XENONnT Low-Energy ER Dataset

• The dataset enables statistically robust exclusion limits for BSM signals including axions, sterile neutrinos, and dark matter–electron interactions.
• It employs rigorous event selection, energy reconstruction down to 1 keV, and a fully characterized nine-component background model.
• Calibration using multiple radioactive sources ensures precise energy resolution and efficiency determination across 1–30 keV.

The XENONnT low-energy electronic recoil (ER) dataset comprises a high-statistics, low-background compilation of ER events collected with the XENONnT dual-phase liquid xenon time-projection chamber (TPC) during its first science run (SR0). This dataset underpins analyses for physics beyond the Standard Model (BSM), including searches for axions, sterile neutrinos, @@@@1@@@@ photons, dark matter–electron couplings, and low-energy tests of quantum foundations. The key dataset features are: a fiducial exposure of 1.16 t·yr in the central 4.37 t of liquid xenon, energy reconstruction and resolution validated down to 1 keV, a fully characterized background model with nine separate components, and rigorous selection and efficiency determination across 30 logarithmic bins spanning 1–30 keV. This enables statistically robust exclusion limits and discovery potential for a broad range of new-physics signals.

1. Experimental Configuration and Exposure

XENONnT is a dual-phase liquid–gas xenon TPC installed at Laboratori Nazionali del Gran Sasso (LNGS), employing 8.5 t of LXe, of which 5.9 t fill the active TPC. The fiducial mass for the low-energy ER analysis is set at 4.37 ± 0.14 t, typically defined by a radial cut (r < 42 cm) and drift-length cut (z ∈ [–94, +154] mm). Data acquisition for SR0 spanned July 6–November 10, 2021, yielding a net live time of T ≈ 97.1 days (0.266 years) after data-quality selection. The total exposure is thus 4.37 t × 0.266 yr = 1.16 t·yr (or 4.23 × 10^5 kg·days) (Mustamin et al., 1 Nov 2025, Aprile et al., 2022, Collaboration et al., 22 Dec 2025).

2. Event Selection, Energy Reconstruction, and Binning

Events are required to be single scatters (one S1 and one S2), with S1 between 1–100 photoelectrons (phe), S2 > 60–500 phe, and a tight threefold S1 coincidence. Fiducialization is enforced with radial and z cuts, avoiding wall/gate and surface regions. Pulse-shape, waveform-quality, PMT-multiplicity, and S2/S1 ratio cuts discriminate ER from nuclear recoils (NR) at >99.8% ER acceptance. The reconstructed ER energy is computed as

$$E_\mathrm{rec}(\mathrm{cS1},\mathrm{cS2}) = W \left( \frac{\mathrm{cS1}}{g_1} + \frac{\mathrm{cS2}}{g_2} \right)$$

where $W = 13.7\,\mathrm{eV}/\mathrm{quanta}$, $g_1\approx0.14\,\mathrm{phe}/\text{quanta}$, $g_2\approx11.5$–$16.5\,\mathrm{phe}/\text{electron}$ (calibrated periodically with $^{83\rm m}$Kr and tritiated methane) (Aprile et al., 2022, Aprile et al., 5 Jun 2025).

Events are histogrammed in 30 uniform bins of width 1 keV, over the primary window $T_e\in[1\,\mathrm{keV},\,30\,\mathrm{keV}]$, with extensions to 140 keV adopted in X-ray and other BSM searches (Mustamin et al., 1 Nov 2025, Aprile et al., 2022, Aprile et al., 5 Jun 2025).

3. Energy Calibration, Resolution, and Efficiency

Calibration utilizes (i) monoenergetic $^{83\rm m}$Kr (9.4 and 32.1 keV) lines to map $g_1,g_2$, (ii) tritiated methane ($\beta^-$ endpoint 18.6 keV) for linearity, (iii) $^{127}$Xe and $^{131\rm m}$Xe for low-energy lines, and (iv) $^{37}$Ar (2.82 keV EC) for the threshold anchor. The energy resolution is characterized by a Gaussian or skew-Gaussian with $\sigma(E) = a\sqrt{E} \oplus b$, where typically $a = 0.28$–$0.31\,\mathrm{keV}^{1/2}$, $b \approx 0.012$–$0.0037 E$ keV. For example, $\sigma(1\,\mathrm{keV})\approx 0.3\,\mathrm{keV}$, $\sigma(10\,\mathrm{keV})\approx 1.0\,\mathrm{keV}$ (Mustamin et al., 1 Nov 2025, Collaboration et al., 22 Dec 2025, Aprile et al., 5 Jun 2025, Demirci et al., 14 Oct 2025).

The total efficiency $\epsilon(E)$ combines trigger, threshold, event selection, and DAQ acceptance, validated by calibration data and toy-MC. Analytic fits used in the literature include $\epsilon(E) = 1 - \exp(-E/E_0)$ with $E_0=3.2\,\mathrm{keV}$; numerical curves show $\epsilon(1\,\mathrm{keV})\sim 20\%$, $\epsilon(3\,\mathrm{keV})\sim80\%$, plateauing at $\sim90$–$95\%$ above 5 keV. Systematic uncertainty on $\epsilon(E)$ is $5\%$ at threshold, $2\%$ for $E\geq3\,\mathrm{keV}$, dropping to $1\%$ above 10 keV (Aprile et al., 5 Jun 2025, Collaboration et al., 22 Dec 2025, Demirci et al., 14 Oct 2025).

4. Background Model and Systematics

The full background model $B(E)$ is a weighted sum of nine components:

1. $^{214}$Pb and $^{85}$Kr $\beta$-decays (dominant below 40 keV)
2. $^{136}$Xe $2\nu\beta\beta$, $^{124}$Xe $2\nu$ double electron capture (peak at 63.6 keV)
3. Solar neutrino–electron scattering (pp, $^7$Be, CNO)
4. External and internal $\gamma$ rays (simulated by Geant4)
5. $^{83\rm m}$Kr, $^{133}$Xe, and accidental coincidences (AC)

Each background component has a detailed spectral template $S_i(E)$ (analytical or simulated), normalization parameter $\alpha_i$, and a systematic normalization "pull" $\beta_j$ constrained in the fit. Typical systematic uncertainty per component is $5$–$15\%$, with exact values from detector assay and in-situ calibration (Mustamin et al., 1 Nov 2025, Aprile et al., 2022, Aprile et al., 5 Jun 2025, Demirci et al., 14 Oct 2025).

The overall ER background rate in the 1–30 keV window is measured as $(15.8\pm1.3)~\mathrm{events}/$(t·yr·keV), the lowest ever achieved in a dark matter detector (Aprile et al., 2022). All backgrounds are constrained in an unbinned likelihood or Poisson $\chi^2$ framework with nuisance parameters for both normalization and energy-dependent efficiency systematics.

5. Calibration and Response Modeling

Calibration of the ER response at low energies is achieved via two principal approaches:

• $^{220}$Rn source deployments inject $^{212}$Pb, producing a nearly flat, high-statistics $\beta$ spectrum below 30 keV, yielding $\sim$10^6$$ER events per 48 h run. Uniformity is achieved within 2<li>The absolute and relative scintillation yield is benchmarked via dedicated Compton-scattering measurements in LXe cells, with$$L_\mathrm{rel}(E)$$measured down to 1.5 keV and a field‑quenching factor$$q(E)\simeq0.75$$below 10 keV. This underpins accurate conversion of S1 signals to true$$E_\text{er}$$and informs the systematic error budget at threshold in large-scale analyses (<a href="/papers/1303.6891" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Baudis et al., 2013</a>).</li> </ul> <p>These calibration data constrain linearity, energy resolution ($$\sigma_E/E\sim8\%$ at 10 keV), and event-selection efficiency to sub-few percent, supported by full-scale MC and dedicated calibration campaigns (Jörg et al., 2023, Baudis et al., 2013).

  6. Data Release, Spectral Content, and Statistical Analysis

  The SR0 ER data are available in binned and unbinned forms over 1–30 keV, with per-bin observed counts, backgrounds, and uncertainties provided in figures and supplementary material (Mustamin et al., 1 Nov 2025, Aprile et al., 2022). An example of the binned dataset is summarized below:

  Bin (keV)	R_obs^k	σ^k	Bkg^k (SM + B_j)
  1–2	1250	35	1240
  2–3	1120	33	1105
  3–4	980	31	970
  ...	...	...	...
  29–30	45	7	42

  Statistical inference is performed using a Poisson-likelihood-based $\chi^2$ test with nuisance pulls for solar flux (α_i) and backgrounds (β_j):

  $$\chi^2(\mathcal{M}) = \min_{\alpha,\,\beta} \left\{ 2\sum_{k=1}^{30} [R_\mathrm{exp}^k(\mathcal{M};\alpha,\beta) - R_\mathrm{obs}^k + R_\mathrm{obs}^k \ln(R_\mathrm{obs}^k/R_\mathrm{exp}^k)] + \sum_i \left( \frac{\alpha_i}{\sigma_{\alpha_i}} \right)^2 + \sum_j \left( \frac{\beta_j}{\sigma_{\beta_j}} \right)^2 \right\}$$

  Exclusion contours are set by $\Delta\chi^2=2.71$ for 90% CL (one dof) and 4.61 for two dof. The statistical model is compatible with both unbinned extended-likelihood and binned-likelihood fits (Mustamin et al., 1 Nov 2025, Demirci et al., 14 Oct 2025).

  7. Applications and Scientific Impact

  The XENONnT low-energy ER dataset has established world-leading sensitivity to a variety of BSM phenomena:

  • Constraints on active–sterile neutrino transition magnetic moments, neutrino millicharge, and charge radius, improving or confirming the leading limits in the low-mass regime (Mustamin et al., 1 Nov 2025, Demirci et al., 14 Oct 2025).
  • Searches for solar axions, bosonic dark matter, and spontaneous quantum collapse predicted X-ray emission, exploiting the background-free window and high exposure (Aprile et al., 2022, Aprile et al., 5 Jun 2025).
  • Dark matter–electron interaction limits for masses down to 0.2 MeV/c$^2$, driven by the low ER background, rigorous efficiency modeling, and calibration down to 1 keV (Collaboration et al., 22 Dec 2025).
  • Data access to the community via open-source straxen software enables reproducibility and BSM statistical reinterpretation. Binned and unbinned spectra, response models, and selection criteria are available to researchers upon request (Aprile et al., 2022).

  Significance: By combining high-mass LXe exposure, state-of-the-art background suppression, comprehensive calibration, and rigorous statistical methodology, the XENONnT low-energy ER dataset sets a new standard in rare event searches at the eV–keV energy scale. This dataset serves as a critical benchmark for next-generation dark matter, neutrino, and quantum-mechanics–test experiments.