Information-theoretic approach to readout-error mitigation for quantum computers
H. Chau Nguyen · Physical Review A · 2023
We show that the method of iterative Bayesian unfolding for mitigating readout errors in quantum computers can be derived from an information theoretic analysis. This inspires more flexible applications of this error-mitigation scheme. In particular, we distinguish between structural mitigation and unstructural mitigation. Structural mitigation addresses nearly deterministic quantum computation, where the computer is expected to output a single- or few-outcome bit strings. It is shown that the readout errors alone can be corrected by few repetitions of the computation. In contrast, unstructural mitigation is designed for quantum simulation, where the computer outputs bit strings broadly distributed. In this case, one is interested in mitigating certain observables of interest. As most observables of interest are dependent on few bits and not the whole bit string, it is sufficient to mitigate the marginal distributions over these dependent bits. As long as the cross talk of readout errors can be ignored, it is shown that the iterative Bayesian unfolding applied locally for these marginal distributions gives similar results as mitigation using least squared errors. We illustrate our analysis using the data of the preparation of the Greenberger-Horne-Zeilinger state in a 127-qubit quantum computer.