Observation of Seven Astrophysical Tau Neutrino Candidates with IceCube

We report on a measurement of astrophysical tau neutrinos with 9.7 yr of IceCube data. Using convolutional neural networks trained on images derived from simulated events, seven candidate ${\nu }_{\tau }$ events were found with visible energies ranging from roughly 20 TeV to 1 PeV and a median expected parent ${\nu }_{\tau }$ energy of about 200 TeV. Considering backgrounds from astrophysical and atmospheric neutrinos, and muons from ${\pi }^{\pm }/K^{\pm }$ decays in atmospheric air showers, we obtain a total estimated background of about 0.5 events, dominated by non-${\nu }_{\tau }$ astrophysical neutrinos. Thus, we rule out the absence of astrophysical ${\nu }_{\tau }$ at the $5\sigma$ level. The measured astrophysical ${\nu }_{\tau }$ flux is consistent with expectations based on previously published IceCube astrophysical neutrino flux measurements and neutrino oscillations.

Figure 1

Top: Simulated rate of ${\nu }_{\tau }^{\text{astro}}$ CC events binned by the number of p. e. detected by DOMs on the most illuminated string in the event, $Q_{\mathrm{str}}^{\max }$, before any selection criteria (solid) and after the CNN-based criteria (dashed) described in the text. (Downward-going cosmic-ray muons trigger the detector at about 3 kHz, are effectively removed by our selection criteria, and are not shown on the plot; other backgrounds are similarly heavily reduced and also not shown.) Bottom: Ratio of rates (selected or all), showing that signal efficiency grows above about 2000 p.e. The IceCube “GlobalFit” ${\nu }^{\text{astro}}$ flux [53] is assumed. (Error bars are statistical only.).

Figure 2

Candidate ${\nu }_{\tau }^{\text{astro}}$ detected in September 2015. The left plot shows the DOM number (proportional to depth) versus the time of the digitized PMT signal in 3.3 ns bins for the highest-charge string, with the scale giving the signal amplitude in p.e. in each time bin. The total p.e. detected on the string, $Q_{\mathrm{str}}$., is shown. The right plot shows $S(C_1)$, that string’s saliency map for $C_1$, with darker regions indicating where the $C_1$ score is more sensitive to a changing light level (see text). (The Supplemental Material [59] shows three-string views and signed saliencies for all seven ${\nu }_{\tau }^{\text{astro}}$ candidates).

Figure 3

Top: Expected rate vs energy of ${\nu }_{\tau }^{\text{astro}}$ CC events, astrophysical and atmospheric neutrino backgrounds with initial selection criteria applied (dashed) and with final selection criteria then also applied (solid); for ${\nu }^{\text{astro}}$ the IceCube GlobalFit flux [53] was assumed. Bottom: Ratio of ${\nu }_{\tau }^{\text{astro}}$ CC rates after final and initial selection criteria. (Statistical error bars are too small to be visible. Although not shown in the plot, the backgrounds were simulated up to $E_{\nu }=100\mathrm{PeV}$).

Figure 4

Histogram of the $C_3$ vs $C_1$ CNN scores with all selection criteria applied. The color in each bin gives the expected number of signal (left) and background (right) events in that bin, assuming IceCube’s GlobalFit flux [53]. The approximate ($C_1$,$C_3$) values of the seven observed candidate ${\nu }_{\tau }^{\text{astro}}$ are shown by white circles, with the number inside each circle indicating the number of candidate events there.

Figure 5

Reconstructed visible energies (top) and $\cos {\theta }_{\mathrm{zen}}$ (bottom) for simulated ${\nu }_{\tau }^{\mathrm{CC}}$ (solid histogram) and seven candidate events (vertical lines) for the flux in Ref. [53]. The upward-going event with $\cos {\theta }_{\mathrm{zen}}\simeq -0.6$ had a reconstructed energy of $\sim 90\mathrm{TeV}$.