Observation of two nuclear recoil peaks induced by neutron capture on Al2O3

TL;DR

This work demonstrates the Crab calibration method on an Al2O3 cryogenic detector, revealing two nuclear-recoil lines produced by $^{27}$Al neutron capture: a $1145$ eV line from single-$\gamma$ de-excitation and a $575$ eV feature arising from timing in $2$-$\gamma$ cascades. The analysis uses blind peak searches and an energy-scale calibration to extract peak positions and rates, finding a significant second peak and notable non-linearity between electron and nuclear recoil energy scales. Comparisons with Geant4/THULLIEZ-based simulations show qualitative agreement but quantify discrepancies in absolute rates and the $575$/1145 ratio, highlighting stopping-time modeling and detector response as key uncertainties. The results validate the CRAB approach for in situ nuclear-recoil calibration and motivate MD-based investigations and higher-flux measurements at future facilities to achieve high-precision sub-keV calibrations for CEvNS and light dark matter experiments.

Abstract

We report the observation of two nuclear recoil peaks induced by neutron capture on aluminum in a cryogenic Al$_2$O$_3$ detector developed by the NUCLEUS collaboration for the detection of reactor neutrinos via coherent elastic neutrino-nucleus (CEvNS) process. Data collected at the Technical University of Munich in 2024 with a $^{252}$Cf source reveal a main recoil line at 1145 eV from single-$γ$ de-excitation of $^{28}$Al and a newly observed structure near 575 eV originating from several two-$γ$ cascades. The latter constitutes the first direct measurement of a nuclear recoil line induced by multi-$γ$ cascades. It is predicted by our simulations when the recoiling nucleus has time to stop before the emission of the next $γ$-ray in the cascade. These results demonstrate the potential performance of the CRAB (Calibration Recoil for Accurate Bolometry) method for in situ nuclear recoil calibration and highlight the importance of accurately modeling recoil stopping and nuclear de-excitation times in cryogenic detectors of CEvNS and dark matter interactions.

Figures (9)

• Figure 1: Schematic of the experimental setup. The dilution cryostat is surrounded by 14 cm thick lead shielding, against which the source and its shielding are placed. A 5 mm thick boron-loaded rubber mat ($\mathrm{B_4C}$) can be inserted between the source and the lead shielding to block the thermal neutrons. The inset shows the copper housing of the cryogenic detector.
• Figure 2: Energy spectra in voltage units of the neutron source (gray shaded histogram) and background data (red) in the region of interest. For the neutron source data, an additional data set with a boron carbide shielding ($\mathrm{B_4C}$), blocking all thermal neutrons but only a small fraction of fast neutrons, is shown (green).
• Figure 3: Pulse height spectra reconstructed with the truncated pulse shape fit showing the Mn K$_\alpha$(5.89 keV), Mn K$_\beta$ (6.49 keV) and Cu K$_\alpha$ (8.04 keV) lines in the neutron source and background data sets. The color code is the same as in Figure \ref{['fig:spectrum']}.
• Figure 4: Results of the blind peak search applied to the neutron source spectrum of Fig. \ref{['fig:spectrum']}. Each point corresponds to a given position in the scan of the 60 mV-wide window. Only two peaks, with mean position around 106 mV and 192 mV (blue dashed lines), are clearly identified. When the sliding window contains one of these two positions, the null hypothesis (background only) has a p-value $\le10^{-4}$ (left vertical axis) or equivalently a rejection significance $\ge 4\sigma$ (right vertical axis).
• Figure 5: Top row: pulse height spectra in voltage units of the neutron source data after background subtraction, showing the low (left) and high energy (right) peaks identified with the blind peak search. The source related background is modeled as an exponential function plus a constant. The solid and dashed lines show the best fits with and without a Gaussian signal, respectively. The obtained count rates contained in the Gaussian peaks are 11.6(0.9)and 28.7(1.4)counts per day. The bottom row shows the fit residuals, compatible with a normal distribution in the signal+background hypothesis.