arXiv:2610.18776
Quantum-enhanced mechanical sensing with machine-learning estimation
Finn Müßigbrodt, Andraž Omahen, Q. Rumman Rahman, Giulio Corazza, Yiwen Chu, Matteo Fadel
Abstract
Massive mechanical oscillators couple directly to forces, accelerations and gravity, making them attractive probes of new physics and of quantum mechanics at macroscopic scales. Pushing sensitivities beyond the standard quantum limit for signals of unknown phase requires non-Gaussian quantum states, whose metrological advantage has so far been confined to lighter systems such as trapped ions. Here we operate a microgram-scale acoustic mode of a high-overtone bulk acoustic wave resonator as a phase-insensitive quantum-enhanced sensor. A transmon qubit prepares the mode in Fock states of up to seven phonons by optimal control, and resolves its phonon-number distribution after interaction with a weak resonant signal. Bayesian analysis yields a Fisher information five times the vacuum value, a 7.0 dB gain in signal-amplitude sensitivity over any classical probe state, irrespective of its energy, currently limited by state preparation infidelities. We then train a neural network on the measured sensor response to perform Bayesian estimation without any analytical model of the system, showing the same sensitivity gain with half the systematic bias. Our results establish acoustic resonators prepared in non-Gaussian states as quantum-enhanced probes for high-frequency gravitational waves, dark matter and wave-function collapse.
AI Summary
The paper investigates phase-insensitive displacement sensing with non-Gaussian states of a microgram-scale mechanical oscillator, and whether calibration-trained inference can reduce estimation bias without an analytical response model. A transmon coupled to a $5.088\,\mathrm{GHz}$ high-overtone bulk acoustic-wave resonator prepares Fock states through optimal-control pulses. The mode has an effective mass of approximately $16\,\mu\mathrm{g}$ and an energy lifetime of $196\,\mu\mathrm{s}$. States up to $|7\rangle$ are prepared, with fidelities ranging from about $95\%$ to $67\%$. A calibrated resonant microwave drive emulates the signal, and Jaynes–Cummings measurements reconstruct the displaced phonon-number distributions. Hellinger-distance curvature and model-based Bayesian analysis give displacement Fisher information $F\simeq20$, peaking near $d=0.3$ for the $|4\rangle$ probe. This exceeds the bound $F_Q\leq4$ for mixtures of coherent states by a factor of five: a $7.0\,\mathrm{dB}$ gain, or approximately $2.2$-fold reduction in displacement uncertainty at fixed sample count. A neural network trained on calibration records reproduces this sensitivity while reducing mean absolute estimator bias from $6.1\times10^{-3}$ to $3.2\times10^{-3}$. The gain saturates above roughly four phonons, attributed primarily to preparation infidelity rather than sensing-window loss. Indirect number-distribution reconstruction incurs sampling overhead—approximately tenfold in a representative calculation—and reset dominates the duty cycle. The quoted force sensitivity is $0.77\,\mathrm{pN}/\sqrt{\mathrm{Hz}}$. Gravitational-wave, dark-matter and collapse-model applications remain prospective; the experiment senses an emulated drive.
Central claim may not hold: The reported fivefold Fisher-information enhancement is obtained from reconstructed phonon-number distributions rather than the binary outcomes of actual RPN repetitions, leaving the implemented sensor's operational enhancement unestablished. The neural-network sensitivity and bias tests use phonon-number records resampled from reconstructed distributions, so they do not directly demonstrate estimation performance on raw RPN measurement records. Matching variance upper bounds to define an effective sample count does not establish statistical equivalence to direct phonon counting and cannot by itself justify the quoted time-normalized sensitivity.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.