Speaker
Description
We improve the existing direct fidelity estimation protocol for quantum channels by introducing a grouped, context-preserving variant tailored to circuit-level gate calibration. The method groups input-output Pauli pairs whose input and output operators are simultaneously qubit-wise commuting, so that a single preparation-measurement configuration estimates multiple Pauli-transfer coefficients. We prove finite-sample guarantees showing that the channel-use complexity is governed by the Rényi-(1/2) effective support of the compatible groups, rather than the raw Pauli support size. Grouping therefore reduces the number of experimental configurations and can also reduce the required channel evaluations when the target weights are non-uniform within groups. We also study the protocol under readout-error mitigation, showing how the REM-induced variance amplification is affected by grouping. We demonstrate the method on parameterized $\mathrm{fSim}(\theta,\varphi)$ gates and use the grouped estimator as the reward in a reinforcement-learning calibration loop, achieving high-fidelity calibration with fewer hardware shots.
| Topic | Quantum state and process tomography with learning-based approaches |
|---|