Maximized Quantum Fisher Information

• MQFI is a family of optimization problems that maximizes quantum Fisher information over admissible control strategies, defining the ultimate sensitivity in quantum metrology.
• It provides a closed-form framework to quantify the sensitivity of both pure and mixed quantum states, guiding state preparation and measurement optimization.
• MQFI finds applications in multiparameter estimation, many-body physics, and variational quantum algorithms, proving essential for both theoretical insights and practical implementations.

Searching arXiv for the cited MQFI papers to ground the article in current arXiv records. {"query":"id:(Fiderer et al., 2019) OR id:(Jing et al., 2015) OR id:(Chen et al., 2017) OR id:(Liu et al., 2024) OR id:(Cheraghi et al., 2020) OR id:(Cerezo et al., 2021) OR id:(Beckey et al., 2020) OR id:(Erol, 19 Sep 2025) OR id:(Maleki et al., 2 Jun 2026)","max_results":10,"sort_by":"relevance"} Maximized Quantum Fisher Information (MQFI) denotes the largest quantum Fisher information attainable after optimizing over a specified admissible set of controls, such as probe-state preparation unitaries, local basis changes, measurement orbits, or, in multiparameter settings, over probe states for the quantum Fisher information matrix. In the mixed-state unitary-metrology formulation that has become a canonical reference, MQFI is defined for a fixed mixed state $\rho$ and sensor dynamics $U_\alpha$ by

$$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$

with $I_\alpha$ the QFI of the encoded state and $U$ an arbitrary preparation unitary (Fiderer et al., 2019). Other strands of the literature use the same term for optimization over local unitaries in bipartite systems, over measurement orbits for entanglement witnesses, or over probe states in channel and many-body metrology. This suggests that MQFI is best understood not as a single universal functional, but as a family of optimization problems built on the same QFI objective (Erol et al., 2015, Liu et al., 2024, Erol, 2017).

1. Formal basis in quantum estimation theory

For a smooth family of states $\rho_\theta$, the quantum Fisher information $F_Q[\rho_\theta]$ sets the ultimate single-parameter precision through the quantum Cramér–Rao bound,

$$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$

where $M$ is the number of independent repetitions. The QFI is defined by

$$F_Q[\rho_\theta]=\operatorname{Tr}[\rho_\theta L_\theta^2],$$

with $U_\alpha$0 the symmetric logarithmic derivative solving

$U_\alpha$1

For unitary encoding,

$U_\alpha$2

the local generator is

$U_\alpha$3

If $U_\alpha$4, then

$U_\alpha$5

with

$U_\alpha$6

For a pure state, this reduces to

$U_\alpha$7

These identities provide the starting point for essentially all MQFI constructions (Fiderer et al., 2019).

2. Mixed-state MQFI under unitary state preparation

In the mixed-state formulation, one assumes a fixed, freely available mixed state $U_\alpha$8 of finite dimension $U_\alpha$9, full unitary control for state preparation, and a fixed parameter-dependent sensor dynamics $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$0. Ordering the eigenvalues of $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$1 as $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$2 and those of $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$3 as $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$4, the maximal attainable QFI is

$$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$5

This formula depends only on the spectrum of the probe state and the spectrum of the dynamical generator. It is therefore a closed-form characterization of the metrological potential available from a given mixed-state spectrum under unrestricted unitary preparation (Fiderer et al., 2019).

The optimal prepared state has the same eigenvalues as $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$6, but its eigenvectors are reorganized into pairings of extremal eigenvectors of $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$7: $$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$8 with

$$I_\alpha^*=\max_{U\in U(d)} I_\alpha(U\rho U^\dagger),$$9

where $I_\alpha$0 are arbitrary real phases. The construction pairs largest and smallest generator eigenvalues, thereby maximizing variance-like contributions within each relevant eigensubspace.

Rank plays a decisive role. For a pure state, $I_\alpha$1, attained by the equal superposition of extremal generator eigenvectors. For $I_\alpha$2, the MQFI becomes a convex combination of pure-state maximal QFIs associated with the nonzero eigenvalues of $I_\alpha$3. For higher rank, the coefficients $I_\alpha$4 mix contributions nontrivially. In all cases,

$I_\alpha$5

so the usual pure-state channel bound remains valid, but it is generally unattainable when only mixed probes are available. Equality requires the pure-state limit or access to distillation or post-selection that discards probes, which is outside the “freely available state” setting (Fiderer et al., 2019).

3. Time-dependent dynamics, Hamiltonian control, and analytic subclasses

For time-dependent Hamiltonians, the generator admits the representation

$I_\alpha$6

If $I_\alpha$7 has ordered instantaneous eigenvalues $I_\alpha$8, then the mixed-state QFI obeys the upper bound

$I_\alpha$9

This bound depends only on the eigenvalues of $U$0 and avoids the explicit evaluation of the full time-ordered propagator. It is attainable if the initial state is prepared with the same pairing structure as in the static case, now in the eigenbasis of $U$1, and if a parameter-independent control Hamiltonian $U$2 enforces the tracking condition

$U$3

The control therefore maintains the probe in the locally most sensitive instantaneous eigenbasis (Fiderer et al., 2019).

A complementary analytic branch concerns pure-state su(2) parametrization processes. For Hamiltonians of the form

$U$4

the maximal QFI over input states is

$U$5

where $U$6. One term is quadratic in time and arises from radial changes in the effective field amplitude; the other is oscillatory and comes from directional changes of the field. In a driven-field problem, the maximal QFI for the driving frequency is optimized at resonance, $U$7 (Jing et al., 2015).

4. Variant definitions, admissible optimizations, and conceptual boundaries

The term MQFI is used in several technically distinct ways.

In two-qubit studies, MQFI often means QFI optimized over local basis changes for a fixed generator: $U$8 This version was introduced to compare QFI with local-unitary-invariant entanglement measures such as concurrence, negativity, and relative entropy of entanglement. It produces new state orderings: there are pairs of states with equal maximized QFI but different concurrence, negativity, and REE, and conversely pairs with equal entanglement but different maximized QFI. Accordingly, MQFI is not an entanglement monotone and cannot be identified with an entanglement measure (Erol et al., 2015).

A related thesis-scale study proposed LOCC-based local Euler-angle optimization of QFI and found that the resulting quantity has a particularly strong ordering relation with the relative entropy of entanglement in two-qubit systems; the optimization domain there is operationally local unitaries, i.e. deterministic reversible LOCC, rather than general measurement-based LOCC (Chen et al., 2017). More recent two-qubit numerical work treated MQFI as intrinsic metrological capacity and reported that optimization over local unitaries makes the QFI–entanglement relationship far more predictable than fixed-generator QFI, while generator independence emerges after optimization up to numerical precision (Erol, 19 Sep 2025).

Another usage arises in entanglement criteria based on maximizing QFI over a measurement orbit. For local observable sets $U$9 and $\rho_\theta$0, one defines

$\rho_\theta$1

where the orbit is generated by orthogonal rotations of the observable family. In that setting, the separability bound remains orbit-invariant, while the left-hand side can be optimized. For local orthonormal observables and SIC-POVM-induced observable families, the resulting MQFI-based criteria detect entanglement, and the SIC-POVM version is reported to be superior in some cases (Liu et al., 2024).

In multiparameter estimation, the relevant object is the maximal quantum Fisher information matrix $\rho_\theta$2, defined—when it exists—as a matrix upper bound dominating all attainable QFIMs in the positive-semidefinite order. Its existence is tied to a single input state maximizing the Bures distance between channel outputs for all infinitesimal parameter directions. In unitary and noisy-channel models, $\rho_\theta$3 can be extracted directly from the underlying dynamics, and it yields universal tradeoff relations and scaling bounds in multiparameter metrology (Erol, 2017).

These variants delimit an important misconception. “Maximized” in MQFI does not refer to a universal optimization over everything at once. It always depends on a declared admissible class—global preparation unitaries, local unitaries, orbit rotations, generator choices, or probe states—and different classes define different resource theories and different operational meanings.

5. Metrological significance in mixed states, many-body systems, and phases of matter

A central implication of the mixed-state theory is that initial mixedness is quantitatively limiting but not qualitatively fatal for unitary metrology. With optimal unitary preparation, and with Hamiltonian control when dynamics are time-dependent, one can extract the full metrological potential inherent in the spectrum of the mixed state. In spin ensembles with thermal single-spin states,

$\rho_\theta$4

and total Hamiltonian

$\rho_\theta$5

the maximal QFI for field-amplitude estimation is

$\rho_\theta$6

For any fixed $\rho_\theta$7, $\rho_\theta$8, and for any fixed $\rho_\theta$9, $F_Q[\rho_\theta]$0, so Heisenberg scaling survives at any finite temperature; only the maximally mixed limit $F_Q[\rho_\theta]$1 annihilates QFI. For frequency estimation with $F_Q[\rho_\theta]$2, the same framework yields

$F_Q[\rho_\theta]$3

hence $F_Q[\rho_\theta]$4 scaling, extending pure-state control results to mixed probes (Fiderer et al., 2019).

In many-body physics, MQFI is also used as a measure of macroscopic coherence. For pure states in spin chains, one defines the set of additive observables

$F_Q[\rho_\theta]$5

and the normalized maximal QFI

$F_Q[\rho_\theta]$6

In the transverse-field Ising model after a sudden quench, this quantity exhibits universal revival times

$F_Q[\rho_\theta]$7

independent of the initial state and quench size, as well as “dynamical MQFI transitions,” namely nonanalytic changes caused by a switch in the local observable direction that realizes the maximum (Cheraghi et al., 2020).

A further many-body application appears in spin-1 chains. There, QFI with respect to strictly non-local string observables is used as a witness of multipartite entanglement and as a phase diagnostic. In the Haldane phase of bilinear-biquadratic and XXZ spin-1 chains, the QFI of suitable string operators scales maximally, while its critical scaling reproduces the scaling dimensions of local and string order parameters. This places MQFI in direct contact with topological order, conformal data, and multipartite entanglement depth (Dell'Anna et al., 2023).

6. Computational surrogates, variational optimization, and empirical modeling

Because direct QFI evaluation for mixed states generally requires spectral resolution of $F_Q[\rho_\theta]$8, several later developments target efficient approximations while preserving the location of the optimum.

One route is sub-Quantum Fisher Information, defined through the super-fidelity. For unitary families, the resulting lower bound $F_Q[\rho_\theta]$9 satisfies

$$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$0

but, crucially, QFI and sub-QFI are maximized by the same optimal state: $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$1 This makes sub-QFI a faithful surrogate for MQFI and allows efficient estimation on a quantum computer using $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$2 qubits for an $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$3-qubit state (Cerezo et al., 2021).

A closely related variational strategy is the Variational Quantum Fisher Information Estimation algorithm. VQFIE estimates lower and upper bounds on QFI from fidelity bounds between $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$4 and $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$5, without requiring explicit knowledge of the sensor dynamics. The lower bound can then be used as a cost function to variationally prepare states with large QFI for sensing tasks such as magnetometry, thereby operationalizing MQFI on NISQ hardware (Beckey et al., 2020).

Machine-learning approaches pursue a different approximation route. For multipartite states with collective-spin generator $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$6, support vector regression was used to predict QFI from restricted sets of experimentally accessible features. The main conclusion is that QFI is governed primarily by the interplay between collective covariance and low-order spectral moments: collective moments alone lose predictive power as Hilbert-space dimension grows, whereas adding $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$7 and $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$8 restores high accuracy (Maleki et al., 2 Jun 2026).

Empirical two-qubit studies have also modeled MQFI directly as a function of entanglement. In a dataset of 20,000 random mixed states from the Hilbert–Schmidt ensemble, MQFI—defined via optimization over local unitaries—showed strong positive correlations with concurrence, negativity, and REE. After binning and model comparison, cubic fits reached $$\Delta\theta \ge \frac{1}{\sqrt{M\,F_Q[\rho_\theta]}},$$9–$M$0, while exponential saturation fits reached $M$1, supporting the view that MQFI behaves as a smooth, saturating function of entanglement in that restricted two-qubit setting (Erol, 19 Sep 2025).

A final caveat concerns measurement. QFI is already optimized over measurements, but fixed informationally complete POVMs generally access only a fraction of it. In a frame-operator formulation, the ratio between classical and quantum Fisher information for a fixed IC-POVM is a Rayleigh quotient bounded by nontrivial eigenvalues of the associated frame operator, and no IC-POVM can saturate the QFI for all parameter directions. This delineates a distinction between state-level or control-level MQFI and measurement-limited achievable Fisher information (Saini et al., 17 Dec 2025).

7. Scope, assumptions, and open directions

Across its main formulations, MQFI is derived under explicit assumptions: finite-dimensional Hilbert spaces, known parameter dependence of the dynamics, unitary encoding for the basic mixed-state theorems, and, often, full unitary control for preparation. The mixed-state results are single-parameter and closed-system; decoherence, non-unitary dynamics, and continuous-variable systems are not covered by the exact formulas of the core theorem (Fiderer et al., 2019).

This leaves several open directions that recur throughout the literature. One is the extension of closed-form MQFI theory to noisy channels and genuinely multiparameter settings, where compatibility of SLDs and matrix ordering of QFIMs become central (Erol, 2017). Another is the systematic comparison of different admissible optimization domains: global unitaries, local unitaries, generator classes, and measurement orbits do not define the same quantity, even when all are called MQFI. A further direction is the search for experimentally tractable surrogates—sub-QFI, variational bounds, Monte Carlo lower bounds for explicit many-body wavefunctions, and learned predictors—that preserve either the value or at least the argmax structure of the exact MQFI (Cerezo et al., 2021, Beckey et al., 2020, Musso et al., 21 May 2026).

Taken together, the literature establishes MQFI as a unifying metrological notion: the maximal statistical sensitivity extractable from a quantum state once the admissible controls are specified. In mixed-state metrology it yields exact attainable formulas; in two-qubit entanglement studies it clarifies the non-monotone relation between QFI and entanglement; in many-body systems it diagnoses macroscopic coherence, criticality, and topological order; and in computational work it motivates a growing ecosystem of faithful bounds and variational approximations (Fiderer et al., 2019, Erol et al., 2015, Cheraghi et al., 2020, Dell'Anna et al., 2023).