Detection of Cascades induced by Atmospheric Neutrinos in the 79-string IceCube Detector

Detection of Cascades induced by Atmospheric Neutrinos in the 79-string IceCube Detector


Neutrino production and oscillation physics can be studied by utilizing the very high flux of atmospheric neutrinos observed with IceCube. In a Cherenkov medium such as ice, atmospheric muon neutrino interactions create tracks while cascades (showers) are produced by atmospheric electron neutrinos and by neutral current interactions of all flavors. We present the first detection of atmospheric neutrino-induced cascades at energies between 30 GeV and 10 TeV using the DeepCore array of the IceCube detector. Using 281 days of data, 1029 events are observed with 59% predicted to be cascades.

DeepCore, IceCube, neutrino, cascades
95.55.Vj, 14.60.Lm, 29.40.Ka, 95.85.Ry, 25.30.Pt


address=Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA ,altaddress=Dept. of Physics, University of California, Berkeley, CA 94720, USA

High energy neutrino telescopes, like the IceCube neutrino observatory are sensitive to all active flavors of neutrinos. The detector, using the deep Antarctic ice sheet as a Cherenkov medium, observes many atmospheric -induced muons through charged current (CC) interactions in the ice at energies as high as 400 TeV Abbasi et al. (2011a). However, searches for atmospheric  CC,  CC, and all-flavor neutral current (NC) interactions have so far only resulted in upper limits for the flux Abbasi et al. (2011b, c). The main signature for these interactions is a compact shower, producing a roughly spherical light distribution, called a “cascade.” We report the observation of atmospheric cascades in IceCube, using the first data with the DeepCore low-energy extension at energies between 30 GeV and 10 TeV. At these energies, the cascade detection channel is well suited for studying kaon production in air showers and detecting neutrino oscillation signatures such as .

DeepCore takes advantage of compact sensor spacing, high quantum efficiency photomultiplier tubes, deployment in the clearest ice, and a lower trigger threshold than the surrounding IceCube detector to observe neutrinos as low as 10 GeV Abbasi et al. (2012). In this analysis, we consider the data collected between May 31, 2010 and May 13, 2011. The IceCube detector consisted of 79 strings where the DeepCore subarray was defined as six densely instrumented strings optimized for low energies Abbasi et al. (2012) and the seven adjacent standard strings. This fiducial volume contains 454 digital optical modules (DOMs) deployed at depths below 2100 m. The raw data, after applying quality criteria (90% of data retained), is composed of light signals (“hits”) in the DOMs. The hits are collected in two modes - with and without a local coincidence (LC) requirement on signals in neighboring DOMs. Details of the LC logic can be found in Ref. Abbasi et al. (2009). With cuts designed to remove noise hits, the extra non-LC hits enhance reconstruction, background rejection, and particle identification especially for the lowest energy events.

A low threshold trigger, requiring minimum three LC hits within a time window of 2500 ns, is applied in the fiducial region. A special filter is run on the triggered event sample at the South Pole to reject cosmic-ray muons which penetrate the fiducial region, resulting in a passing rate of 17.5 Hz. A factor of 10 reduction in data compared to the triggered events is achieved by this algorithm while maintaining 99% of the atmospheric neutrinos that interact in the fiducial volume. The filtered data are sent north for further processing. The details of the trigger and filter algorithms are in Ref. Abbasi et al. (2012).

Atmospheric neutrinos are the decay products of charged pions, kaons, and muons created in cosmic-ray interactions with nucleons in the atmosphere. At high energies, above 100 GeV, atmospheric are produced mainly by semileptonic kaon decays like mode. The muons from meson decays with higher energies are more likely to interact before they decay, not producing in the atmosphere. Thus, production is highly suppressed compared to  Agrawal et al. (1996). The theoretical uncertainties associated with the flux are large due to the model uncertainties in the kaon production and the lack of measurements at these energies Barr et al. (2006); Sanuki et al. (2007).

The backgrounds for the atmospheric -induced cascades consist of cosmic-ray muons that mimic signal events and  CC events with low energy muons. The cascade analysis identifies the topology of the light pattern and enforces containment of the signal to reject those backgrounds. Veto techniques in the DeepCore fiducial volume remove more than six orders of magnitude of the background events while retaining reasonable efficiency for atmospheric -induced cascades Ha (2011). Following the veto step, the background rejection is performed in three stages, each with a Boosted Decision Tree (BDT) Hocker et al. (2007). The first stage is formed from five variables that characterize cascade signals in the detector and are built from the spherical hit pattern, localized time structure, and quick charge deposition. This selection reduces the data rate to 0.1 Hz, a factor of 1800 with respect to the trigger. The simulation predicts the atmospheric rate at this stage to be  Hz, corresponding to 63% efficiency with respect to the trigger. In the next stage, we carry out more sophisticated likelihood reconstructions on the reduced dataset, utilizing the scattering and absorption of Cherenkov photons in the ice Ahrens et al. (2004). From these reconstruction results, a seven variable BDT is formed and its selection reduces the atmospheric muon background to  Hz, rejecting an additional factor of 200 ( cumulatively) and retaining 40% of the signal ( Hz) compared to the first stage. At the final stage, tight cuts are made on the previous sample which contains a large fraction of atmospheric neutrinos. The cuts aim for high purity cascade detection by rejecting  CC events where possible. The background rejection techniques are discussed in more detail in Refs. Ha (2011, 2012).

We observe in total 1029 events in 281 days of data and expect 651 (550) cascades and 455 (415) tracks from simulations using the Bartol (Honda) atmospheric neutrino model Barr et al. (2004); Honda et al. (2007). The observation is consistent with either model, as shown in Fig. 1. Approximately half of the cascades are predicted to be events and the other half  NC events. The residual  CC background events have short outgoing muons with simulations indicating a median track length of 80 m where the tracks are not easily detected. In the currently available 28 hours of atmospheric muon simulation, no events remain. The 90% upper limit on this prediction is 554 events in 281 days. The simulated statistics are being increased. Note that neutrino oscillations of have a small (3%) effect due to relatively high energy of the cascades (180 GeV). The rate predicted by the Honda model is lower than that from the Bartol model because the two models treat the normalization of the cosmic-ray spectrum and the kaon production in the atmosphere differently Agrawal et al. (1996); Sanuki et al. (2007). Systematic uncertainties are not included.

Figure 1: The event rate as a function of the number of hit DOMs (left). The total simulated rate is consistent with the measurement from 281 days of data. Errors are statistical only. The bar histogram (right) shows simulation predictions for different interactions with two atmospheric flux models (Bartol Barr et al. (2004) and Honda Honda et al. (2007)) and the observed data rate.

We have reported on the observation of atmospheric neutrino-induced cascade events in IceCube. The observations are consistent with models of atmospheric neutrinos between approximately 30 GeV and 10 TeV. Systematic errors arise from light sensitivity of DOMs, ice modeling, cosmic-ray composition, neutrino-nucleon cross section, and atmospheric neutrino flux model. The background contamination from atmospheric muons and systematic uncertaities are currently under evalution. We expect results using a similar veto technique, aiming at neutrino oscillation measurements and WIMP dark matter searches in near future.


  1. see the full author list at


  1. R. Abbasi, et al., Phys. Rev. D 83, 012001 (2011a).
  2. R. Abbasi, et al., in Proc. of the 32nd ICRC, Beijing, China (2011b), arXiv:1111.2736.
  3. R. Abbasi, et al., Phys. Rev. D 84, 072001 (2011c).
  4. R. Abbasi, et al., Astropart. Phys. 35, 615 (2012).
  5. R. Abbasi, et al., Nucl. Instrum. Meth. A601, 294 (2009).
  6. V. Agrawal, T. K. Gaisser, P. Lipari, and T. Stanev, Phys. Rev. D 53, 1314 (1996).
  7. G. D. Barr, S. Robbins, T. K. Gaisser, and T. Stanev, Phys. Rev. D 74, 094009 (2006).
  8. T. Sanuki, M. Honda, T. Kajita, K. Kasahara, and S. Midorikawa, Phys. Rev. D 75, 043005 (2007).
  9. C. Ha, Ph.D. dissertation, Penn State University (2011).
  10. A. Hocker, et al., PoS ACAT, 040 (2007), arXiv:physics/0703039.
  11. J. Ahrens, et al., Nucl. Instrum. Meth. A524, 169 (2004).
  12. C. Ha, J. Phys.: Conf. Ser. 375, 052034 (2012), arXiv:1201.0801.
  13. G. D. Barr, T. K. Gaisser, P. Lipari, S. Robbins, and T. Stanev, Phys. Rev. D 70, 023006 (2004).
  14. M. Honda, T. Kajita, K. Kasahara, S. Midorikawa, and T. Sanuki, Phys. Rev. D 75, 043006 (2007).
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minumum 40 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description