Thermal Quantum Fisher Information

• Thermal quantum Fisher information is a metric for Gibbs and mixed thermal states, linking quantum metrology with response-theoretic and resource-theoretic concepts.
• It quantifies sensitivity in parameter estimation by leveraging variance bounds, non-commutation, and linear response functions in thermal equilibrium settings.
• Applications include detecting multipartite entanglement, probing quantum phase transitions, and guiding the design of thermal sensing protocols in experimental systems.

Searching arXiv for recent and foundational work on thermal quantum Fisher information. Thermal quantum Fisher information is the quantum Fisher information (QFI) of thermal equilibrium states, thermally prepared probes, or states studied under explicitly thermodynamic constraints. In the literature, it is used to quantify distinguishability of Gibbs states, metrological sensitivity of thermal probes, coherence with respect to time translations, and multipartite entanglement at finite temperature. For a mixed state $\rho=\sum_n p_n |\psi_n\rangle\langle\psi_n|$ and a unitary encoding generated by $\hat O$, the standard mixed-state QFI is

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$

with the pure-state reduction $F_Q=4(\Delta \hat O)^2$; for thermal states $\rho_T=e^{-\hat H/T}/\mathcal Z$, this quantity acquires direct thermodynamic, response-theoretic, and resource-theoretic interpretations (Gabbrielli et al., 2018, Marvian, 2021, Shitara et al., 2015).

1. Definitions, geometric meaning, and thermal-state specialization

In quantum metrology, the QFI is defined through the symmetric logarithmic derivative (SLD) $L_\theta$,

$$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$

and it sets the quantum Cramér–Rao bound $\Delta\theta^2\ge F_\theta^{-1}$. This is the form used throughout the thermal metrology literature for Gibbs states, thermally mixed probes, and thermal many-body systems (Meng et al., 7 Nov 2025, Wójtowicz et al., 12 Jun 2025).

A further geometric point is that the quantum Fisher information is not a single object, but a family indexed by operator monotone functions $f$, with the SLD Fisher information corresponding to $f(x)=(x+1)/2$, the BKM Fisher information to $\hat O$0, and the RLD Fisher information to $\hat O$1. In thermal equilibrium, this family can be related to generalized covariances and linear response functions (Shitara et al., 2015). In the SLD case emphasized by most thermal-state studies, the QFI is also the Bures metric up to the usual factor, and for unitary encodings it satisfies

$\hat O$2

with equality in the variance bound for pure states (Gabbrielli et al., 2018).

For time translations generated by a Hamiltonian $\hat O$3, the QFI is commonly written as

$\hat O$4

where $\hat O$5. For pure states $\hat O$6, this reduces to

$\hat O$7

This pure-state identity underlies both the thermodynamic resource interpretation of QFI and its role as a finite-temperature entanglement witness (Marvian, 2021, Gabbrielli et al., 2018).

2. Quantum thermodynamic and resource-theoretic interpretation

A distinctive thermodynamic interpretation arises in the resource theory of thermodynamics when work is treated as free. In that setting, preparing a general quantum state requires two distinct resources: work, quantified by free energy, and coherence, identified with superpositions across energy levels. Once work is unrestricted, the relevant free operations are time-translation-invariant (TI) operations, i.e. CPTP maps satisfying the covariance condition with respect to the input and output Hamiltonians. TI operations cannot create energetic coherence from incoherent states, so coherence must be supplied as a resource (Marvian, 2021).

Using as the standard unit a “c-bit,” a two-level system in the pure coherent state

$\hat O$8

the coherence cost of a state $\hat O$9 is defined as the minimum asymptotic rate of c-bits needed to prepare $$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$0 by TI operations with vanishing trace-distance error. The central theorem is

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$1

because for the chosen c-bit

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$2

This provides a direct operational meaning for the QFI of time translations: the asymptotic coherence cost is exactly its QFI measured in c-bit units (Marvian, 2021).

The same framework yields asymptotic conversion laws. For pure states $$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$3 and $$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$4 with the same period,

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$5

is possible with vanishing error iff

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$6

equivalently a QFI-ratio statement because $$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$7. More generally, any TI conversion $$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$8 with vanishing trace-distance error must satisfy

$$F_Q[\rho,\hat O]=2\sum_{n,m}\frac{(p_n-p_m)^2}{p_n+p_m}\,|\langle \psi_n|\hat O|\psi_m\rangle|^2,$$9

The same paper also shows

$F_Q=4(\Delta \hat O)^2$0

and

$F_Q=4(\Delta \hat O)^2$1

thereby connecting the minimum-variance purification property and the convex-roof-of-variance property of QFI in a single resource-theoretic setting (Marvian, 2021).

3. Equilibrium linear response, fluctuation spectra, and thermodynamic uncertainty

For thermal equilibrium states, a major theme is that QFI is not only a state-space metric but also a response-theoretic quantity. In canonical equilibrium, the generalized fluctuation-dissipation theorem gives

$F_Q=4(\Delta \hat O)^2$2

where $F_Q=4(\Delta \hat O)^2$3 is the $F_Q=4(\Delta \hat O)^2$4-generalized covariance and $F_Q=4(\Delta \hat O)^2$5 is the linear response function. This establishes that the full continuous family of QFI metrics can be determined from measurable linear-response data (Shitara et al., 2015).

For the standard thermal SLD-QFI, one widely used formula is

$F_Q=4(\Delta \hat O)^2$6

and equivalently

$F_Q=4(\Delta \hat O)^2$7

These formulas make thermal QFI a frequency-filtered measure of quantum fluctuations, with the thermal kernel suppressing high-temperature sensitivity (Gabbrielli et al., 2018).

A complementary equilibrium formulation considers a classical intensive parameter $F_Q=4(\Delta \hat O)^2$8 and its thermodynamically conjugate extensive quantum operator

$F_Q=4(\Delta \hat O)^2$9

For this Gibbs family, the QFI admits an exact integral representation in terms of the autocorrelation spectrum $\rho_T=e^{-\hat H/T}/\mathcal Z$0 of $\rho_T=e^{-\hat H/T}/\mathcal Z$1, obtained by combining the Kubo formula and the fluctuation-dissipation theorem. The analysis emphasizes that a naive Callen–Welton relation misses a singular $\rho_T=e^{-\hat H/T}/\mathcal Z$2 contribution, and that the zero-frequency correction is essential to recover the full QFI, including the purely statistical term (Meng et al., 7 Nov 2025).

From that exact representation, the following chain of inequalities is derived: $\rho_T=e^{-\hat H/T}/\mathcal Z$3 The upper bound

$\rho_T=e^{-\hat H/T}/\mathcal Z$4

is the new, tighter bound, while the lower bound is obtained from Cauchy–Schwarz and is credited to Holevo. Combining the new upper bound with the quantum Cramér–Rao inequality gives the thermodynamic uncertainty relation

$\rho_T=e^{-\hat H/T}/\mathcal Z$5

Here $\rho_T=e^{-\hat H/T}/\mathcal Z$6 is not the ordinary fluctuation of $\rho_T=e^{-\hat H/T}/\mathcal Z$7, but the uncertainty in the inferred $\rho_T=e^{-\hat H/T}/\mathcal Z$8 when one estimates $\rho_T=e^{-\hat H/T}/\mathcal Z$9 and then uses the known response function $L_\theta$0. The same analysis shows $L_\theta$1 for an optimal estimator saturating the Cramér–Rao bound (Meng et al., 7 Nov 2025).

4. Thermal probes under dynamic encoding and open-system evolution

Thermal QFI also appears when the probe itself is a Gibbs state and the unknown parameter is encoded dynamically. In the dynamic sensing framework,

$L_\theta$2

For thermal states, the exact QFI can be expressed in terms of the commutator $L_\theta$3, and the universal upper bound

$L_\theta$4

shows that non-commutation between the probe Hamiltonian and the transformed local generator is the resource enabling sensitivity (Zhang et al., 2 Dec 2025).

For time-independent encoding $L_\theta$5, one has

$L_\theta$6

and therefore

$L_\theta$7

This is the central $L_\theta$8 scaling result. In the high-temperature limit, $L_\theta$9, while in the low-temperature regime a complementary gap-dependent estimate,

$$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$0

makes the spectral gap $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$1 decisive, especially near criticality (Zhang et al., 2 Dec 2025).

Open-system settings provide further versions of thermal QFI. For a qubit in a squeezed thermal bath, the reduced dynamics is a pure-dephasing channel with decoherence factor $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$2, and the QFI for temperature estimation is

$$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$3

It is maximized by the equatorial initial state

$$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$4

and the numerics show an optimal temperature window, an optimal interaction time, strong dependence on the bath ohmicity, and robustness against squeezing (Ullah, 2020).

In the Unruh setting, the QFI of an accelerated Unruh–DeWitt detector separates global from local thermality. The equilibrium detector state is a Gibbs state at the Unruh temperature, and the asymptotic QFI for estimating $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$5 is

$$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$6

which depends only on $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$7. By contrast, the full time-dependent QFI retains dependence on spacetime dimension, field mass, and detector response details (Feng et al., 2021).

5. Finite-temperature entanglement, criticality, ETH, and thermalization

A major use of thermal QFI in many-body physics is as a witness of multipartite entanglement and a probe of quantum criticality. For local generators, if the state is at most $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$8-partite entangled, one has approximately $$\frac{\partial \rho_\theta}{\partial \theta}=\frac{1}{2}(L_\theta \rho_\theta+\rho_\theta L_\theta), \qquad F_\theta=\operatorname{Tr}(\rho_\theta L_\theta^2),$$9 for collective spin operators, so $\Delta\theta^2\ge F_\theta^{-1}$0 witnesses $\Delta\theta^2\ge F_\theta^{-1}$1-partite entanglement. Near a quantum phase transition, the finite-temperature behavior organizes into a universal phase diagram with four regimes: quantum plateau, thermal plateau, critical plateau, and maximum entropy plateau (Gabbrielli et al., 2018).

In the critical plateau, the scaling law is

$\Delta\theta^2\ge F_\theta^{-1}$2

while in the maximum entropy plateau

$\Delta\theta^2\ge F_\theta^{-1}$3

In gapped phases, the gap protects the low-temperature QFI; in symmetry-breaking phases with low-energy degeneracy, a thermal plateau appears; and in topological systems the survival of large multipartite entanglement can be much more robust. For the Kitaev chain, the paper shows that Heisenberg-limit scaling $\Delta\theta^2\ge F_\theta^{-1}$4 can survive at finite temperature up to $\Delta\theta^2\ge F_\theta^{-1}$5 (Gabbrielli et al., 2018).

In chaotic systems satisfying ETH, the QFI distinguishes a thermal density matrix from a pure thermal-looking eigenstate. For a thermal Gibbs state,

$\Delta\theta^2\ge F_\theta^{-1}$6

whereas for a thermal pure eigenstate,

$\Delta\theta^2\ge F_\theta^{-1}$7

so that

$\Delta\theta^2\ge F_\theta^{-1}$8

This means that states that are thermally equivalent in local observables can still differ sharply in their multipartite entanglement structure, with the difference amplified near thermal phase transitions (Brenes et al., 2019).

The same contrast persists across protocols. For ETH-satisfying chaotic systems, a universal relation

$\Delta\theta^2\ge F_\theta^{-1}$9

suppresses the thermal-state QFI relative to the energy-eigenstate QFI. In particular, the thermal adiabatic QFI remains finite, whereas the corresponding eigenstate QFI is exponentially large, and thermal quench QFI saturates to a constant rather than growing linearly in time (Iniguez et al., 2023). A related, more formal viewpoint states that von Neumann ergodicity and ETH can be reformulated as conditions on a QFI functional $f$0, with thermalization corresponding to the QFI reaching its absolute minimum under an entropy bound (Gomez, 2019).

Recent frustrated-magnet work extends this finite-temperature program to momentum-resolved spectroscopy. In quantum spin ice, the QFI density for a spin operator $f$1 is obtained from the dynamical structure factor as

$f$2

and its temperature and momentum dependence distinguishes the ferromagnetic ordered phase, the thermal critical region above it, and the zero-flux and $f$3-flux quantum spin ice phases (Zhou et al., 16 Oct 2025).

6. Computation, state manifolds, and experimental access

Because mixed-state QFI is nonlinear in $f$4, thermal many-body calculations are often limited by the cost of diagonalization. A recent tensor-network approach rewrites the SLD equation as a continuous Lyapunov equation with formal solution

$f$5

and therefore

$f$6

This formulation avoids explicit eigendecomposition and can be implemented with standard MPO/MPS time-evolution methods such as TDVP, Krylov methods, and Trotter-like schemes; the paper explicitly notes compatibility with YASTN, TeNPy, and ITensor (Wójtowicz et al., 12 Jun 2025).

The same work shows that convergence becomes harder for highly mixed states and near criticality. The physical message stated there is that higher temperature and higher entropy make the Lyapunov integral harder to converge, whereas low-temperature and gapped regimes are numerically easier. In the transverse-field Ising thermal-state example used for magnetic-field estimation, convergence is fast away from criticality, slows near the critical point where the gap closes, and worsens with larger system size because the relevant spectral gap shrinks (Wójtowicz et al., 12 Jun 2025).

Experimental access to thermal QFI is available through response spectroscopy and inelastic scattering. The generalized fluctuation-dissipation framework reconstructs the full family of thermal QFI metrics from admittance or susceptibility measurements under weak ac driving (Shitara et al., 2015). In quantum magnets, the QFI can be extracted from the finite-temperature dynamical structure factor and is therefore measurable in inelastic neutron scattering experiments, a point used explicitly in the quantum spin ice analysis (Zhou et al., 16 Oct 2025).

State-manifold geometry provides another established thermal-QFI setting. For two-mode Gaussian optical states, the QFI matrices on the manifolds of mode-mixed thermal states and squeezed thermal states are diagonal in the natural parameters. For mode-mixed thermal states,

$f$7

while for squeezed thermal states

$f$8

The associated Bures scalar curvatures depend only on the thermal occupancies $f$9, which the paper interprets as a geometric encoding of local statistical distinguishability (Marian et al., 2016).

A related finite-temperature information-theoretic extension appears in the thermal SSH chain, where an optimized QFI is used as a quantifier of quantum information processing capacity. In that setting, the optimized QFI is finite in the topological region and tends to vanish in the trivial region; in the extended SSH model with long-range hopping, it captures topological-to-topological transitions $f(x)=(x+1)/2$0 in the low-temperature limit (Navarro-Labastida, 2024).

Thermal quantum Fisher information therefore occupies a technically diverse but conceptually coherent position: it is simultaneously a metric of thermal-state distinguishability, a response-theoretic observable, a thermodynamic resource monotone, a finite-temperature entanglement witness, and a practical design criterion for thermal sensing protocols. The literature shows that its specific form depends strongly on which thermal structure is being probed—equilibrium Gibbs variation, time-translation asymmetry, dynamic unitary encoding, or open-system thermalization—but in each case the QFI isolates how temperature, spectral gaps, response functions, and non-commutation constrain the best achievable precision (Marvian, 2021, Meng et al., 7 Nov 2025, Zhang et al., 2 Dec 2025, Gabbrielli et al., 2018).