- Title: The paper outlines for a 140m vertical atom interferometer at CERN’s PX46 access shaft to search for bosonic ultralight dark matter (ULDM) in scalar, vector B-L, and axion-like channels over the mass range 10^{-22}-10^{-12}eV.
- The facility employs three operating configurations: differential gradiometer, dual-species full-baseline Mach-Zehnder, and full-baseline single-species clock, using single-photon interferometry to cancel laser phase noise.
- Projected sensitivities show that AICE can surpass existing limits by several orders of magnitude, with key performance targets including $|d_{m_e}| extgreater 2.4 imes10^{-8}$ and $|G_{an}| extgreater 9.5 imes10^{-10}GeV^{-1}$ for baseline, achieving new scientific milestones.
Overview
This document is a technical proposal for the Atom Interferometer CERN Experiment (AICE), a 140 m vertical atom interferometer to be installed against the wall of the PX46 access shaft at LHC Point 4 (2608.18743). The facility is conceived as a configurable platform rather than a single-purpose experiment, whose primary scientific goal is the search for bosonic ultralight dark matter (ULDM) in scalar, vector B−L, and axion-like channels over the mass range 10−22–10−12 eV, with secondary programmes in equivalence-principle tests, fine-structure-constant determination, and gravitational-wave (GW) pathfinding in the ∼0.03–3 Hz band. AICE is endorsed by the Terrestrial Very-Long-Baseline Atom Interferometry (TVLBAI) proto-collaboration of 57 institutions in 22 countries and is positioned as the key European step toward km-scale detectors.
Scientific motivation and measurement principle
ULDM fields in the target mass range behave on laboratory scales as coherently oscillating classical backgrounds with frequencies 10−7–1 Hz, producing oscillating perturbations of fundamental constants or composition-dependent forces. AICE transduces these effects into differential interferometric phase through three operating configurations of a single infrastructure:
- Differential gradiometer mode: two vertically separated 87Sr interferometers interrogated by a common 698 nm clock laser; sensitive to scalar ULDM via clock-frequency modulation, with response scaling as keff​Δr, where keff​=nk0​.
- Dual-species full-baseline Mach-Zehnder mode: co-located species (87Sr/88Sr or 10−220Sr/10−221Yb) interrogated over the full one-way drop (10−222 s); sensitive to vector 10−223 dark matter and to violations of the weak equivalence principle through species-differential acceleration.
- Full-baseline single-species clock mode: a single 10−224Sr cloud with two simultaneous opposite-spin-state interferometers, isolating axion-induced spin-dependent phases while cancelling laser phase noise in common mode.
The choice of single-photon interferometry on the ultra-narrow 10−225 transitions is deliberate: it permits complete cancellation of laser phase noise in gradiometric operation, whereas two-photon schemes retain residual phase noise that grows with LMT order and baseline separation. The proposal argues that this makes single-photon clock interferometry essential rather than merely convenient for the targeted sensitivity levels.
The facility follows a three-level performance ladder — Initial (commissioning), Baseline, and Stretch — with interrogation times fixed at 10−226 s and 10−227 s throughout, and LMT order rising from 10−228 initially to 10−229 (Baseline) and 10−120 (Stretch). Phase-noise targets fall from 10−121 to 10−122 rad10−123.
The headline projections are substantial. In the scalar channel, AICE surpasses MICROSCOPE by several orders of magnitude in both 10−124 and 10−125 couplings across 10−126–10−127 eV, reaching Baseline plateaus of 10−128 and 10−129 near the optimal mass ∼0 eV, improving to ∼1 at Stretch. In the axion channel, the Baseline plateau ∼2 approaches the nominal SN 1987A bound to within a factor of two; Stretch surpasses it by roughly five and the 1 km extension by fourteen. In the vector channel with the ∼3Sr/∼4Sr pair, the Baseline reaches ∼5, approximately three orders of magnitude beyond current torsion-balance limits. These are SNR = 1, one-year, shot-noise-limited figures under a single consistent statistical prescription; the proposal candidly notes that overlaid external constraints derive from heterogeneous statistical conventions, and that stochastic-amplitude corrections of order 1.5 are discussed but not applied.
The precision-physics programme is embedded within the same timeline. An Ultimate WEP campaign targets ∼6, competitive with MICROSCOPE but using a systematically distinct atom-interferometric technique; notably, the measurement is explicitly systematics-limited rather than statistics-limited, since the formal shot-noise-limited sensitivity reaches ∼7–∼8. The fine-structure-constant programme culminates in ∼9 using single-photon Ramsey-Bordé recoil measurements on Sr and Yb — an independent family that would probe the more-than-10−70 tension between the 2018 caesium and 2020 rubidium determinations, and which would push the QED contribution to the electron anomalous magnetic moment below the hadronic uncertainty floor.
Detector systems
The vacuum system comprises a modular main interferometer tube of twenty-four DN200 sections with thirteen interconnecting chambers, NEG-coated for distributed pumping, targeting 10−71 mbar in the beam pipe. Atom sources build on demonstrated high-flux compact Sr sources (up to 10−72 atoms/s at the MOT stage) combined with matter-wave lensing toward pK temperatures; the Initial configuration targets 10−73–10−74 atoms/s, with continuous BEC and steady-state reservoir technologies identified as candidate upgrade paths. Laser infrastructure separates master lasers and frequency references in a surface laboratory in SX4 from local power amplification in the shaft, with phase-stabilised fibre delivery; reaching Stretch parameters requires an 8 W → 100 W upgrade of the 698 nm clock laser that is explicitly not achievable with commercially available technology and constitutes a key R&D item shared with the TVLBAI programme. Magnetic shielding adopts double-layer octagonal mu-metal shields providing a static shielding factor of 1000, sufficient against measured ambient noise and LHC machine-cycle field variations.
Site integration and civil engineering
Two dedicated CERN studies established that PX46 — the deepest CERN shaft at ~143 m height and 10.1 m internal diameter — can host the instrument without fundamental technical obstacles and without impacting HL-LHC operations. Roughly 17 m² of free cross-sectional area suffices for the experiment, leaving nearly 60 m² for equipment transport. Radiation protection analysis shows ambient dose rates well below Supervised Area limits behind a proposed 0.8 m concrete shielding wall, including under accidental HL-LHC beam loss. Fire safety relies on a custom elevator platform achieving evacuation to surface or shaft base within two minutes, with battery backup and mechanical failsafe descent. The LS3 enabling works — shielding wall, elevator, access-safety system modifications, fire detection, ventilation room — are costed at 1.5 MCHF (Class 4), with the experiment itself estimated at 11.6–17.8 MCHF including contingency for the Initial strontium configuration. The critical scheduling argument is that completing site preparation during LS3 decouples installation from accelerator operations; deferring these works would leave the programme dependent on a later shutdown window.
Limitations and open questions
The proposal is explicit about several constraints bearing directly on its projections. Gravity gradient noise (GGN) is excluded from all ULDM sensitivity curves; the unmitigated model predicts seismic GGN exceeding atom shot noise below approximately 0.8, 2.5, and 5 Hz for Baseline, Stretch, and km-scale configurations respectively. At the optimum this costs roughly a factor of ten in coupling reach for Baseline and 10−75 for Stretch, and — importantly — the GGN-limited reach is independent of LMT order because signal and GGN phase share the same 10−76 dependence, so increased momentum transfer alone cannot recover the shot-noise-limited projections in the affected band. Without mitigation, the Baseline scalar reach would largely lie within already-excluded territory in the GGN-dominated region. The estimate also rests on assumptions acknowledged as uncertain in either direction: the half-space Newtonian prefactor, power-law extrapolation of seismic spectra below 0.1 Hz, and above all the unmeasured cross-correlation between the shaft ends, which quadrature addition treats conservatively only for positively correlated far-field disturbances and potentially optimistically for near-field anthropogenic sources such as the cooling and ventilation plant. A characterisation programme of synchronised seismometry, witness-sensor subtraction, multi-gradiometer combinations, and atmospheric monitoring is defined but not yet demonstrated at the required scale.
Further open items include the quantification of airflow-induced turbulence in the shaft, validation of wavefront-aberration suppression toward the 10−77 Stretch target (which requires 10−78 source position control and spatially resolved subtraction techniques validated only in simulation), and the high-power clock-laser R&D. The Baseline and Stretch performance levels will require R&D in atom flux, squeezing-enhanced readout, and LMT pulse efficiency; the proposal correctly identifies the step from existing 10 m prototypes as one of integration and scale rather than new principle, but the sensitivity claims remain conditional on that R&D succeeding. The GW pathfinder role depends on GGN mitigation to be scientifically useful at the low-frequency end of the mid-band, where the unmitigated benchmark degrades strain sensitivity.
Conclusion
AICE presents a technically mature, staged proposal for a 140 m vertical atom interferometer exploiting a uniquely prepared site at CERN, with projected sensitivities to scalar, vector, and pseudoscalar ultralight dark matter that would substantially exceed existing laboratory bounds, alongside competitive equivalence-principle tests and an independent fine-structure-constant determination. The scientific case is internally consistent and the statistical conventions transparently stated. The principal quantitative caveat is gravity gradient noise at low frequencies, whose mitigation is identified as the central experimental challenge and whose resolution will determine how much of the shot-noise-limited reach survives in the mass range most relevant to the lowest-mass ULDM candidates.