Measurements using the inelasticity distribution of multi-TeV neutrino interactions inIceCube

M. G. Aartsen,16 M. Ackermann,52 J. Adams,16 J. A. Aguilar,12 M. Ahlers,20 M. Ahrens,44 I. Al Samarai,25

D. Altmann,24 K. Andeen,34 T. Anderson,49 I. Ansseau,12 G. Anton,24 C. Arguelles,14 J. Auffenberg,1 S. Axani,14

P. Backes,1 H. Bagherpour,16 X. Bai,41 A. Barbano,25 J. P. Barron,23 S. W. Barwick,27 V. Baum,33 R. Bay,8

J. J. Beatty,18, 19 J. Becker Tjus,11 K.-H. Becker,51 S. BenZvi,43 D. Berley,17 E. Bernardini,52 D. Z. Besson,28

G. Binder,9, 8 D. Bindig,51 E. Blaufuss,17 S. Blot,52 C. Bohm,44 M. Borner,21 F. Bos,11 S. Boser,33 O. Botner,50

E. Bourbeau,20 J. Bourbeau,32 F. Bradascio,52 J. Braun,32 M. Brenzke,1 H.-P. Bretz,52 S. Bron,25

J. Brostean-Kaiser,52 A. Burgman,50 R. S. Busse,32 T. Carver,25 E. Cheung,17 D. Chirkin,32 K. Clark,29

L. Classen,36 G. H. Collin,14 J. M. Conrad,14 P. Coppin,13 P. Correa,13 D. F. Cowen,49, 48 R. Cross,43 P. Dave,6

M. Day,32 J. P. A. M. de Andre,22 C. De Clercq,13 J. J. DeLaunay,49 H. Dembinski,37 K. Deoskar,44 S. De Ridder,26

P. Desiati,32 K. D. de Vries,13 G. de Wasseige,13 M. de With,10 T. DeYoung,22 J. C. Dıaz-Velez,32

V. di Lorenzo,33 H. Dujmovic,46 J. P. Dumm,44 M. Dunkman,49 E. Dvorak,41 B. Eberhardt,33 T. Ehrhardt,33

B. Eichmann,11 P. Eller,49 P. A. Evenson,37 S. Fahey,32 A. R. Fazely,7 J. Felde,17 K. Filimonov,8 C. Finley,44

A. Franckowiak,52 E. Friedman,17 A. Fritz,33 T. K. Gaisser,37 J. Gallagher,31 E. Ganster,1 S. Garrappa,52

L. Gerhardt,9 K. Ghorbani,32 W. Giang,23 T. Glauch,35 T. Glusenkamp,24 A. Goldschmidt,9 J. G. Gonzalez,37

D. Grant,23 Z. Griffith,32 C. Haack,1 A. Hallgren,50 L. Halve,1 F. Halzen,32 K. Hanson,32 D. Hebecker,10

D. Heereman,12 K. Helbing,51 R. Hellauer,17 S. Hickford,51 J. Hignight,22 G. C. Hill,2 K. D. Hoffman,17

R. Hoffmann,51 T. Hoinka,21 B. Hokanson-Fasig,32 K. Hoshina,32, ∗ F. Huang,49 M. Huber,35 K. Hultqvist,44

M. Hunnefeld,21 R. Hussain,32 S. In,46 N. Iovine,12 A. Ishihara,15 E. Jacobi,52 G. S. Japaridze,5 M. Jeong,46

K. Jero,32 B. J. P. Jones,4 P. Kalaczynski,1 W. Kang,46 A. Kappes,36 D. Kappesser,33 T. Karg,52 A. Karle,32

U. Katz,24 M. Kauer,32 A. Keivani,49 J. L. Kelley,32 A. Kheirandish,32 J. Kim,46 T. Kintscher,52 J. Kiryluk,45

T. Kittler,24 S. R. Klein,9, 8 R. Koirala,37 H. Kolanoski,10 L. Kopke,33 C. Kopper,23 S. Kopper,47 J. P. Koschinsky,1

D. J. Koskinen,20 M. Kowalski,10, 52 K. Krings,35 M. Kroll,11 G. Kruckl,33 S. Kunwar,52 N. Kurahashi,40

A. Kyriacou,2 M. Labare,26 J. L. Lanfranchi,49 M. J. Larson,20 F. Lauber,51 K. Leonard,32 M. Leuermann,1

Q. R. Liu,32 E. Lohfink,33 C. J. Lozano Mariscal,36 L. Lu,15 J. Lunemann,13 W. Luszczak,32 J. Madsen,42

G. Maggi,13 K. B. M. Mahn,22 Y. Makino,15 S. Mancina,32 I. C. Maris,12 R. Maruyama,38 K. Mase,15 R. Maunu,17

K. Meagher,12 M. Medici,20 M. Meier,21 T. Menne,21 G. Merino,32 T. Meures,12 S. Miarecki,9, 8 J. Micallef,22

G. Momente,33 T. Montaruli,25 R. W. Moore,23 M. Moulai,14 R. Nagai,15 R. Nahnhauer,52 P. Nakarmi,47

U. Naumann,51 G. Neer,22 H. Niederhausen,45 S. C. Nowicki,23 D. R. Nygren,9 A. Obertacke Pollmann,51

A. Olivas,17 A. O’Murchadha,12 E. O’Sullivan,44 T. Palczewski,9, 8 H. Pandya,37 D. V. Pankova,49 P. Peiffer,33

J. A. Pepper,47 C. Perez de los Heros,50 D. Pieloth,21 E. Pinat,12 A. Pizzuto,32 M. Plum,34 P. B. Price,8

G. T. Przybylski,9 C. Raab,12 M. Rameez,20 L. Rauch,52 K. Rawlins,3 I. C. Rea,35 R. Reimann,1 B. Relethford,40

G. Renzi,12 E. Resconi,35 W. Rhode,21 M. Richman,40 S. Robertson,9 M. Rongen,1 C. Rott,46 T. Ruhe,21

D. Ryckbosch,26 D. Rysewyk,22 I. Safa,32 S. E. Sanchez Herrera,23 A. Sandrock,21 J. Sandroos,33 M. Santander,47

S. Sarkar,20, 39 S. Sarkar,23 K. Satalecka,52 M. Schaufel,1 P. Schlunder,21 T. Schmidt,17 A. Schneider,32

J. Schneider,24 S. Schoneberg,11 L. Schumacher,1 S. Sclafani,40 D. Seckel,37 S. Seunarine,42 J. Soedingrekso,21

D. Soldin,37 M. Song,17 G. M. Spiczak,42 C. Spiering,52 J. Stachurska,52 M. Stamatikos,18 T. Stanev,37 A. Stasik,52

R. Stein,52 J. Stettner,1 A. Steuer,33 T. Stezelberger,9 R. G. Stokstad,9 A. Stoßl,15 N. L. Strotjohann,52

T. Stuttard,20 G. W. Sullivan,17 M. Sutherland,18 I. Taboada,6 F. Tenholt,11 S. Ter-Antonyan,7 A. Terliuk,52

S. Tilav,37 P. A. Toale,47 M. N. Tobin,32 C. Tonnis,46 S. Toscano,13 D. Tosi,32 M. Tselengidou,24 C. F. Tung,6

A. Turcati,35 C. F. Turley,49 B. Ty,32 E. Unger,50 M. A. Unland Elorrieta,36 M. Usner,52 J. Vandenbroucke,32

W. Van Driessche,26 D. van Eijk,32 N. van Eijndhoven,13 S. Vanheule,26 J. van Santen,52 M. Vraeghe,26 C. Walck,44

A. Wallace,2 M. Wallraff,1 F. D. Wandler,23 N. Wandkowsky,32 T. B. Watson,4 A. Waza,1 C. Weaver,23

M. J. Weiss,49 C. Wendt,32 J. Werthebach,32 S. Westerhoff,32 B. J. Whelan,2 N. Whitehorn,30 K. Wiebe,33

C. H. Wiebusch,1 L. Wille,32 D. R. Williams,47 L. Wills,40 M. Wolf,35 J. Wood,32 T. R. Wood,23 E. Woolsey,23

K. Woschnagg,8 G. Wrede,24 D. L. Xu,32 X. W. Xu,7 Y. Xu,45 J. P. Yanez,23 G. Yodh,27 S. Yoshida,15 and T. Yuan32

(IceCube Collaboration)†

1III. Physikalisches Institut, RWTH Aachen University, D-52056 Aachen, Germany2Department of Physics, University of Adelaide, Adelaide, 5005, Australia

3Dept. of Physics and Astronomy, University of Alaska Anchorage,3211 Providence Dr., Anchorage, AK 99508, USA

4Dept. of Physics, University of Texas at Arlington, 502 Yates St.,

arX

iv:1

808.

0762

9v2

[he

p-ex

] 2

4 Fe

b 20

19

2

Science Hall Rm 108, Box 19059, Arlington, TX 76019, USA5CTSPS, Clark-Atlanta University, Atlanta, GA 30314, USA6School of Physics and Center for Relativistic Astrophysics,Georgia Institute of Technology, Atlanta, GA 30332, USA

7Dept. of Physics, Southern University, Baton Rouge, LA 70813, USA8Dept. of Physics, University of California, Berkeley, CA 94720, USA9Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA

10Institut fur Physik, Humboldt-Universitat zu Berlin, D-12489 Berlin, Germany11Fakultat fur Physik & Astronomie, Ruhr-Universitat Bochum, D-44780 Bochum, Germany

12Universite Libre de Bruxelles, Science Faculty CP230, B-1050 Brussels, Belgium13Vrije Universiteit Brussel (VUB), Dienst ELEM, B-1050 Brussels, Belgium

14Dept. of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA15Dept. of Physics and Institute for Global Prominent Research, Chiba University, Chiba 263-8522, Japan

16Dept. of Physics and Astronomy, University of Canterbury, Private Bag 4800, Christchurch, New Zealand17Dept. of Physics, University of Maryland, College Park, MD 20742, USA18Dept. of Physics and Center for Cosmology and Astro-Particle Physics,

Ohio State University, Columbus, OH 43210, USA19Dept. of Astronomy, Ohio State University, Columbus, OH 43210, USA

20Niels Bohr Institute, University of Copenhagen, DK-2100 Copenhagen, Denmark21Dept. of Physics, TU Dortmund University, D-44221 Dortmund, Germany

22Dept. of Physics and Astronomy, Michigan State University, East Lansing, MI 48824, USA23Dept. of Physics, University of Alberta, Edmonton, Alberta, Canada T6G 2E1

24Erlangen Centre for Astroparticle Physics, Friedrich-Alexander-Universitat Erlangen-Nurnberg, D-91058 Erlangen, Germany25Departement de physique nucleaire et corpusculaire,Universite de Geneve, CH-1211 Geneve, Switzerland

26Dept. of Physics and Astronomy, University of Gent, B-9000 Gent, Belgium27Dept. of Physics and Astronomy, University of California, Irvine, CA 92697, USA28Dept. of Physics and Astronomy, University of Kansas, Lawrence, KS 66045, USA

29SNOLAB, 1039 Regional Road 24, Creighton Mine 9, Lively, ON, Canada P3Y 1N230Department of Physics and Astronomy, UCLA, Los Angeles, CA 90095, USA

31Dept. of Astronomy, University of Wisconsin, Madison, WI 53706, USA32Dept. of Physics and Wisconsin IceCube Particle Astrophysics Center,

University of Wisconsin, Madison, WI 53706, USA33Institute of Physics, University of Mainz, Staudinger Weg 7, D-55099 Mainz, Germany

34Department of Physics, Marquette University, Milwaukee, WI, 53201, USA35Physik-department, Technische Universitat Munchen, D-85748 Garching, Germany

36Institut fur Kernphysik, Westfalische Wilhelms-Universitat Munster, D-48149 Munster, Germany37Bartol Research Institute and Dept. of Physics and Astronomy,

University of Delaware, Newark, DE 19716, USA38Dept. of Physics, Yale University, New Haven, CT 06520, USA

39Dept. of Physics, University of Oxford, 1 Keble Road, Oxford OX1 3NP, UK40Dept. of Physics, Drexel University, 3141 Chestnut Street, Philadelphia, PA 19104, USA

41Physics Department, South Dakota School of Mines and Technology, Rapid City, SD 57701, USA42Dept. of Physics, University of Wisconsin, River Falls, WI 54022, USA

43Dept. of Physics and Astronomy, University of Rochester, Rochester, NY 14627, USA44Oskar Klein Centre and Dept. of Physics, Stockholm University, SE-10691 Stockholm, Sweden45Dept. of Physics and Astronomy, Stony Brook University, Stony Brook, NY 11794-3800, USA

46Dept. of Physics, Sungkyunkwan University, Suwon 440-746, Korea47Dept. of Physics and Astronomy, University of Alabama, Tuscaloosa, AL 35487, USA

48Dept. of Astronomy and Astrophysics, Pennsylvania State University, University Park, PA 16802, USA49Dept. of Physics, Pennsylvania State University, University Park, PA 16802, USA

50Dept. of Physics and Astronomy, Uppsala University, Box 516, S-75120 Uppsala, Sweden51Dept. of Physics, University of Wuppertal, D-42119 Wuppertal, Germany

52DESY, D-15738 Zeuthen, Germany(Dated: February 26, 2019)

Inelasticity–the fraction of a neutrino’s energy transferred to hadrons–is a quantity of interest inthe study of astrophysical and atmospheric neutrino interactions at multi-TeV energies with IceCube.In this work, a sample of contained neutrino interactions in IceCube is obtained from 5 years of dataand classified as 2650 tracks and 965 cascades. Tracks arise predominantly from charged-currentνµ interactions, and we demonstrate that we can reconstruct their energy and inelasticity. Theinelasticity distribution is found to be consistent with the calculation of Cooper-Sarkar et al. acrossthe energy range from ∼ 1 TeV to ∼ 100 TeV. Along with cascades from neutrinos of all flavors,we also perform a fit over the energy, zenith angle, and inelasticity distribution to characterize

3

the flux of astrophysical and atmospheric neutrinos. The energy spectrum of diffuse astrophysicalneutrinos is well-described by a power-law in both track and cascade samples, and a best-fit indexγ = 2.62 ± 0.07 is found in the energy range from 3.5 TeV to 2.6 PeV. Limits are set on theastrophysical flavor composition that are compatible with a ratio of

(13

: 13

: 13

)⊕. Exploiting the

distinct inelasticity distribution of νµ and νµ interactions, the atmospheric νµ to νµ flux ratio in theenergy range from 770 GeV to 21 TeV is found to be 0.77+0.44

−0.25 times the calculation by Honda et al.Lastly, the inelasticity distribution is also sensitive to neutrino charged-current charm production.The data are consistent with a leading-order calculation, with zero charm production excluded at91% confidence level. Future analyses of inelasticity distributions may probe new physics that affectsneutrino interactions both in and beyond the Standard Model.

I. INTRODUCTION

The observation of astrophysical neutrinos [1, 2] wasa landmark in high-energy astrophysics. It introduced anew probe that is directionally sensitive to high-energyhadronic particle accelerators in the universe. Neutrinosprovide both good directional information, unaffected bymagnetic fields, and extremely long range, allowing us toprobe accelerators at cosmologically significant distances.Measurements of the flux and energy spectrum [3, 4] andflavor composition [5, 6] are so far fully compatible withconventional acceleration models, but more exotic pro-duction mechanisms cannot be ruled out.

At the same time, the observation of high-energy as-trophysical and atmospheric neutrinos by detectors likeIceCube has opened up the study of neutrino interactionsat energies orders of magnitude above those accessible atterrestrial accelerators. Already, the 1 km3 IceCube neu-trino observatory has used atmospheric and astrophysicalneutrinos to measure neutrino absorption in the Earth,and from that determined the neutrino-nucleon cross sec-tion at energies from 6.3 TeV to 980 TeV to be in agreee-ment with the Standard Model prediction [7].

In this paper, we report on a new study of high-energycharged-current (CC) νµ interactions contained withinIceCube’s instrumented volume. These interactions pro-duce a cascade of hadrons and a muon, an event topologyknown as a starting track. By estimating the hadroniccascade and muon energies separately, we can estimatethe inelasticity of each interaction – the ratio of hadroniccascade energy to total neutrino energy [8]. The central90% of neutrinos have estimated energies in the rangefrom 1.1 TeV to 38 TeV, energies far beyond the reachof terrestrial accelerators. For example, the NuTeV datawere used to measure inelasticity distributions at ener-gies up to 250 GeV [9], while earlier experiments werelimited to lower energies [10].

The starting track data, together with a similarly ob-tained set of cascades due to all neutrino flavors arebinned by reconstructed energy and zenith angle and (forthe tracks) inelasticity. This data is fitted to a neu-trino flux model containing both atmospheric and as-

∗ Earthquake Research Institute, University of Tokyo, Bunkyo,Tokyo 113-0032, Japan† Email: analysis@icecube.wisc.edu

trophysical neutrinos. From this, we present measure-ments of neutrino inelasticity and estimate the fractionof neutrino interactions that produce charmed particles.We also compare the track and cascade samples, studywhether they have the same astrophysical neutrino fluxspectral indices, and constrain the flavor composition ofastrophysical neutrinos. Finally, we measure the ratio ofneutrinos to antineutrinos in the atmospheric neutrinoflux.

II. ν INTERACTIONS AND INELASTICITY

The inelasticity distribution of neutrinos is expected tobe well described by the Standard Model for weak inter-actions. At TeV energies, the interactions are dominatedby Deep Inelastic Scattering (DIS). Neutrinos interactwith quarks in nuclear targets (hydrogen and oxygen)via charged-current (CC) and neutral-current (NC) re-actions. In NC interactions, the neutrino interacts viaZ0 exchange, leaving the quark flavor unchanged. InCC ν interactions via W± exchange, the quark chargechanges by one, turning a charge +2/3 quark into acharge −1/3 one, while for ν, the reverse reactions occur,but with a different inelasticity distribution. W± also in-teract with sea quarks and antiquarks in the nucleus insimilar charge-changing reactions, so the differences be-tween neutrinos and antineutrinos largely disappear atvery high energies. The relative probability for produc-ing a given final state quark depends on the Cabibbo-Kobayashi-Maskawa (CKM) matrix elements, so reac-tions involving u→ d and s→ c predominate.

The neutrino-nucleon cross section involves four mainkinematic variables: s is the neutrino-nucleon centerof mass energy squared, Q2 is the negative of the 4-momentum transfer squared, Bjorken−x, the fraction ofthe struck nucleon’s momentum carried by the struckquark, and the inelasticity, y. These quantities are re-lated through x = Q2/2MEνy, where M is the nucleonmass and Eν is the incoming neutrino energy. Most ofthe interactions discussed in this paper involve struckpartons with 10−2 < x < 10−1, and the typical Q2 is∼M2

W,Z ≈ 6× 103 GeV2.

The double-differential cross section for a neutrino

4

with energy Eν is

dσν,ν

dxdy=G2FMEν2π

(M2V

M2V +Q2

)2 [(1+(1−y)2)F2(x,Q2)

− y2FL(x,Q2)± (1− (1− y)2)F3(x,Q2)]

(1)

where the + sign is for neutrinos, and the− is for antineu-trinos. MV is the vector boson mass (MW for chargedcurrent interactions and MZ for neutral current interac-tions) and GF is the Fermi coupling constant [11]. Thecross sections depend on three nucleon structure func-tions: F2, F3, and FL.

Assuming standard vector minus axial vector (V −A)coupling and an isoscalar target, at leading order thesestructure functions are related to the quark parton distri-bution functions (PDFs) of the target nucleon, qi(x,Q

2),according to [11]

F2(x,Q2) = 2xΣi(qi(x,Q2) + qi(x,Q

2)), (2)

F3(x,Q2) = 2xΣi(qi(x,Q2)− qi(x,Q2)), (3)

FL(x,Q2) = 0. (4)

The longitudinal structure function FL becomes non-zeroin higher order calculations. It should be noted thatmeasurements of the nuclear structure function F2(x,Q2)inferred from neutrino-DIS on an iron target may beslightly different from those observed in charged leptonDIS at low Q2 [12].

This paper uses as a baseline a next-to-leading ordercalculation (‘CSMS’) which uses the DGLAP formalismfor parton evolution [13]. The calculation uses the HER-APDF1.5 [14] parton distribution functions, with theMSTW2008 and CT10 [15] distributions used as a stan-dard for comparison. In the relevant energy range, thecalculation has an expected uncertainty of a few percentand is consistent with an independent calculation thatused the MSTW2008 PDFs [16].

Equation 1 does not account for threshold behaviorswhich are important for heavy quark production. Forthis analysis, charm production is most important. Inthe relevant energy range, charm production is about10% of the total production. Because of the quark mass,heavy quark production near its production threshold oc-curs at a larger average inelasticity than for light quarksat the same energy [17]. In the case of charm quarks,the inelasticity also tends to be higher even at energiesfar above the production threshold since they originateprimarily from scattering off sea s-quarks. In addition,heavy quarks may decay semileptonically, transferringsome of their energy to a muon. This muon will notbe separately visible in IceCube. Instead, it will increasethe apparent brightness of the primary muon, leading toa higher measured muon energy and lower measured in-elasticity. However, due to the small ∼ 10% branchingratio of muonic charm decays, IceCube is primarily sen-sitive to the signature of larger inelasticity rather thanthe di-muon signal.

The CSMS calculation is for neutrino DIS on a sin-gle nucleon target; there are a few other contributionsto consider. Water contains hydrogen, so is not a per-fectly isoscalar target. The MINERvA collaboration hasseen that nuclear shadowing reduces the cross section forneutrino-heavy ion interactions with x < 0.1 [19]. Thesuppression is expected to increase with decreasing x val-ues [20]. However, oxygen is a small nucleus, and weexpect the cross section reduction to be small for themoderate x values probed here.

An additional contribution to the cross section is due toneutrino electromagnetic (diffractive) interactions withthe Coulomb field of the nuclei, but this is small for low-Z nuclei like hydrogen and oxygen [21, 22]. These effectsare not expected to be significant for this analysis.

III. NEUTRINO SOURCES AND SIMULATION

This analysis uses atmospheric and astrophysical neu-trinos as sources. The signals observed in IceCube de-pend on the neutrino fluxes incident on the Earth, theirabsorption in the Earth, their interactions in and (forbackgrounds) around IceCube, their propagation throughthe detector and the detector response. We will brieflydiscuss these factors, with a special emphasis on the in-teractions, which determine the inelasticity.

Since most of the results in this paper are based oncomparisons of data with various simulations, this sectionwill focus on the physics models used in the simulations.These models are implicit in the results presented, andthe systematic errors depend on the assumptions usedin the simulations; these uncertainties will be discussedwhen the individual results are presented.

A. Neutrino Sources

Conventional atmospheric neutrinos come from pionsand kaons that are produced in cosmic-ray air showers.At the energies relevant for this analysis, the flux roughlyfollows a power-law spectrum dN/dEν ∝ E−(γ+1), whereγ is the spectral index for the cosmic-ray energy spec-trum. Below the cosmic-ray knee, γ ≈ 2.7, while athigher energies γ ≈ 3.0. The flux is highest for near-horizontal incidence. This analysis uses the HKKMS [23]flux calculations extrapolated upward in energy, with amodification to account for the knee of the cosmic-rayspectrum [24]. This calculation is in good agreement withprevious IceCube measurements [25–27].

Prompt atmospheric neutrinos come from the decayof charmed mesons produced in cosmic-ray air showers.They have yet to be observed but are expected to have ahard spectrum that follows that of the primary cosmic-rays. This analysis uses as a baseline the BERSS [28]perturbative QCD calculation of the prompt flux, whichis tied to recent data from RHIC and the LHC, and con-sistent with similar independent calculations [29, 30].

5

The number of observed prompt and atmospheric neu-trinos requires an important adjustment to account forthe IceCube ‘self-veto’ - a downward-going atmosphericneutrino will be accompanied by an air shower and muonbundle, which may overshadow the neutrino and causethe event to fail the event containment cuts, and so notregister as a starting event. This probability for a self-veto depends on whether the neutrino is prompt or con-ventional and on the neutrino energy and zenith angle.This analysis uses the probabilities calculated in Ref. [31].The muon threshold is taken to be 100 GeV; this is theminimum muon energy which is likely to trigger the self-veto. The appropriateness of the 100 GeV threshold wasverified using detector simulations using the CORSIKAprogram [32].

Previous IceCube measurements have found that, inthis energy range, the astrophysical flux is consistent witha single power law. Our fit is based on this single powerlaw.

The neutrinos observed in IceCube may pass throughthe Earth before reaching the detector; the absorptionin the Earth is simulated following the Standard Modelcross sections [13]. The Earth’s density profile is as-sumed to follow the Preliminary Reference Earth Model(PREM) [33].

B. Neutrino Interactions and Cherenkov lightemission

Neutrino interactions in and around the detector aremodeled following the CSMS calculation, as described inthe previous section. For NC interactions, the inelasticityis the fraction of the neutrino energy that is transferredto the struck nucleus; the remaining energy escapes fromthe detector. For CC interactions, the remainder of theenergy is transferred to a charged lepton. IceCube ob-serves the Cherenkov light emitted by the lepton andits secondary relativistic charged particles. The hadroniccascade from the struck nucleus also produces Cherenkovlight. Each type of lepton produces light very differentlyin the detector.

Electrons produce electromagnetic cascades; the lightoutput is proportional to the electron energy, with therelationship determined from detailed simulations [34].At low energies, they are treated as point sources, whileat energies above 1 TeV, the longitudinal profile of acascade is approximated using a sequence of uniformlyspaced point sources. At higher energies above 1 PeV,the longitudinal profile includes the LPM effect [35].

As they traverse the detector, muons radiate energy viaionization, bremsstrahlung, direct pair production andphotonuclear interactions; these are modeled followingRefs. [36, 37].

Tau leptons are propagated through the detector ina manner similar to muons, with adjustments for theirhigher mass. IceCube simulations allow the taus to decayinto ντeνe, ντµνµ or ντ plus hadrons. For the leptonic

decays, the leptons are propagated through the detectorstarting at the τ decay point, while the neutrinos escape.For the hadronic decays, the hadronic energy is summed,producing another hadronic cascade, at the point wherethe τ decays. Because of the energy carried off by theescaping neutrinos, ντ will deposit less energy in the de-tector and appear similar to lower energy νe and νµ. Inmuonic decays, the outgoing muon has a lower averageenergy than in corresponding νµ interactions. This canaffect the measured inelasticity distribution [38].

This analysis is sensitive to charm production inneutrino interactions. DIS interactions produce charmquarks, which then hadronize, forming charmed hadronsand baryons, with lifetimes of order 10−12 s. We usethe hadronization fractions into D+, D0, D+

s , and Λ+c

from the calculation in Ref. [39] at 100 GeV, which isalso used in the GENIE event generator [40]. If thecharmed hadrons have energies above about 10 TeV, theymay interact in the ice and lose energy before they de-cay. The interaction probability depends on the individ-ual hadron-ice cross sections and hadron lifetimes. Thisanalysis used parametrizations of the charm cross sec-tions in CORSIKA [32]. Since the parametrizations areonly valid above 1 PeV, we extrapolate downward in en-ergy using the kaon-nucleon and nucleon-nucleon crosssections scaled by 88% and 122%, respectively, to matchCORSIKA parametrization at 1 PeV for charm mesonsand baryons. This leads to critical energies for the D+,D0, D+

s and Λ+c of 22 TeV, 53 TeV, 47 TeV and 104

TeV, respectively. When the charmed hadrons interact,they lose energy. We use the approach of Ref. [18] toparametrize their energy loss distribution. The observ-able effect of multiple charm interactions is the produc-tion of a low energy muon after a semileptonic decay,mimicking a high inelasticity track for νe or ντ interac-tions. However, due to the 10% muonic branching ratio,this is not a large effect, and the approximate treatmentof charm interactions here does not significantly influencelater results on charm production.

The conversion between hadronic cascade energy andlight follows Ref. [41]. Hadronic cascades produce lesslight than electromagnetic cascades of the same energy,with larger cascade-to-cascade variation. At 100 TeV, ahadronic cascade produces an average of 89% of the lightof an equivalent energy electromagnetic cascade. Thedifference drops with increasing energy. Since electro-magnetic and hadronic cascades cannot be readily dis-tinguished in IceCube, we will refer to the visible cas-cade energy as the energy of an electromagnetic cascadeproducing an equivalent amount of light as the hadroniccascade.

C. Optical transmission through Antarctic ice

The emitted Cherenkov light travels through the ice,where it may scatter or be absorbed before reaching anoptical sensor. Because the sensor array is sparse, only

6

a tiny fraction of the produced Cherenkov photons areobserved, and they are likely to scatter multiple timesbefore reaching a sensor. Because of this, the signalsseen by IceCube are sensitive to the optical properties ofthe ice. The scattering and absorption lengths in the icedepend on the position in the ice (largely, but not entirelyon the depth below the surface) and on the direction ofphoton propagation [42, 43]. These optical propertieshave been measured using a variety of means, includinglaser and light emitting diode (LED) signals and cosmic-ray muons. The optical properties of the ice vary stronglywith depth, with certain depth ranges containing “dustlayers” with very large absorption and scattering, andother depths providing good transmission. This posi-tional dependence is accounted for with an ice model.

This analysis used as a baseline the “SPICE Mie” icemodel [42]. It is based on measurements of the ice prop-erties using LEDs mounted in the detector housings, sup-plemented with parameterizations to give the wavelengthdependence of the scattering and absorption lengths. Itdivides the ice into 10 m thick layers and determinesthe scattering and absorption lengths separately for eachdepth range. It does not account for tilts of the dustlayers (i.e. variation of optical properties depending onhorizontal position), or for anisotropy in the ice.

Near the optical sensors, there is additional scatteringin the ‘hole ice’ – the refrozen column of melted watercreated by the drill during deployment. This ice con-tains a central column of bubbles [44]. In our baselinesimulations, it is treated as having a scattering length of50 cm.

IV. DETECTOR AND DATA

The Cherenkov light is detected with an array of5,160 digital optical modules (DOMs), spread over 1 km3

[45, 46]. The DOMs are deployed in 86 vertical strings,each containing 60 DOMs. Seventy-eight of the stringsare laid out on a 125 m triangular grid. On those strings,the DOMs are deployed with 17 m vertical spacing be-tween 1450 and 2450 m below the surface. The remain-ing strings are laid out in the middle of the array, withsmaller string-to-string spacing [47]. On those strings,the DOMs are deployed closer together at depths betweenabout 2,000 and 2450 m.

Each DOM collects data independently, sending digi-tized data to the surface [48, 49]. The optical sensor isa 10 inch photomultiplier tube (PMT) [50]. The PMTis read out with a data-acquisition system comprisingtwo waveform digitizer systems which are triggered by adiscriminator with a threshold of about 1/4 of a typicalphotoelectron pulse height. One records data with 14-bitdynamic range at 300 megasamples/s for 400 nsec, andthe other collects data with 10-bit dynamic range at 40megasamples/s for 6.4 µs.

All of the digitized data is sent to the surface where atrigger system monitors the incoming data and creates an

event when certain conditions are satisfied. For this anal-ysis, the main trigger required 8 hits within a sliding 5 µswindow. When a trigger occurs, all of the data within 4µs before or 6 µs after the trigger time is saved and sentto an on-line computer farm for further processing.

The farm applies a number of different selection algo-rithms to each event. Each algorithm tests for differentclasses of interesting events, albeit with significant over-lap. This analysis used as input all events that passedeither the cascade or muon track filters. These filtershave very loose cuts. Simulation studies show that theycapture more than 99.5% of the events that pass the othercuts that are applied here.

This analysis uses data collected between May 2011and May 2016, a total live time of 1734 days during whichthe detector was in its complete 86-string configuration.During the design of the analysis, 10% of the data wasused for testing and the remaining 90% was kept blind.

V. EVENT SELECTION

The analysis aims to select a high-purity sample of neu-trino interactions that occur within the detector. Start-ing tracks are of the greatest interest here, but cascadesare more numerous for an astrophysical flux with equalflavor ratio and provide important constraints for manyof the fits discussed here. For both channels, the firststep of the analysis is to select a clean sample of start-ing events. The initial cut selects events with more than100 observed photoelectrons (PE), which captures mostcontained neutrino interactions above 1 TeV.

The next selection applies an outer layer veto to rejectevents that come from charged particles that enter fromoutside the detector. The veto uses DOMs on the top,bottom and sides of the detector. It also includes DOMsin a horizontal layer that passes through the detector,below the “dust layer” that allows charged particles tosneak in undetected. The veto is similar to that in Ref. [5]but with some changes to reduce the energy threshold,as discussed in Ref. [51, 52].

First, the number of photoelectrons, Qstart, required ina rolling 3 µs time window to define the start of the event,is chosen to depend on the total number of photoelectrons(NPE) of the event, Qtot,

Qstart =

3 Qtot < 72 PE

Qtot/24 72 PE ≤ Qtot < 6000 PE

250 Qtot ≥ 6000 PE

. (5)

The horizontal veto layer that is placed just below thedust layer is made thicker at 120 m. The veto layer atthe bottom is expanded to include the lowest DOM oneach string. This leaves one remaining significant back-ground, from two nearly-coincident air showers. Some-times, a low-energy muon can sneak into the detector,producing only a small signal in the veto. Then, a higher

7

0.00 0.25 0.50 0.75 1.00Atmospheric µ BDT Score

2

3

4

5

log

10

NP

EAtmospheric µ

0.00 0.25 0.50 0.75 1.00Atmospheric µ BDT Score

2

3

4

5

log

10

NP

E

Cascade-like ν

0.00 0.25 0.50 0.75 1.00Atmospheric µ BDT Score

2

3

4

5

log

10N

PE

Track-like ν

0.00 0.25 0.50 0.75 1.00Atmospheric µ BDT Score

2

3

4

5lo

g10

NP

E

Data

10−2

10−1

100

101

102

103

104

105

106

Eve

nts

FIG. 1. Two-dimensional distributions of atmospheric muonBDT score and number of photoelectrons (NPE), after theouter-layer veto, for atmospheric muons (top left), cascade-like neutrinos (top right), track-like neutrinos (bottom left)and 10% of the data (bottom right). Events above the blacklines were accepted as signal events. To access ∼ 1 TeV neu-trinos that typically produce ∼ 100 PE, one must overcomemuon background with a rate 1,800 times higher.

energy muon can enter, producing a high enough num-ber of photoelectrons to satisfy the Qtot requirement. Toavoid this, an algorithm is used to count the number ofcausally disconnected clusters of light in the detector andreject events where more than one cluster is found. Thesecuts reduce the event rate to 0.36 Hz in the 10% testingsample, compared to an expected starting neutrino rateof 0.20 mHz.

A. BDT background rejection

The remaining background is harder to reject, particu-larly at lower muon energies where less light is produced.To further clean up the signal, a boosted decision tree(BDT) is used to classify the passing events as atmo-spheric muons, starting tracks or cascades. The BDTuses 15 variables as input, including log10(Qtot) and thenumber of photoelectrons in the outer layer veto. Theother variables are from the output of the track andcascade reconstructions. For the wrong hypothesis (i.e.tracks reconstructed as cascades), the output may varygreatly, providing considerable separating power.

The track-reconstruction variables are: cosine zenithof the reconstructed direction, the distance of the track’sfirst visible energy loss from the edge of the detector, thedistance between the first and last visible energy loss, thenumber of direct hits (photoelectrons with DOM arrivaltime consistent with zero scattering), the track lengthbetween the first and last DOMs that registered a directhit, the estimated angular uncertainty of the track recon-struction, and the angle between the track direction and

a simplified reconstruction where first arrival times arefit to a propagating plane wave.

For the cascade reconstruction, the variables were thedepth of the cascade within the detector, the horizon-tal distance between the cascade and the edge of thedetector, the reduced log likelihood of the fit, and thelog likelihood ratio of the track and cascade fits. Thereare also two variables that use multiple track fits to thesame event. The first considers 104 different in-comingdown-going track directions, ending at the reconstructedcascade, and counts the number of photoelectrons in thetime window from -15 ns to +1000 ns around the geomet-ric first arrival time from the tracks. The largest numberof photoelectrons among all the tracks is selected for usein the BDT, and this helps to discriminate against down-going muons that do not have a well-reconstructed track.The second considers 192 outgoing track hypotheses fromall directions and counts the number of photoelectrons inthe time window from -30 to +500 ns. This is importantto identify starting tracks with a low-energy outgoingmuon [51, 52].

The BDT was trained with a sample of 4.5 millionsimulated atmospheric muon events from CORSIKA [32]that pass the outer-layer veto using SIBYLL2.1 [53] tomodel hadronic interactions. The spectrum was weightedto the H3a cosmic-ray flux model [54]. The neutrino in-put was a total of 734,000 simulated neutrinos, weightedto the sum of the HKKMS conventional atmospheric flux[23], the BERSS prompt neutrino flux [28] and an astro-physical flux with a spectral index γ = 2.5, using theflux from Ref. [3]. Several quality criteria were applied tothe training sample: neutrinos labeled as starting trackswere required to have a vertex that was contained withinthe detector and produce a muon having a path lengthmore than 300 m within the detector and energy above100 GeV. Further, the reconstructed muon direction wasrequired to be within 5o of the simulated direction.

Figure 1 shows the number of events as a function ofBDT atmospheric muon score and number of photoelec-trons, for atmospheric muons, cascade-like and track-likeneutrinos, and data. There are an estimated 1,800 timesas many cosmic-ray muons as neutrinos in the sample,concentrated at low NPE and fairly high BDT muonscores. To optimally select neutrino events, an ellipti-cal cut was used,(

sµ − 1

a

)2

+

(log10(Qtot)− 2

b

)2

> 1, (6)

where sµ is the BDT atmospheric muon score and (a, b)are parameters describing the semi-major and semi-minor axes of the ellipse. Values of a = 0.75 and b = 2.5were chosen to eliminate muon background, and the re-sulting ellipse is shown in Fig. 1. The total event ratefor data, atmospheric muons, and neutrinos as a func-tion of a in Eq. 6 while keeping b fixed is shown in Fig. 2and shows good agreement between data and simulation.The chosen value of a = 0.75 reduces the data event rateto 0.024 mHz or 3615 events in the final data sample.

8

0.0 0.2 0.4 0.6 0.8 1.0a

10−9

10−7

10−5

10−3

10−1

Eve

nt

rate

in10

%d

ata

sam

ple

[Hz]

Atmospheric µ

Cascade-like ν

Track-like ν

Total ν

Total MC

Data

FIG. 2. The total event rate for data (black), atmosphericmuons (red), cascade-like neutrinos (green) and track-likeneutrinos (orange) in the 10% testing sample as a functionof the parameter a appearing in Eq. 6 describing the ellipticalcut. The chosen value of a = 0.75 is shown with a verticalblack line.

The CORSIKA Monte Carlo statistics are inadequateto estimate the number of surviving cosmic-ray muons inthe final sample; an extrapolation based on the distancefrom the cut line in Fig. 2 indicates that the number ofevents should be negligible. To check this, an additionalsimulation was made using a parametrization of the fluxof single muons in the deep ice calculated from the H3acosmic-ray model [52]. The simulation contains 26.8 mil-lion single muons passing the outer layer veto, but sinceit does not contain multi-muon bundles, it underpredictsthe total muon background rate at this level by a factorof 3. However, most events near the BDT cut thresholdare single muons, as can be verified by looking at COR-SIKA events, so the simulation can still be used to calcu-late a background estimate. The predicted backgroundof single muons from the simulation is 2.7± 1.0 events inthe final sample. Even conservatively assuming that anunexpectedly large contribution from bundles increasesthis by a factor of 3 as observed for the outer layer veto,the resulting muon background of 8.1 events is negligiblecompared to the full sample size of 3615 events and willno longer be considered.

B. Cascade/track classification

The same BDT was used to split the neutrino sampleinto cascades and tracks. The BDT track score, strack,and cascade score, scasc. are combined into a variable,strack = strack/(strack + scasc.), which runs from 0 to 1and is an estimate of an event’s “trackness,” the likeli-hood that it contains a track. The final selection betweentracks and cascades depends on the threshold chosen forstrack. Figure 3 shows the purity, efficiency and signal to

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.00.00.10.20.30.40.50.60.70.80.91.0

Fra

ctio

n

Track Purity

Track Efficiency

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0Normalized BDT Track Score

0

5

10

15

SN

R

FIG. 3. Top: The track purity and efficiency as a functionof normalized BDT score strack. Bottom: The signal-to-noiseratio as a function of strack. A BDT track score cut strack ≥0.52 (black line) maximized the signal-to-noise ratio, yieldinga 92% purity and 98% efficiency. The fluctuations in thepurity curve are due to low statistics in that region.

noise ratio (SNR) as a function of strack, as determinedusing simulated track and cascade samples. The SNRis defined as the ratio of the true track rate to the sizeof Poisson fluctuations in the total rate of true tracksand misidentified cascades. Optimizing SNR, the crite-rion strack ≥ 0.52 is used to identify tracks. Events notsatisfying this criterion are identified as cascades.

With this classification, there were 965 cascades and2650 track events in the final sample. The larger numberof tracks is largely due to the dominance of νµ over νe inthe atmospheric neutrino flux.

VI. DIRECTION AND ENERGYRECONSTRUCTIONS

IceCube has previously developed algorithms for recon-structing the direction and energy of cascades and trackevents. Events classified as cascades are reconstructedusing a maximum likelihood fit as in Ref. [5], using thefull photoelectron timing recorded by each DOM. Simu-lations predict that the median angle between the simu-lated and reconstructed direction is 16◦. However, start-ing tracks have received much less attention and call fornew approaches.

9

0 200 400 600 800 1000 1200100

102

104

En

ergy

Los

s(G

eV/1

0m

) Reconstructed: Ecasc. = 64 TeV, Etrack = 724 TeV

Detector Entry

Detector Exit

Excluded

0 200 400 600 800 1000 1200Distance from Detector Entry (m)

0.0

0.5

1.0

Fra

ctio

nof

Dep

osit

edE

ner

gy Percentiles

FIG. 4. Top: An event display showing the most energeticstarting track found in the data. DOMs are represented bycolored spheres with a radius corresponding to the numberof photoelectrons detected and with a color showing the firstphotoelectron time going from red (earliest) to green (latest).The larger blue spheres show the reconstructed sequence ofelectromagnetic cascades along the track, and their size isproportional to the reconstructed energy of each cascade on alogarithmic scale. The event originates in the top half of thedetector. Middle: The reconstructed energy loss profile as afunction of distance along the reconstructed track. The detec-tor boundaries are shown by green and red dashed lines. En-ergy losses outside the detector, shown in grey, are excludedfrom the energy and inelasticity reconstruction. Bottom: Thecumulative fraction of the total deposited energy within thedetector. Percentiles of the energy loss distribution, shownwith blue points, are features for the random forest regressionof cascade and muon energy. The cascade and muon energiesare estimated to be Ecasc. = 64 TeV and Eµ = 724 TeV, re-spectively, leading to Evis. = 788 TeV and yvis. = 0.08 forthe visible energy and inelasticity, respectively. The total de-posited energy is 135 TeV, and the muon escapes the detectorwith most of the neutrino’s energy.

A. Starting track reconstruction and inelasticity

Most IceCube analyses use track reconstructions thatfit the track to a straight line by maximizing the likeli-hood for the reconstruction, based on functions that givethe light amplitude and arrival time distribution func-

tion for a given DOM, given a track hypothesis [55]. Forstarting tracks, these reconstructions have two significantlimitations. First, they assume that the track is infinite,originating outside the detector and traversing entirelythrough it. Second, they assume that the muon energyloss is continuous, rather than stochastic. The latter as-sumption does not hold for through-going tracks withenergies above 1 TeV. However, it is much more prob-lematic for starting tracks, which are accompanied by alarge hadronic cascade from the recoiling nuclear target.Nevertheless, it is the best reconstruction that we havefor finding the directions of starting tracks.

The directional resolution depends on both the neu-trino energy and the inelasticity; the higher the inelas-ticity, the more the cascade dominates the event. Thatsaid, the median angular error is less than 2◦ degrees forevents with a visible inelasticity up to 0.9, rising to 5◦ ata visible inelasticity of 0.99. Overall, simulations predicta 1.5◦ median angular error for the entire starting tracksample. These resolutions do not significantly impact thecurrent analysis.

After the direction is found, the energy loss profile isunfolded as a sequence of electromagnetic cascades alongthe reconstructed track [41], as is shown in the middlepanel of Fig. 4 for the most energetic starting track foundin the sample. Generally, the largest cascade is at theinteraction vertex.

This unfolding places some of the cascades outside thedetector; these cascades have significantly larger uncer-tainties than those within the detector and are not usedfor the reconstruction. The energy loss profile is thenintegrated to give the cumulative fraction of the energyloss as a function of position along the track, as shownin the bottom panel of Fig. 4. The quantiles where thefirst 0%, 1%, 2%, etc. of total energy loss are determined,where 0% corresponds to the first observed loss.

Another machine learning technique, a random forest,is used to estimate the energy of the starting tracks. Ittakes the positions of the 101 energy loss quantiles as in-put, along with the total deposited energy, the total tracklength contained in the detector and the normalized trackBDT score – a total of 104 inputs. The random forestis trained using simulated events and validated using anindependent sample. For each event, the forest producestwo outputs: the estimated visible cascade energy at thevertex, Ecasc., and the estimated track energy, Etrack.These are combined to produce two new variables: thetotal visible energy,

Evis. = Ecasc. + Etrack (7)

and the visible inelasticity,

yvis. =Ecasc.

Evis.. (8)

Since the visible cascade energy is less than the hadronicenergy, Evis. and yvis. tend to be lower than the actualneutrino energy and inelasticity, for CC νµ interactions.

10

2 4 6True

2

4

6

Rec

onst

ruct

edlog10(Ecasc./GeV) RMS Error: 0.19

2 4 6True

2

4

6

Rec

onst

ruct

ed

log10(Etrack/GeV) RMS Error: 0.36

2 4 6True

2

4

6

Rec

onst

ruct

ed

log10(Evis./GeV) RMS Error: 0.18

0.0 0.5 1.0True

0.00

0.25

0.50

0.75

1.00R

econ

stru

cted

yvis. RMS Error: 0.19

10−4

10−3

10−2

10−1

100

101

Eve

nts

FIG. 5. Joint distributions of reconstructed quantity versustrue quantity for cascade energy, Ecasc., muon energy, Eµ,total visible energy, Evis., and visible inelasticity, yvis.. Theroot mean square (RMS) error for each quantity is shown atthe top of each panel.

Figure 5 shows the resolutions for these variables as quan-tified through the root mean square (RMS) error. Theresolution for log10Evis. is better than either of its com-ponents, because there is some anti-correlation betweenthe cascade and track energies. With this algorithm, forstarting tracks, the RMS error on log10Evis. is 0.18, bet-ter than the typical resolution of 0.22 for through-goingmuons [56]. The resolution for the visible inelasticity is0.19. These performance metrics are mildly dependenton the assumed neutrino energy spectrum, which here isassumed to follow from the HKKMS conventional atmo-spheric flux, BERSS prompt atmospheric flux, and thebest-fit power-law astrophysical flux from Ref. [3].

B. Cascade angular resolution check

Starting track events offer an opportunity to study cas-cade directional reconstruction, using the track as an in-dicator of true direction. This is possible because themuon and cascade are boosted to nearly the same di-rection, and the track angular resolution is much betterthan the cascade resolution. Track events with visible in-elasticity yvis. > 0.75 are chosen to minimize the effect ofthe outgoing muon on the cascade reconstruction. Thiscomparison is sensitive to both systematic offsets, as maybe caused by improper modeling of optical scattering inthe bulk ice, hole ice, DOMs, or other detector phenom-ena. Figure 6 shows the distribution of the differencein zenith angle between cascade and track reconstruc-tions for a sample of high-inelasticity tracks. The dataare reconstructed using the SPICE Mie ice model andshows a significant mismatch between the cascade andtrack zenith angle distributions, with the cascades be-

−60 −40 −20 0 20 40 60θcasc. − θtrack

0.00

0.01

0.02

0.03

0.04

0.05

Pro

bab

ilit

yD

ensi

ty

yvis. > 0.75

SPICE Mie Baseline, Median: −1.3◦

Bulk Ice Scattering +10%, Median: −4.9◦

Hole Ice Scattering +67%, Median: −5.3◦

Data, Median: −8.1◦

FIG. 6. The distribution of the difference in zenith anglebetween cascade and track reconstructions for tracks withyvis. > 0.75. All reconstructions use the baseline SPICE Mieice model. 192 events meeting this condition were found in thefull 5-year data sample. The data (black) show a mismatchbetween cascade and track zenith angles, with cascades be-ing reconstructed with a median angle of 8.1 degrees moredowngoing than the corresponding tracks. When events sim-ulated using the SPICE Mie model are reconstructed (red),the median zenith angle difference is only 1.3 degrees. Thisdiscrepancy may be caused by systematic errors in the icemodel since the cascade reconstruction is much more sensitiveto the ice model than the track reconstruction. Distributionsfor two separate simulations increasing the bulk ice scatteringglobally by +10% (blue) and increasing hole ice scattering by+67% (green) are shown. Increased bulk ice scattering andhole ice scattering produce a shift that can explain the down-going bias seen in the data.

ing reconstructed as more downward-going. This zenithangle distribution is sensitive to the amount of opticalscattering in the bulk ice and in the hole ice, and we findincreased scattering can produce a downward bias in thedistribution. Our fits, discussed below, also confirm thisobservation and find somewhat higher levels of opticalscattering than the IceCube baseline.

VII. INELASTICITY FIT

With the reconstruction results from Section VI, wecan characterize the distribution of visible inelasticityacross energy. The visible inelasticity distribution isshown in Fig. 7 for four half-decade energy bins, from1 TeV to 100 TeV; a 5th bin is used for energies above100 TeV. The data are compared to predictions based onthe CSMS cross section calculation, weighted by the ex-pected neutrino and antineutrino fluxes. The flux modelsused are the best-fit atmospheric and astrophysical mod-els to be described in Sec. VIII. The data are in goodagreement with the predictions.

To further characterize the inelasticity distribution in

11

0.0 0.2 0.4 0.6 0.8 1.00

50

100

150

200

250

300

Nu

mb

erof

Tra

cks

103.0 GeV < Evis. < 103.5 GeV

0.0 0.2 0.4 0.6 0.8 1.0Reconstructed yvis.

0

50

100

150

200

250

103.5 GeV < Evis. < 104.0 GeV

0.0 0.2 0.4 0.6 0.8 1.00

20

40

60

80

104.0 GeV < Evis. < 104.5 GeV

0.0 0.2 0.4 0.6 0.8 1.0Reconstructed yvis.

0.0

2.5

5.0

7.5

10.0

12.5

15.0

17.5

20.0

22.5

Nu

mb

erof

Tra

cks

104.5 GeV < Evis. < 105.0 GeV

0.0 0.2 0.4 0.6 0.8 1.0Reconstructed yvis.

0

1

2

3

4

5

6Evis. > 105.0 GeV

νµ CC

νµ CC

CC Charm

Total

Best-fit Parametrization

Data

FIG. 7. The reconstructed visible inelasticity distribution in five different bins of reconstructed energy. Observed data areshown in black. Error bars show 68% Feldman-Cousins confidence intervals for the event rate in each bin [57]. The resultof fitting the distribution to the parameterization of Eq. 10 is shown with dashed green lines. The prediction of the CSMSdifferential CC cross section are shown for neutrinos with solid blue lines and antineutrinos with dashed blue lines. The totalCC charm contribution is shown in magenta, illustrating its flatter inelasticity distribution. The best-fit flux models of Tab. IIare assumed for all predictions.

a model-independent fashion, ideally we would use thesevisible inelasticity distributions to unfold dσ/dy, butthere are several complications. The detection efficiencydrops at low energies for very large y because the track isno longer visible. It also drops for small y at low energiesbecause there is not enough light visible in the detector.Because of these strongly varying efficiencies, the limitedstatistics and the limited resolution, we do not attemptto unfold the data to present dσ/dy distributions. In-stead, we parameterize the true inelasticity distribution,reweight simulation to the parametrized distribution, andfit the visible inelasticity distribution to find the param-eters. These parameters can be compared with the Stan-dard Model distribution, and also used to test alternative

theories.To motivate the parameterization, recall the differen-

tial CC cross section can be written schematically at lead-ing order as

dσ

dxdy=

2G2FMEνπ

(M2W

Q2 +M2W

)2

[xq(x,Q2) + (1− y)2xq(x,Q2)

], (9)

where q and q represent sums of quark PDFs [16].High-energy neutrinos probe the PDFs at low valuesof Bjorken−x ∼ 3 × 10−3(Eν/PeV), where sea quarksdominate, and they should have a power-law behavior,xq(x,Q2) ∼ A(Q2)x−λ with λ ∼ 0.4. Following Ref. [58],

12

log10 (Evis./GeV) log10 (Eν/GeV) Events 〈y〉 λ

[3.0, 3.5) 3.33+0.20−0.22 1111 0.42+0.06

−0.09 1.06+0.74−1.90

[3.5, 4.0) 3.73+0.22−0.22 1107 0.42+0.02

−0.02 1.09+0.25−0.40

[4.0, 4.5) 4.18+0.22−0.22 310 0.38+0.03

−0.03 0.97+0.26−0.30

[4.5, 5.0) 4.65+0.22−0.22 72 0.37+0.05

−0.05 0.75+0.44−0.75

> 5.0 5.23+0.50−0.33 11 0.28+0.15

−0.24 0.13+0.78−0.13

TABLE I. The best-fit parameters when reconstructed in-elasticity distributions are fit to Eq. 10 in five bins of re-constructed energy. The energy range containing the central68% of simulated neutrino energies for each bin is shown inthe second column. Because of the limited energy resolution,sometimes the 68% central range extends outside the nominalbin boundaries.

102 103 104 105 106 107

Eν (GeV)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

〈y〉

ν CSMS

ν CSMS

Flux-averaged CSMS

FIG. 8. The mean inelasticity obtained from the fit to Eq. 13in five bins of reconstructed energy. Vertical error bars in-dicate the 68% confidence interval for the mean inelasticity,and horizontal error bars indicate the expected central 68% ofneutrino energies in each bin. The predicted mean inelasticityfrom CSMS is shown in blue for neutrinos and in green forantineutrinos. The height of the colored bands indicates theo-retical uncertainties in the CSMS calculation. A flux-averagedmean inelasticity per the HKKMS calculation is shown in red.

when transforming variables from (x, y) to (Q2, y), theQ2-dependence of Eq. 9 can be separated from the y-dependence and integrated out to give a two-parameterfunction,

dσ

dy∝(1 + ε(1− y)2

)yλ−1, (10)

where the parameter ε indicates relative importance ofthe term proportional to (1 − y)2 in Eq. 10. This pa-rameterization also works for antineutrinos, but ε takeson a different value since q and q are interchanged. Ourmeasurement represents an average over neutrinos andantineutrinos. The normalized inelasticity distribution

can then be written as

dp

dy= N

(1 + ε(1− y)2

)yλ−1, (11)

where N is the normalization

N =λ(λ+ 1)(λ+ 2)

2ε+ (λ+ 1)(λ+ 2). (12)

This simple parameterization can accurately representsophisticated calculations of inelasticity distributions.For example, a fit of Eq. 10 to the full CSMS calculationproduces no more than a 1% root mean square devia-tion (averaged over y) for neutrino energies from 1 TeVto 10 PeV.

In practice, the parameters ε and λ are highly corre-lated when fitting Eq. 11 to realistic inelasticity distribu-tions. To avoid this correlation, it is convenient to fit forthe mean of the distribution, 〈y〉, and λ instead, whichshow far less correlation. The mean inelasticity can befound analytically,

〈y〉 =

∫ 1

0

ydp

dydy =

λ(2ε+ (λ+ 2)(λ+ 3))

(λ+ 3)(2ε+ (λ+ 1)(λ+ 2))(13)

It is then straightforward to substitute

ε = − (λ+ 2)(λ+ 3)

2

〈y〉(λ+ 1)− λ〈y〉(λ+ 3)− λ (14)

into Eq. 11 so that dp/dy can be found as a function of〈y〉 and λ only.

The visible inelasticity distribution in each energyrange from Fig. 7 can then be fit to the parametriza-tion of Eq. 11 using a binned Poisson likelihood fit of the10 bins. The goodness-of-fit test statistic is

−2 ln Λ = 2∑i

[µi(θ)− ni + ni ln

niµi(θ)

]+∑j

(θj − θ∗j )2

σ2j

(15)where µi(θ) is the expected event count in each bin de-pending on parameters θ and ni is the observed eventcount per bin [59]. The second sum accounts for a Gaus-sian prior distribution on a parameter θj with mean θ∗jand standard deviation σj . The expected event rate isderived from weighted Monte Carlo simulations. To ac-count for the parametrized inelasticity distribution fromEq. 11, each simulated event receives a reweighting fac-tor, dp

dy (y; 〈y〉, λ)/ dpdyCSMS, where dp

dyCSMSis the inelastic-

ity distribution calculated by CSMS that is used in thesimulation. A total event rate scaling factor is also in-cluded to account for uncertainties in the flux normaliza-tion. The neutrino flux is assumed to follow the best-fitflux models in Sec. VIII, but the flux model and its uncer-tainties have negligible effect since the size of each energyrange is comparable to the energy resolution. Detectorsystematic uncertainties on bulk ice scattering and ab-sorption, DOM optical efficiency, and hole ice scattering

13

Parameter 68% CI Prior

Φconv 1.05+0.07−0.06 Flat [0,∞)

∆γcr 0.04+0.03−0.03 Gaussian 0.00± 0.05

RK/π 1.11+0.35−0.28 Gaussian 1.00± 0.50

Φprompt 0.00+1.10−0.00 Flat [0,∞)

Φ0/10−18 GeV−1s−1cm−2sr−1 2.04+0.23−0.21 Flat [0,∞)

γ 2.62+0.07−0.07 Flat

εDOM/ε0 1.08+0.03−0.03 Gaussian 1.00± 0.10

αScat. 1.07+0.01−0.01 Gaussian 1.00± 0.10

αAbs. 1.02+0.02−0.02 Gaussian 1.00± 0.10

αHole Ice 1.46+0.12−0.12 Flat [1.00, 1.67]

Test Statistic −2 ln Λ 175.54 –

TABLE II. The best-fit parameters including all neutrino fluxand detector systematic uncertainties. 68% confidence inter-vals are shown as calculated by the profile likelihood method.The prior distribution for each parameter is shown in the lastcolumn. The last row is the goodness-of-fit test statistic.

are included through the use of 4 additional nuisance pa-rameters in the fit. They are constrained by Gaussianpriors and are further described in Sec. VIII.

The best-fit parameters for describing these inelastic-ity distributions are shown in Tab. I. Figure 8 compares〈y〉 as a function of energy with the predictions of theCSMS calculation for neutrinos and antineutrinos. Themeasured values of 〈y〉 agree well with the flux-weightedaverage of neutrinos and antineutrinos. The downwardtrend in 〈y〉 is due to the W -boson propagator.

VIII. LIKELIHOOD FIT RESULTS ANDSTARTING TRACK/CASCADE COMPARISON

Inelasticity introduces a new dimension into the stud-ies of high-energy astrophysical neutrinos, providing sen-sitivity to a number of new phenomena. In addition, themuch more precise measurement of the starting tracksallows new tests, such as comparing the energy spectraof astrophysical νµ with that from cascades, a mixturethat is mostly νe and ντ . The inelasticity distribution isalso sensitive to the ν/ν ratio and to neutrino interac-tions that produce charm quarks. In this section, we willpresent a baseline maximum likelihood fit and compareit with previous analyses.

The fit is done jointly over both cascades and startingtracks. For cascades, data is binned in two dimensionswith half decade bins in energy ranging from 102.5 GeVto 107 GeV and 5 bins in cosine zenith angle. For tracks,data is binned in three dimensions with the same en-ergy and zenith binning but additionally 5 bins in visibleinelasticity. The same binned Poisson test statistic inEq. 15 is used.

We first fit the data to a model that is similar to pre-vious IceCube analyses [1–3, 5]. It includes three com-ponents: the conventional atmospheric neutrino flux, theprompt atmospheric flux and the astrophysical flux. Af-ter describing this standard fit here, Sections VIII.A-VIII.D discuss some additional fits which each add oneadditional degree of freedom to explore additional aspectsof the physics.

The conventional atmospheric flux is based on theHKKMS calculation, extrapolated in energy to above 10TeV and modified to include the knee of the cosmic-rayspectrum following the H3a cosmic-ray model. To ac-count for the uncertainties in this flux model, we includeseveral nuisance parameters in the fit. The first is theoverall normalization, Φconv. The second, ∆γcr, accountsfor uncertainty in the energy spectrum by allowing thespectral index to vary with a prior. A third parameter,RK/π, accounts for uncertainties in the kaon to pion ra-tio in cosmic-ray air showers [25]. Neutrinos from kaonshave a somewhat different zenith angle distribution thanthose from pions; RK/π accounts for that possible vari-ation. The prompt atmospheric flux follows the BERSScalculation, an update of the ERS calculation [60] usedin many previous IceCube works. It is incorporated intothe analysis with a single parameter, the normalizationfor the overall amplitude. The self-veto probability isincluded for both atmospheric flux calculations.

Astrophysical neutrinos are initially assumed to beisotropic. In this section the νe : νµ : ντ ratio is taken tobe(13 : 1

3 : 13

)⊕, an approximation expected from almost

any conventional source model, after accounting for os-cillation en-route to Earth. The ν : ν ratio is taken tobe 1 : 1. The flux per flavor is assumed to follow a singlepower-law:

Φα(Eν) = 3Φ0fα,⊕

(Eν

100 TeV

)−γ(16)

where fα,⊕ ≈ 1/3 is the fraction of each flavor at Earth, γis the power-law index, and Φ0 is a normalization factorthat corresponds to the average flux of ν and ν per flavorat 100 TeV.

Detector systematic uncertainties are incorporated inall results through the use of four more nuisance pa-rameters that describe uncertainties in the detection andtransmission of light through ice. The first, εDOM, ac-counts for uncertainties in the overall optical sensitivityof the DOMs; the prior on this is ±10%. Two parame-ters, αScat. and αAbs., account for uncertainties of opticalscattering and absorption in the bulk ice. These param-eters linearly scale the inverse scattering and absorptionlengths uniformly over all ice layers. Finally, αHole Ice

accounts for uncertainties on the overall scattering inthe hole ice, the columns of refrozen drill water that thestrings are emplaced in. Because of the presence of vis-ible air bubbles [44] and possible impurities, the opticalquality of this ice is expected to be much worse than thatof the rest of the ice. The baseline ice model assumes a 50cm scattering length in hole ice, but calibration data are

14

103 104 105 106 107

10−1

100

101

102

103

Nu

mb

erof

Tra

cks

0.0 0.2 0.4 0.6 0.8 1.0

10−1

100

101

102

103

−1.0 −0.6 −0.2 0.2 0.6 1.0

10−1

100

101

102

103

103 104 105 106 1070.5

1.0

1.5

Dat

a/M

C

0.0 0.2 0.4 0.6 0.8 1.0

Reconstructed Inelasticity

0.5

1.0

1.5

−1.0 −0.6 −0.2 0.2 0.6 1.00.5

1.0

1.5

103 104 105 106 10710−2

10−1

100

101

102

103

Nu

mb

erof

Cas

cad

es

−1.0 −0.6 −0.2 0.2 0.6 1.010−2

10−1

100

101

102

103

103 104 105 106 107

Reconstructed Energy (GeV)

0.5

1.0

1.5

Dat

a/M

C

−1.0 −0.6 −0.2 0.2 0.6 1.0

cos(Reconstructed Zenith)

0.5

1.0

1.5

Best-fit Total

Astrophysical ν

Prompt Atm. ν

Conventional Atm. ν

Data

FIG. 9. Best-fit distributions with all neutrino flux and detector parameters. Top: The distribution of reconstructed visibleenergy, visible inelasticity, and cosine zenith for the sample of starting tracks. Bottom: the distribution of reconstructedenergy and cosine zenith for the sample of cascades. Contributions of conventional atmospheric and astrophysical neutrinosare shown in blue and red, respectively, and the total predicted distribution is shown in maroon. The prompt atmosphericneutrino contribution is not shown since its best-fit value is zero. The black error bars show the data. The inclusion of detectorparameters describing bulk and hole ice scattering substantially improves the model description of the cascade cosine zenithdistribution. The best-fit parameters are shown in Tab. II.

also consistent with a scattering length of 30 cm, or 1.67times more scattering. Uncertainties in both the hole iceand bulk ice scattering can lead to a bias in the zenithangle distribution of cascades as shown in Fig. 6. Thefit finds values for these parameters that are in line withexpectations. The larger value of αHole Ice is comparablewith other recent IceCube measurements [4].

The fit results are shown in Tab. II, and the fit iscompared with the data in Fig. 9. Based on a set ofcomputed pseudo-experiments, the goodness-of-fit teststatistic, −2 ln Λ = 175.54, corresponds to a probability(p-value) of 0.04. This value indicates the fit model maynot be a complete description of the data, perhaps due to

inadequately simulated systematic uncertatinties. Still,the p-value is acceptable, and there are no obvious prob-lem areas visible in the distributions in Fig. 9. We alsofit the data with the optical properties of the ice fixed totheir default values. The test statistic worsened signifi-cantly, but, except for RK/π, the other fit parameters didnot change significantly. The inelasticity measurementsare quite robust against systematic errors.

15

1.8 2.0 2.2 2.4 2.6 2.8 3.0γ

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0Φ

0/1

0−18

GeV−

1s−

1cm−

2sr−

1

68%

95%

68%

95%

68%

95%

68%95

%

Starting Tracks Only

Cascades Only

Cascades and Starting Tracks

Through-going Tracks

FIG. 10. Confidence regions for the astrophysical power-lawindex, γ and flux normalization, Φ0. The blue contours showthe confidence region for the joint fit of the cascade and start-ing track samples, which is the main result obtained here. Thered contours show the confidence region for a fit of startingtracks only, and the green contours show the confidence re-gion for a fit of cascades only, and all are consistent with eachother. The confidence region from the IceCube analysis ofthrough-going tracks [24] is shown in orange, which is in ten-sion with the cascade confidence region. The contours fromthe starting track sample are consistent with both cascadeand through-going samples.

A. Astrophysical neutrino energy spectrum

The baseline fit finds an astrophysical power law in-dex of γ = 2.62 ± 0.07, with a normalization Φ0 =2.04+0.23

−0.21 × 10−18 GeV−1s−1cm−2sr−1. This is in agree-ment with earlier IceCube studies of contained events[1], but softer than the most recent measurements ofcontained events [4], and considerably steeper than themeasurement using through-going tracks, which foundγ = 2.13±0.13 [24]. There are a couple of possible expla-nations for the latter tension. First, the through-goingtracks have considerably higher neutrino energies thanthe current sample, with 90% of the sample sensitivityin the energy range from 194 TeV to 7.8 PeV. Using thesame method as in Ref. [24], the current sample has a90% central range of 3.5 TeV to 2.6 PeV. For reference,90% of the selected contained events are in the energyrange from 3.3 TeV to 220 TeV; this range is much lowerthan for the sensitivity because the most energetic eventshave the largest effect on the astrophysical flux measure-ment. If the astrophysical flux is not a single power law,then one might measure different spectral indices in dif-ferent energy ranges. Or, with difficulty, one might finddifferent spectral indices for tracks and cascades. Oneway to test this hypothesis is to repeat the fit, allowingthe astrophysical spectral indices and normalizations tovary between the starting tracks and cascades. Whenthis is done, we find a power-law index of γ = 2.43+0.28

−0.30

for starting tracks and γ = 2.62 ± 0.08 for the cascades.The two indices are compatible within uncertainties. Fig-ure 10 shows the two-dimensional confidence regions forthe cascade and track measurements, the combined mea-surement, and the previous through-going track fit. Con-fidence regions are derived from the profile likelihood overall nuisance parameters, and it is assumed the test statis-tic follows a χ2 distribution throughout. The cascadesample drives the combined-sample index of 2.62± 0.07,by virtue of the much better energy resolution and loweratmospheric background compared to tracks. Within theuncertainty, the starting track power-law index is alsocompatible with that from the through-going tracks. Weconsidered alternate scenarios with a double power-lawor a power law with a cutoff that could explain a harderpower-law index found at high energies, but we found noevidence for either when fitting our sample alone. Alllater results will continue the assumption of an unbrokenpower-law spectrum.

The other parameters in the fit in Tab. II are inline with expectations. The best-fit prompt flux iszero, in agreement with many previous IceCube stud-ies [1, 6, 24, 51], but the 1σ upper limit is compatiblewith the expected BERSS flux. The conventional flux,cosmic-ray spectral index and RK/π are all in line withexpectations.

For the remainder of this section, we will add param-eters one at a time to this baseline fit to independentlymeasure the flavor composition of astrophysical neutri-nos, the atmospheric νµ : νµ ratio, and neutrino charmproduction. The results are little affected if we chooseallow all of these parameters float simultaneously, andnone of these measured parameters show strong correla-tion with another.

B. Astrophysical neutrino flavor composition

A related test of the astrophysical flux is to measurethe flavor composition of the contained event sample.Compared with the previous contained event analysis [5],this analysis benefits from much better track energy res-olution and also the presence of the inelasticity distri-bution; the inelasticity distribution has some sensitivityto the presence of ντ , since ντ interactions, followed byτ → µνµντ decays, will lead to events with larger visibleinelasticity than νµ interactions of the same energy. Aglobal fit combining results from contained events andthrough-going muons found tighter limits that enabledconstraints on the source flavor composition, however itcompared cascades and tracks in different energy rangeswhere the energy spectrum may differ [6].

Figure 11 shows confidence levels for various νe : νµ :ντ ratios obtained by fitting the data with the same pa-rameters in Tab. II as nuisance parameters. The linesand points show the expectation from different produc-tion models and standard neutrino oscillations, includingconventional pion decay with source flavor composition

16

0.00

0.17

0.33

0.50

0.67

0.83

1.00

0.00

0.17

0.33

0.50

0.67

0.83

1.00

0.00

0.17

0.33

0.50

0.67

0.83

1.00

fe,⊕

fτ,⊕ fµ,⊕

68%

68%

95%

95%

0

20

40

60

80

100

Con

fid

ence

Lev

el(%

)

FIG. 11. Confidence regions for astrophysical flavor ratios(fe : fµ : fτ )⊕ at Earth. The labels for each flavor refer to thecorrespondingly tilted lines of the triangle. Averaged neutrinooscillations map the flavor ratio at sources to points within theextremely narrow blue triangle diagonally across the center.The ≈

(13

: 13

: 13

)⊕ composition at Earth, resulting from a(

13

: 23

: 0)S

source composition, is marked with a blue circle.The compositions at Earth resulting from source compositionsof (0 : 1 : 0)S and (1 : 0 : 0)S are marked with a red triangle anda green square, respectively. The updated best-fit neutrinooscillation parameters from [61] are used here. Though thebest-fit composition at Earth (black cross) is (0 : 0.21 : 0.79)⊕,the limits are consistent with all compositions possible underaveraged oscillations.

(13 : 2

3 : 0)S

, neutron decay with (1 : 0 : 0)S , and a modelwhere muons lose their energy via synchrotron radiationbefore they decay with (0 : 1 : 0)S [62]. All of these con-ventional scenarios are along a narrow line, and, unfor-tunately, all are within the 68% confidence range for theanalysis. The best-fit composition, (0 : 0.21 : 0.79)⊕, ison the left side of the triangle. However, because mostντ produce cascades, there is a relatively high degener-acy between ντ and νe, so the confidence levels nearer themiddle of the triangle are high; the break in the degener-acy from the inelasticity distribution of starting tracks isinadequate to statistically separate the ντ and νe compo-nents. In this analysis, astrophysical cascades and trackshave a similar energy range, with 68% central ranges of11 TeV to 410 TeV and 8.6 TeV to 207 TeV, respectively.This is the tightest limit using samples with a similar en-ergy range; a previous global fit [6] used contained cas-cades and through-going muons; the latter had a muchhigher energy range for through-going tracks (330 TeVto 1.4 PeV) than for cascades, where energies above 30TeV were probed [63]. With track and cascade sampleshaving different energies, if the astrophysical spectrum isnot a perfect power law, then the confidence regions onthe flavor triangle will be shifted.

The extreme compositions of 100% νe and 100% νµare ruled out with high confidence, at 5.8σ and 7.4σ,respectively. In fact, any composition with more than 2/3νµ is ruled out at 95% confidence level. These constraintscan be used to put limits on exotic models.

C. Atmospheric neutrino/antineutrino ratio

At energies below about 10 TeV, neutrinos and an-tineutrinos have substantial differences in their inelas-ticity distributions, since neutrino interactions are moresensitive to quarks, while antineutrinos are more attunedto antiquarks. At large Bjorken−x values where valencequarks dominate, the differences are substantial, lead-ing to roughly a factor of 2 difference in cross section[64, 65] as well as a difference in inelasticity distribu-tion. Unfortunately, as Fig. 12 shows, the differenceslowly disappears above 10 TeV. So, there is little sen-sitivity to the νµ : νµ ratio of astrophysical neutrinos.However, inelasticity can be used to measure the νµ : νµratio at lower energies for atmospheric neutrinos. Theatmospheric νµ : νµ flux ratio varies with both energyand zenith angle, and we choose to measure an overallscaling factor of the νµ : νµ flux ratio from the HKKMScalculation, Rνµ/νµ

. At 1 TeV, the direction-averagedνµ : νµ flux ratio is 1.55 and rises slowly to an asymptoicvalue of 1.75 above 100 TeV.

When the parameter Rνµ/νµis added to the list of

parameters in Tab. II, the best fit value is Rνµ/νµ=

0.77+0.44−0.25. A flux of 100% neutrinos (no ν) is disfavored

at 3.8σ, while the reverse is excluded at 5.4σ. It shouldbe noted that these limits are also dependent on the cal-culated angular distributions of ν and ν in addition toinelasticity. The sensitive range for this analysis is 770GeV to 21 TeV; at higher energies, there is little ν : ν dis-crimination. This is the first ν : ν measurement in thisenergy range. Along with measurements of the atmo-spheric muon charge ratio [66–69], it is a useful diagnos-tic of particle production in cosmic-ray interactions [54].However, since this is an overall scaling factor, the samefor all energies and zenith angles, it should not be directlyinterpreted in terms of hadronic interaction models sincethey also change the energy and zenith distribution as-sumed.

D. Neutrino charm production

Inelasticity measurements can also be used to probecharm production in neutrino interactions. The fractionof CC neutrino interactions that produce charm quarksrises slowly, from 10% at 100 GeV to 20% at 100 TeV.Charm quarks are produced primarily when a neutrinointeracts with a strange sea quark; these sea quarks havelower mean Bjorken−x values than valence quarks, so theinteractions have flatter inelasticity distributions thanneutrino interactions with valence quarks. There is also a

17

103 104 105 106 1070.0

0.2

0.4

0.6

0.8

1.0R

econ

stru

cted

y vis.

0.23

0.22

0.22

0.32

0.26

0.23

0.20

0.19

0.32

0.27

0.24

0.21

0.20

0.32

0.29

0.26

0.24

0.24

0.35

0.33

0.31

0.29

0.29

0.40

0.39

0.37

0.38

0.39

0.47

0.46

0.47

0.47

0.48

0.48

0.49

0.48

0.47

0.49

0.48

0.50

0.51

0.48

0.48

103 104 105 106 107

Reconstructed Evis. (GeV)

0.0

0.5

1.0

Atm

osp

her

icF

ract

ion

0.20

0.25

0.30

0.35

0.40

0.45

0.50

ν µF

ract

ion

FIG. 12. The predicted fraction of νµ contributing to thetotal νµ + νµ event rate in bins of reconstructed energy andinelasticity. At energies below ∼ 10 TeV, there are substan-tial differences in the inelasticity distribution that enable theatmospheric neutrino to antineutrino ratio to be measured.The bottom panel shows the fraction of atmospheric neutrinoscontributing to the total event rate in bins of reconstructedenergy. At energies above ∼ 100 TeV where astrophysicalneutrinos begin to dominate the event rate, differences in theinelasticity distribution vanish, and it is not possible to mea-sure the neutrino to antineutrino ratio for the astrophysicalflux. An equal neutrino and antineutrino composition is as-sumed for the astrophysical flux here.

roughly 10% chance for a charm quark to decay semilep-tonically and produce an extra muon. These muons willnot be distinguishable from the primary muon from theCC interaction; the energy loss from the two muons willadd, and they will be reconstructed as a single, higher-energy muon, leading to a lower apparent inelasticity.For a νe or ντ interaction, the presence of a muon fromsemileptonic charm decay may lead to a track event withan apparent high inelasticity.

The contribution of charm production to different in-elasticity events at different energies is shown in Fig. 13.Charm is most visible at energies above 100 TeV andat high inelasticity. For energies between 100 TeV and1 PeV, more than 1/3 of the events with reconstructedyvis. > 0.8 produce charm. This shape difference canbe used to search for charm production in a maximumlikelhood fit.

An additional parameter, RCC,charm, scaling the CCcharm production event rate is added to the parametersin Tab. II. Fitting cascades and tracks jointly, we findthe 68% interval, RCC,charm = 0.93+0.73

−0.59. The test statis-tic for a null hypothesis with zero charm production is−2∆ ln(L) = 2.8, so zero charm production is excludedat 91% confidence level. The 90% upper limit is 2.3 timesthe leading-order HERAPDF1.5 prediction.

The central 90% of neutrino energies contributing tothis test statistic is from 1.5 to 340 TeV. This is a widerenergy range than the central 90% of charm production

103 104 105 106 107

Reconstructed Evis. (GeV)

0.0

0.2

0.4

0.6

0.8

1.0

Rec

onst

ruct

edy v

is.

0.13

0.16

0.13

0.10

0.11

0.12

0.14

0.17

0.11

0.13

0.15

0.16

0.17

0.12

0.15

0.17

0.19

0.21

0.14

0.18

0.21

0.23

0.26

0.15

0.22

0.24

0.29

0.34

0.17

0.23

0.27

0.32

0.36

0.18

0.26

0.28

0.32

0.29

0.20

0.25

0.32

0.31

0.13

0.15

0.20

0.25

0.30

0.35

ν+ν

CC

Ch

arm

Fra

ctio

n

FIG. 13. The predicted fractional contribution of all-flavorneutrino CC charm production events to the total event ratein bins of reconstructed visible energy and inelasticity. Theincreased charm production fraction at high visible inelastic-ity and high energy (up to 36%) provides a shape differencethat allows the presence of charm production to be identifiedin a likelihood fit to the data.

events, 1.3 TeV to 44 TeV, because charm production islarger at higher energy. This is the highest energy mea-surement of charm production yet, and the upper end ofthe energy range is above the critical energies of charmhadrons where interactions in ice must occur. Similarmethods could be used to search for other special typesof neutrino interactions beyond charm production, in-cluding those beyond the Standard Model.

IX. CONCLUSIONS

We have developed a tool to measure neutrino inelas-ticity in Gigaton-scale H2O based detectors and pre-sented the first measurements of neutrino inelasticity invery high-energy (above 1 TeV) neutrino interactions, us-ing a sample of starting track events collected by the Ice-Cube Neutrino Observatory. The measured inelasticitydistributions are in good agreement with the predictionsof a modern NLO calculation. More data is needed toreach anticipated theoretical uncertainties in these cal-culations.

We have made a global fit to these neutrino data, fit-ting cascades in two dimensions: energy and zenith angle,and starting tracks in three dimensions: energy, zenithangle and inelasticity, to extract information about theastrophysical and atmospheric neutrino fluxes. This fitfinds an astrophysical power-law spectral index of γ =2.62± 0.07, in good agreement with previous fits to con-tained events and cascades, but in tension with previousresults based on through-going muons, a sample that isgenerally higher in energy than the contained event sam-ples. To explore this tension, we performed a fit where

18

we allowed the astrophysical flux to float separately forcascades and starting tracks, with different spectral in-dices. Unfortunately, this leads to a spectral index forthe tracks, γ = 2.43+0.28

−0.30, intermediate between the com-bined result and that for through-going tracks, with anerror that is consistent with either.

We then relaxed the requirement that the astrophysi-cal neutrino flavor ratio be

(13 : 1

3 : 13

)⊕, and calculated

the confidence level for other compositions. We found abest fit point consisting of 79% ντ , 21% νµ, but with abroad allowed contour that encompasses all of the mod-els that invoke conventional acceleration mechanisms andstandard neutrino oscillations. More exotic models maybe ruled out.

We also set limits on the νµ : νµ ratio in atmosphericneutrinos and exclude zero production of charm quarksin neutrino interactions at 91% confidence level. This isonly the second study, after measurements of the crosssection using neutrinos with energies above 1 TeV.

Using the indirect signature in the inelasticity distri-bution, we observe, at greater than 90% confidence level,CC charm production in neutrino interactions, at ener-gies between 1.5 and 340 TeV, more than an order ofmagnitude higher in energy than accelerator measure-ments.

Looking ahead, we expect that IceCube-Gen2 [70] andKM3NeT2.0 [71] will collect larger samples of containedevents, which can be used to make more precise measure-ments of inelasticity. These detectors could collect sub-stantial samples of events with energies above 100 TeV.With the increased precision, it will also be possible tostudy several new topics. Tau neutrinos are one example;ντ interactions have a distinctive inelasticity distributionwhich could be used to detect a ντ signal. Top quark pro-duction may also be accessible if enough energetic neu-trinos are available. One calculation found that, for 10PeV neutrinos, top quarks are produced in 5% of theinteractions [17]. It may also be possible to study otherStandard Model neutrino interactions, such as diffractiveproduction of W bosons in the Coulomb field of oxygennuclei [21, 22]; the cross section for ν+O → l+W++X isabout 8% of the charged-current cross section for 1 PeVneutrinos. Even the first phase of IceCube-Gen2 shouldenable improved calibrations of the existing data, reduc-ing the systematic uncertainties. With moderately im-proved calibrations, the precision of the inelasticity mea-surements should scale as the square root of the effectivevolume times the live time.

With an improved surface veto to reject atmosphericneutrinos, it might also be possible to measure the ν : νratio of astrophysical neutrinos. If one could use the self-veto and a surface air-shower-array veto to reject atmo-spheric neutrinos with energies in the 1-10 TeV energyrange, the inelasticity distribution could be used to de-termine the ν : ν ratio of astrophysical neutrinos.

The data could also be used to search for beyond-standard-model (BSM) physics, such as supersymmetry[75], leptoquarks [73] or quantum gravity with a relatively

low scale [74]. These phenomena also produce cross sec-tion enhancements which could be visible via increasedneutrino absorption in the Earth, but the inelasticity dis-tribution has a higher diagnostic utility than a simple in-crease in neutrino absorption. The use of inelasticity al-lows for a more sensitive search than by merely countingcascades and tracks [72]. A combined fit to cross sectionand inelasticity measurements would provide even betterconstraints on new physics.

For most of these phenomena, the LHC provides bet-ter limits compared to IceCube Gen2 and KM3NeT 2.0.However, experiments that aim to record the coherentradio Cherenkov emission from ultra-energetic neutrinoswith energies above 1017 eV can reach supra-LHC ener-gies. The ARA [76] and ARIANNA [77] collaborationsboth propose to deploy large (> 100 km3) arrays thatwill, unless ultra-high energy cosmic-rays are primarilyiron, collect useful (order 100 events) samples of cosmo-genic neutrinos. The challenge here is that these exper-iments are primarily sensitive to cascades, while the en-ergy deposition from tracks is too diffuse to be observ-able. However, it may be possible to take advantage ofthe Landau-Pomeranchuk-Migdal (LPM) effect to sep-arate electromagnetic showers, which at energies above1020 eV are elongated, from the hadronic showers fromthe target nucleus, which are less subject to the LPMeffect. This leads to a moderately elongated electromag-netic shower following a compact hadronic shower [78].With this, it might be possible to measure the inelasticityof charged-current νe interactions [79].

ACKNOWLEDGMENTS

The IceCube Collaboration designed, constructed andnow operates the IceCube Neutrino Observatory. Dataprocessing and calibration, Monte Carlo simulations ofthe detector and of theoretical models, and data analy-ses were performed by a large number of collaborationmembers, who also discussed and approved the scien-tific results presented here. The main authors of thismanuscript were Gary Binder and Spencer Klein. Itwas reviewed by the entire collaboration before publi-cation, and all authors approved the final version of themanuscript.

We acknowledge the support from the following agen-cies: USA – U.S. National Science Foundation-Officeof Polar Programs, U.S. National Science Foundation-Physics Division, Wisconsin Alumni Research Founda-tion, Center for High Throughput Computing (CHTC)at the University of Wisconsin-Madison, Open ScienceGrid (OSG), Extreme Science and Engineering Discov-ery Environment (XSEDE), U.S. Department of Energy-National Energy Research Scientific Computing Cen-ter, Particle astrophysics research computing center atthe University of Maryland, Institute for Cyber-EnabledResearch at Michigan State University, and Astropar-ticle physics computational facility at Marquette Uni-

19

versity; Belgium – Funds for Scientific Research (FRS-FNRS and FWO), FWO Odysseus and Big Science pro-grammes, and Belgian Federal Science Policy Office (Bel-spo); Germany – Bundesministerium fur Bildung undForschung (BMBF), Deutsche Forschungsgemeinschaft(DFG), Helmholtz Alliance for Astroparticle Physics(HAP), Initiative and Networking Fund of the HelmholtzAssociation, Deutsches Elektronen Synchrotron (DESY),and High Performance Computing cluster of the RWTHAachen; Sweden – Swedish Research Council, SwedishPolar Research Secretariat, Swedish National Infrastruc-

ture for Computing (SNIC), and Knut and Alice Wallen-berg Foundation; Australia – Australian Research Coun-cil; Canada – Natural Sciences and Engineering ResearchCouncil of Canada, Calcul Quebec, Compute Ontario,Canada Foundation for Innovation, WestGrid, and Com-pute Canada; Denmark – Villum Fonden, Danish Na-tional Research Foundation (DNRF); New Zealand –Marsden Fund; Japan – Japan Society for Promotion ofScience (JSPS) and Institute for Global Prominent Re-search (IGPR) of Chiba University; Korea – National Re-search Foundation of Korea (NRF); Switzerland – SwissNational Science Foundation (SNSF).

[1] M. G. Aartsen et al. [IceCube Collaboration], Phys. Rev.Lett. 113, 101101 (2014).

[2] M. G. Aartsen et al. [IceCube Collaboration], Science342, 1242856 (2013).

[3] L. Mohrmann [IceCube Collaboration], PoS ICRC 2015(2016) 1066.

[4] H. M. Niederhausen and Y. Xu [IceCube Collaboration],PoS ICRC 2017 (2018) 968.

[5] M. G. Aartsen et al. [IceCube Collaboration], Phys. Rev.Lett. 114, no. 17, 171102 (2015).

[6] M. G. Aartsen et al. [IceCube Collaboration], Astrophys.J. 809, no. 1, 98 (2015).

[7] M. G. Aartsen et al. [IceCube Collaboration], Nature551, 596 (2017).

[8] G. Binder, PhD dissertation, University of Cal-ifornia, Berkeley, (2017). Available at https:

//docushare.icecube.wisc.edu/dsweb/Get/

Document-82240/thesis.pdf

[9] I. Schienbein, J. Y. Yu, C. Keppel, J. G. Morfin, F. Olnessand J. F. Owens, Phys. Rev. D 77, 054013 (2008).

[10] B. Aubert et al., Phys. Rev. Lett. 33, 984 (1974).[11] R. Devinish and A. Cooper-Sarkar, Deep Inelastic Scat-

tering, Oxford University Press, 2004.[12] N. Kalantarians, C. Keppel and M. E. Christy, Phys.

Rev. C 96, no. 3, 032201 (2017).[13] A. Cooper-Sarkar, P. Mertsch and S. Sarkar, JHEP 1108,

042 (2011).[14] V. Radescu, [H1 and ZEUS Collaborations], Proceedings,

35th Int. Conf. High energy Phys. (ICHEP 2010) Paris,Fr. July 22-28, 2010.

[15] H. L. Lai, M. Guzzi, J. Huston, Z. Li, P. M. Nadolsky,J. Pumplin and C.-P. Yuan, Phys. Rev. D 82, 074024(2010).

[16] A. Connolly, R. S. Thorne and D. Waters, Phys. Rev. D83, 113009 (2011).

[17] V. Barger, E. Basso, Y. Gao and W. Y. Keung, Phys.Rev. D 95, no. 9, 093002 (2017).

[18] A. Bueno and A. Gascn, Comput. Phys. Commun. 185,638 (2014).

[19] J. Mousseau et al. [MINERvA Collaboration], Phys. Rev.D 93, no. 7, 071101 (2016).

[20] I. Schienbein, J. Y. Yu, K. Kovarik, C. Keppel,J. G. Morfin, F. Olness and J. F. Owens, Phys. Rev.D 80, 094004 (2009).

[21] D. Seckel, Phys. Rev. Lett. 80, 900 (1998).[22] I. Alikhanov, Phys. Lett. B 741, 295 (2015).

[23] M. Honda, T. Kajita, K. Kasahara, S. Midorikawa andT. Sanuki, Phys. Rev. D 75, 043006 (2007).

[24] M. G. Aartsen et al. [IceCube Collaboration], Astrophys.J. 833, 3 (2016).

[25] M. G. Aartsen et al. [IceCube Collaboration], Phys. Rev.D 91, 122004 (2015).

[26] M. G. Aartsen et al. [IceCube Collaboration], Eur. Phys.J. C 77, 692 (2017).

[27] M. G. Aartsen et al. [IceCube Collaboration], Phys. Rev.Lett. 110, no. 15, 151105 (2013).

[28] A. Bhattacharya, R. Enberg, M. H. Reno, I. Sarcevic andA. Stasto, JHEP 1506, 110 (2015).

[29] M. V. Garzelli, S. Moch and G. Sigl, JHEP 1510, 115(2015).

[30] R. Gauld, J. Rojo, L. Rottoli, S. Sarkar and J. Talbert,JHEP 1602, 130 (2016).

[31] T. K. Gaisser, K. Jero, A. Karle and J. van Santen, Phys.Rev. D 90, no. 2, 023009 (2014).

[32] D. Heck et al., CORSIKA a Monte Carlo code to simulateextensive air showers., T.. Forschungszentrum KarlsruheGmbH, Karlsruhe (Germany)., Feb 1998, V + 90 p., TIBHann. D-30167 Hann. (Germany). (1998).

[33] A. M. Dziewonski and D. L. Anderson, Phys. Earth Plan-etary Interiors 25, 297 (1981).

[34] L. Radel and C. Wiebusch, Astropart. Phys. 44, 102(2013).

[35] S. Klein, Rev. Mod. Phys. 71, 1501 (1999).[36] D. Chirkin and W. Rhode, hep-ph/0407075.[37] J. H. Koehne, K. Frantzen, M. Schmitz, T. Fuchs,

W. Rhode, D. Chirkin and J. Becker Tjus, Comput. Phys.Commun. 184, 2070 (2013).

[38] T. DeYoung, S. Razzaque, and D. Cowen, Astropart.Phys. 27, 238 (2007).

[39] G. De Lellis, F. Di Capua, and P. Migliozzi, Phys. Lett.B 550, 16 (2002).

[40] C. Andreopoulous et al., arXiv:1510.05494[41] M. G. Aartsen et al. [IceCube Collaboration], JINST 9,

P03009 (2014) doi:10.1088/1748-0221/9/03/P03009.[42] D. Chirkin [IceCube Collaboration], in Proc 33rd Inter-

national Cosmic Ray Conference (ICRC2013): Rio deJaneiro, Brazil, 2013, p. 0580.

[43] M. G. Aartsen et al. [IceCube Collaboration], Nucl. In-strum. Meth. A 711, 73 (2013).

[44] M. Rongen [IceCube Collaboration], EPJ Web Conf.116, 06011 (2016).

[45] F. Halzen and S. R. Klein, Rev. Sci. Instrum. 81, 081101

20

(2010).[46] M. G. Aartsen et al. [IceCube Collaboration], JINST 12,

no. 03, P03012 (2017).[47] R. Abbasi et al. [IceCube Collaboration], Astropart.

Phys. 35, 615 (2012).[48] A. Achterberg et al. [IceCube Collaboration], Astropart.

Phys. 26, 155 (2006).[49] R. Abbasi et al. [IceCube Collaboration], Nucl. Instrum.

Meth. A 601, 294 (2009).[50] R. Abbasi et al. [IceCube Collaboration], Nucl. Instrum.

Meth. A 618, 139 (2010)[51] M. G. Aartsen et al. [IceCube Collaboration], Phys. Rev.

D 91, 022001 (2015).[52] J. van Santen, Ph. D. thesis, University of Wisconsin,

Madison (2014).[53] E. Ahn et al. Phys. Rev. D 80, 094003 (2009).[54] T. K. Gaisser, Astropart. Phys. 35, 801 (2012).[55] J. Ahrens et al. [IceCube Collaboration], Astropart.

Phys. 20, 507 (2004).[56] R. Abbasi et al. [IceCube Collaboration], Nucl. Instrum.

Meth. A 703, 190 (2013).[57] G. J. Feldman and R. D. Cousins, Phys. Rev. D 57, 3873

(1998).[58] J.A. Castro Pena, G. Parente, E. Zas, Phys. Lett. B, 50,

12, 125-132, (2001).[59] M. Tanabashi et al. (Particle Data Group), Phys. Rev.

D 98, 030001 (2018).[60] R. Enberg, M. H. Reno and I. Sarcevic, Phys. Rev.

D 78, 043005 (2008) doi:10.1103/PhysRevD.78.043005[arXiv:0806.0418 [hep-ph]].

[61] I. Esteban, M. C. Gonzalez-Garcia, M. Maltoni, et al., J.High Energy Phys. 2017, 87 (2017).

[62] T. Kashti and E. Waxman, Phys. Rev. Lett. 95, 181101(2005).

[63] M. G. Aartsen et al. [IceCube Collaboration],Phys. Rev. Lett. 115, no. 8, 081102 (2015)

doi:10.1103/PhysRevLett.115.081102.[64] H. Paukkunen and C. A. Salgado, JHEP 1007, 032

(2010) doi:10.1007/JHEP07(2010)032 [arXiv:1004.3140[hep-ph]].

[65] J. A. Formaggio and G. P. Zeller, Rev. Mod. Phys. 84,1307 (2012).

[66] N. Agafonova et al. [OPERA Collaboration], Eur. Phys.J. C 74, 2933 (2014).

[67] V. Khachatryan et al. [CMS Collaboration], Phys. Lett.B 692, 83 (2010) doi:10.1016/j.physletb.2010.07.033[arXiv:1005.5332 [hep-ex]].

[68] P. Adamson et al. [MINOS Collaboration], Phys. Rev. D76, 052003 (2007).

[69] J. Singh and J. Singh, arXiv:1709.01064 [physics.ins-det].[70] M. Aartsen et al. [IceCube Gen2 Collaboration],

arXiv:1710.01207 [astro-ph.IM].[71] S. Adrian-Martinez et al. [KM3Net Collaboration], J.

Phys. G 43, no. 8, 084001 (2016).[72] D. Becirevic, B. Panes, O. Sumensari and R. Zukanovich

Funchal, arXiv:1803.10112 [hep-ph].[73] L. A. Anchordoqui, C. A. Garca Canal, H. Goldberg, et

al., Phys. Rev. D 74, 125021 (2006).[74] L. A. Anchordoqui, M. M. Glenz, and L. Parker, Phys.

Rev. D 75, 024011 (2007).[75] M. Carena, D. Choudhury, S. Lola, et al., Phys. Rev. D

58, 095003 (1998).[76] P. Allison et al. [ARA Collaboration], Phys. Rev. D 93,

no. 8, 082003 (2016).[77] S. W. Barwick et al. [ARIANNA Collaboration], As-

tropart. Phys. 70, 12 (2015).[78] L. Gerhardt and S. R. Klein, Phys. Rev. D 82, 074017

(2010).[79] J. Alvarez-Muniz, R. A. Vazquez, and E. Zas, Phys. Rev.

D 61, 023001 (1999).