and numerically verify its correctness by simulating up to 20 qubits with different noise models. Third, we develop a theoretical model of the total variance of RCS benchmarking, which generalizes previous statistical analysis of cross entropy by considering the variance across different circuits. Finally, we experimentally demonstrate RCS benchmarking on IBM Quantum hardware with up to 20 superconducting qubits, which verifies our theoretical analysis. We expect RCS benchmarking to be widely applicable across different hardware platforms due to its flexibility with the gate set and architecture.