Fault-Tolerant Heisenberg-Limited Quantum Metrology with Bosonic Probes

Abstract

Quantum sensors have been shown to enable more precise measurements than classical sensors. By using entanglement and coherent evolution, the estimation error can be made to scale better in interrogation time, photon count, and other resources, achieving Heisenberg-limited scaling. However, environmental noise and hardware errors typically destroy the coherence of the sensitive quantum states. This motivates the use of fault-tolerant techniques to create more robust quantum sensing protocols. Such fault-tolerant quantum sensing approaches have been shown to enhance sensitivity for qubit systems, but many practical sensors require bosonic probes. In this paper, we propose a fault-tolerant protocol using bosons to estimate displacements in the presence of dephasing noise. We construct an optimal error-correcting code under a mean photon-number constraint. Controlled displacements couple the probe to a repetition-encoded qubit register, while repeated boson-parity measurements and resets remove probe noise without erasing the stored logical phase. Under circuit-level bit-flip noise, a distance-$d$ code suppresses logical failures to order $p^{(d+1)/2}$. We show that, for a decay rate $γ_A$, a total sensing time $T$ and a mean photon number budget $N$, and for sufficiently small residual signals, the Fisher information scales as $NT^2$ over interrogation times of order $γ_A^{-(d+1)/2}$. This is a significant improvement over the non-fault-tolerant approach that achieves Heisenberg scaling only for a total time $\mathcal O(γ_A^{-1})$.

The paper asks whether quantum-error-corrected bosonic displacement sensing can retain Heisenberg scaling when ancilla operations and recovery measurements are faulty. It considers $H=\theta(a+a^\dagger)$, photon-number dephasing $L=\sqrt\kappa\,n$, and a mean photon-number budget $N$ at the start of each sensing interval. The protocol couples coherent states $|\pm\sqrt N\rangle$ to a distance-$d$ qubit repetition register using controlled displacements. After sensing for time $\tau$, it uncomputes the coupling, measures oscillator parity, applies a conditional logical phase correction, and resets the oscillator while retaining accumulated phase in the register. The code satisfies the error-correction conditions for $\{I,n\}$ and maximizes the projected signal gap, $4\sqrt N$, among two-dimensional codes correcting $n$ under the energy constraint. Repeated parity measurements and time-resolved repetition-code decoding suppress logical failures to $O_d(p^{(d+1)/2})$ under independent circuit-level qubit bit flips. Combining explicit recovered-channel calculations with memory-decoding bounds gives local classical Fisher information $\mathcal F_C(0)\ge16NT^2f(\kappa\tau)^2(1-2\eta_M)^2$, where $T=M\tau$, $f(x)=(1-e^{-x/2})/(x/2)$, and $\eta_M\le b_dp+a_d(M+1)p^{(d+1)/2}$. Thus a fixed fraction of ideal sensitivity survives for $M=O(p^{-(d+1)/2})$, versus $O(p^{-1})$ for unprotected parity reporting. The guarantees assume ideal oscillator reset, reliable classical processing, and strongly biased ancilla noise; phase faults remain uncorrected. Finite-signal bounds apply on a shrinking residual-signal interval, sufficiently $|\theta|\lesssim1/(\sqrt N T)$, without providing global parameter localization. Recovery latency reduces the sensing duty cycle, distance-dependent constants are unspecified, and no uniform large-distance threshold is established.