First observation of isolated nuclear recoils following neutron capture for dark matter calibration

Low-energy nuclear recoils (NRs) are hard to measure, because they produce few ${\mathrm{e}}^{-}/{\mathrm{h}}^{+}$ pairs in solids—i.e., they have low “ionization yield.” A silicon detector was exposed to thermal neutrons over 2.5 live days, probing NRs down to 450 eV. The observation of a neutron capture-induced component of NRs at low energies is supported by the much-improved fit upon inclusion of a capture NR model. This result shows that thermal neutron calibration of very low recoil energy NRs is promising for dark matter searches, coherent neutrino experiments, and improving understanding of ionization dynamics in solids.

Figure 1

Top-down view of experimental setup. Cartoon of thermalizing neutron shown in green.

Figure 2

PuBe source spectrum. The neutrons from the source were measured in the early reference [27], and the gammas come from the accompanying deexcitation of the residual ${}^{12}\mathrm{C}$ system. The neutrons are emitted in broad groups corresponding to the excitation level of the residual system and the amount of stopping experience by the $\alpha$ prior to the ($\alpha ,\mathrm{n}$) reaction.

Figure 3

Calibration lines in the background energy spectrum with selection of events near the Am-241 calibration source.

Figure 4

Energy-dependent efficiencies for PuBe (upper) and background (lower) datasets. Black curves show smooth functional forms of the total cut efficiency used for further analysis.

Figure 5

Overlaid histograms comparing the yielded energy PDFs for Sorensen [11] and Lindhard [7] models, including the resolution of the current detector. The histograms are comprised of approximately ${10}^6$ simulated cascades. The orange (front) filled histogram represents the Lindhard model, while the blue (back) filled histogram represents the Sorensen model. In the Sorensen model, many points are pushed to zero due to the presence of a cutoff energy, leading to a peak in the first bin that is not present in the Lindhard model. For both models, we use $k=0.178$ and $q=0.00075$. The solid-line unfilled histogram represents the Lindhard model with a $5\times$ better resolution, and the dashed unfilled histogram represents the Sorensen model with that resolution. Both of these show much taller, narrower peaks than their counterparts.

Figure 6

MCMC fitting results for Lindhard (top) and Sorensen (bottom) yield models. Left: Best-fit yield curves using the specified yield model are shown with statistical and systematic errors, in comparison with multiple published measurements. The detector threshold level is shown as the low magenta dashed curve. Right: range of best-fit background-subtracted reconstructed spectra. Shaded bands represent the 1-sigma equivalent range of rates in each energy bin. On each of the left-hand plots, the result of the integral method without including the neutron-capture-induced NRs is shown (gray dashed lines and gray bands); the resulting ionization yield function is poorly constrained and oddly shaped, implying the necessity of the neutron-capture-induced NR contribution.