Unified strategy for non-invertible Fisher information matrix in quantum metrology

Min Namkung, Changhyoup Lee, Hyang-Tag Lim · 2024

Read Unified strategy for non-invertible Fisher information matrix in quantum metrology by Min Namkung, Changhyoup Lee, Hyang-Tag Lim (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-01117.txt.

Introduction

Min Namkung, Changhyoup Lee, and Hyang-Tag Lim tackle a snag in quantum measurement: sometimes the Fisher information matrix cannot be inverted. That matrix sets the best precision any estimate can reach, through the Cramér-Rao bound. In settings like distributed quantum sensing, repeated parameters make the matrix singular, and the usual fix uses a weaker bound that can overstate the error.

The authors propose a cleaner route. They add equality constraints that remove the repeated parameters, then define the bound through a generalized inverse of the matrix. The scheme handles joint estimation and distributed sensing in one frame. They revisit several known examples under the new view and find the results easier to read and closer to practice. The paper gives a concrete guideline for a recurring problem in multi-parameter quantum sensing, where several unknowns must be measured at once with the highest precision the physics allows.

Catalog record

CREATORMin Namkung, Changhyoup Lee, Hyang-Tag Lim
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-01117

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Unified strategy for non-invertible Fisher information matrix in quantum metrology Min Namkung, Changhyoup Lee, Hyang-Tag Lim Abstract In quantum multi-parameter estimation, the precision of estimating unknown parameters is bounded by the Cramér-Rao bound (CRB), defined via the inverse of the Fisher information matrix (FIM). However, in certain scenarios—such as distributed quantum sensing—the FIM becomes non-invertible due to parameter redundancy, which depends on the probe state and measurement. This issue is often handled using a weaker form of the CRB, potentially overestimating the uncertainty and underrepresenting achievable precision. Here, we propose an alternative approach by introducing equality constraints to remove redundancy and define the CRB via the Moore-Penrose pseudoinverse of the FIM. This framework enables systematic treatment of both simultaneous estimation and distributed sensing cases. We demonstrate its utility by reanalyzing several known examples within this unified perspective, highlighting improved interpretability and practical relevance. Our results offer a concrete guideline for addressing non-invertible FIMs and enhancing the precision of quantum multi-parameter estimation in realistic scenarios. Preprint: APS/123-QED I Introduction Quantum systems are typically characterized by multiple physical parameters [ 1 , 2 , 3 ] , making simultaneous measurement of these parameters essential across various quantum technologies. Particularly, it has attracted significant interest in quantum sensing [ 4 , 5 , 6 , 7 , 8 , 9 ] , which aims to measure multiple parameters with reduced uncertainty, i.e., high precision. In estimation theory, the uncertainty in estimating the multiple parameters is quantified by the covariance matrix of the estimated parameters \breve{x} , which satisfies the Cramér-Rao inequality (CRI) [ 11 , 13 , 12 , 14 ] \mathrm{cov}\left(\breve{x}\right)\succeq{F}_{x}^{-1}. (1) Here the lower bound is called the Cramér-Rao bound (CRB) and is formulated in terms of the Fisher information matrix (FIM) {F}_{x} for given true parameters {x} . Note that the CRB is calculated as the inverse of the FIM, which is feasible in many cases of quantum metrology [ 15 , 16 , 17 , 18 , 20 , 19 , 26 , 27 , 21 , 22 , 23 , 24 , 25 ] . In particular, it is known that the FIM is less than or equal to the quantum FIM (QFIM) F_{x}^{\rm(Q)} , meaning that the quantum CRB formulated as F_{x}^{\rm(Q)-1} lower-bounds the CRB of Eq. ( 1 ) [ 14 , 20 , 19 ] . However, the FIM is not always invertible and its invertibility depends on a chosen probe state and measurement scheme that defines the FIM for a given parameter encoding scheme. Non-invertible FIMs have occasionally been observed in previous studies. For example, in the case of distributed quantum sensing [ 28 ] , where the CRI of Eq. ( 1 ) is modified by a weight vector {w} under study, a weaker form of the CRB can be algebraically defined [ 29 ] . This efficiently calculable weak form has been used as the bound in the previous studies [ 7 , 8 , 9 ] , rather than directly addressing non-invertible FIMs [ 42 , 43 , 41 , 44 ] . However, this weak form is lower than or equal to the exact CRB [ 29 , 30 ] , and it may provide an optimistic bound on uncertainty. To understand the weaker form when the FIM is non-invertible, a recent work in Ref. [ 31 ] studied an example of a distributed quantum sensing scheme [ 9 ] , where the inverse FIM is calculated by removing redundancy in the set of parameters being estimated. However, a rigorous parameter-reduction process still needs to be further studied such that it applies to arbitrary quantum multi-parameter estimation scenarios. Also, one could consider changing the physical setting such as the probe state [ 32 ] , measurement scheme [ 33 ] , and parameter encoding [ 34 ] . On the other hand, one can focus on methods to address the non-invertibility of the FIM without alternating those physical configurations in case it might not be an option in limited scenarios [ 34 , 31 ] . In this work, we discuss a strategy using a parameter function that is used for rigorously and efficiently constraining redundant parameters, based on revisited classical parameter estimation theory [ 38 , 37 , 39 , 40 , 36 , 34 ] . This redundant process eventually allows an unbiased estimator to attain the CRB formulated by Moore-Penrose (MP) pseudoinverse of the FIM [ 35 ] . It consistently applies to both simultaneous estimation and distributed quantum sensing, depending on the invertibility of the FIM. We show that when the FIM is non-invertible, the key issue is to remove redundancy from the parameter set, thereby reducing it and making the FIM invertible in the reduced parameter set. We emphasize that this strategy can unify all the aforementioned methodologies without underestimating the uncertainty, and eventually apply to evaluating the QCRB from the non-invertible QFIM. Our approach further provides how to realize the effect of the projection onto FIM’s support space, which has been concisely discussed in Refs. [ 42 , 43 , 41 , 44 ] , by just tailoring an unbiased estimator. It thus supports that our scope is beyond calculation of the precision limit. To illustrate the proposed strategies, we investigate various cases in simultaneous estimation [ 20 , 19 , 21 , 23 , 22 ] and distributed quantum sensing [ 6 , 7 , 8 , 9 ] . We believe our study offers simple and clear strategies applicable to any multi-parameter estimation scenario. II Proposed approach II.1 Preliminaries Let us consider parameter-encoded states |\psi_{x}\rangle=\hat{U}_{x}|\psi\rangle for real parameters {x} and a unitary operator \hat{U}_{x} . To extract information of the parameters, a measurement represented by a positive-operator-valued-measure (POVM) \{\hat{M}_{y}\} is performed on the parameter-encoded state, where y denotes measurement outcomes. From the measured outcome {y} , one can use an estimator that yields estimates \breve{x} of multiple parameters for given true but unknown {x} . Here the POVM is considered to have an element \hat{M}_{y} whose support space spans the whole Hilbert space, such that all the parameter information is extracted from the outcome y . Otherwise, some parameter may not be extracted by the POVM. The probability to obtain y is given by p({y}|{x})=\langle\psi_{x}|\hat{M}_{y}|\psi_{x}\rangle from Born’s rule, and the associated FIM elements are defined by [F]_{jk}=\sum_{{y}}\frac{1}{p({y}|{x})}\frac{\partial p({y}|{x})}{\partial y_{j}}\frac{\partial p({y}|{x})}{\partial y_{k}} [ 11 , 13 , 12 , 14 ] . For invertible {F}_{x} , the uncertainty in estimating multi-parameters {x} , quantified by the covariance matrix \mathrm{cov}(\breve{{x}})=\mathbb{E}\left[(\breve{{x}}-\left\langle\breve{{x}}\right\rangle)(\breve{{x}}-\left\langle\breve{{x}}\right\rangle)^{\rm T}\right] , for \breve{{x}} being estimated parameters with an unbiased estimator \breve{x}={\theta}({y}) such that \langle\breve{{x}}\rangle={x} [ 19 ] . In this case, an unbiased estimator yielding \langle\breve{{x}}\rangle={x} exists to reach the CRB in Eq. ( 1 ) [ 38 ] . When the measurement is repetitively performed, the maximum likelihood estimator can be employed to asymptotically achieve the CRB in the limit of a large sample size [ 45 ] . On the other hand, when the FIM is not invertible in particular cases of quantum multi-parameter estimation [ 8 , 7 , 9 ] , one should pay attention to using the CRB in Eq. ( 1 ). Parameter estimation theory suggests that,

Read the complete text