Jet Finding Algorithmと Jet Energy Calibration の課題研究 田中 礼三郎 (岡山大学) アトラス日本夏の学校@CERN 2003年8月22日 内容 1. 2. 3. TevatronとLHC Wボソン,トップクォークの質量測定 Jet Finding Algorithm Cone, kT algorithm 4. Jet Energy Correction LEP, HERA, Tevatron, LHC 5. 纏め 1.TevatronとLHC Tevatron collider in Run II The Tevatron is a protonantiproton collider with 980 GeV/beam s 1.96TeV in RunII (1.8TeV RunI) 36 p and p bunches 396 ns between bunch crossing Increased from 6x6 bunches with 3.5ms in Run I Increased instantaneous luminosity: Run II goal 30 x 1031 cm–2 s-1 Current: 3~4.5 x 1031 cm–2 s-1 Run II Data Taking Status Lint~300 pb-1 delivered by the Tevatron Good quality data since Spring 2002 Data collection efficiency 85~90% Next Year projection: additional 310~380pb-1 delivered Tevatron Collaborations 12 countries, 62 institutions 767 physicist 19 countries 83 institutions, 664 physicists The CDFII Detector RETAINED FROM CDF in RUN I Solenoidal magnet (1.4 Tesla) Central calorimeters Central muon detectors NEW FOR CDF in RUN II Tracking system Silicon vertex detector (SVXII) Intermediate silicon layers (ISL) Central outer tracker (COT) Scintillating tile end plug calorimeter Intermediate muon detectors Scintillator time of flight system Front-end electronics (132 ns) Trigger system (pipelined) DAQ system All detectors inside the solenoid are new Others are from Run 1 with enhancements wire drift chamber (96 layers) m TOF System 2.0 A new scintillating tile plug calorimeter covering |η| out to 3.6. END PLUG HADRON CALORIMETER • END WALL HADRON CAL. SOLENOID 1.5 1.0 END PLUG EM CALORIMETER A new 3d tracking system and vertex detector covering |η|out to 2.0. n = 1.0 COT .5 30 n = 2.0 n = 3.0 3 0 0 0 .5 1.0 Inner silicon 6 layers 1.5 2.0 2.5 3.0 Intermediate silicon 1 or 2 layers 0 m D0 Silicon Microstrip Tracker (SMT) ATLAS (A Toroidal LHC ApparatuS) Liq.Ar EM calorimeter good e/g id, energy, ETmiss muon spectrometer air-core troidal magnet Bdl = 2~6Tm (4~8Tm) inner tracking system pixel, silicon strip, TRT 2T solenoid magnet good e/g id, t/b-tag ATLAS detector tracker |h| < 2.5 calorimeter |h| < 4.9 CMS (Compact Muon Solenoid) 4T solenoid Compact muon spectrometer EM calorimeter PbWO4 for Hgg ABC at hadron collider We never know total longitudinal momentum in any event. Total transverse momentum of all particles is zero. transverse momentum pT = |p| sinq transverse energy ET = E sinq pseudo-rapidity h = -ln tan(q/2) missing transverse energy ETmiss = En Distance in pseudorapidity-azimuthal angle space (used in jet cone algorithm) DR=(D h)2 +(D)2 Existence of minimum bias events. LHC: inelastic, non-diffractive cross section: s70mb 23 pile up/crossing@1034 Tevatron RUN-II: 6 pile-up/crossing (Poisson) Troid + 2T solenoid 4T solenoid Challenge for tracking HZZ4m Jet Energy Flow Particle type g e m Jet Et miss Tracking ECAL HCAL Muon μのエネルギーロス b-tag Vertex detector Soft lepton tagging b-quarks have a long lifetime: t(b) ~ 1.5ps (ct~450mm) B-tagging using displaced vertices identifies lepton in semi-leptonic b(or c) decays leptons are softer less isolated than from W/Z decay. CDF RUN2a: b = 60% , c = 25%, j = 0.2% ATLAS: b = 60(50)% for low (high) lumi. RUN2b: b = 70% , c = 10%, j = 0.02% c = 10%, j = 1% 2.Wボソン,トップ クォークの質量測定 MW measurement W transverse mass mTW 2 pTl pTn (1 cos D ) l n pT | pT u | major uncertainty source E, p scale & resolution use Ze+e-/m+m-, , J/, (2s) Recoil modelling ISR(QCD), spectator quarks,min. bias exploit similar production mechanism for W and Z pTW use of lepton pseudorapidity distributions in W and Z decays PDF (parton distribution) estimate from Z data MW measurement at RUN-I/LHC Energy and momentum scale/resolution Ze+e-, Zm+m- , J/, (2s) Recoil modelling neutrino PT imbalance recoil from ISR(QCD) spectator quarks additional minimum bias Exploit similar production mechanism for W and Z. Parton distribution functions (PDF) x-region of W production asymmetry u(x)>d(x) W+(W-) boosted along p(p-bar) use of W charge asymmetry data to constrain PDF such an asymmetry does not exist at LHC(pp) ! use of lepton pseudorapidity distributions in W and Z decays constraint PDF to few % DMW < 10 MeV W production model pTW pTW is estimated from Z data d 2s W 2 2 d s d s dpT dy dpTW dy dpTZ dy data d 2s Z dpT dy theory error DMW=20 MeV dominated by Z statistics theoretical error (5 MeV) Top production cross section top factory s 2TeV s NLO 7 pb s 14TeV s NLO 800pb stot=70mb for LHC 109 interactions/sec@1034cm2s-1 Interesting physics W production: ~2kHz Top production: 10Hz Higgs production: 0.1(0.01)Hz for MH=100(500) GeV s ( s) ij 1 dx dx 0 1 ^ ^ f ( x1 ) f i ( x2 ) s ij ( s, mt ) 2 i PDF: fi(x1),fi(x2) xi is momentum fraction of parton i. ^ s x1 x2 s Tevatron qq(90%), gg(10%) RUN-I qq(85%), gg(15%) RUN-II LHC qq( 5%), gg(95%) enhanced gluon structure function. LO CDF実験 RUN1 トップクォーク質量 最大の系統誤差は, Jet Energy Calibration →LHCでも同じ。 Top, Higgs, SUSY… PRD63(2001)032003 Double b-tagged dilepton event @ CDF 69.7 First look at top mass in Run II CDF RunII preliminary, 108 pb-1 CDF RunII preliminary, 126 pb-1 Data 22 evts 6 events Mass in lepton+jets channel with a b-tagged jet 2 177.5 12.7 (stat) 7.1(syst) GeV/c 9.4 Mass in dilepton channel 2 175.0 17.4 (stat) 7.9(syst) GeV/c 16.9 3.Jet Finding Algorithm Jet Finding Calorimeter jet (cone) jet is a collection of energy deposits with a given cone R: R Δ 2 Δη2 cone direction maximizes the total ET of the jet various clustering algorithms correct for finite energy resolution subtract underlying event add out of cone energy Particle jet a spread of particles running roughly in the same direction as the parton after hadronization Jet Algorithms Fixed Cone (RunI) kT (Ellis-Soper) • Iterative • Recombinant • Fixed cone of radius R • Distance parameter D • Overlapping cones: split/merge parameter • Sensitivity to soft radiation • Infrared and collinear safe in principle kT Algorithm d ij min( PT2,i , PT2, j ) Ellis-Soper PRD48(1993) 3160 DRij2 D2 preclusters KT jet Cone jet A cone jet is just the highest-ET stable cone… final jets hep-ex/0005012 Jet Algorithms: kT For each object and pair of objects: dii k 2T,i order all dii and dij: If dmin=dij 2 ΔR ij dij min(k 2T,i ,k 2T, j ) 2 D Soft Collinear merge particles (if DR<<1 ) Resolution parameter (D=1) theoretically favored, no split-merge to reduce computation time, start with 0.2 x 0.2 pre-clusters x-section measurement differ from cone-jet If dmin=dii jet (JETRAD) DØ Subjet multiplicity of gluon and quark jets reconstructed using the kT algorithm in pbarp collisions Phys. Rev. D65 052008 (2002) hep-ex/0108054 The inclusive jet cross section in pbarp collisions at sqrt(s)=1.8 TeV using the kT algorithm Phys. Lett. B {525}, 211 (2002) hep-ex/0109041 Inclusive Jet Cross Section First analysis you do…count jets in pT bins Central region has large cross section, well-controlled systematic uncertainties d s N Csmear dET dh L DpT Dh 2 versus pT Results for kT and cone D0 Each distribution is compared to its own prediction Uncertainties highlycorrelated from one bin to the next normalization not well-determined, but shape is Important deviation from cone and from predictions at low-pT Jet Finding Algorithmの歴史 UA2: fixed size cones, R=1.3 ± 30 % cross section uncertainty TeVatron Run 1: cones R=0.7, merge/split factor of 2 or more improvement in precision End of Run 1: DØ looks at kT algorithm, similar to ones used by HERA experiments CDF/DØ attempt to improve both algorithms and achieve consistency 4.Jet Energy Correction ① LEP Energy Flow Total energy in the Jets ETOT=pe+ pm + pcharged hadron + Eg + Eneutral hadron [ tracks only] [calorimeter only] to improve the energy flow resolution, the neutral particle id such as g(0), neutron, K0L is most important, this is achieved with fine granular and hermetic calorimeter design. e/ ratio can be corrected to unity with software correction (i.e. don't need to construct Scinti:Pb=1:4 calorimetre for hardware compensation). s Energy Flow 0.59 E (GeV ) (1.2 E (GeV ) if ETOT EECAL EHCAL ) E-flow algorithm M-N. Minard http://3w.hep.caltech.edu/calor02/ - Good track resolution - Calorimeter segmentation is used to identify differents contribution: * Charged tracks & identified lepton * g ( and 0) * hadronic neutral * Residual from g, charged hadron Energy resolution from Z->qqg events s(E)/E = (0.59+-0.03)/E + (0.6+-0.3) GeV Expected resolution at high energy is derived Jet reconstruction Algorithm used Durham: - E-flow object with yij = 2 min(Ei2,Ej2)(1-cosqij)/Ecm2 < ycut associated in the same jet - WW-> q1q2q ’1q ’2 forced into 4 jets Jet performances studied from Z data - Energy response and resolution d(Ejet)/Ejet = 0.67/Ejet (10% at 45 GeV- perfect detector:6-7%) - Angular resolution 0.9° with Eflow charged track only : 1.6°, calorimeters : 1.4° ② HERA H1/ZEUS Jet Algorithm hep-ph/0211298 ③ Tevatron Jet Energy Correction Relative correction 検出器の相対的な補正 Absolute correction 検出器の絶対的な補正 Underlying event subtraction ミニマムバイアス,マルチプル散乱 Out-of-cone addition コーン外側のエネルギー補正 Physics effects & Detector response Phycics effects Natural W width Underlying event fluctuation Final State Radiation (FSR) Initial State gluon Radiation (ISR) "Halloween" Photon + Jet Event (seen October 24,1994) This event has a 311 GeV photon opposite a 295 GeV jet. The photon + jet mass is at least 0.76 TeV, not bad for a 1.8 TeV pbar-p collision! Energy Flowとγ+Jetによる補正 Jet Energy Scale jet g correct Jet Energy to the particle level ptcl E jet E meas jet E offset shower R calo R jet jet Eoffset energy offset from underlying event, pile-up, noise determined from Min. Bias Events Rcalo calorimeter response using g-jet events: Missing ET Projection Fraction Method Rshower energy contained in jet corrections from MC - energy in cones around the jet axis depending on jet algorithm! Determination of the Absolute Jet Energy Scale in the D0 Calorimeters. NIM A424, 352 (1999), hep-ex/9805009 Run II: g+jet / Z+jet g/Z+jet QCD (udsg) signal:152 M evts bkgd: 47.1 M evts signal: 64.8k evts bkgd: 650 evts QCD (cbt) W+jet, Z+X, nn • g+jet: Run I method – jet calibration possible up to 250 GeV • Z+jet: lower statistics, but clean sample, useful at low energies, x-check! b-jet calibration naïve reconstruction of Z-mass shows a lower mass for selected b-jets than light quark jets. energy losses from semi-leptonic b decays (n, m) wider b-jets (due to the large b-mass) Z bb peak: 82.6 Z qq peak: 86.8 Z bb vs g + b-jet g + b-jet : Zbb: high statistics, allows for a tight b-jet selection (btagging). expected number of tagged events: 1.2 M but: sensitive fractional imbalance I= (pT(g) - ET(jet))/ pT(g) systematics closer to physics processes (H or Top) at high pT resonance mass independent of multiple interactions. but: signal/noise~10-3 requires special trigger (Silicon Track Trigger – operational end 2002) CDF Run 1: Z bb Signal after cuts: S/N=1/6 at the Z mass peak select/antiselect w.r.t. the 2 variables to determine the tagging probability 3.2 s exces ④ LHC ATLAS ATLAS実験 Jet / ET miss / Tau Combined Performance WG Martine Bosman,Donatella Cavalli,Frank Paigeら。 Cone and KT jet algorithms がAthenaに入っている。 JetRec, TauRec, EmisRec and EflowRec in Athena。 H1の較正方法を採用している。 検出器のノイズやpile-upの効果の研究がなされている。 SUSYグループと協力。 Calibration Workshop at Ringberg Castle (Germany) 21-24th July http://wwwatlas.mppmu.mpg.de/ringberg2002/ D.Cavalli et al., Athens 2003 ETmiss critical for invariant tt mass reconstruction : Z/A/H t1 t2 prod1 n1 prod2 n2 prod1(2) = jet , lept Assumptions : mt = 0 the two neutrino system directions are coincident with the ones of the measured t-decay products ( u1, u2 ) t-decay products are not back to back mtt = 2(E1+ En1 )(E2+ En2)(1 - cosq) E1,E2 = t-decay products energies q = angle between t-decay products directions En1, En2 = energies of the two neutrino systems : pxmiss (pymiss) = (En1 u1 ) x(y) + (En2 u2 ) x(y) En1, En2 must be physical ( > 0 ) s (mtt) s (ETmiss) / |sin (D) prod1 prod2 | DC1 data : bbA tt , Z tt No Noise added D.Cavalli et al., Athens 2003 PHYS TDR Noise added : Ecell > 1.5 s(Noise) Calibration : different sets of calibration constants for hadronic cluster cells, em cluster cells and cells outside clusters in different calorimeters H1-Style calib in GOOD agreement with PHYS TDR !! ETmiss Resolution = s ( Ex(y)miss Rec |h|< 5 - Ex(y)miss_NonInt ) SumET = ET calo cells within |h|< 5 ETmiss Resolution = k SumEt 5.纏め LHCにおいても,Jet Energy Calibrationはとても重要。 Jet Finding algorithm – cone, kTなど。 実際のデータを用いて較正する。実験屋のアイデア次第。 J/ΨやΥデータ,γ+Jets,W/Z+Jets,Z→bb ATLAS測定器はfine granularである。 Energy Flowの研究をしてはどうか? おわり