Measurement of atmospheric neutrino mixing with improved IceCube DeepCore calibration and data processing

We describe a new data sample of IceCube DeepCore and report on the latest measurement of atmospheric neutrino oscillations obtained with data recorded between 2011–2019. The sample includes significant improvements in data calibration, detector simulation, and data processing, and the analysis benefits from a sophisticated treatment of systematic uncertainties, with significantly greater level of detail since our last study. By measuring the relative fluxes of neutrino flavors as a function of their reconstructed energies and arrival directions we constrain the atmospheric neutrino mixing parameters to be ${\sin }^2{\theta }_{23}=0.51\pm 0.05$ and $\mathrm{\Delta }{m}_{32}^2=2.41\pm 0.07\times {10}^{-3}{\mathrm{eV}}^2$, assuming a normal mass ordering. The errors include both statistical and systematic uncertainties. The resulting 40% reduction in the error of both parameters with respect to our previous result makes this the most precise measurement of oscillation parameters using atmospheric neutrinos. Our results are also compatible and complementary to those obtained using neutrino beams from accelerators, which are obtained at lower neutrino energies and are subject to different sources of uncertainties.

Figure 1

Muon neutrino survival probability as function of energy and $\cos ({\theta }_{\text{zenith}})$, the latter of which is a proxy for the distance traveled, $L$. The oscillation probability is calculated within a three-neutrino framework where oscillation parameters used here are taken from a recent global fit of experimental data assuming the normal mass ordering [19]. We also account the Earth’s matter profile [20], which impacts the oscillations [21, 22, 23] most prominently at the core-mantle boundary as seen below 15 GeV at $\cos {\theta }_z\approx -0.8$.

Figure 2

Top and side projections of IceCube. DeepCore DOMs are depicted with red circles while IceCube DOMs are shown as green circles. String 36 is depicted as a black circle in the top projection for reference. The green region represents the DeepCore analysis region. The absorption length for Cherenkov light vs. depth is shown at the bottom left of the figure. An example corridor created by the hexagonal geometry is illustrated with a purple arrow (see Sec. 3d).

Figure 3

Distribution of the average charge observed over all triggered DOMs for events at Level 5 of the selection (see Sec. 3), compared to expectation from simulation of neutrinos, muons and detector noise. Data are shown for the same time period in 2014, before (Pass 1) and after the SPE calibration described in the text (Pass 2).

Figure 4

Relative optical efficiency of IceCube DOMs as a function of the photon incident angle $\eta$ ($\cos \eta =1$ means facing the PMT). (Left) Efficiency as measured in the lab before deployment (dashed line) and various calibration curves (solid gray lines) are shown. (Right) Two-parameter model used in this analysis with its baseline curve and example variations of the two parameters, $p_0$ and $p_1$ within their allowed range (N.B. parameters in this figure are varied one at a time for visualization, while in the analysis both are varied simultaneously). See main text for details of the model. The wide allowed range at $\cos \eta =1$ is due to the lack of controlled sources that illuminate the DOM directly face-on. The best fit from the analysis discussed in Sec. 5 is shown as the bold, red line in both panels.

Figure 5

Distribution of the time difference between the last and first pulse observed in an event, after the pulse series has been cleaned. This variable is used at L3 to reject events that contain more than one muon in a single readout window, coming from multiple air showers. The simulation does not contain these events, and therefore only describes the data up to values of $4\mathrm{\mu }\mathrm{s}$.

Figure 6

Distribution of the Level 4 noise classifier score, where larger values close to 1 indicate neutrinolike events and lower values reflect noiselike events.

Figure 7

Distribution of the Level 4 muon classifier score for events passing the cut on the L4 noise classifier score. Larger values close to 1 indicate neutrinolike events and lower values reflect muonlike events. The classifier was trained on detector data, which is about 99% atmospheric muons at this level, to avoid issues coming from limited statistics on the simulation.

Figure 8

Number of DOMs found in L5 corridors (top); cosine of an angle between the brightest corridor and SPEFit11 direction (bottom).

Figure 9

Summary of the rates obtained after each level of selection. Neutrinos are weighted to an atmospheric spectrum with oscillations included.

Figure 10

Cosine zenith resolutions for different classes of neutrino events at final level. The solid lines show the median resolutions and dashed lines indicate the central 68% containment region.

Figure 11

Energy resolutions for different classes of neutrino events at final level. The solid lines show the median resolutions and dashed lines indicate the central 68% containment region. All events are reconstructed using a track-plus-cascade hypothesis.

Figure 12

Output probability score distribution from the PID algorithm for the different interaction types. Distributions are normalized to better visualize shape differences. A dashed line at 0.55 indicates scores where tracklike event topologies begin to diverge from the cascadelike topologies.

Figure 13

Reduced ${\chi }^2$ from the SANTA track hypothesis fit.

Figure 14

Proxy for direction of travel calculated using the uppermost 15 layers of IceCube DOMs. Only events with at least 4 hits in those layers are included in the histograms.

Figure 15

Observed rates as a function of reconstructed energy for the data taking periods 2017–2018 (blue) and 2018–2019 (orange). The agreement for this observable between these years, as quantified by a KS-test p-value, is 1.1%. The black histogram shows the average rate across all 8 data taking periods for reference. See text for more details.

Figure 16

Observed rates as a function of the L4 muon classifier score for the data taking periods 2013–2014 (blue) and 2015–2016 (orange). The agreement for this observable between these years, as quantified by a KS-test p-value, is 1.2%. The black histogram shows the average rate across all 8 data taking periods for reference. See text for more details.

Figure 17

Observed number of data events in the analysis binning for the full 8 years of livetime.

Figure 18

Change in the expected number of events in the analysis when independently varying the values of $\mathrm{\Delta }{m}_{32}^2$ (top) and ${\sin }^2{\theta }_{23}$ (bottom) to the 90% confidence level. The mass splitting changes the position of the oscillation probability, while the mixing angle modifies the amplitude. The largest change is observed in the track channel.

Figure 19

Example of a hypersurface function in one bin projected on the dimension of the DOM efficiency correction. Each data point corresponds to one systematic set. Translucent datapoints are from sets where one or more systematic parameter besides the DOM efficiency correction is off-nominal. Those points are projected along the fitted plane to the nominal point. Several systematic sets have a nominal DOM efficiency of 1.0. The band shows to the standard deviation of the fitted function.

Figure 20

Difference between the fit hypersurface gradient for DOM optical efficiency in each analysis bin, comparing the gradient fit assuming $\mathrm{\Delta }{\mathrm{m}}_{32}^2=3.0\times {10}^{-3}{\mathrm{eV}}^2$ minus the gradient fit for $\mathrm{\Delta }{\mathrm{m}}_{32}^2=2.5\times {10}^{-3}{\mathrm{eV}}^2$.

Figure 21

Impact of the variations of the flux parameters used as a function of reconstructed energy (left figure) and reconstructed zenith angle (right figure). The parameters where changed by +$1\sigma$, one at a time. By construction, the variations are symmetric around the nominal expectation. The phase space that each parameter affects follows that given in [72].

Figure 22

Inclusive total neutrino-nucleon cross section for neutrinos (left) and antineutrinos (right) on an isoscalar target (black line) from GENIE, compared to measurements from CCFR [93], NUTEV [94] and NOMAD [95]. The GENIE cross section with its DIS fraction fully converted to “CSMS mode” (brown line) is also shown.

Figure 23

Observed $\mathrm{\Delta }{\chi }_{\text{mod}}^2$ (solid) compared to the expectation (dashed) and the distribution of 1000 pseudodata trials (yellow and green bands) produced at the best fit point of the analysis for the atmospheric (top) mixing angle and (bottom) mass splitting. The red lines indicate the 68% C.L. derived using the Feldman-Cousins method [103].

Figure 24

Distributions of reconstructed (top) energy, (middle) cosine zenith and (bottom) PID score for data compared to the best-fit simulation. Background includes atmospheric $\mu$ and all neutrino types besides ${\nu }_{\mu }+{\overline{\nu }}_{\mu }$ CC events for which this analysis was optimized.

Figure 25

The L/E distribution for the best-fit expectations overlaid with the observed data. Background includes atmospheric $\mu$ and all neutrino types besides ${\nu }_{\mu }+{\overline{\nu }}_{\mu }$ CC events. The expectation at the best fit but without oscillations is shown as a dashed line.

Figure 26

Contours showing the 90% C.L. allowed region for $\mathrm{\Delta }{m}_{32}^2$ and ${\sin }^2{\theta }_{23}$ from this study (blue) compared to previous IceCube DeepCore results [25, 26]. All confidence intervals shown are derived assuming Wilks’ theorem for a consistent comparison.

Figure 27

Contours showing the 90% C.L. allowed region for the atmospheric neutrino oscillation parameters from this study (blue) compared to results from MINOS [106], NOvA [108], Super-Kamiokande [109] and T2K [107]. Daya Bay also measures $\mathrm{\Delta }{m}_{32}^2$ in conjunction to ${\theta }_{13}$, but the results cannot be displayed in the format above [6]. The DeepCore confidence interval is derived assuming Wilks’ theorem.

Figure 28

DOM optical efficiency.

Figure 29

Hole ice p0.

Figure 30

Hole ice p1.

Figure 31

Ice absorption.

Figure 32

Ice scattering.

Figure 33

Atm. flux $\mathrm{\Delta }\gamma$.

Figure 34

Atm. flux A-F parameters.

Figure 35

Atm. flux G parameter.

Figure 36

Atm. flux H parameter.

Figure 37

Atm. flux W parameter ($K^{+}$).

Figure 38

Atm. flux W parameter ($K^{-}$).

Figure 39

Atm. flux Y parameter.

Figure 40

Axial mass CCQE.

Figure 41

Axial mass RES.

Figure 42

Deep inelastic scattering correction to CSMS.

Figure 43

NC normalization.

Figure 44

Atmospheric $\mu$ scale.

Figure 45

Agreement between years for the reconstructed quantities used in the fit. Top panels show the distribution for every year, compared with the average. Bottom panels display the Kolmogorov-Smirnov p-values calculated between each season of data.

Figure 46

Agreement between years for two of the variables used in the event selection (L4 muon classifier score and depth of the event vertex) and one low-level control variable that shows the typical charge per hit observed. For a description of the content of the panels, see Fig. 45.