30TH INTERNATIONAL COSMIC RAY CONFERENCE

A New Background Rejection Technique for the Milagro Gamma-Ray Detector

A.A.A BDO1 FOR THEM ILAGRO COLLABORATION.1 Department of Physics and Astronomy, Michigan State University, 3245 BioMedical Physical SciencesBuilding, East Lansing, MI 48824abdo@pa.msu.edu

Abstract: A new background rejection technique that significantly increases the sensitivity of the Milagrodetector has been developed. This technique, combined withother improvements, improves the sensitivityof the Milagro detector by more than a factor of 2.5 over the previous technique [4]. This new techniquedifferentiates between hadronic and gamma-ray showers by looking at the fundamental differences inthe shower parameters between these two types of showers andhow they register in the detector. Theseshower parameters include the number of Muons presented in the EAS, the size of the EAS, and someshower reconstruction parameters. This technique resulted in discoveries of localized TeV gamma-raysources from the Galactic plane [3]. Details of the new technique along with an all-sky TeV gamma-raymap –using this technique– will be presented.

Introduction

Milagro is a water-Cherenkov detector at an alti-tude of 2630m capable of continuously monitoringthe overhead sky. The Milagro detector is com-posed of a central 80m x 60m x 8m (depth) pondwith a sparse 200m x 200m array of 175 “outrig-ger” tanks surrounding it. The central pond is in-strumented by 723 photo multiplier tubes (PMTs)that are submerged in 24 million liters of puri-fied water which acts as the detection mediumfor the secondary particles in an air shower. The723 PMTs are split into two layers: the top “airshower” layer consists of 450 PMTs and is under1.4m of water, and the bottom “muon” layer has175 PMTs and is located 6m under the surface ofthe water. The outrigger array extends the phys-ical area of Milagro from 5000m2 to 40,000m2,thus improving the core location and angular res-olution of the detector by providing a longer leverarm with which to reconstruct events. Extensiveair showers initiated by charged cosmic-ray parti-cles out number those of cosmic gamma rays by afactor of∼1000 at the altitude of Milagro. A suc-cessful detection of a celestial gamma-ray signalrequires the proper identification and rejection ofthis large cosmic-ray background.

Identification and Rejection of HadronicEvents

It is well known that extensive air showers inducedby hadronic cosmic rays contain many more muons(from pion decay) and hadrons than EAS inducedby gamma rays of comparable energies. The at-mospheric overburden at the Milagro site is 20.5radiation lengths and 8.3 interaction lengths. Themuon layer in Milagro is located under 6m of wa-ter (corresponding to 17 more radiation lengths and7.2 interaction lengths). This means that all EMcharged particles that enter the pond get absorbedbefore reaching this layer. On the other hand,muons with energies as low as 1.2 GeV can pen-etrate and radiate Cherenkov light near the PMTsof the muon layer. These penetrating muons andhadrons will result in bright compact clusters oflight in this layer. Using Monte Carlo simulationswe estimate that 79% of all proton showers thattrigger Milagro contain a muon and/or a hadronthat enters the pond, while only 6% of gammaray induced air showers contain a muon and/or ahadron that enters the pond.

A new parameter has been developed to parame-terize the “clumpiness” in the muon layer. Thisparameter, A4, is given by[1, 2]:

ICR

C 2007 P

roceedings - Pre-C

onference Edition

BACKGROUND REJECTION INM ILAGRO

A4 =(ftop + fout) × Nfit

mxPE(1)

where

• ftop is the fraction of the air shower layerPMTs hit in an event.

• fout is the fraction of the outriggers hit in anevent.

• Nfit is the number of PMTs that entered inthe angle fit.

• mxPE is the number of photoelectrons(PEs) in the muon layer PMT with the high-est hit.

The first part in the numerator of A4 carries infor-mation about the size of the shower, while Nfitcarries information about how well the shower wasreconstructed. mxPE carries information about theclumpiness in the muon layer. Figure 1 shows A4distributions for different types of showers. TheA4 distribution of Monte Carlo cosmic-ray eventsare shown in red, Monte Carlo gamma-ray eventsare shown in blue, and data in black. The greenline represents the Quality Factor1 as a function ofA4. The A4 distribution for gamma-ray events ex-tends to larger values of A4 because these showerscontain fewer penetrating particles, such as muonsand hadrons, which produce localized light in thebottom layer to give large values of mxPE withoutgreatly increasing the number of hit PMTs. Re-quiring events to have A4 ≥ 3.0 rejects96.5%of the simulated cosmic-ray-induced air showersthat trigger Milagro and96.3% of the data (for thisdata sample), and retains32% of the gamma-ray-induced air showers. This results in a predictedQ-factor of 1.7 comparing Monte Carlo cosmic-ray events to Monte Carlo gamma-ray events, and1.66 comparing the data to Monte Carlo gamma-ray events.

Tests of A4 on the Crab Nebula

The Crab Nebula acts as a standard candle forgamma-ray astronomy due to its long-term, un-changing energy emission through many wave-

4A0 2 4 6 8 10 12

4F

ract

ion

of

Eve

nts

wit

h >

A

0

0.2

0.4

0.6

0.8

1

Qu

alir

y F

acto

r

0

0.5

1

1.5

2

2.5

3

3.5MC Gamma RaysMC Cosmic RaysDataQuality Factor

Figure 1: Distribution of A4 for different typesof showers. The A4 distribution of Monte Carlocosmic-ray events are shown in red, Monte Carlogamma-ray events are shown in blue, and data inblack. The green line represents the Quality Factoras a function of A4.

lengths. Thus, the performance of A4 is best veri-fied on this source. 1.7 years of data was searchedfor a TeV gamma-ray signal from the Crab Neb-ula where the A4 ≥ 3.0 was applied. 2074 sig-nal events were observed over the 60637 back-ground events. This corresponds to 3.42% signal-to-background ratio and statistical significance of8.02σ. when a harder A4 cut of 12 was applied onthe same data set, 44 signal events where observedon top of the 75 background events. This corre-sponds to 60% signal-to-background ratio and sta-tistical significance of 5.6σ. Although there wasa∼ 30% loss in statistical significance of the CrabNebula when the harder A4 cut was applied, themain advantage of applying this cut is the higherS/B ratio achieved with this cut compared to thatwith the soft A4 cut.

1. The Quality Factor (Q) is a measure of the relativeincrease in significance of a signal for a given selectioncriterion. For a large number of events, Q is given by:

Q =ǫs√

ǫb(2)

Whereǫs andǫb are the efficiencies for retaining thesignal and background events, respectively.

ICR

C 2007 P

roceedings - Pre-C

onference Edition

30TH INTERNATIONAL COSMIC RAY CONFERENCE

A4 Weighting Analysis

The fact that one can achieve higher S/B ratios withno major loss in statistical significance when ap-plying harder A4 cuts led to the development ofan alternative analysis method that weights eventsbased on the relative probability that the event wasdue to a gamma-ray. Therefore, instead of countingall the events in an angular bin equally, a weightedsum of events is used in which events with highervalues of A4 are assigned higher weights. The sig-nificance is computed using the method describedin Li & Ma [5], where background fluctuationsare similarly estimated for the sum of the weightsof the background sample, rather than the back-ground event count. The statistical significance de-rived from this analysis was verified both by MonteCarlo simulation and through the study of statis-tical fluctuations in the background data sample.The values of the weights used in this analysis aredetermined from the predicted S/B ratio as cal-culated a priori from our detector simulation foran incident Crab-like spectrum.With this weight-ing scheme, this analysis is equivalent to a likeli-hood ratio method estimation in the limit that thebackground is large, which is true for the Milagrodata.

Figure 2 shows a map of the statistical significancearound the Crab Nebula with the new A4 weight-ing analysis method applied. The statistical signif-icance at the location of the Crab in this map is10.55σ.

The weighted analysis enhances the contribution ofhigher A4 events that are, on average, higher en-ergy gamma rays. The median energy with thisanalysis is 12 TeV for a Crab-like spectrum witha differential photon spectral index of E−2.6. Bycomparison, a previously published Milagro anal-ysis had a median energy of 3-4 TeV [4]. The an-gular resolution improves from 0.72◦ to 0.5◦ whenthis analysis method is used.

Conclusions

The combination of the installation of the outrig-ger array, the adoption of the A4 discriminant, andthe event weighting increases the sensitivity of Mi-

RA (deg)72 74 76 78 80 82 84 86 88 90 92 94

Dec

linat

ion

(d

eg)

10

15

20

25

30

-4

-2

0

2

4

6

8

10

Map of Significances

Figure 2: Map of the statistical significance aroundthe Crab Nebula with the new A4 weighting anal-ysis method applied. The statistical significance atthe location of the Crab in this map is 10.55σ.

lagro as compared to previous analysis by∼ 2.5times as confirmed by observation of the Crab.

Acknowledgments

We acknowledge Scott Delay and Michael Schnei-der for their dedicated efforts in the construc-tion and maintenance of the Milagro experiment.This work has been supported by the NationalScience Foundation (under grants PHY-0245234,-0302000, -0400424, -0504201, -0601080, andATM-0002744) the US Department of Energy (Of-fice of High-Energy Physics and Office of Nu-clear Physics), Los Alamos National Laboratory,the University of California, and the Institute ofGeophysics and Planetary Physics.

References

[1] A. A. Abdo. Detection of tev gamma-rayemission from the cygnus region of the galaxywith milagro using a new background rejectiontechnique.AIP Conf. Proc. 867, Calorimetryin High Energy Physics: 12th Int. Conf., ed.

ICR

C 2007 P

roceedings - Pre-C

onference Edition

BACKGROUND REJECTION INM ILAGRO

S. R. Magill and R. Yoshida (New York: AIP),199.

[2] A. A. Abdo. Discovery of Localized TeVGamma-Ray Sources and Diffuse TeV Gamma-Ray Emission from the Galactic Plane withMilagro using a new Background RejectionTechnique. PhD. Thesis, Michigan StateUnieversity, 2007.

[3] A. A. et al. Abdo. Discovery of tev gamma-rayemission from the cygnus region of the galaxy.Astrophysical Journal Letters 658 (2007) L33-L36.

[4] R. et al. Atkins. Observation of tev gammarays from the crab nebula with milagro usinga new background rejection technique.Astro-physical Journal 595 (2003) 803-811.

[5] T.P. Li and Y.Q. Ma. Analysis methods for re-sults in gamma-ray astronomy.ApJ, 272, 317.

ICR

C 2007 P

roceedings - Pre-C

onference Edition