Boundary Quantum Fisher Information

• Boundary QFI is a family of boundary-sensitive metrological constructions that leverage physical edges, operational bounds, and spectral hard edges to characterize quantum sensitivity.
• It reveals how localized modes such as Majorana zero modes create nonzero QFI plateaus, emphasizing encoding-axis asymmetry and robustness under practical perturbations.
• Various formulations—from fidelity-based bounds to Krylov spectral analysis—offer rigorous tools for quantifying and certifying metrological performance in quantum systems.

Boundary quantum Fisher information (QFI) is not a single universally standardized formal object. In current arXiv literature, the phrase appears in several technically distinct senses: as the QFI of a boundary-local reduced state in an open many-body system, as a lower or upper boundary on QFI obtained from experimentally accessible data, and as a spectral-edge formulation in which QFI is controlled by the behavior of a Liouville-space measure near the boundary point $\lambda=0$ (Płodzień et al., 1 May 2026). Related work also uses bounded parameter domains or temporal inequalities to place rigorous constraints on QFI, so the topic is best understood as a family of boundary-sensitive metrological constructions rather than a single definition (Górecki et al., 2024).

1. Terminological scope

Several usages of “boundary QFI” coexist in the literature.

Usage	Boundary object	Representative result
Physical boundary	Single boundary site in an open chain	Majorana zero mode fixes boundary QFI at a nonzero plateau (Płodzień et al., 1 May 2026)
Operational boundary	Lower/upper QFI bounds from data	Certified interval $H_\delta \le I_\delta \le J_\delta$ (Beckey et al., 2020)
Truncation boundary	Support-restricted or subnormalized state	TQFI lower-bounds standard QFI (Sone et al., 2020)
Temporal boundary	LGI violation as a lower bound	$F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$ for stationary pure states (Abboud et al., 10 Apr 2026)
Spectral boundary	Hard edge at $\lambda=0$	QFI becomes a singular inverse moment $\int d\mu(\lambda)/\lambda^2$ (Alishahiha et al., 23 Feb 2026)
Parameter-domain boundary	Finite interval $[a,b]$	$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$ (Górecki et al., 2024)

A recurrent point is that several papers explicitly do not introduce a formal object called “boundary QFI.” Instead, they develop QFI lower bounds, upper bounds, detectability thresholds, or generalized QFI measures that become boundary-sensitive because of physical edges, bounded supports, truncated spectra, or spectral hard edges (Fröwis et al., 2015). Accordingly, the expression functions as an umbrella term spanning topological boundary metrology, experimentally accessible QFI certification, and edge-controlled spectral geometry.

2. Boundary-local QFI in topological many-body systems

The most literal use of the term occurs in the study of local QFI at a physical boundary. In the open Kitaev chain, a parameter is encoded by a local spin rotation,

$$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$

the system evolves under the open XY/Kitaev Hamiltonian, and the readout is performed on the left boundary site $j=1$. The object of interest is the QFI of the single-site reduced density matrix $\hat\varrho_j(\theta,t)$, written in Bloch form as

$H_\delta \le I_\delta \le J_\delta$0

with single-qubit QFI

$H_\delta \le I_\delta \le J_\delta$1

In this setting, “boundary QFI” means $H_\delta \le I_\delta \le J_\delta$2, the locally accessible metrological sensitivity of the boundary qubit after encoding and evolution (Płodzień et al., 1 May 2026).

The exact free-fermion analysis yields a particularly simple result at the optimal operating point $H_\delta \le I_\delta \le J_\delta$3: $H_\delta \le I_\delta \le J_\delta$4 where

$H_\delta \le I_\delta \le J_\delta$5

and $H_\delta \le I_\delta \le J_\delta$6 are the normal and anomalous Bogoliubov propagators. The long-time averaged plateau is

$H_\delta \le I_\delta \le J_\delta$7

For boundary encoding and boundary readout, $H_\delta \le I_\delta \le J_\delta$8, the plateau term is

$H_\delta \le I_\delta \le J_\delta$9

The plateau forms once bulk contributions have dephased on the timescale $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$0, and for finite size it persists up to times

$F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$1

because the Majorana zero mode acquires only an exponentially small splitting $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$2. Near the transition, the plateau amplitude behaves as

$F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$3

This formulation differs sharply from global many-body QFI. The full evolution remains unitary, but local accessibility is the issue: in generic dynamics, single-site metrological sensitivity decays because local information disperses into nonlocal correlations. In the topological phase $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$4, the Majorana zero mode prevents complete local information loss and pins the boundary QFI to a nonzero plateau. The protocol requires only product-state initialization, Hamiltonian evolution, and single-site readout.

3. Majorana separation, axis asymmetry, and robustness

The mechanism behind the plateau is the spatial separation of the two Majorana quadratures. Writing the zero-mode envelopes as

$F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$5

the left and right Majorana components localize at opposite ends of the chain. The local encoding channels are

$F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$6

At the left boundary and for $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$7, $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$8-axis encoding probes the left Majorana with $F_Q \ge 8[K(\tau)-\langle Q^2\rangle]$9 weight, whereas $\lambda=0$0-axis encoding probes the right Majorana, whose weight at the left edge is exponentially small. Consequently, for left boundary readout,

$\lambda=0$1

The paper identifies this as a boundary encoding-axis asymmetry (Płodzień et al., 1 May 2026).

This asymmetry is significant because a plateau alone does not prove topological origin: a trivial localized boundary mode can also produce a residual long-time local signal. The paper therefore emphasizes that the asymmetry, not merely the existence of a nonzero plateau, distinguishes topological boundary memory from a generic localized subgap signal. If the sign of the real pairing is reversed, $\lambda=0$2, the preferred boundary axis is interchanged.

Robustness was analyzed in two perturbative directions. Under quenched on-site disorder,

$\lambda=0$3

the asymmetry remains close to the clean topological value for disorder strengths $\lambda=0$4 below the clean bulk-gap scale $\lambda=0$5. Under parity-preserving interactions,

$\lambda=0$6

the exact free-fermion identity $\lambda=0$7 no longer applies, but finite-size real-time simulations still show a visible boundary plateau. At the Kitaev sweet spot $\lambda=0$8, the boundary QFI retains about $\lambda=0$9 of its free-fermion value at $\int d\mu(\lambda)/\lambda^2$0. Outside the topological phase, by contrast, no protected nonzero plateau survives in the thermodynamic limit; the long-time boundary QFI is controlled only by extended bulk modes and scales as $\int d\mu(\lambda)/\lambda^2$1.

4. Operational QFI boundaries: fidelity, truncation, and finite precision

A second usage of boundary QFI concerns experimentally accessible lower and upper bounds on QFI. One route is the finite-time distinguishability protocol based on the Bhattacharyya coefficient

$\int d\mu(\lambda)/\lambda^2$2

where $\int d\mu(\lambda)/\lambda^2$3 and $\int d\mu(\lambda)/\lambda^2$4 are outcome probabilities before and after a small unitary $\int d\mu(\lambda)/\lambda^2$5. The central inequality is

$\int d\mu(\lambda)/\lambda^2$6

This gives a direct lower-bound estimator for QFI from measurement statistics, but finite detector resolution imposes a practical detectability boundary: to certify $\int d\mu(\lambda)/\lambda^2$7, one needs $\int d\mu(\lambda)/\lambda^2$8, and the induced statistical change remains confined to a scale

$\int d\mu(\lambda)/\lambda^2$9

so one requires

$[a,b]$0

For Heisenberg-like scaling $[a,b]$1, $[a,b]$2-level resolution is required. An additional interaction-based readout $[a,b]$3 can shift this boundary and restore sensitivity (Fröwis et al., 2015).

Mixed-state boundary estimation was developed further in the variational algorithm VQFIE. Instead of computing QFI exactly, the method variationally estimates lower and upper bounds derived from fidelity bounds, producing a certified interval

$[a,b]$4

with

$[a,b]$5

The final bounds combine truncated generalized fidelity and sub-/super-fidelity,

$[a,b]$6

$[a,b]$7

The method is dynamics agnostic and is especially useful for mixed states of high purity or low effective rank (Beckey et al., 2020).

A closely related construction is the truncated quantum Fisher information (TQFI),

$[a,b]$8

where the generalized fidelity for subnormalized states is

$[a,b]$9

TQFI satisfies

$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$0

increases monotonically with $$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$1, and becomes exact at full support $$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$2. For unitary families,

$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$3

This is a genuine subnormalized-state QFI geometry, not merely a heuristic truncation (Sone et al., 2020).

5. Temporal nonclassicality as a lower boundary on QFI

Another boundary formulation relates QFI to Leggett–Garg inequality (LGI) violations. For a bounded observable $$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$4 with $$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$5, the symmetrized stationary correlator is

$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$6

and the stationary LGI quantity is

$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$7

For stationary pure states and thermal states, LGI violation yields rigorous lower bounds on the QFI associated with the same generator $$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$8. In the pure-state case,

$$\chi \le \ln\!\left(1+\frac12\int_a^b d\theta\,\sqrt{F_Q[\rho_\theta]}\right)$$9

and since $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$0,

$$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$1

For thermal states,

$$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$2

with a universal thermal factor $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$3, satisfying

$$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$4

Thus a qualitative foundations test becomes a quantitative witness of metrological sensitivity (Abboud et al., 10 Apr 2026).

The proof is spectral: $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$5 is decomposed into off-diagonal matrix elements of $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$6 in the energy basis, and the oscillatory kernel

$$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$7

is bounded above by $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$8. The same transition weights enter the QFI, so LGI violation implies that $$\hat R_{x/y}^{(k)}(\theta)=e^{-i\theta\hat\sigma_k^{x/y}/2},$$9 coherently connects energy levels and therefore generates nonzero QFI. In thermal states the comparison is termwise through the universal ratio

$j=1$0

This temporal lower boundary is experimentally economical. One measures only

$j=1$1

at times $j=1$2, forms $j=1$3, and infers a QFI lower bound. For dichotomic $j=1$4, sequential projective measurements suffice; for bounded non-dichotomic $j=1$5, weak measurements reconstruct the same symmetrized correlator. In collective many-body settings, the resulting QFI lower bound converts, through standard QFI criteria, into a lower bound on multipartite entanglement depth. The GHZ example is especially sharp: at maximal violation $j=1$6,

$j=1$7

so the bound is saturated exactly.

6. Spectral-edge and information-theoretic boundary formulations

In the Krylov–resolvent framework, the decisive boundary is spectral rather than spatial. Writing the symmetric logarithmic derivative problem as

$j=1$8

and introducing the seed $j=1$9, the exact QFI becomes

$\hat\varrho_j(\theta,t)$0

where $\hat\varrho_j(\theta,t)$1 is the scalar spectral measure of $\hat\varrho_j(\theta,t)$2. Because the integrand has a pole at $\hat\varrho_j(\theta,t)$3, the behavior of $\hat\varrho_j(\theta,t)$4 near the boundary point $\hat\varrho_j(\theta,t)$5 controls both the size of the QFI and the convergence of Krylov approximations. If the spectrum is gapped away from zero, convergence is exponential; if the support reaches zero with hard-edge behavior

$\hat\varrho_j(\theta,t)$6

then convergence is algebraic and governed by Bessel universality. For $\hat\varrho_j(\theta,t)$7, the QFI diverges in the thermodynamic limit. The truncation error is exactly the tail of the Krylov distribution,

$\hat\varrho_j(\theta,t)$8

so near-zero modes simultaneously enhance sensitivity and deepen the Krylov tail (Alishahiha et al., 23 Feb 2026).

A different kind of boundary is the finite parameter interval. For a differentiable family $\hat\varrho_j(\theta,t)$9 with prior support in $H_\delta \le I_\delta \le J_\delta$00, the Holevo information obeys

$H_\delta \le I_\delta \le J_\delta$01

For arbitrary differentiable prior $H_\delta \le I_\delta \le J_\delta$02,

$H_\delta \le I_\delta \le J_\delta$03

These inequalities make QFI a measurement-independent upper boundary on accessible information over bounded parameter domains and yield Bayesian mean-square-error lower bounds. They are especially useful for bounded-support priors, where van Trees-type bounds can be trivial. At the same time, the literature stresses a crucial limitation: large QFI does not imply large mutual information globally, as illustrated by the N00N example, which has Heisenberg scaling in QFI for local estimation but satisfies $H_\delta \le I_\delta \le J_\delta$04 globally (Górecki et al., 2024).

Taken together, these formulations show that boundary QFI is best viewed as a family of metrological boundary phenomena. At a physical edge, topology can protect locally accessible QFI for exponentially long times. At an operational edge, fidelity bounds, truncation, and temporal correlators provide certified lower and upper boundaries on QFI without full tomography. At a spectral edge, near-zero Liouville modes control both metrological enhancement and computational difficulty. And on bounded parameter domains, QFI constrains the maximum globally accessible information.