Randomized Benchmarking in the Analogue Setting

TL;DR

This paper introduces an analogue randomized benchmarking protocol (ARB) for quantum simulators, enabling scalable error-rate measurement while accounting for SPAM errors, and demonstrates its effectiveness through case studies.

Contribution

It adapts digital randomized benchmarking to analogue quantum simulators, providing a new scalable error characterization method for these platforms.

Findings

Data fit theoretical predictions for tested noise models

Gained error-rate values for different Hamiltonian unitary sets

Compared ARB with other RB methods, discussing advantages and disadvantages

Abstract

Current development in programmable analogue quantum simulators (AQS), whose physical implementation can be realised in the near-term compared to those of large-scale digital quantum computers, highlights the need for robust testing techniques in analogue platforms. Methods to properly certify or benchmark AQS should be efficiently scalable, and also provide a way to deal with errors from state preparation and measurement (SPAM). Up to now, attempts to address this combination of requirements have generally relied on model-specific properties. We put forward a new approach, applying a well-known digital noise characterisation technique called randomized benchmarking (RB) to the analogue setting. RB is a scalable experimental technique that provides a measure of the average error-rate of a gate-set on a quantum hardware, incorporating SPAM errors. We present the original form of digital RB, the necessary alterations to translate it to the analogue setting and introduce the analogue randomized benchmarking protocol (ARB). In ARB we measure the average error-rate per time evolution of a family of Hamiltonians and we illustrate this protocol with two case-studies of analogue models; classically simulating the system by incorporating several physically motivated noise scenarios. We find that for the noise models tested, the data fit with the theoretical predictions and we gain values for the average error rate for differing unitary sets. We compare our protocol with other relevant RB methods, where both advantages (physically motivated unitaries) and disadvantages (difficulty in reversing the time-evolution) are discussed.