Local Quantum Fisher Information

• Local Quantum Fisher Information is a family of locality-constrained QFI constructions that quantify discord-like correlations in bipartite and many-body quantum systems.
• It is computed by optimizing over local generators, observables, or sensing directions to reveal metrological sensitivity under imposed locality constraints.
• LQFI methods enable analytic solutions in spin models and open systems, capturing sudden transitions and robustness of quantum correlations under decoherence.

Local quantum Fisher information (LQFI) is not a single universally fixed object but a family of locality-constrained quantum Fisher constructions. In the most common bipartite usage, it is the minimum quantum Fisher information (QFI) obtained when the generator is restricted to act on one subsystem only, and it functions as a discord-like quantifier of nonclassical correlations (1711.02323). In other strands of the literature, the same phrase denotes the QFI of a reduced subsystem with respect to an extensive observable on that subsystem, or the QFI associated with a chosen direction in a space of locally encoded parameters (Ferro et al., 27 Mar 2025, Farokhi, 20 May 2026). Across these formulations, the common theme is local metrological sensitivity under a locality constraint, but the constrained object itself—generator, subsystem, measurement class, or parameter direction—depends on context.

1. Definitions and scope

For a state $\rho=\sum_i p_i |\psi_i\rangle\langle\psi_i|$ and Hermitian generator $H$, the QFI used throughout this literature is

$$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$

The best-known LQFI construction specializes $H$ to a local generator on one subsystem, typically

$$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$

and then minimizes the QFI over all directions $\mathbf r$ (1711.02323, Slaoui et al., 2019).

Usage of “local”	Quantity	Representative source
Local generator on one party	$\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$	(1711.02323, Slaoui et al., 2019)
Local subsystem	$F(\rho_A,O_A)$ or $\chi(\rho_A;O_A)$	(Ferro et al., 27 Mar 2025)
Local sensing direction	$$F_Q(\boldsymbol u^\top\boldsymbol\theta)=\boldsymbol u^\top F_Q(\boldsymbol\theta)\boldsymbol u$$	(Farokhi, 20 May 2026)
Local parameter displacement	QFIM element $H$0 or flow $H$1	(Parlato et al., 22 Aug 2025)

In the discord-oriented formulation, the optimized quantity for a qubit–qudit state can be written as

$H$2

where $H$3 is a real symmetric $H$4 matrix with entries

$H$5

This quantity vanishes exactly for zero-discord states in the relevant direction, and for pure bipartite states it coincides with geometric discord, taking the form $H$6 in terms of Schmidt coefficients $H$7 (1711.02323).

A complementary notion is measurement-induced Fisher information under restricted measurement classes. For bipartite systems, local measurements recover the QFI of the reduced states,

$H$8

while larger measurement classes interpolate between local accessibility and the full global QFI (Lu et al., 2012). This distinction is central to avoiding a common terminological confusion: LQFI may refer either to locally generated sensitivity or to locally accessible sensitivity, and these are not the same construction.

2. Analytic structure and relation to local quantum uncertainty

The formal similarity between LQFI and local quantum uncertainty (LQU) is one of the defining features of the subject. LQU is based on Wigner–Yanase skew information rather than QFI, but for $H$9 systems both quantities reduce to optimization over local Pauli directions and admit closed forms based on $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$0 matrices (Slaoui et al., 2019). The basic inequality

$$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$1

implies

$$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$2

so LQFI upper-bounds LQU in the standard local-phase-estimation setting (Slaoui et al., 2019).

For two-qubit $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$3 states, the structure becomes especially explicit. Several papers show that the matrix entering the LQFI optimization is diagonal in the natural basis, so the optimization reduces to comparing a small number of branches (Haseli, 2020, Fedorova et al., 2021, Yurischev et al., 2023). In the general $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$4-state treatment,

$$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$5

with analytic formulas for $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$6, $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$7, and $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$8 directly in terms of the matrix elements and eigenvalues of the state (Yurischev et al., 2023). This closed form is technically important because it turns LQFI from a variational definition into a directly computable correlation functional for broad classes of thermal and dissipative spin states.

The branch structure also explains sudden but continuous changes in LQFI. In Bell-diagonal situations, the switching condition can be parameter-only and temperature independent, while for general $$\mathcal{F}(\rho,H)=\frac12\sum_{i\neq j}\frac{(p_i-p_j)^2}{p_i+p_j}\,|\langle\psi_i|H|\psi_j\rangle|^2.$$9 states with field asymmetry the equal-branch conditions become temperature dependent, allowing one or more abrupt branch changes during a thermal sweep (Fedorova et al., 2021, Yurischev et al., 2023). This suggests that many reported “sudden transitions” in LQFI are optimization-branch phenomena rather than singularities of the density matrix itself.

3. Equilibrium spin systems

A large part of the LQFI literature is devoted to thermal two-spin or dimerized spin models, where the density matrix is often of $H$0 form and analytical evaluation is possible. In the two-qubit Heisenberg $H$1 model, LQFI and LQU both decrease with temperature, are largest near the isotropic $H$2 point, and vanish in the Ising limit $H$3; in the isotropic $H$4 model with magnetic field, the field produces a sudden change in LQFI around $H$5, with low-temperature enhancement for $H$6 (Slaoui et al., 2019).

For the two-qubit Heisenberg $H$7 chain with a $H$8-directed Dzyaloshinskii–Moriya term, the thermal state also has $H$9 structure, enabling a closed LQFI formula in terms of diagonal entries of the corresponding $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$0 matrix. In that model, $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$1, $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$2, $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$3, and $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$4 all enhance LQFI, while temperature suppresses it; at very low temperature the LQFI is equal to one and starts to decay only after a threshold temperature, with the threshold increasing with $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$5 (Haseli, 2020).

Related conclusions appear in the dipolar two-spin system with DM interaction. There, lQFI is again treated as a discord-like correlation measure, is always larger than LQU for the parameter ranges studied, decreases monotonically with temperature, and is strongly enhanced by the DM coupling, which acts as a catalyst for quantum correlations in that model (Muthuganesan et al., 2020).

In the fully anisotropic Heisenberg $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$6 chain with both DM and KSEA interactions, LQFI, LQU, and entropic discord share the same branch boundary,

$$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$7

show the same high-temperature decay law $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$8, and can all display local increases with temperature near the branch boundary (Fedorova et al., 2021). In the more general $$H_A=\boldsymbol{\sigma}\cdot\mathbf r\otimes \mathbb I,\qquad |\mathbf r|=1,$$9-state extension with inhomogeneous magnetic field, the field asymmetry makes the branch structure richer and allows a sequence of sudden transitions as temperature varies (Yurischev et al., 2023).

Taken together, these results establish a stable pattern for equilibrium spin dimers: LQFI is analytically tractable in broad $\mathbf r$0-state families, typically exceeds LQU, is suppressed by thermal mixing, and is often enhanced by anisotropy, spin-orbit-type antisymmetric exchange, or parameter regimes that favor Bell-like eigenvectors.

4. Dissipation, decoherence, and non-Markovian dynamics

In dissipative settings, LQFI is often used to track nonclassical correlations that survive when entanglement becomes fragile. The Ising–XYZ diamond-chain study provides a representative example. There the system consists of two interstitial Heisenberg spins coupled to nodal Ising spins and to two independent zero-temperature reservoirs, with the reduced two-qubit state remaining $\mathbf r$1 shaped under the Lindblad evolution. For the maximally entangled initial state $\mathbf r$2, the LQFI initially matches the maximal entanglement and then decays only mildly with small oscillations before reaching a steady state, while concurrence undergoes stronger decay and sudden death/birth effects. For the initially separable state $\mathbf r$3, the environment can generate entanglement after a delay, whereas LQFI develops oscillations and saturates, remaining nonzero even when entanglement is absent or intermittently vanishing (Carrion et al., 2024).

The same work also sharpens the distinction between LQFI and basis-dependent coherence. In the $\mathbf r$4 basis, the $\mathbf r$5 coherence contains only quantum correlations and LQFI is more robust than coherence; in the $\mathbf r$6 basis, the coherence can exceed both LQFI and concurrence because local coherences contribute in addition to genuine correlations (Carrion et al., 2024). This directly addresses a recurrent misconception: comparisons between LQFI and coherence are meaningful only after specifying the reference basis.

LQFI has also been promoted as a non-Markovianity monotone for bipartite $\mathbf r$7 systems. In that formulation,

$\mathbf r$8

and Markovian evolution implies

$\mathbf r$9

Positive derivative is therefore taken as a witness of information backflow, and the corresponding non-Markovianity measure is the total accumulated increase of LQFI over all revival intervals (Dakir et al., 2024). The construction is worked out for phase damping, amplitude damping, and depolarizing channels, and is compared with the analogous LQU-based quantity, with the reported ordering

$\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$0

(Dakir et al., 2024).

A more restrictive open-system relation appears in the two-coupled-qubit model with independent non-Markovian reservoirs. There the phase-estimation QFI is independent of the encoded phase parameter and is protected by stronger inter-qubit coupling or stronger memory effects, but the closed-system-style bound between LQU and QFI survives only in special cases: when the qubit–qubit coupling is switched off or when the initial state is Bell-like (Wu et al., 2018). This suggests that LQFI–LQU inequalities that hold in unitary settings need not transfer unchanged to generic dissipative dynamics.

5. Subsystem LQFI in many-body and topological settings

A distinct many-body usage defines local QFI as the QFI of a reduced density matrix $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$1 with respect to an extensive observable on a subsystem,

$\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$2

In equilibrium, if correlations cluster exponentially, then $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$3 and $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$4 for large subsystems. At criticality the scaling can become superlinear—for the critical Ising chain, $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$5—but $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$6 still vanishes in the large-subsystem limit (Ferro et al., 27 Mar 2025). In this formulation, nonzero asymptotic $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$7 is the signature of macroscopic multipartite entanglement, and ordinary equilibrium local systems do not generate it.

The central nonequilibrium mechanism in that work is localized kicking. Starting from a low-temperature ferromagnetic phase, a local finite-support unitary perturbation creates quasiparticles or domain walls whose moving front causes cluster decomposition to fail at Euler scales inside a subsystem. During the window $\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$8, the local QFI becomes quadratic,

$\min_{H_A}\mathcal F(\rho,H_A\otimes\mathbb I)$9

whereas at late times and fixed $F(\rho_A,O_A)$0 one recovers $F(\rho_A,O_A)$1 (Ferro et al., 27 Mar 2025). Multiple kicks and periodic localized driving extend this mechanism and provide control over the subsystem size that remains macroscopically entangled.

Topological protection supplies a different route to persistent local metrological sensitivity. In the open Kitaev chain, a single-site encoding followed by Hamiltonian evolution and single-site boundary readout yields a boundary qubit QFI

$F(\rho_A,O_A)$2

In the topological phase, a Majorana zero mode pins the boundary QFI to a nonzero plateau that persists for times exponentially long in system size. The key diagnostic is an encoding-axis asymmetry: one local axis couples to the boundary Majorana quadrature and produces the plateau, while the orthogonal axis couples to the remote quadrature and is exponentially suppressed (Płodzień et al., 1 May 2026). The plateau survives moderate quenched disorder and remains visible under parity-preserving interactions in finite-size simulations (Płodzień et al., 1 May 2026).

For large many-body mixed states, direct QFI evaluation is difficult. A tensor-network method based on the Lyapunov integral representation of the symmetric logarithmic derivative rewrites

$F(\rho_A,O_A)$3

and computes QFI using MPO/MPS time evolution rather than eigendecomposition (Wójtowicz et al., 12 Jun 2025). That work does not introduce a separate local-QFI formalism, but its framework is explicitly compatible in principle with local generators, local encodings, and reduced density matrices (Wójtowicz et al., 12 Jun 2025).

6. Operational extensions and neighboring constructions

Beyond correlation quantification, locality-constrained Fisher information enters several operational frameworks. In distributed quantum sensing with local phase encoding,

$F(\rho_A,O_A)$4

the relevant scalar quantity is the QFI along a sensing direction,

$F(\rho_A,O_A)$5

For any two orthogonal unit directions $F(\rho_A,O_A)$6 and $F(\rho_A,O_A)$7,

$F(\rho_A,O_A)$8

with equality for all equatorial probes when $F(\rho_A,O_A)$9 and for GHZ states for all $\chi(\rho_A;O_A)$0 (Farokhi, 20 May 2026). Heisenberg-limited precision in one direction therefore forces zero QFI in every orthogonal direction, which is interpreted there as intrinsic parameter privacy (Farokhi, 20 May 2026).

In the local-estimation theory of measurements, informationally complete POVMs are analyzed through a state-dependent frame operator $\chi(\rho_A;O_A)$1, leading to

$\chi(\rho_A;O_A)$2

All nontrivial estimation directions are strictly attenuated: the top eigenvalue of the frame operator is exactly $\chi(\rho_A;O_A)$3 with unique eigenvector $\chi(\rho_A;O_A)$4, while physically admissible tangent directions lie in the orthogonal mean-zero subspace (Saini et al., 17 Dec 2025). This is not the discord-type LQFI of bipartite systems, but it clarifies a broader local-metrological theme: locality constraints are often encoded spectrally as attenuations of the globally optimal QFI.

Two nearby correlation measures reverse or modify the standard LQFI optimization. QFI-based measurement-induced nonlocality defines

$\chi(\rho_A;O_A)$5

where $\chi(\rho_A;O_A)$6 commutes with $\chi(\rho_A;O_A)$7, so the local marginal is undisturbed; for pure two-qubit states this equals $\chi(\rho_A;O_A)$8 in terms of concurrence (Muthuganesan, 29 Jun 2026). By contrast, the standard LQFI minimizes over local generators and vanishes on zero-discord states (1711.02323). The two constructions therefore probe opposite extremal responses of a bipartite state to local unitary perturbations.

A further geometric extension views Fisher information through mixed-state Fubini–Study metrics. A purification-based, locally $\chi(\rho_A;O_A)$9-gauge-invariant metric for mixed states reduces under monotonicity to the square-root derivative Fisher information metric and satisfies a quantum Cramér–Rao bound (Mondal, 2015). This line of work does not define LQFI in the discord sense, but it supplies a geometric basis for interpreting QFI as a local distinguishability metric on state space.

The cumulative picture is therefore plural rather than singular. LQFI most often denotes a minimized local-generator QFI that measures discord-like quantum correlations, but the same phrase also names subsystem QFI in many-body physics, directional QFI in sensor networks, and parameter-local sensitivity in open systems. The unifying content is always local distinguishability, while the object with respect to which locality is imposed changes from one research program to another.