Neutrino source searches and a realtime neutrino alert stream in the southern sky with IceCube starting tracks
NNeutrino source searches and a realtime neutrinoalert stream in the southern sky with IceCubestarting tracks
The IceCube Collaboration ∗ http://icecube.wisc.edu/collaboration/authors/icrc19_icecubeE-mail: [email protected], [email protected] IceCube analyses which look for an astrophysical neutrino signal in the southern sky face a largebackground of atmospheric muons and neutrinos created by cosmic ray air showers. By selectingstarting events in the southern sky, atmospheric muons and neutrinos with accompanying muonsare rejected, producing a sample with high astrophysical neutrino purity at lower energiesthan northern sky samples. Our new selection method looks for muon tracks from a neutrinointeraction with a vertex contained inside the detector volume by using the good pointingresolution of the track morphology to create an event specific veto region in the detector to rejectentering tracks. This starting track event selection has a high astrophysical neutrino purity above10 TeV at declinations less than -30 ◦ which makes it ideal for use as a southern sky realtimeneutrino alert stream. We will discuss neutrino point source searches using this event selectionand look at the advantages of the starting track alert stream for multimessenger astrophysics. Corresponding authors:
Sarah Mancina † , Manuel Silva University of Wisconsin-Madison36th International Cosmic Ray Conference -ICRC2019-July 24th - August 1st, 2019Madison, WI, U.S.A. ∗ For collaboration list, see PoS(ICRC2019) 1177. † Speaker. c (cid:13) Copyright owned by the author(s) under the terms of the Creative CommonsAttribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). http://pos.sissa.it/ a r X i v : . [ a s t r o - ph . H E ] A ug ceCube Southern Sky Starting Tracks Sarah Mancina
1. Introduction
The IceCube South Pole Neutrino Observatory looks for astrophysical neutrinos that lie amonga large background of muons created in cosmic ray air showers in the atmosphere. IceCube is anarray of 86 strings–each with 60 digital optical modules (DOMs)–drilled into over a cubic kilome-ter of ice at the geographic south pole. The earth can be used to reduce the muon background bylooking for up-going events which pass through the interior of the earth which is nearly transpar-ent to neutrinos. This technique is used in the realtime neutrino alert stream that discovered theIceCube 170922A event which was found in the direction of the flaring blazar TXS 0506+056 [1].However, with this method of background reduction, there is still a significant atmospheric neu-trino background even at energies up to 100TeV [2]. To study the astrophysical neutrino sky atenergies below 100TeV, it is necessary to reduce our atmospheric neutrino background even fur-ther. By looking for neutrino events with their interaction vertex inside the detector, we have theability to reject atmospheric neutrinos accompanied by muons from the same cosmic ray showersin the southern sky [3]. With this technique, we can obtain a sample of higher purity astrophysicalneutrinos in the 1TeV - 100TeV neutrino energy range.For a neutrino to be observed, it must first interact through the weak force to produce a highenergy, relativistic charged particle. Different particle interactions can create different morphologi-cal shapes in the IceCube detector. A charged-current muon neutrino interaction will create a muonthat will travel in a relatively straight path before it decays or is absorbed. We refer to muon-likemorphologies as tracks since they can traverse over a kilometer, leaving a long trace of Chernkovemission in the detector. Charged-current electron neutrino interactions and neutral current inter-actions create electromagnetic and hadronic showers. These shower morphologies appear muchmore isotropic in their light deposition than tracks and are referred to as cascades.Previous IceCube starting track event selections have used the outer layers of the detector asa veto region, in which if light is deposited, the event is rejected [4]. To obtain higher purity atneutrino energies below 100TeV, the size of the veto layers can be adjusted based on the totalamount of charge deposited by the event [5]. These veto region selections have been used toidentify a diffuse astrophysical flux [4, 5]. However, they greatly reduce the fiducial volume of thedetector at lower energies and create a cascade dominated sample for which the angular resolutionis around 5 ◦ - 20 ◦ . The event selection technique presented here uses the better angular resolutionof tracks to evaluate each event uniquely and assess if the track is starting inside the detector.This work consists of our new starting track event selection and its application to searches forneutrino sources in the southern sky at energies of 1TeV - 200TeV. The following sections willfirst focus on the event selection technique, then discuss sensitivities of time-integrated astrophys-ical neutrino source searches, and finally cover the proposed near-realtime neutrino alert streamusing this selection. This event selection has interesting implications for diffuse astrophysical neu-trino spectrum analyses [6]; however, we will focus on the strengths of this selection for studyingsouthern sky sources at medium neutrino energies.2 ceCube Southern Sky Starting Tracks Sarah Mancina
2. Starting Track Event Selection
Figure 1:
Diagram of the dark region definitionfor an atmospheric neutrino being vetoed due tothe incoming muon it is accompanied by.
To reject background atmospheric neutrinoevents, our incoming muon veto technique usesthe expected light deposition as a function of timeof minimum ionizing muons to calculate the prob-ability that an event could have been a cosmic rayair shower muon. An IceCube event is composedof photon hit information from the DOMs, whichcan be used to reconstruct the path of the parti-cle through the ice. Given a track reconstructionand set of DOM hits, we are able to define tworegions around the track: the dark region, wherethere is no light consistent with the muon tracktiming, and the muon region, where there are hitsobserved consistent with the timing of the muonhypothesis. Then, we can use light in the muonregion to scale the expected average luminosity ofthe track and calculate the probability, p miss , thatdark region DOMs did not observe charge if theevent was an incoming muon.We first delineate the muon region and dark regions as illustrated in figure 1 and 2. Given areconstructed track hypothesis, we can calculate the expected photo-electron (PE) yield over timeseen by DOMs from a minimum-ionizing muon. The photon expectation includes modeling ofthe Antarctic ice properties. A time window is constructed for each DOM around the peak of theexpected charge to contain 99% of the total charge. If a DOMs observed pulses during its timewindow, we categorize that DOM as seeing hits consistent with the track hypothesis. The pathbetween DOMs and the track are traced along the Cherenkov angle to find the point at which un-scattered light would have been emitted. We select the first point along the track where Cherenkovlight could have been emitted and hit one of the DOMs which observed light consistent with thetrack hypothesis to determine the start of the track. To define our two regions we first considerDOMs within 350m of the track. Then, we construct a Cherenkov cone for light radiating out ofthe reconstructed interaction vertex. The muon region is all DOMs that within of the cone and350m of the track, and the dark region is all DOMs that lie behind the cone and within 350m of thetrack.Second, we must calculate the probability that the DOMs in the dark region did not observecharges from our track hypothesis. For a given DOM, we expect the probability of observing anumber of PE, k , to be poisson distributed with an mean of λ : p ( λ , k ) = e − λ λ k / k !. We can definethe probability that dark region DOMs did not see charges, p miss , as the product of probabilities3 ceCube Southern Sky Starting Tracks Sarah Mancina
Figure 2:
A two-dimensional diagram of our starting track event selection technique. Each DOM has anexpected charge distribution as a function of time (red curve). For all DOMs within 350m of the track, ifthe observed photo-electrons (blue lines) are seen within the time window (grey box) they are classified asconsistent with the muon track hypothesis. that those DOMs saw zero charge from our track hypothesis: p miss = Dark Region DOMs ∏ i log ( p ( λ i , k = )) (2.1)where λ i is defined as λ i ( a ) = a × λ muon + λ noise . The scale factor, a , is calculated by maximizingthe likelihood in equation 2.2. This allows us to use the hit information in the muon region to inferthe average luminosity of the muon with this simplified model of the stochastic energy losses.LLH = Muon Region DOMs ∑ i log ( p ( λ i ( a ) , k i )) (2.2)With the scale factor, a , determined we can calculate p miss in equation 2.1, which is used later inthe event selection. The starting track event selection is built around the p miss parameter; however, we must testseveral track hypothesis and use a boosted decision tree to obtain the desired purity. The IceCubedetector strings are spaced 125m apart on a triangular gird to create a hexagonal shaped detector.The large spacing between strings allows muons to sneak past the external layers of the detector.Therefore, the selection must check if the tracks could have entered into the holes in the detectorby testing several possible paths through the alleyways of the detector.First, we select events which pass through simple filters that look for muon-like events or start-ing events and require a minimum amount of charge deposited in the detector. We then calculate p miss for an pre-run track reconstruction and reject events with a p miss greater than 10 − .After the quick initial cuts, the event selection generates test paths that a muon could takethrough alleyways of the detector. A point 300m down the track from the event center of charge is4 ceCube Southern Sky Starting Tracks Sarah Mancina
Atmospheric µ Atmospheric ν Astrophysical ν Up-going ( δ > − ◦ ) 0 127 8Down-going ( δ ≤ − ◦ ) 0.8 33 8 Table 1:
Final level expected events per year assuming an astrophysical flux of 2 . × − (cid:0) E TeV (cid:1) − . [GeV − cm − s − sr − ] from [5]. The atmospheric neutrino self-veto effect in the southern sky is modeledby forcing neutrinos in CORSIKA simulation to interact [7]. defined as the shifted center of charge. We take a table of positions along the edge of the detectorin between the strings of the detector and trace a path to the shifted center of charge to create a testtrack. The reduced log likelihood (rLLH) of the reconstruction is calculated for all of the test tracksand select tracks with an rLLH within 2% of the maximum rLLH of all the tracks to calculated their p miss . If the minimum p miss of all the tracks is less than 10 − the event is kept in the next stage.These same best rLLH test tracks are split into 131 segments, and for each segment the ex-pected photon yield as a function time is calculated for each DOM to create the time windows usedin calculating p miss . By splitting the track into segments, the new timing information provides amore accurate estimation for the start of the muon region and affects λ muon of λ i in equation 2.1. Acut is made on the minimum p miss obtained in this step requiring it to be less than 10 − .The final test of alleyways is a finer scan of track directions around the top rLLH test tracksfrom the coarser scan. The shifted center of charge, zenith, and azimuth of the top test tracks areslightly shifted around to test a total of 1625 fine search tracks around each initial test track fromthe coarser search. Again, tracks with an rLLH within 2% of the max rLLH are kept and their p miss calculated. The track is kept if, after splitting the tracks into 131 segments like above, the minimum p miss is less that 10 − .The initial track reconstruction is then fed to a stochastic muon energy loss reconstruction.For events with a zenith of less than 80 ◦ ( δ < − ◦ ), the energy loss information, p miss values, andother quality information are fed to a boosted decision tree trained to distinguish between muonand starting neutrino events [6]. The parameters that are most important in the BDT are the fractionof energy lost in the first reconstructed loss and the distance of the reconstructed start of the trackto the edge of the detector. Some simple quality cuts are run on the up-going events. This bringsthe event selection to its final level.At the final level of our selection we are left with a relatively pure sample of astrophysicalneutrinos in the southern sky. We expect less than one atmospheric muon per year in the eventselection as seen in Table 1. This event selection has the largest neutrino effective area at declina-tions of less that 30 ◦ between 8TeV and 200TeV (Figure 3). Other IceCube event selections areforced to cut strongly on energy in the southern sky in order to reduce the muon background [8].The northern sky starting track event selection has a similar effective area as in the southern sky,which is not competitive with the regular through-going point source tracks event selection andthere is a large overlap of events in the two selections in the northern sky. The starting track eventselection also suppresses atmospheric neutrinos in the 10TeV - 100TeV energy range. This resultsin a higher purity of astrophysical neutrinos at these energies.The direction and energy reconstruction resolution are affected by the nature of starting tracks.The average angular error is 1.7 ◦ . The angular error is highly correlated with the length of the5 ceCube Southern Sky Starting Tracks Sarah Mancina
Figure 3:
Left: Effective area of the starting track event selection in the southern sky (red) binned in trueneutrino energy. The effective area is greater than the IceCube through-going event selection (blue) [8] inthe southern sky at energies below 200TeV. Right: Cummulative and differential distribution of events peryear as a function of true neutrino energy from − ◦ to − ◦ assuming the flux from [5]. The atmosphericself-veto is responsible for the suppression of atmospheric neutrinos at TeV scales. track in the detector, and southern events ( δ < -70 ◦ ) tend to have shorter lengths due to stricterrequirements on the distance of the start of the track to the edge of the detector from the BDT.Therfore, the angular resolution is slightly worse in these most southern declinations with a medianangular error of 3.0 ◦ . The neutrino energy reconstruction resolution is 0.25 in the log ( E ν ) spaceat all neutrino energies from 1TeV to 1PeV. The through-going track selection uses a reconstructionthat has a muon energy resolution of 0.22 in the log ( E µ ) space [9]. The improved neutrino energyresolution for starting tracks is due to the addition of information from the hadronic cascade at thestart of events.
3. Neutrino Source Searches
For our time-integrated neutrino source searches, we plan to do an all-sky search, catalogsearch, stacking search, and galactic plane template fit. To search for point-like sources we use aunbinned maximum likelihood method to estimate the number of signal neutrinos at the potentialsource location in the sky like previous neutrino searches [8].In the all-sky search, we scan a HEALPix grid of pixels that are 1 . × − steradians in area.To calculate the pre-trial sensitivities, first, we create a background test statistic distribution andfind the median test statistics for each test declination. Then, we inject signal events with a powerlaw energy distribution and find the flux normalization for which 90% of the injected trials arelarger than the median test statistic.In figure 4 the sensitivities are shown for a source with a spectral index of 2 for the normal-ization at 100TeV. As expected, the starting track sensitivity becomes competitive in the southernsky where the through-going tracks become less sensitive. The improved sensitivity is more ap-parent when the source spectral index increases to 2.5 or 3 because we then expect more neutrinosat energies below 100TeV and the starting track event selection’s effective area is better than the6 ceCube Southern Sky Starting Tracks Sarah Mancina
Fermi π [12] KRA γ [13] KRA γ [13] ( TeV − cm − s − ) (5 PeV Cutoff) (50 PeV Cutoff)Starting Tracks (8 Years) 2 . × − . × KRA γ . × KRA γ Through-Going Tracks (7 Years) [11] 2 . × − - 79% × KRA γ Table 2:
Comparison of the galactic plane template model sensitivities for this starting track event selectionand the published icecube through-going tracks event selection. For the Fermi π template, a power lawwith a spectral index of 2.5 is assumed and for the KRA γ template the spectrum shown in [13] is used. through-going tracks sample in the mid-TeV range (figure 3). Similarly the event selection alsoimproves current limits in the southern sky when an exponential energy cutoff is applied to thesource energy spectrum. There are plans to combine both event selections to increase sensitivity inthe future. Figure 4:
Pre-trial sensitivities for the start-ing tracks compared to the sensitivities from [8]and [10].
For the galactic template analysis we usethe unbinned likelihood with signal subtraction asused in [11]. In this method we test for a sig-nal from two models of the diffuse galactic planeneutrino emission: the Fermi π model [12] andthe KRA γ model [13]. A comparison to the pre-viously published results with the 7 year templatefit is shown in table 2. The starting tracks havea greater sensitivity due to the location of denserregions of the galaxy in the southern sky and theshape of the energy spectrum of the galactic planeemission models.
4. Realtime Alert Stream
Running the full selection at the South Polein realtime is not possible due to its time andmemory requirements, so a modified version isrun instead. The modified selection only looks forneutrinos from the southern sky ( δ ≤ − ◦ ). Ifan event passes the modified version at the southpole, the candidate event is then sent North to be processed by the full selection. The modified se-lection requires a larger amount of energy deposited in the detector and a longer track length thanthe offline version; therefore, events that pass both the modified and full selection have a higherpurity and quality than the regular full offline sample.The modified online selection triggers on average 16.8 times per day. Most of these events areexpected to be muons. From simulation, we estimate approximately 17.9 atmospheric neutrinosper year and 5.5 astrophysical neutrinos per year with 50% signalness or greater to pass throughthe modified selection assuming the astrophysical flux from [5]. The signalness is calculated bytaking the percentage of astrophysical neutrinos to the total number of neutrinos from simulation7 ceCube Southern Sky Starting Tracks Sarah Mancina that lie in the same reconstructed energy and declination at the final event selection level. The 50%signalness events have energies in the 10TeV to 200TeV range, which is a lower range than thecurrently running alerts [14]. Due to the low energy range, these events could be of great interestfor galactic transient events, especially since this stream is looking at the southern sky where someof the most active parts of the galaxy are located.
5. Concluding Remarks
The starting track event selection takes advantage of the pointing resolution of IceCube muontracks to evaluate event by event the probability that the track is an incoming cosmic ray muon.This allows us to reject not only muons from cosmic rays, but also atmospheric neutrinos withaccompanying muons. Our new selection can probe the southern sky at lower neutrino energies dueto the supression of atmospheric neutrinos. The high purity of our event selection makes it a greatcandidate for a realtime neutrino alert stream. The realtime starting track events are anticipated tobe added to IceCube’s community alerts soon.
References [1]
IceCube
Collaboration, M. G. Aartsen et al.,
Science (2018) 147–151.[2]
IceCube
Collaboration, M. G. Aartsen et al.,
Phys. Rev. Lett. (2015) 081102.[3] C. A. Argüelles, S. Palomares-Ruiz, A. Schneider, L. Wille, and T. Yuan,
JCAP (2018) 047.[4]
IceCube
Collaboration, M. G. Aartsen et al.,
Science (2013) 1242856.[5]
IceCube
Collaboration, M. G. Aartsen et al.,
Phys. Rev.
D91 (2015) 022001.[6]
IceCube
Collaboration,
PoS(ICRC2019)1010 (these proceedings).[7] D. Heck, J. Knapp, J. N. Capdevielle, G. Schatz, and T. Thouw.[8]
IceCube
Collaboration, M. G. Aartsen et al.,
Astrophys. J. (2017) 151.[9]
IceCube
Collaboration, R. Abbasi et al.,
Nucl. Instrum. Meth.
A703 (2013) 190–198.[10]
ANTARES, IceCube
Collaboration, S. Adrian-Martinez et al.,
Astrophys. J. (2016) 65.[11]
IceCube
Collaboration, M. G. Aartsen et al.,
Astrophys. J. (2017) 67.[12]
Fermi-LAT
Collaboration, M. Ackermann et al.,
Astrophys. J. (2012) 3.[13] D. Gaggero, D. Grasso, A. Marinelli, A. Urbano, and M. Valli,
Astrophys. J. (2015) L25.[14]
IceCube
Collaboration, M. G. Aartsen et al.,
Astropart. Phys. (2017) 30–41.(2017) 30–41.