Characterizing the performance of NISQ devices with random Clifford circuits

There has been recent interest in sampling the output distributions of pseudo-random quantum circuits as a benchmark for large quantum system [1,2]. However, based on the work of the Datta group on circuit accreditation [3] we suggest that for characterizing device performance, it is sufficient to measure the outputs of only pseudo-random Clifford circuits, a classically-simulatable subset of all quantum circuits. We require a few reasonable assumptions about our devices at IBM, namely that both Clifford and non-Clifford Z-rotations are done in software and therefore have the same error rates. Given this result, we note that a) all of the classical pre/post-processing is efficient and b) we can deal with the variational distance of the output distributions directly without proxies such as the cross entropy or heavy output generation.

[1] Cross et. al. PRA 100, 032328 (2019)
[2] Arute et. al. Nature 574, 505-510 (2019)
[3] S. Ferracin, T. Kapourniotis and A. Datta, NJP (2019)