Optomechanical Accelerometer Search for Ultralight Dark Matter

TL;DR

The paper reports a resonant search for ultralight dark matter with vector couplings using a cavity optomechanical accelerometer. It uses a cryogenic Si$_3$N$_4$ membrane cavity attached to a 4 K copper plate, with photothermal tuning to scan a 39 kHz mechanical resonance, achieving shot-noise-limited displacement readout and radiation-pressure feedback cooling to a near-thermally-limited acceleration sensitivity of $\sim 10\,\mathrm{n}g_0/\sqrt{\mathrm{Hz}}$. A Bayesian, matched-filter search over a scanned spectral window yields upper bounds on vector couplings (notably $g_{\mathrm{B-L}}$) that are consistent with thermal noise and currently weaker than equivalence principle tests, with no detected signal. The results validate a scalable resonant-detection paradigm and, with improvements in test mass, temperature, and sensor arrays, offer a path to competitive constraints on vector-mediated dark-matter interactions in the optomechanical band.

Abstract

Cavity optomechanical systems have recently been proposed as detectors for ultralight dark matter, leveraging their ability to cool and probe mechanical oscillators at the quantum limit. Here we present a resonant search for ultralight dark matter using a cavity optomechanical accelerometer. The detector consists of a cryogenic Si$_3$N$_4$-membrane cavity mounted to a 4 K copper plate, with photothermal tuning used to scan its 39 kHz mechanical resonance. Shot-noise-limited displacement readout and radiation-pressure feedback cooling yield an acceleration sensitivity of $\sim 10\;\text{n}g_0/\sqrt{\text{Hz}}$ over 30 Hz near resonance. The detector's material inhomogeneity gives access to direct vector coupling to the dark-matter field. We conduct a Bayesian search based on matched-filter statistics, yielding upper bounds consistent with thermal noise and above those set by equivalence principle tests. No signal is observed, but the experiment demonstrates stable, quantum-limited operation and validates a scalable approach to resonant detection. With optimized test masses, lower temperature, and multiplexed arrays, the platform offers a path toward competitive constraints on vector-mediated dark-matter interactions.

Figures (5)

• Figure 1: Dual-membrane optomechanical dark matter detector. (a) Si chip with suspended Si$_3$N$_4$ membranes fixed to a Cu plate. Inset: Microscope image of trampoline (foreground) and square membrane (background) serving as acceleration test and reference mass, respectively. Concept: (b) UDM differentially accelerates the Cu plate, Si$_3$N$_4$ membranes, and Si chip. (c) In the reference frame of the base, the membranes experience a base excitation. Their different stiffnesses $k_{1,2}$ result in a relative displacement $x$.
• Figure 2: Cryogenic operation enabled by vibration isolation. (a) Photos of Bluefors LD-4K cryostat (left) and custom vibration isolation system (VIS) based on thin Cu strips (right) suspended from the mixing stage. (b) Cavity alignment module (photo, bottom) and readout: probe and feedback laser beams are fiber-coupled and aligned to the dual-membrane cavity; the assembly is mounted on the VIS-platform (top, cartoon). A photodetector placed outside the cryostat records the cavity's transmission. (c) Finite-element simulations of the fundamental modes of the VIS along $x$ (left) and $y$ (right). (d) Transmission fringes obtained from a laser detuning sweep before and after cooldown, showing cavity alignment remains stable. The green circle highlights the detuning used for side-of-fringe readout. (e) Energy ringdown of the trampoline's fundamental mode at 4 K reveals $Q_\mathrm{m}$ of 60 million. (f) Broadband spectrum showing a reduction of vibration background above $f\sim 1\;\mathrm{Hz}$.
• Figure 3: Displacement and acceleration measurements. (a) Calibrated membrane-trampoline displacement PSD near trampoline's fundamental mode frequency $f_\mathrm{m}$, for a weak probe with transmitted power $P_\mathrm{out} \approx 1\;\upmu\mathrm{W}$. The red (gray) trace is obtained with (without) optical damping. (b) Corresponding closed-loop acceleration PSD after inverting the mechanical susceptibility. (c) Plots of acceleration noise versus frequency detuning from resonance at $P_\mathrm{out} = 1\;\upmu\mathrm{W}$ (red) and $100\;\upmu\mathrm{W}$ (blue), illustrating the trade-off between bandwidth and sensitivity due to photothermal heating. The gray-shaded region corresponds to the UDM-signal linewidth $\gamma_\mathrm{DM}\approx 2\pi\times 0.08$ Hz. (d) Photothermal frequency tuning of $f_\mathrm{m}$. Increasing the probe power decreases $f_\mathrm{m}$ at the expense of increased thermal noise $S_a^\mathrm{th}$.
• Figure 4: Search for B-L UDM. (a) Estimates of UDM acceleration power $\langle a_\mathrm{DM}^2\rangle \equiv D$ obtained from the data in Fig. \ref{['fig:3']}d using a matched-filter and weighted averaging, Eq. \ref{['eq:D']}. Dashed lines are thresholds for confidence level $\mathrm{CL} = 68\;\%$ ($1\sigma$) and $95\;\%$ ($2\sigma$). The solid red line is our threshold, which includes the look-elsewhere effect (Eq. \ref{['eq:CL']}). (b) Histogram of normalized power estimates $\hat{D}/\sigma_D$. Solid and dotted curves show measured and ideal distributions expected from periodogram averaging (Eq. \ref{['eq:sigmaSa']}). (c) Constraints on $g_\mathrm{B-L}$ from data in (a). Light and bold traces are statistical and analytic estimates, respectively. Gray traces assume a rigid chip; red traces include the estimated chip transfer function SI. (d) Comparison to current constraints from LIGO guo2019searchingabbott2022constraints, the Eöt-Wash experiment wagner2012torsion, MICROSCOPE berge2018microscope, and POLONAISE amaral2025first. Gray dashed line is an extension of the analytical model in (c) assuming a displacement imprecision of $S_x^\mathrm{imp}=(5\times 10^{-14}\;\mathrm{m}/\sqrt{\mathrm{Hz}})^2$SI.
• Figure 5: Projected performance of contemporary optomechanical accelerometers as cryogenic UDM detectors, assuming an integration time of $\tau = 10^5$, a differential charge ratio of $\Delta_{12}=0.04$, and experimental parameters as described in Table 1. Orange markers assume coherent averaging of 10 thermal-noise-limited sensors.