[Paper Review] On Quantum Detection and the Square-Root Measurement

Yonina C. Eldar, G. David Forney|ArXiv.org|May 31, 2000

Quantum Information and Cryptography17 references4 citations

TL;DR

This paper proposes the least-squares measurement (LSM) as an optimal POVM for distinguishing non-orthogonal quantum states by minimizing the sum of squared errors between state vectors and measurement vectors. It proves the square-root measurement (SRM) is equivalent to the LSM under equal priors and shows the SRM minimizes detection error for geometrically uniform state sets, establishing its optimality in a least-squares sense and under symmetry constraints.

ABSTRACT

In this paper we consider the problem of constructing measurements optimized to distinguish between a collection of possibly non-orthogonal quantum states. We consider a collection of pure states and seek a positive operator-valued measure (POVM) consisting of rank-one operators with measurement vectors closest in squared norm to the given states. We compare our results to previous measurements suggested by Peres and Wootters [Phys. Rev. Lett. 66, 1119 (1991)] and Hausladen et al. [Phys. Rev. A 54, 1869 (1996)], where we refer to the latter as the square-root measurement (SRM). We obtain a new characterization of the SRM, and prove that it is optimal in a least-squares sense. In addition, we show that for a geometrically uniform state set the SRM minimizes the probability of a detection error. This generalizes a similar result of Ban et al. [Int. J. Theor. Phys. 36, 1269 (1997)].

Motivation & Objective

• To develop a systematic method for constructing POVMs that minimize the squared error between given quantum states and measurement vectors.
• To address the lack of non-asymptotic optimality proofs for the square-root measurement (SRM), a widely used but heuristic measurement.
• To generalize the SRM framework by introducing a weighted least-squares measurement (WLSM) for unequal prior probabilities.
• To establish conditions under which the SRM minimizes detection error, particularly for geometrically uniform state sets.
• To provide a new characterization of the SRM as the solution to a least-squares optimization problem.

Proposed method

• Formulates the detection problem as minimizing the sum of squared norms of error vectors between target states and measurement vectors.
• Derives the least-squares measurement (LSM) as the optimal POVM with rank-one operators minimizing this error criterion.
• Introduces a weighted least-squares measurement (WLSM) to accommodate unequal prior probabilities by assigning different weights to state errors.
• Uses singular value decomposition (SVD) of the state matrix to characterize the optimal measurement vectors and residual error.
• Proves that the SRM coincides with the LSM when prior probabilities are equal and with the WLSM otherwise.
• Applies Ostrowski's theorem on eigenvalue perturbations to analyze error sensitivity under linear transformations of the state set.

Experimental results

Research questions

• RQ1 Is the square-root measurement (SRM) optimal in a non-asymptotic, least-squares sense for quantum state discrimination?
• RQ2 Under what conditions does the SRM minimize the probability of detection error?
• RQ3 How can a general measurement be constructed to minimize the squared error between target states and measurement vectors?
• RQ4 What is the relationship between the SRM and the least-squares measurement (LSM) under equal prior probabilities?
• RQ5 How does the residual squared error change under linear transformations of the state set?

Key findings

• The SRM is proven to be optimal in a least-squares sense, coinciding with the least-squares measurement (LSM) when prior probabilities are equal.
• For geometrically uniform state sets, the SRM minimizes the probability of detection error, generalizing a result by Ban et al. [7].
• The SRM is equivalent to the weighted least-squares measurement (WLSM) when prior probabilities are unequal.
• The minimal achievable squared error depends only on the singular values of the state matrix, making it invariant under unitary transformations of the state set.
• Under general linear mixing of states, the change in minimal squared error is bounded by the extreme singular values of the transformation matrix.
• The residual squared error is invariant under unitary mixing of the states, as the singular values of the state matrix remain unchanged.