Improving probabilistic error cancellation in the presence of non-stationary noise

Abstract

We investigate the stability of probabilistic error cancellation (PEC) outcomes in the presence of non-stationary noise, which is an obstacle to achieving accurate observable estimates. Leveraging Bayesian methods, we design a strategy to enhance PEC stability and accuracy. Our experiments using a 5-qubit implementation of the Bernstein-Vazirani algorithm and conducted on the ibm_kolkata device reveal a 42% improvement in accuracy and a 60% enhancement in stability compared to non-adaptive PEC. These results underscore the importance of adaptive estimation processes to effectively address non-stationary noise, vital for advancing PEC utility.