Hyukgun Kwon, Kento Tsubouchi, Chia-Tung Chu, Liang Jiang (Mar 24 2025).
Abstract: We establish the necessary and sufficient conditions for unbiased estimation in multi-parameter estimation tasks. More specifically, we first consider quantum state estimation, where multiple parameters are encoded in a quantum state, and derive two equivalent necessary and sufficient conditions for an unbiased estimation: one formulated in terms of the quantum Fisher information matrix (QFIM) and the other based on the derivatives of the encoded state. Furthermore, we introduce a generalized quantum Cramér-Rao bound, which provides a fundamental achievable lower bound on the estimation error even when the QFIM is non-invertible. To demonstrate the utility of our framework, we consider phase estimation under unknown Pauli noise. We show that while unbiased phase estimation is infeasible with a naive scheme, employing an entangled probe with a noiseless ancilla enables unbiased estimation. Next, we extend our analysis to quantum channel estimation (equivalently, quantum channel learning), where the goal is to estimate parameters characterizing an unknown quantum channel. We establish the necessary and sufficient condition for unbiased estimation of these parameters. Notably, by interpreting unbiased estimation as learnability, our result applies to the fundamental learnability of parameters in general quantum channels. As a concrete application, we investigate the learnability of noise affecting non-Clifford gates via cycle benchmarking.
