Tau Lepton Reconstruction and Identification at ATLAS FelixFriedricha onbehalfoftheATLASCollaboration Institutfu¨rKern-undTeilchenphysik,TechnicalUniversityofDresden,Dresden,Germany Abstract. TauleptonsplayanimportantroleinthephysicsprogramattheLHC.Theyareusedinsearchesfor 2 newphenomenaliketheHiggsbosonorSupersymmetryandinelectroweakmeasurements.Identifyinghadron- 1 0 icallydecayingtauleptonswithgoodperformanceisanessentialpartoftheseanalyses.Wepresentthecurrent 2 statusof thetaureconstruction and identificationat theLHC withtheATLASdetector. Thetauidentification efficienciesandtheirsystematicuncertaintiesaremeasuredusingW → τνandZ → ττevents,andcompared n withthepredictionsfromMonteCarlosimulations. a J 6 1 Introduction 2 4 ] Tau leptons are important signatures for Standard Model n / 0.00 0.1 2Z0→11ττ d+iWjet→ dτaνta ∫dt L = 130 pb-1 A3 TpLroAnSgs PprTe>li2m0in GareyV x processes and searches for new physics. With a mass of ctio 0.08 a p-e s1h.7o7rt7lGifeeVtim, tehoeft2a.u9i×s1th0e−1h3sea(vcτies=t 8le7pµtomn),atnhdetdauueletoptoitns mple Fr 0.06 e decaysinsidethebeampipeoftheLHC[1].Thetaulepton Sa 0.04 h istheonlyleptonthathasahadronicdecaymode.Whileit 0.02 [ decaysin35%ofthetimeleptonically,thehadronicdecay 1 mode occurs 65% of the time. The majority of hadronic 00 0.020.040.060.08 0.1 0.120.140.160.18 0.2 v tau decays are characterized by one or three charged pi- ∆R max 6 ons usually accompanied by neutral pions. The kinemat- Fig.1.Maximaldistancebetweenatrackandthetauaxis,∆R . 6 ics of QCD jets are similar to that ofhadronicallydecay- max 4 Only tracks inside a cone of ∆R < 0.2 around the tau axis are ing τ leptons, leading to a high potential probability for 5 considered[3]. misidentifyingthemastauleptons.Inaddition,thecross- . 1 section of most of the Standard Model and new physics 0 processeswithtauleptonsinthefinalstatearesmallcom- energyiscalculatedusingallcalorimeterclusterswithina 2 pared to the overwhelming background from QCD pro- core of∆R < 0.2 aroundthe 4-vectorsum of clustersas- 1 cesses at LHC. Therefore well performing tau identifica- sociated with the jet seed. Calibration factors are derived : v tioniscrucial.InATLAS[2],taureconstructionandiden- from response functions using Monte Carlo simulations, i tification[3]concentratesonthehadronicdecaymodesof which comefromthe ratioof reconstructedtau energyto X ataulepton.Theyareclassifiedaccordingtothenumberof true visible tau energy. Response functions are functions ar reconstructedchargeddecayparticles(prongs).Thesede- dependentonthetautransversemomentumpT,andcalcu- cayscanbedifferentiatedfromQCD jetsbytheircharac- latedseparatelyforsingle-andmulti-prongtauleptons,as teristics,suchaslowtrackmultiplicity,collimatedenergy well as for different detector regions. The systematic un- deposits, and in case of 3-prongtau leptonsthe displace- certainties on the tau energy scale are fully derived from mentofthesecondaryvertex. MonteCarloandwerefoundtobe4%–7%[3]. 3 Identification 2 Reconstruction SincethereisnoattempttoseparateQCDjetsandtaulep- Calorimeterjetswithatransverseenergylargerthan10GeV tons in the reconstruction process a dedicated identifica- and within the detector acceptanceare used as a seed for tionstepisneeded.Itisbasedonvariableswhichprovide thereconstructionoftaucandidates.Trackswithinacone discrimination power between QCD jets and tau leptons. of∆R = p(∆φ)2+(∆η)2 < 0.4aroundthetau axispass- Whilethechargedtracksfromtheτleptondecayarecol- ingcertainqualitycriteriaareassociatedtothetaucandi- limated in a narrow cone, tracks from QCD jets are dis- dateandusedtocalculatethediscriminatingvariables.The tributedmorewidely(Figure1).Theenergydepositinthe numberof tracks within ∆R < 0.2 is used to classify the calorimeter is also collimated in a small area around the taucandidateintosingle-ormulti-prongcategories.Vari- tauaxis,whileforQCDjets,alargerareaisaffected(Fig- ablesbasedoncalorimeterinformationarecalculatedfrom ure2). Therearethreeindependentmethodsfortauiden- calorimetercellsin∆R < 0.4aroundthetauaxis.Thetau tification in ATLAS: a cut-based approach, placing rect- a e-mail:[email protected] angular cuts on variables, a projective likelihood (LLH) EPJWebofConferences 8 y mple Fraction / 0.00 0000....00111.82461 Z20→11ττ d+iWjet→ dτaνta ∫dt L = 130 pb-1 A1 TpLroAnSg pPT>re2li0m iGnaerVy ackground Efficienc 110023 CBLiDukteTslihood ATLAtaSu pPerrefolirmmiannacrey Sa 0.06 erse B 10 0.04 v In 2011 dijet data ∫dt L = 130 pb-1 0.02 1-prong, 20 GeV < p ≤ 40 GeV 0 1 T 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R Signal Efficiency Cal Fig. 2. Energy weighted shower width in the calorimeter, R , Fig.4.Signalefficiencyversusinversebackgroundefficiencyfor Cal for tau signal Monte Carlo (red) and compared to QCD di-jet the different tau identification methods shown for 1-prong tau data(black)[3]. candidateswith20GeV< p <40GeV[3]. T 8 action / 0. 0.01.21 ATLAZ20→S11ττ Pd+irWjeetl→ imdτaiνtnaa ∫rydt L = 130 pb-1 3 prongs pT>20 GeV Efficiency 103 CBLiDukteTslihood ATLAtaSu pPerrefolirmmiannacrey Fr d mple 0.08 groun 102 Sa 0.06 ack B 0.04 erse 10 v 0.02 In 2011 dijet data ∫dt L = 130 pb-1 -020 -15 -10 -5 0 5 10 15 20 1 3-prong, 40 GeV < pT ≤ 100 GeV 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Likelihood Score Signal Efficiency Fig.3.Outputscoreoftheprojectivelikelihoodtauidentification Fig.5.Signalefficiencyversusinversebackgroundefficiencyfor method[3]. the different tau identification methods shown for 3-prong tau candidateswith40GeV< p <100GeV[3]. T method,usingthelog-likelihood-ratioofsignalandback- ground,andboosteddecisiontrees(BDT),tofindtheop- timal separation in a multi-dimensionalphase space. The ency 104 BDT-based electron veto ci Cut-based electron veto amreetsheopdasrautseelydtirffaeinreendtfsoertssinogflied-eanntdifimcautilotin-pvroarnigabtaleuscaannd- nd Effi 103 u o didates.Inaddition,thelikelihoodandBDTaretrainedfor gr differentnumbersofreconstructedverticesinordertotake ack 102 B ewviethntspiiglne-aulpeiffintcoieancccioeusnot.fT∼hre6e0d%ed,i∼cat4e5d%woarnkding∼po3i0n%ts nverse 10 AtaTuL pAerSforPmrealnimcienary I 1-prong, p > 20 GeV, | η| ≤ 1.37 (loose, medium, tight) are provided for all tau identifica- T tionmethods.ThelikelihoodoutputscoreisshowninFig- 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 ure 3) for 3-prong tau candidates. For the training of the Signal Efficiency identification algorithms, the QCD background was ob- Fig.6.Signalefficiencyversusinversebackgroundefficiencyfor tainedfromdata,whilethetaudecaysignalwassimulated the different tau electron veto methods shown for 1-prong tau inW →τνandZ →ττMonteCarlosamples.Theinverse candidates with p > 20GeV inthecentral (barrel) part of the T background efficiency [3] versus signal efficiency for all detector[3]. threemethodsisshownfor1-prong(Figure4)low-p and T 3-prong(Figure5)high-p taucandidates. T Electronscanalsobemisidentifiedasataulepton.Due approachusingtheeventselectionfromtheATLASZ → tothesignatureoftheelectroninthedetector,theywillbe ττcross-sectionmeasurement[4].Eventsaretaggedwitha reconstructed mostly as a 1-prong tau-candidate. To dis- muonfromataudecay,andtheothertauleptonintheevent tinguish between electrons and such tau leptons two ve- isrequiredtodecayhadronically,formingtheprobethatis toes–acut-basedandboosteddecisiontree(BDT)-based used to measurethe identificationefficiency.The electro- – are available. The performanceof these electron vetoes weakbackgroundisdominatedbyW → µνandwasesti- isshowninFigure6. matedfromMonteCarlosimulation,whiletheQCDmulti- jetbackgroundwasobtainedbyadata-drivenmethod.The visible mass of the muon and the hadronic tau is shown 4 Identification Efficiency Measurements fordatabefore(Figure7)andafter(Figure8)applyingthe tight BDT tau identification and agrees well with Monte The performance and systematic uncertainties of the tau Carlo predictions. The tau identification efficiency was identificationmethodsareevaluatedondatausingtwodif- alsomeasuredusingW →τνeventscollectedin1.37fb−1 ferentsignalchannels.ThefirstmethodusesZ →ττevents ofATLASdata.Variablesbasedonthemissingtransverse in 800pb−1 ofATLASdata andrelieson a tag-and-probe energywereusedtoselecttheevents.Thenumberofhadronic 2011HadronColliderPhysicssymposium(HCP-2011) nts / 5 GeV 111468000000 Before tau ID ∫Ads tT= L L 7=A T 8eS0V0 Pprbe-l1iminary Events3500 Dτ a(Wta→ 20 τ1ν1) (1.37 fb-1) Eve 1200 DZ→ataττ 2 (0tr1u1th-matched) 3000 e (W→ eν) 1000 Z→ττ (non-truth-matched) Jet background Multijet 2500 800 W → µ ν W → τ ν ATLASPreliminary 600 Z → µ µ 2000 400 tt Before tau ID 200 1500 0 0 20 40 60 80 100 120 140 160 180 200 1000 m (µ,τ) [GeV] vis h 500 Fig.7.Visiblemassoftheselectedmuonandtaucandidatefor data and Monte Carlo simulation after full event selection, but 0 0 2 4 6 8 10 12 14 16 beforeapplyinganytauidentification.TheQCDmulti-jetback- Number of tracks groundwasobtainedbyadata-drivenmethod[3]. Fig. 9. Number of charged tracks for tau candidates after full GeV After tight BDT ID ∫dt L = 800 pb-1 event selection, but before applying any tau identification. The 5 250 s = 7 TeV threedifferenttemplatesareshown[3]. nts / ATLASPreliminary Eve 200 DZ→ataττ 2 (0tr1u1th-matched) 150 MWZ→u →ltτiτj eµ (t nνon-truth-matched) ents1400 Data 2011 (1.37 fb-1) 100 Z → µ µ Ev1200 τ (W→ τν) tWt → τ ν e (W→ eν) 50 1000 Jet background 00 20 40 60 80 100 120 140 160 180 200 800 ATLASPreliminary m (µ,τ) [GeV] vis h 600 After tau ID Fig.8.Visiblemassoftheselectedmuonandtaucandidatefor (Tight BDT working point) dataandMonteCarlosimulationafterfulleventselectionandap- 400 plyingthetightBDTtauidentification.TheQCDmulti-jetback- groundwasobtainedbyadata-drivenmethod[3]. 200 0 0 2 4 6 8 10 12 14 16 tau candidates are derived by a template fit of the track Number of tracks multiplicityofthetaucandidates.Threedifferenttemplates were used: real hadronic tau decays, electrons misidenti- Fig. 10. Number of charged tracks for tau candidates after full fiedastauleptons,andQCDmulti-jetsmisidentifiedastau event selection and applying tight BDT tau identification. The leptons.WhilethefirsttwoareobtainedfromMonteCarlo threedifferenttemplatesareshown[3]. simulation,theQCDmulti-jettemplatewasestimatedfrom a controlregionrich in QCD events.The track multiplic- ity distributionisshownfor dataandMonte Carlobefore 2. ATLASCollaboration,JINST3,(2008)S08003 (Figure9)andafter(Figure10)applyingthetightBDTtau 3. ATLAS Collaboration, Conference identification. Note, ATLAS-CONF-2011-152, 2011, Themeasuredefficienciesinbothmethodsareingood https://cdsweb.cern.ch/record/1398195 agreementwithMonteCarlopredictionswithin5%(8%- 4. ATLASCollaboration,Phys.Rev.D84,(2011)112006, 12%)fortheW →τν(Z →ττ→µτ )method. http://arxiv.org/abs/1108.2016 had 5 Summary and Conclusion ATLAS has a large physics program with tau lepton fi- nal states, and a well performing tau identification is a essential part of these analyses. Different techniques are usedtoseparatetauleptonsfromthequarkandgluonini- tiated jet background. The multivariate methods perform betterthanasimplecut-basedapproach,especiallyfortau leptonswith a transverse momentumlarger than 40 GeV. The corresponding efficiencies and systematic uncertain- tiesofthetauidentificationmethodshavebeenstudiedus- ingStandardModelprocesses. References 1. (ed.)Evansand(ed.)Bryant,JINST3,(2008)S08001