Minimax estimation of qubit states with Bures risk

TL;DR

This paper develops measurement strategies for qubit state estimation that achieve the optimal $1/n$ convergence rate in Bures risk, improving upon traditional methods that scale as $1/ oot n$ near the Bloch sphere boundary.

Contribution

It introduces adaptive and collective measurement schemes that attain minimax optimal $1/n$ scaling for the maximum Bures risk in qubit estimation.

Findings

Proposed estimators achieve $1/n$ Bures risk scaling.

Collective measurement approach bounds the maximum Bures risk.

Estimator based on collective measurements achieves $O(n^{-1}\log n)$ rate for quantum relative entropy.

Abstract

The central problem of quantum statistics is to devise measurement schemes for the estimation of an unknown state, given an ensemble of $n$ independent identically prepared systems. For locally quadratic loss functions, the risk of standard procedures has the usual scaling of $1/n$. However, it has been noticed that for fidelity based metrics such as the Bures distance, the risk of conventional (non-adaptive) qubit tomography schemes scales as $1/\sqrt{n}$ for states close to the boundary of the Bloch sphere. Several proposed estimators appear to improve this scaling, and our goal is to analyse the problem from the perspective of the maximum risk over all states. We propose qubit estimation strategies based on separate and adaptive measurements, that achieve $1/n$ scalings for the maximum Bures risk. The estimator involving local measurements uses a fixed fraction of the available resource $n$ to estimate the Bloch vector direction; the length of the Bloch vector is then estimated from the remaining copies by measuring in the estimator eigenbasis. The estimator based on collective measurements uses local asymptotic normality techniques which allows us derive upper and lower bounds to its maximum Bures risk. We also discuss how to construct a minimax optimal estimator in this setup. Finally, we consider quantum relative entropy and show that the risk of the estimator based on collective measurements achieves a rate $O(n^{-1}\log n)$ under this loss function. Furthermore, we show that no estimator can achieve faster rates, in particular the `standard' rate $1/n$.