---
title: General Heisenberg Limit in Quantum Metrology
url: https://www.emergentmind.com/topics/general-heisenberg-limit-ghl
type: topic
---

# General Heisenberg Limit in Quantum Metrology

Searching arXiv for recent and foundational uses of “General Heisenberg Limit” in quantum metrology.
In the literature considered here, the expression **General Heisenberg Limit** (GHL) is used in several non-identical but closely related ways. In each case it extends the standard Heisenberg scaling beyond the simplest noiseless, time-independent, locally unbiased scenario, but the extension depends on what is treated as the relevant resource and which constraints are imposed on the sensing protocol. Across these formulations, the recurrent themes are the quantum Cramér–Rao bound (QCRB), the quantum Fisher information (QFI), the role of the parameter-translation generator, and the fact that locality, control, prior information, time dependence, noise, and global resource accounting can all change the precise form of the ultimate precision law [1201.2225], [1803.01724], [2308.10929], [2507.15348].

## 1. Multiple meanings of the term

At the baseline level, the QCRB for a parameter $\theta$ encoded in $\rho(\theta)$ and measured $\nu$ times is
$$
\mathrm{Var}(\hat{\theta}) \ge \frac{1}{\nu F_Q(\rho(\theta))},
$$
with pure-state QFI
$$
F_Q(\theta)=4\big[(\partial_\theta\psi|\partial_\theta\psi)-|\psi|\partial_\theta\psi|^2\big].
$$
For time-independent unitary encoding, this yields the familiar Heisenberg scaling $F_Q \propto T^2$ and $\Delta \omega \propto 1/T$ for a single probe at fixed $\nu$ [1803.01724]. For $N$ probes interrogated for time $t$, standard presentations distinguish SQL scaling $F_Q \sim N$ and Heisenberg scaling $F_Q \sim N^2$, equivalently $\Delta\theta \sim 1/\sqrt{N}$ versus $\Delta\theta \sim 1/N$ up to the interrogation-time factor [2308.10929].

The term GHL appears when this baseline is generalized. In some papers it means a universal resource-counting bound in terms of the expectation value of the generator above its ground state or in terms of query complexity [1201.2225], [1004.3944]. In others it means a global, prior-averaged, or minimax single-shot bound that corrects naive QFI-based constants [1111.0788], [2304.14370]. In still others it denotes the best achievable scaling under time-dependent generators, realistic locality constraints, perturbing interactions, noise, or nonlinear encoding [1803.01724], [2308.10929], [1903.07888], [2507.15348].

| Usage of GHL | Representative formula | Representative papers |
| --- | --- | --- |
| Universal resource-counting bound | $\Delta\phi \ge \frac{1}{2|\mathcal{H}|}$; $\delta\phi \ge \frac{1}{|\langle H\rangle|}$ | [1201.2225], [1004.3944] |
| Global single-shot or prior-averaged bound | $\delta \hat{\Phi} > \frac{k_A}{\langle N+1\rangle}$; $\Delta\theta_{\minimax} \ge \frac{\pi}{N\lambda}$ | [1111.0788], [2304.14370] |
| Time-dependent encoding with control | $F_Q(\omega)\le \left[\int_0^T(\mu_{\max}-\mu_{\min})dt\right]^2$ | [1803.01724] |
| Locality- and LOCC-constrained attainability | $F_Q=\Theta(N^2)$ under local perturbations | [2308.10929] |
| Nonlinear and multiparameter scaling | $\sigma^{(k)}_{GHL}\ge \frac{1}{N^k}$ | [2507.15348] |
| Global resource accounting under losses | $\Delta\theta_{GHL}\ge \frac{1}{N_{\mathrm{global}}}$ | [2505.03290] |

## 2. Universal resource counts and global single-shot formulations

A central line of work reformulates the Heisenberg limit by identifying a universal resource count. In the network-based treatment of parameter encoding $U(\phi)=\exp(-i\phi\mathcal{H})$, the universal resource is
$$
R:=\langle \mathcal{H}\rangle-h_{\min}\equiv |\mathcal{H}|,
$$
and for optimal balanced-superposition probe states the resulting universal Heisenberg limit is
$$
\Delta \phi \ge \frac{1}{2|\mathcal{H}|}.
$$
In the same framework, query complexity $Q$ unifies linear, $k$-body, exponential, and sequential strategies, so that apparent “super-Heisenberg” scalings in $N$ are re-expressed as ordinary $1/Q$ behavior once the interaction order is counted properly [1201.2225]. A related general optimality proof formulates the single-shot bound as
$$
\delta\phi \ge \frac{1}{|\langle H\rangle|},
$$
with $\langle H\rangle$ the expectation value of the generator above its ground energy. That proof identifies the Heisenberg limit as an information-theoretic interpretation of the Margolus–Levitin bound rather than of the variance-based uncertainty relation, and treats multimode, nonlinear, adaptive, and multipass networks within the same query-complexity language [1004.3944].

A second line of work argues that QFI-based local bounds are not yet the correct global single-shot statement. For a completely unknown phase with uniform prior, the average phase error obeys the non-asymptotic bound
$$
\delta \hat{\Phi} > \frac{k_A}{\langle N+1\rangle},\qquad
k_A:=\sqrt{2\pi/e^3}\approx 0.559,
$$
and the conjectured asymptotically optimal constant is
$$
\delta \hat{\Phi} > \frac{k_C}{\langle N+1\rangle},\qquad
k_C\approx 1.37608.
$$
This formulation is explicitly constraint-free, non-asymptotic, and prior-averaged, and it applies to multimode probes, multiple passes, nonlinear phase shifts, arbitrary POVMs, and adaptive strategies, provided the phase is a priori completely unknown and the generator has nonnegative integer eigenvalues [1111.0788].

The Bayesian/minimax formulation sharpens this point further. For noiseless unitary estimation with a strict fixed-$N$ resource constraint and bounded generator width $\lambda$, the asymptotically saturable single-shot bound is
$$
\Delta\theta_{\minimax}\ge \frac{\pi}{N\lambda}
$$
up to finite-prior corrections, rather than the naive local-QFI expression $1/(N\lambda)$. When only the average resource is bounded, the universal constant changes to
$$
c=\frac{4|A_0|^3}{27}\approx 1.89,
$$
with Airy-shaped optimal amplitudes [2304.14370]. Taken together, these papers treat GHL as a global precision law whose exact constant depends on whether the task is local-unbiased, prior-averaged, or minimax.

## 3. Time-dependent generators and control-based generalizations

A distinct formulation of GHL arises when the parameter is encoded through a time-dependent Hamiltonian. For
$$
H(t;\omega)=\omega G(t)+H_{\mathrm{ctrl}}(t),\qquad
U(\omega,T)=\mathcal{T}\exp\!\left[-i\int_0^T H(t;\omega)\,dt\right],
$$
the relevant bound is expressed through the instantaneous extremal eigenvalues $\mu_{\max}(t)$ and $\mu_{\min}(t)$ of $\partial_\omega H_0(t)$:
$$
F_Q(\omega)\le \left[\int_0^T\big(\mu_{\max}(t)-\mu_{\min}(t)\big)\,dt\right]^2.
$$
In the single-ion experiment based on a laser-cooled ${}^{138}\mathrm{Ba}^+$ ion, the engineered Hamiltonian was
$$
H_0(t;\omega)=-\hbar \Omega_0\sigma_z+\hbar \Omega_d\sin(\omega t)\,\sigma_z,
$$
for which the frequency-encoding extremal eigenvalues are
$$
\mu_{\max,\min}(t)=\pm \hbar \Omega_d\, t\,\cos(\omega t).
$$
Without control, periodic eigenvalue crossings produce cancellations in the integral and recover only the time-independent-Hamiltonian behavior $\Delta\omega\propto 1/T$. With optimal level-crossing control,
$$
H_{LC}(t)=\sum_{n=1}^{N}\delta\!\left(t-\frac{(2n+1)\pi}{\omega_c}\right)\sigma_x^\pi,
$$
these cancellations are removed, the ideal QFI scales as $F_Q\propto T^4$, and the ideal uncertainty becomes $\Delta\omega\propto 1/T^2$ [1803.01724].

The experimental result was a controlled scaling
$$
\Delta\omega \propto \frac{1}{T^{1.75\pm 0.03}},
$$
compared with
$$
\Delta\omega \propto \frac{1}{T^{0.87\pm 0.02}}
$$
without control, up to the coherence limit of approximately $80\,\mu\mathrm{s}$. The deviation from the ideal exponent $-2$ was attributed to imperfect equal-superposition preparation and finite control-pulse durations, which consumed about $30\%$ of the evolution time. The same framework was proposed as a proof-of-principle route to detecting oscillatory weak couplings from axion-like dark matter, with an inferred bound $g_{aee}\lesssim 400\,\mathrm{GeV}^{-1}$ for $m_a\approx 2\pi\times 50\,\mathrm{kHz}$ in the single-ion setting [1803.01724].

Control also restores Heisenberg scaling for general noncommuting but time-independent dynamics. In the sequential-control scheme for $U_T(\theta)=e^{-iH(\theta)T}$, the generator
$$
S_T=iU_T^\dagger \partial_\theta U_T
$$
can fail to grow linearly with $T$ when $[H(\theta),\partial_\theta H(\theta)]\neq 0$. By splitting the evolution into $N$ segments and interspersing controls so that $[U_{ct},S_t]=0$, one obtains
$$
S_T^{(N)}=N S_t
$$
and therefore $F_Q\propto N^2$. For the qubit model
$$
H(x)=\sin(2x)\sigma_1+\cos(2x)\sigma_3,
$$
the controlled QFI becomes
$$
J_T^{(N)}=16N^2\sin^2(T/N),
$$
approaching $16T^2$ as $N\to\infty$. At the sweet spots $t=\pi/2$, the optimal controls can be fixed as $U_c=i\sigma_3$, yielding the exact Heisenberg scaling $J=16N^2$ at $T=(\pi/2)N$ without adaptation [1902.01097].

## 4. Locality, perturbing interactions, and preparation complexity

Another influential use of GHL concerns realistic many-body sensing under locality constraints. In the model
$$
H_\omega=V+\omega Z,\qquad Z=\sum_{j=1}^N Z_j,
$$
the perturbation $V$ is spatially local on a bounded-degree graph and may be strong, but is assumed known exactly. For GHZ-like input states with an extensive $Z$-polarization difference and short-range correlations, the central result is that for short times $t<c_{\rm in}/J$ one still has
$$
F_Q(\lvert\psi_\omega\rangle)\ge 4t^2N^2(c_{\rm in}-Jt)^2+O(N^{3/2}),
$$
so that $F_Q=\Theta(N^2)$ and
$$
\Delta\omega\simeq \frac{1}{\sqrt{\nu}\,Nt}.
$$
This is called the GHL in the sense of preserving Heisenberg scaling under locality, perturbing interactions, LOCC measurements, and polynomial-time classical post-processing. The explicit protocol uses adaptive local measurements, a parity observable $\mathbf P$, a prior window $\mathcal I_{\omega'}$, and efficient classical computation via matrix product states in one dimension and cluster expansion in higher dimensions [2308.10929].

This attainability result is sharply qualified by work that counts state-preparation complexity as part of the metrological cost. For a $d$-dimensional lattice with $k$-local Hamiltonian and bounded one-site energy, the growth of metrologically useful entanglement during resource-state preparation is bounded by the Lieb–Robinson light cone:
$$
F_Q(t)\le \kappa\,c_{\mathrm{WY}}\big[1+\gamma\,2^d\,R_{\mathrm{L}}(t)^d\big]\,N.
$$
For short-range interactions, $R_{\mathrm{L}}(t)\simeq v_{\mathrm{LR}}t$, so preparing an $F_Q\sim N^2$ state requires at least
$$
t_{\mathrm{HL}}(N)\gtrsim \frac{1}{v_{\mathrm{LR}}}\Big(\frac{N}{C}\Big)^{1/d}.
$$
If total experimental time is fixed and preparation time is counted, the resulting strong precision limit becomes
$$
\Delta\theta\gtrsim \frac{1}{\sqrt{T}\,N^{\,1-\frac{1}{2d}}}
$$
for short-range systems, rather than the usual $1/N$ [2301.12113].

These two lines of work are not contradictory; they encode different resource models. One proves that $F_Q=\Theta(N^2)$ survives known local perturbations for sufficiently short interrogation times with feasible LOCC extraction [2308.10929]. The other shows that if the many-body preparation time itself is treated as a limiting resource, Lieb–Robinson propagation can prevent asymptotic Heisenberg scaling within fixed total time [2301.12113].

## 5. Noise, error correction, and attainable Heisenberg scaling

Under Markovian noise, GHL often means the condition under which Heisenberg scaling can be restored by control or error correction. For probes obeying a Lindblad equation with Hamiltonian $H(\omega)=\omega \hat H$, define the Lindblad span
$$
\mathcal S=\mathrm{span}_{\mathbb R}\left\{\openone,\;L_j^{\mathrm H},\;L_j^{\mathrm {AH}},\;(L_j^\dagger L_k)^{\mathrm H},\;(L_j^\dagger L_k)^{\mathrm {AH}}\right\},
$$
and decompose
$$
\hat H=\hat H_\parallel+\hat H_\perp,\qquad \hat H_\parallel\in\mathcal S,\qquad \langle \hat H_\perp,X\rangle=0\ \forall X\in\mathcal S.
$$
If the Hamiltonian-not-in-Lindblad-span condition holds, $\hat H_\perp\neq 0$, then full and fast ancilla-free control yields an effective Hamiltonian
$$
\hat H_{\mathrm{eff}}=\sum_{n=1}^N \hat H_\perp^{(n)},
$$
and the QFI scales as
$$
F_Q(\omega)=4N^2\|\hat H_\perp\|^2T^2.
$$
If $\hat H_\perp=0$, Heisenberg scaling is unattainable even with ancillas [1903.07888].

A different noisy-sensing mechanism achieves Heisenberg scaling by shifting the target parameter from a coherent phase to a collective noise rate. For the master equation
$$
\frac{d\rho}{dt}=\Gamma_c[M_z,[M_z,\rho]]+\sum_{j=1}^N \gamma[\sigma_z^{(j)},[\sigma_z^{(j)},\rho]],
$$
a GHZ probe has coherence
$$
\langle 0^{\otimes N}|\rho(t)|1^{\otimes N}\rangle=\frac{1}{2}e^{-i\omega t}e^{-N^2\Gamma_c t-N\gamma t}.
$$
The QFI for estimating $\Gamma_c$ is
$$
F_Q(\Gamma_c;t,N)=\frac{N^4 t^2 e^{-2N^2\Gamma_c t-2N\gamma t}}{1-e^{-2N^2\Gamma_c t-2N\gamma t}},
$$
and choosing $t=t_0/N^2$ gives
$$
\Delta\Gamma_c(t^\ast)\sim \frac{1}{N}.
$$
Product states remain at SQL, so the Heisenberg scaling here is specific to collective-dephasing estimation under independent dephasing [1809.00176].

The fault-tolerant extension of this theme treats noisy QEC operations themselves as part of the metrological model. For a Pauli-$Z$ signal under bit-flip noise with state-preparation and measurement errors in all QEC operations, a repetition-code protocol with repeated syndrome measurements and a fault-tolerant logical readout yields explicit thresholds. The logical-measurement threshold is
$$
q_m^{(\mathrm{th})}=\frac14,
$$
the state-preparation threshold found numerically is
$$
p_{\mathrm{th}}^{(s)}\approx 0.067,
$$
and below threshold the classical Fisher information satisfies
$$
F_{cl}^{(\mathrm{QEC})}(0)\approx [\text{constant}]^2 n^2.
$$
Because the overhead grows only polylogarithmically, the protocol retains $\Delta\theta\sim 1/(nT\sqrt{\nu})$ in the fully fault-tolerant setting [2601.05457].

## 6. Nonlinear, multiparameter, and networked formulations

In nonlinear metrology, the GHL is explicitly generalized from $1/N$ to $1/N^k$. For a single parameter $\chi$ encoded through a $k$-th order nonlinearity, the proposed definition is
$$
\sigma_{GHL}^{(k)}\ge \frac{1}{N^k}.
$$
For a vector of parameters $\boldsymbol{\chi}=\{\chi_1,\ldots,\chi_d\}$ with balanced multipartite $N00N$ probes, the overall accuracy obeys
$$
\sigma_{\boldsymbol{\chi}} \ge \frac{1}{N^k}\sqrt{\frac{d(d+1)}{2}}.
$$
The relevant phase encoding is
$$
\hat U_{PS}=\exp\Big[i\chi_1(\hat a_1^\dagger \hat a_1)^k+i\chi_2(\hat a_2^\dagger \hat a_2)^k\Big],
$$
so that $F_Q\sim N^{2k}$ and $\sigma_\theta\sim 1/N^k$. In the bright-soliton setting, $k=1$ corresponds to linear metrology and $k=3$ to the cubic phase accumulation associated with soliton interactions. The three-mode soliton Josephson junction is proposed as an architecture whose phase transition near $\Lambda_{\mathrm{cr}}$ produces tripartite $N00N$-like ground states and near-GHL performance even under weak losses [2507.15348].

A further nonlinear reformulation uses the parameter-space canonical momentum
$$
\hat p_\theta\equiv \hat{\mathcal K}=i\hat U^\dagger(\theta)\partial_\theta \hat U(\theta),
$$
together with the uncertainty relation
$$
\Delta\theta\,\Delta p_\theta \ge \frac12.
$$
In the standard scheme with $N$ sequential uses,
$$
\Delta\theta \ge \frac{1}{2N\Delta\mathcal V_S T}.
$$
For an indefinite-time-direction generating process implemented by a quantum switch, the bound becomes
$$
\Delta\theta \ge \frac{1}{2\sqrt{\Delta\mathcal V_S^2N^2T^2+\frac{(N^2+N)^2}{4}T^4}},
$$
which asymptotically gives a nonlinear improvement $\sim 1/(N^2T^2)$ when the quadratic term dominates. This construction treats noncommutativity and superposition of time directions as additional resources contributing to the canonical momentum dispersion [2510.09216].

Networked and global-accounting formulations push the terminology in a different direction. In the quantum-switch experiment on conjugate displacement processes, the fair global resource count is
$$
N_{\mathrm{global}}=n\,\frac{m(1+\xi)}{\eta},
$$
and the global Heisenberg benchmark is
$$
\Delta\theta_{\mathrm{GHL}}\ge \frac{1}{N_{\mathrm{global}}}
=\frac{\eta}{nm(1+\xi)}.
$$
For the indefinite-causal-order protocol, the control-qubit probabilities are
$$
P_\pm=\frac12\big(1\pm \cos(n^2A)\big),
$$
with Fisher information
$$
F_A=n^4
$$
and precision
$$
\delta A=\frac{1}{\sqrt{m}\,n^2}.
$$
The experiment reports unconditional violation of the global benchmark once losses, visibility, and multi-pair emission are fully included [2505.03290].

## 7. Conceptual tensions and interpretive boundaries

The surveyed literature indicates that GHL is not a single universally fixed theorem. In one family of papers it is a **universal lower bound** derived from resource counting, query complexity, prior averaging, or minimax single-shot analysis [1201.2225], [1004.3944], [1111.0788], [2304.14370]. In another it is an **attainability statement** showing that $F_Q=\Theta(N^2)$ or faster time scaling can still be reached under time-dependent encoding, locality, perturbing interactions, or structured noise, provided suitable control, prior estimates, or QEC conditions are available [1803.01724], [2308.10929], [1903.07888], [2601.05457]. In yet another it is a **generalized scaling law** for nonlinear generators or globally accounted network resources, such as $\sigma_{GHL}^{(k)}\ge 1/N^k$ or $\Delta\theta_{GHL}\ge 1/N_{\mathrm{global}}$ [2507.15348], [2505.03290].

Several recurring misunderstandings are therefore addressed directly by the literature. First, “surpassing the Heisenberg limit” usually means surpassing a narrower benchmark, such as the time-independent $1/T$ law, the fixed-order $1/N$ law, or a detected-photon-only benchmark, not violating a fully accounted universal bound [1803.01724], [2505.03290]. Second, QFI-based local bounds need not coincide with single-shot global performance; the explicit $\pi$ factor in minimax bounds and the prior-averaged constants $k_A$ and $k_C$ were introduced precisely to correct that mismatch [2304.14370], [1111.0788]. Third, Heisenberg scaling under noise is not generic: it depends on structural conditions such as $\hat H_\perp\neq 0$, on estimating the correct parameter such as a collective dephasing rate, or on operating below explicit fault-tolerance thresholds [1903.07888], [1809.00176], [2601.05457].

A plausible implication is that the expression “General Heisenberg Limit” functions as a family of precision statements parameterized by resource accounting, dynamical model, prior structure, and control assumptions, rather than as a single formula valid in all metrological settings. What remains common across these uses is the attempt to state the best achievable precision only after the relevant resources and constraints have been specified with sufficient generality.

Source: https://www.emergentmind.com/topics/general-heisenberg-limit-ghl