Average circuit eigenvalue sampling on NISQ devices

Average circuit eigenvalue sampling (ACES) was introduced by Steven T. Flammia in arXiv:2108.05803 as a protocol to characterize the Pauli error channels of individual gates across the device simultaneously. The original paper posed using ACES to characterize near-term devices as an open problem. This work advances in this direction by presenting a full implementation of ACES for real devices, along with a more accurate and device-tailored resource estimation obtained through simulations and experiments. Our simulations show that ACES is able to estimate one- and two-qubit non-uniform depolarizing error channels to high accuracy with $10^5$ shots per circuit executed. The question of estimating general error channels through twirling techniques in real devices remains open, as it is dependent on a device's native gates, but simulations with the Clifford set show promising results. Real-hardware results on IBM's Manila and Belem devices are presented, where we approximate their error channels as Pauli channels without twirling. A high-level analysis of IBM's hardware calibration data shows the protocol we implemented outputs sensible results.