Search for low-mass dark matter with CDMSlite using a profile likelihood fit

The Cryogenic Dark Matter Search low ionization threshold experiment (CDMSlite) searches for interactions between dark matter particles and germanium nuclei in cryogenic detectors. The experiment has achieved a low energy threshold with improved sensitivity to low-mass ($<10\mathrm{GeV}/c^2$) dark matter particles. We present an analysis of the final CDMSlite dataset, taken with a different detector than was used for the two previous CDMSlite datasets. This analysis includes a data “salting” method to protect against bias, improved noise discrimination, background modeling, and the use of profile likelihood methods to search for a dark matter signal in the presence of backgrounds. We achieve an energy threshold of 70 eV and significantly improve the sensitivity for dark matter particles with masses between 2.5 and $10\mathrm{GeV}/c^2$ compared to previous analyses. We set an upper limit on the dark matter-nucleon scattering cross section in germanium of $5.4\times {10}^{-42}{\mathrm{cm}}^2$ at $5\mathrm{GeV}/c^2$, a factor of $\sim 2.5$ improvement over the previous CDMSlite result.

Figure 1

The drift in the total phonon energy for events in the $10.37{\mathrm{keV}}_{\mathrm{ee}}$ peak (from ${}^{71}\mathrm{Ge}$ $K$-shell decays) is well modeled by the measured variation of the detector voltage [Eq. (6)]. Early in April the detector voltage stabilized at 75 V. Three ${}^{252}\mathrm{Cf}$ calibrations were performed over the course of the run. The timing of the calibrations (Feb. 2–3, Feb. 20–23, and May 1–5), along with the 11.43 day half-life of ${}^{71}\mathrm{Ge}$, is seen in the variable intensity of the $K$-shell decays. Data points labeled as background simply represent events originating from sources other than $K$-shell decays.

Figure 2

The reconstructed energies of the $10.37{\mathrm{keV}}_{\mathrm{ee}}$ peak events are positively correlated with the base temperature. This dependence is approximated as linear and corrected according to the fitted dashed line.

Figure 3

Calculated voltage map for high radius events, showing the difference in electric potential $\mathrm{\Delta }V$ between the final collection points of the positive and negative charge carriers, as a function of initial position of the pair (plotted as radius squared vs vertical position). Here, the top of the crystal is biased at $+75\mathrm{V}$ and the bottom is grounded. Charge carriers in the outermost (radius $>800{\mathrm{mm}}^2$) detector annulus can experience less than the full detector bias voltage.

Figure 4

Fits of a Gaussian $+$ linear background to the energy spectra of zero-energy (baseline) events and events from each ${}^{71}\mathrm{Ge}$ activation peak. The widths of the Gaussians are the energy resolution $\sigma$.

Figure 5

Two uncorrelated BDT variables are formed based on three sets of parameters that are sensitive to LF noise (see main text). The acceptance region of the bifurcated cuts using the BDT variables is shaded in the upper right.

Figure 6

Variation of the number of background events leaking through the BDT cuts, as a function of the BDT B cut value. The observed number of events, after subtracting the expected contribution from signal events, agrees with the bifurcated analysis estimate.

Figure 7

Resolution ($1\sigma$) for $\xi$ (radial parameter) shown as a function of $\xi$ and energy. At lower energy, the resolution worsens as the increased noise affects the reconstruction of the radial parameter.

Figure 8

Distribution of the radial parameter $\xi$ vs energy in the DM search data. An energy-dependent cut on $\xi$ defines the fiducial volume below $2{\mathrm{keV}}_{\mathrm{ee}}$, while a stricter constant cut is used above $2{\mathrm{keV}}_{\mathrm{ee}}$.

Figure 9

The signal efficiency with successive application of the trigger efficiency, quality cuts efficiency, and fiducial volume cut efficiency. The final data is included with statistical and systematic $1\sigma$ uncertainty. Fitting the efficiency model to these data gives the final (blue) efficiency curve and the corresponding $\pm 1\sigma$ uncertainty band.

Figure 10

Best fit of the Compton scattering spectral model of Eq. (19) to a Geant4 simulation of Compton scatters.

Figure 11

The spectra (normalized to event density) of surface events expected from the three surface background locations (left: germanium; center: housing; right: top lid). For each location, the solid curve represents the mean of the expected event distribution (${\rho }_0$). The shaded band shows the $1\sigma$ uncertainty, where the top and bottom edges of the bands correspond to ${\rho }_{+}$ and ${\rho }_{-}$ in Eq. (20), respectively.

Figure 12

The CDMSlite Run 3 final energy spectrum overlaid with the best-fit background components. The best-fit rates for the ${}^{65}\mathrm{Zn}$ and ${}^{55}\mathrm{Fe}$ components are below the scale of the plot.

Figure 13

The CDMSlite Run 3 90% CL PLR limit (this result, solid black) on the spin-independent WIMP-nucleon cross section, along with the $\pm 1\sigma$ and $\pm 2\sigma$ sensitivity bands (green and yellow, respectively). The CDMSlite Run 3 optimum interval limit (dashed grey) and Run 2 (red) optimum interval limit [18] are overlaid. Examples of limits from other detector technologies are overlaid: DarkSide-50 2018 No Quenching Fluctuations (magenta) [10]; PandaX-II 2016 (blue) [57]; PICO-60 2017 (orange) [58]; CRESST-II 2016 (cyan) [59]; CDEX-10 2018 (purple) [60].