arXiv:2611.03943
Pump quantization changes optimal measurements for quantum sensing
S. A. Wadood, L. Bello, T. Ilias, J. W. Fleischer
Abstract
Standard analyses of quantum phase sensing often neglect the overhead of generating squeezed-light by treating the pump as an undepleted classical resource. Realistically, the pump is a quantum object that undergoes significant depletion and entangles with the produced squeezed light, raising the question of how the crossover from a classical to a quantum pump limits the quantum advantage in sensing. Contrary to prior analyses, we show that quantum advantage persists despite pump quantization, but depends critically on the nonlinear $(χ^{(2)})$ source dynamics. In particular, we demonstrate that Heisenberg scaling emerges in the regime of strong pump depletion, rather than maximum signal squeezing. At maximum squeezing, the quantum Fisher information (QFI) for interferometric sensing of an unknown phase scales as $N^{3/2}$, where $N$ is the pump photon number, surpassing the shot-noise limit but establishing that squeezing itself is insufficient to attain Heisenberg scaling of $N^2$. At maximal pump depletion, where the signal flux is maximized, the QFI scales as $N^2$. In this regime, standard photon counting is suboptimal, while a joint nonlinear measurement of the entire pump-signal system is shown to achieve Heisenberg scaling. Our work shows that pump quantization changes both the optimal operating point and the measurement strategies required to preserve a realistic quantum advantage in phase estimation and, more broadly, many-body quantum dynamics.
AI Summary
The paper asks whether squeezed-light phase sensing retains a quantum advantage when the pump’s finite energy, depletion, and entanglement with the signal are included. It models a coherent pump with mean photon number $N$ and a vacuum signal under a fully quantized, three-wave-mixing $\chi^{(2)}$ Hamiltonian, comparing preparation times optimized for squeezing or signal flux. Quantum Fisher information (QFI) and error-propagation sensitivities are calculated using photon-number-sector simulations for $N=50$–$1000$. At the first maximum of squeezing, the signal mean and number variance scale approximately as $N^{1/2}$ and $N$, respectively. Mixing this signal with a coherent probe containing approximately $N$ photons in a Mach–Zehnder interferometer gives fitted QFI scaling $N^{1.48\pm0.008}$; output photon-number-difference detection gives sensitivity scaling $N^{1.52\pm0.012}$. Without the coherent probe, the QFI at this operating point scales only linearly in $N$. Near maximum pump depletion and signal flux, the QFI instead scales as $N^{2\pm0.00024}$. Photon-number-difference error propagation then gives only shot-noise scaling. A different protocol, encoding phase directly on the signal and applying a second pump–signal nonlinear interaction before signal photon counting, recovers $N^2$ sensitivity scaling near a dark fringe. The analysis assumes lossless, perfectly phase-matched, single-mode dynamics and local single-parameter estimation. The scaling claims rely on finite-resource numerical fits and qualitative distribution arguments. Multimode and open-system dynamics, $\chi^{(3)}$ sources, global estimation, finite-data effects, and longer-time optimization remain outside the analysis.
Central claim may not hold: The claimed asymptotic powers are supported by finite-range numerical fits, while Appendix E's support-spanning argument alone does not establish quadratic photon-number variance. The assertion that ordinary photon counting cannot achieve Heisenberg scaling is false in the stated ideal model: signal-parity information obtainable from output photon counts has Fisher information proportional to N^2 near an identity-output fringe at maximum flux. Appendix A's displayed Hamiltonian matrix is non-Hermitian because its upper off-diagonal coefficient is incorrect, although the paper does not establish whether this transcription error affects the simulations.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.