Search for sterile neutrino mixing in the MINOS long-baseline experiment

A search for depletion of the combined flux of active neutrino species over a 735 km baseline is reported using neutral-current interaction data recorded by the MINOS detectors in the NuMI neutrino beam. Such a depletion is not expected according to conventional interpretations of neutrino oscillation data involving the three known neutrino flavors. A depletion would be a signature of oscillations or decay to postulated noninteracting sterile neutrinos, scenarios not ruled out by existing data. From an exposure of $3.18\times {10}^{20}$ protons on target in which neutrinos of energies between $\sim 500\mathrm{MeV}$ and 120 GeV are produced predominantly as ${\nu }_{\mu }$, the visible energy spectrum of candidate neutral-current reactions in the MINOS far detector is reconstructed. Comparison of this spectrum to that inferred from a similarly selected near-detector sample shows that of the portion of the ${\nu }_{\mu }$ flux observed to disappear in charged-current interaction data, the fraction that could be converting to a sterile state is less than 52% at 90% confidence level (C.L.). The hypothesis that active neutrinos mix with a single sterile neutrino via oscillations is tested by fitting the data to various models. In the particular four-neutrino models considered, the mixing angles ${\theta }_{24}$ and ${\theta }_{34}$ are constrained to be less than 11° and 56° at 90% C.L., respectively. The possibility that active neutrinos may decay to sterile neutrinos is also investigated. Pure neutrino decay without oscillations is ruled out at 5.4 standard deviations. For the scenario in which active neutrinos decay into sterile states concurrently with neutrino oscillations, a lower limit is established for the neutrino decay lifetime ${\tau }_3/m_3>2.1\times {10}^{-12}\mathrm{s}/\mathrm{eV}$ at 90% C.L.

Figure 1

The distribution of time separation between events $\Delta t$ for data (solid points) and Monte Carlo simulated data (solid histogram). A background component arising from poorly reconstructed events (hatched histogram) is confined to a narrow region of low $\Delta t$ values. Events accepted for further analysis are indicated by the arrows.

Figure 2

Section of the distribution of the number of strips with nonzero pulse-height, per event, after all other selections are applied. The region displayed, in which the contribution from poorly reconstructed events (hatched histogram) is significant, corresponds to low strip counts, and represents a small fraction of the total number of events. The event range accepted for the analysis is identified by the arrow.

Figure 3

Comparisons of near-detector data with Monte Carlo predictions for distributions of the variables (a) event length and (b) track extension. The data quality requirements described in Sec. 3 are applied. Systematic uncertainties are displayed as shaded bands on the Monte Carlo expectation. Events selected as neutral-current-like are indicated by the arrows.

Figure 4

Far-detector data versus predictions from the Monte Carlo (MC) simulation including ${\nu }_{\mu }\to {\nu }_{\tau }$ oscillations, for distributions of the classification variables (a) event length and (b) track extension. The data quality requirements of Sec. 3 are applied. The shaded bands show the systematic errors on the MC predictions. The arrows indicate events selected as neutral-current-like.

Figure 5

Selection efficiency and sample purity for Monte Carlo (MC) events selected as neutral-current candidates in the near and far detectors, as a function of reconstructed event energy. Detector selection, fiducial volume, and neutral-current/charged-current separation requirements have been applied. The shapes of both efficiency and purity distributions are observed to be very similar for each of the two MINOS detectors.

Figure 6

Distributions of selection efficiency (lower curves) and sample purity (upper curves) versus reconstructed event energy, for Monte Carlo (MC) events selected as charged-current candidate events in the near detector and in the far detector.

Figure 7

Distribution of reconstructed visible energy for selected neutral-current events in the near detector, for the data (solid points) versus the Monte Carlo prediction (open histogram). The systematic errors ($1\sigma$) for the Monte Carlo are shown by the shaded band. Also shown is the Monte Carlo prediction for the background of misidentified charged-current events in the near-detector sample (hatched histogram).

Figure 8

The reconstructed energy spectrum for selected charged-current events of the near detector, for the data (solid points) versus the Monte Carlo (open histogram). Systematic errors at $1\sigma$ for the Monte Carlo are indicated by the shaded band; the hatched histogram (lower left) shows the Monte Carlo expectation for the small background contribution by misidentified neutral-current events to this sample.

Figure 9

The reconstructed energy spectrum of neutral-current selected events at the far detector (points with statistical uncertainties). The Monte Carlo prediction assuming standard three-flavor oscillations is also shown, both with (dashed line) and without (solid line) ${\nu }_e$ appearance at the CHOOZ limit. The shaded region indicates the 1 standard deviation systematic uncertainty on the prediction. The hatched region shows the Monte Carlo prediction for the background of misidentified charged-current events in this sample.

Figure 10

Schematic representation of the possible mass spectra for models including four mass eigenstates. In the left panel, the first and fourth eigenstates are degenerate, the center panel has the fourth mass eigenstate much heavier than the third, and the right panel has the third and fourth eigenstates as degenerate. Only the scenarios illustrated in the left and center panels are considered in this analysis.

Figure 11

The reconstructed energy spectrum of charged-current selected events at the far detector (points with statistical uncertainties). The Monte Carlo prediction assuming standard three-flavor oscillations (solid line) is shown with the 1 standard deviation systematic uncertainty on the prediction indicated by the shaded region. The prediction for the small background of misidentified neutral-current events in this sample (hatched region) is also shown.

Figure 12

Projections of $\Delta {\chi }^2$ as a function of the mixing angles for the $m_4=m_1$ model. The solid line contours are obtained with null ${\nu }_e$ appearance, whereas the dashed line contours include ${\nu }_e$ appearance at the CHOOZ limit. The ranges of values allowed at 68% and 90% confidence levels lie within contours below the horizontal dashed lines.

Figure 13

Contours representing 90% confidence level for the $m_4=m_1$ model. The solid line and best-fit point (solid symbol) are obtained assuming null ${\nu }_e$ appearance, whereas the dashed line and corresponding best-fit point (open symbol) are obtained with ${\nu }_e$ appearance set at the CHOOZ limit.

Figure 14

Projections of $\Delta {\chi }^2$ as a function of the mixing angles for the $m_4\gg m_3$ model. The solid line is obtained for the case of null ${\nu }_e$ appearance whereas the dashed line represents solutions with ${\nu }_e$ appearance at the CHOOZ limit. The ranges of values allowed at 68% and 90% confidence levels lie within contours below the horizontal dashed lines.

Figure 15

Contours representing 90% confidence level for the $m_4\gg m_3$ model. The solid line and best-fit point (solid symbol) are obtained for the case of null ${\nu }_e$ appearance, whereas the dashed line and corresponding best-fit point (open symbol) is obtained with ${\nu }_e$ appearance included with ${\theta }_{13}$ at the CHOOZ limit.

Figure 16

The best-fit point and 90% C.L. contour for the two parameters of the neutrino oscillations-with-decay model, the neutrino mass-lifetime ratio $\alpha$ and the mixing angle $\theta$.

Figure 17

Projections in $\Delta {\chi }^2$ for fit solutions for $\alpha$ and $\theta$ mixing angle for the oscillations-with-decay model. Parameter ranges allowed at 68% and 90% confidence levels lie below the corresponding dashed horizontal lines.