IAS Program on High Energy Physics, Jan 12, 2017 Top-TaggingattheEnergyFrontier MinhoSon KAIST WorkinprogresswithZhenyuHanandBrockTweedie jet q jet q’ q q’ jet b Slowly moving top Fastly moving top p" ∼ O(sub − TeV) b jet q q’ Almost QCD-like jet b Super-Fastly moving top p" ∼ O(afew×10TeV) q q’ jet b Fastly moving top p" ∼ O(sub − TeV) Sub-TeV top-tagging is sort of well-established q How would top-tagging performance evolve with much higher 𝑝6 ? q’ Almost QCD-like jet b Super-Fastly moving top p" ∼ O(afew×10TeV) Many challenges arise as tops enter into hyper-boosted regime Instability from soft radiation QCD-jet Pert.Emission neartheboundary QCD-jet 𝒑𝑻 ∼ 𝟔𝐓𝐞𝐕 𝒑𝒔𝒐𝒇𝒕 ∼ 𝟓𝐆𝐞𝐕 𝑻 𝑹∼𝟏 Spuriousmassscale 𝒔𝒐𝒇𝒕 𝒎𝟐 ∼ 𝒑𝑻 𝒑𝑻 𝑹𝟐 ∼ 𝒎𝟐𝒕𝒐𝒑 Borrowed a discussion from Larkoski, Maltoni, Selvaggi 2015 Instability from soft radiation Top-jet Top-jet 𝒑𝑻 ContaminationfromISR 𝒑𝑰𝑺𝑹 𝑻 Borrowed a discussion from Larkoski, Maltoni, Selvaggi 2015 𝑹∼𝟏 Fluctuationoftop-jetmassforFIXED cone 𝟐 𝟐 𝒎𝟐 ∼ 𝒎𝟐𝒕𝒐𝒑 + 𝒑𝑻 𝒑𝑰𝑺𝑹 𝑻 𝑹 ∼ 𝑶 𝟏 ×𝒎 𝒕𝒐𝒑 If 𝒑𝑰𝑺𝑹 𝑻 𝒎𝟐𝒕𝒐𝒑 ∼ 𝒑𝑻 𝑹𝟐 ∼ 𝟓𝟎𝐆𝐞𝐕if 𝒑𝑻 ∼ 𝐟𝐞𝐰×𝒎𝒕𝒐𝒑 Moderately boosted ∼ 𝟓𝐆𝐞𝐕if 𝒑𝑻 ∼ 𝐟𝐞𝐰×𝒎𝒕𝒐𝒑 ×𝟏𝟎 Hyper-boosted Instability from soft radiation vs Shrinking jet size Top-jet Top-jet 𝒑𝑻 ContaminationfromISR 𝒑𝑰𝑺𝑹 𝑻 𝑹 ∼ #/𝒑𝑻 Krohn, Thaler,Wang2009 Han2014 Larkoski,Maltoni,Selvaggi 2015 Fluctuationoftop-jetmassforSHRINKING cone 𝟐 𝟐 𝒎𝟐 ∼ 𝒎𝟐𝒕𝒐𝒑 + 𝒑𝑻 𝒑𝑰𝑺𝑹 𝑻 𝑹 ∼ 𝒎 𝒕𝒐𝒑 𝑰𝑺𝑹 𝒑 𝟏 + 𝜷𝟐𝑹 𝑻 𝒑𝑻 Withshrinkingcone 𝒎𝒕𝒐𝒑 𝑹 ∼ 𝜷𝑹 , 𝒑𝑻 𝐞. 𝐠.𝜷𝑹~𝟒 EW-strahlung at high pT QCD-jet 𝒑𝑻 W/Z-emission QCD-jet 𝑾/𝒁 E.g. ∼ 6% at 5 TeV quark-jet (unpolarized) Splitting functions (momentum integrated) 𝜶 𝟏+ 𝟏−𝒙 𝑷𝒒→𝑾𝑻 ∼ 𝑬𝑾 𝝅 𝒙 𝑷𝒒→𝑾𝑳 ∼ 𝜶𝑬𝑾 𝟏 − 𝒙 𝝅 𝒙 𝟐 𝒑𝟐𝑻 𝐋𝐨𝐠 𝟏 − 𝒙 𝒎𝟐𝑾 Dead cone, FSR, and shrinking cone 𝑹𝒅𝒆𝒂𝒅𝒄𝒐𝒏𝒆 ∼ 𝟐𝒎𝒕𝒐𝒑 𝒑𝑻 Deadconecapturestopdecayproducts ,butnoradiationfromtop.Relevantforsuccessfultop tagging 𝐸. 𝑔. 𝑋 → 𝑡𝑡̅ G. Salam, talk given at LHC New Physics Forum, IWH, Heidelberg, 23-26 Feb, 2009 𝑋 𝑡 CapturingFSRbeforedecayisimportantto reconstructthecorrectresonancemass wheretopsweredecayedfrom ∼ 0.43𝛼k ln 𝑅opq forquark-jet Salam2010 ‘TowardsJetography’ Instrumental challenge Detectorgranularityisbecomingabigproblem. ATLAS/CMS hasthreelayersofmainsub-detectors One𝐇𝐂𝐀𝐋 cell 0.1x0.1 Typical jet R=0.5 One𝐄𝐂𝐀𝐋 cell 0.02x0.02 Instrumental challenge Detectorgranularityisbecomingabigproblem. Cluster of ’many ’ CAL cells 0.5 jet from W 0.5 jet from W jet Top-quark 0.5 b-jet b-jet q b jet q’ Slowly moving top Instrumental challenge Detectorgranularityisbecomingabigproblem. ATLAS/CMS hasthreelayersofmainsub-detectors 0.5 jet from etal13’ W Spannowsky One𝐇𝐂𝐀𝐋 cell 0.1x0.1 0.5 jet from W Top-quark 0.5 b-jet One𝐄𝐂𝐀𝐋 cell 0.02x0.02 Instrumental challenge Detectorgranularityisbecomingabigproblem. ATLAS/CMS hasthreelayersofmainsub-detectors 0.5 jet from W One𝐇𝐂𝐀𝐋 cell 0.1x0.1 0.5 jet from W Top-quark 0.5 b-jet One𝐄𝐂𝐀𝐋 cell 0.02x0.02 FCC (Future Circular Collider) FCC-hh (CERN),CEPC(China) 100 TeV collider ∼ 7x upgrade of CM energy E.g.3TeV topatthe LHC wouldexpect~20 TeV topsat100TeV Current proposal for the future detector 𝐄𝐂𝐀𝐋, 𝐇𝐂𝐀𝐋 𝐓𝐫𝐚𝐜𝐤𝐞𝐫 2x 4x 14TeV Looks like the detector technology can not catch up with the energy upgrade Capability attheLHC + FCC proposal Understanding our current detector better is a KEY-ingredient to predict our future capability → Capability attheFCC ExistingLiterature Katz,MS,Tweedie,Spethmann 2011,2012 Snowmass2013 Schaetzel,Spannowsky 2013 CMSPASJME-14-0022014 Spannowsky,Stoll2015 Larkoski,Maltoni,Selvaggi 2015 …. *Listedonly studiesonW/Z/H/tops-taggers Outline WewillfocusonJHUTopTagger +N-subjettiness 1. Optimize JHU TopTagger + N-subjettiness at particle level We will newly show that N-subjettiness is not just an alternative to other top-taggers, but it adds a new information to improve top/gluon discrimination 2. Introduce various detector models We will illustrate how one can combine information, scattered in here and there in subdetectors, to extract a meaningful result 3. Optimize JHU TopTagger + N-subjettiness in various detector models This step will establish the “robustness of shape variables vs declustering variables against different detector configurations” Top-jetsize JHU TopTagger with CMS type cuts 𝑅opq = 𝛽Ž × Top Ateachbranch, 𝑚q 𝑝6 €‚ oƒ ‚ q„€opq ‡ 𝛽… × € ˆ ∶minangulardist. ‚ 𝛿€ = € 𝛿… = cell 𝒋𝒕𝒐𝒑 𝒋𝟏 Firstiteration 𝒋𝟏 𝒋𝟐 𝒋𝟐𝟏 𝒋𝟐𝟐 Hardsplitting Softsplitting cell 𝒋𝟏𝟏 Seconditeration 𝒋𝟏𝟏 𝒋𝟏𝟐 𝒋𝟐 cell cell cell 𝒋𝟏𝟐 cell cell cell cell cell cell cell subjet 3 subjet 1 subjet 2 3 (or 4)subjets Cutsontwovariables:𝑚‡Œ• ,𝑚q„€ *Insteadofm• andcosθ intheoriginalJHUTopTagger N-subjettiness • exploitsradiationpatternaround Nsubjet axes min ∑ 𝑝6 𝑖 min Δ𝑅 𝑖, 𝑗œ› , … . , Δ𝑅 𝑖, 𝑗œ• 𝜏• ≡ ∑𝑝6 𝑖 𝑅 Weightedsumoverconsitituent’s pT :smallerweight,Δ𝑅(𝑖, 𝚥̂),forsmaller distance.Preferstobesmallforthe rightnumber ofaxesfound Thaler,TilburgJHEP1103 Top:threelocalizedenergyclusters N-subjettiness Thaler,TilburgJHEP1103 • exploitsradiationpatternaround Nsubjet axes min ∑ 𝑝6 𝑖 min Δ𝑅 𝑖, 𝑗œ› , … . , Δ𝑅 𝑖, 𝑗œ• 𝜏• ≡ ∑𝑝6 𝑖 𝑅 Weightedsumoverconsitituent’s pT :smallerweight,Δ𝑅(𝑖, 𝚥̂),forsmaller distance.Preferstobesmallforthe rightnumber ofaxesfound Thaler,TilburgJHEP1103 Top:threelocalizedenergyclusters N-subjettiness Thaler,TilburgJHEP1103 • exploitsradiationpatternaround Nsubjet axes min ∑ 𝑝6 𝑖 min Δ𝑅 𝑖, 𝑗œ› , … . , Δ𝑅 𝑖, 𝑗œ• 𝜏• ≡ ∑𝑝6 𝑖 𝑅 Thaler,TilburgJHEP1103 ü N-subjettiness is qualitatively different from other top taggers based on mass/pT-drops and it has been introduced as an alternative for top tagger ü We observe that combining other top taggers with N-subjettiness can give 𝑂(1) improvement in top/gluon discrimination Forsimilardiscussion,CMSPASJME-13-007,Adamsetal15’ Optimization JHU Top-tagger with CMS-type cuts & N-subjettiness Clustering/declustering/cutparameter 𝑅¡¢£¤¥¦£ = 1.0 ‡ 𝑅opq ≡ 𝛽Ž × € ˆ:Shrinking jet-conesize ‚ 𝛿€ :pT asymmetrycut 𝑚‡Œ• :minjetpairmass ‡ ,𝛿… ≡ 𝛽… × € ˆ :minangularseparation ‚ 𝑚q„€ :reco- topmass 𝜏¾¿ ≡ 𝜏¾ /𝜏¿:N-subjettiness Optimizationoversevenparameters Tag/mistag Rate≡ #§¨©ª¤ª«¬£-£®««¢¬ #¯«¢«©°£«¬±¤£®›%³"±¤¢¬-± Signal:continuum 𝑡𝑡̅ → 𝜇 + 𝑗𝑒𝑡𝑠 Quark/gluon: 𝑞𝑍 → 𝑞 𝜈𝜈̅ , 𝑔𝑍 → 𝑔(𝜈𝜈̅ ) :samplesarerestrictedto 𝜂 < 1.0,𝑝6 = [𝑝6 − 1%, 𝑝6 + 1%] GeV Top/gluon/quark discrimination at particle level Withgluon-optimized parameters Flavordependentoptimizations: Withquark-optimized parameters N-subjettiness dominates JHU/CMS tagger dominates 𝛽Ž ∼ 4, 𝛽… ∼ 0.7, 𝛿 … ∼ 0.03 forrelevanttagefficiencies • Simultaneous optimizations of the quark and gluon jets are possible • N-subjettiness adds extra discriminating power for gluon-jets, not quark-jets Top/gluon discrimination at particle level 𝑝6 -dependentoptimizationsontop/gluon-jets: Twoseparateoptimizations: JHUwithCMS-typecutswithout vswithN-subjettiness (combined tagger) Nearly scale-invariant Better performance due to N-subjettiness for gluon-jets Optimizedparametersareroughlyunchanged,e.g.optimized𝛽Ž and𝛽… stayfixed,simple ∼ 1/𝑝6 scalingworks Introducingdetectoreffect Whatisagooddetectormodel? Itistheonethatminimallybreaksthe‘scaleinvariance’and bringstheresultbacktoourexpectationatthe‘particle-level’ Introducingdetectoreffect Whiletherealdetectorsareinsanely complicated,ourtoydetectormodel wouldcatchthe leading effects.However, weareaiming tobeasclosetotherealityaspossible HCALcellsizeservesasacut-off inmanypheno- studies q RawHCAL Final state particles Schaetzel, Spannowsky 2013 Introducingdetectoreffect Whiletherealdetectorsareinsanely complicated,ourtoydetectormodel wouldcatchthe leading effects.However, weareaiming tobeasclosetotherealityaspossible Finalstateparticlesintothreedifferentsub-detectors :rawcalorimetercellsasinputsfor jetclustering q RawECAL&HCAL Hadrons Photons, nonisolated electrons Muons, isolated electrons Toydetectormodels Cartoonic picture of our toy detector model 𝜋 𝜋 𝜋 ü Use every bit of information from three layers of sub-detectors 𝜋 Repeat with trackers (track flow,particle flow) Katz,MS,Spethmann,Tweedie,2011,2012(SeeAppendicesof1010.5253/1204.0525) Toydetectormodels Combining information is not unique Toydetectormodels Combining information is not unique q EM-flow ECALsarelocallyrescaledtotheenergy ofthefull calorimeter, andHCALcellsdiscarded Katz,MS,Spethmann,Tweedie2011,2012 (SeeAppendicesof1010.5253/1204.0525) RescaleECALcellsby ÂÃÄÅÆ ÇÂÈÄÅÆ ÂÃÄÅÆ EÊË¡Ì › EÂË¡Ì ¿ EÂË¡Ì Detectorwith ECAL&HCAL E › ′ÂË¡Ì E ¿ ′ÂË¡Ì Detectorwithonly rescaledECAL *RescaledECALcellsareinputforthejetclustering Toydetectormodels Combining information is not unique q EM-flow ECALsarelocallyrescaledtotheenergy ofthefull calorimeter, andHCALcellsdiscarded Katz,MS,Spethmann,Tweedie2011,2012 q Track-flow RescaleECALcellsby Similarlyrescaletracksby Schatzel, Spannowsky 2014 Larkoski,Maltoni,Selvaggi 2015 q Particle-flow Rescaletracksby ÂÃÄÅÆ ÇÂÈÄÅÆ ÂÃÄÅÆ ÂÃÄÅÆ ÇÂÈÄÅÆ ÂÎÏÐÑÒÓ ÂÈÄÅÆ andleaveEÂË¡Ì ÂÎÏÐÑÒÓ as-is *PERFECTtrackingefficiencyisassumed.Realityisworsethanthisperfectcase Cartoon Pseudo-CMStypeEvent q EM-flow q Track-flow q Particle-flow Two crucial detector effects added to be more realistic 1.Energy-smearingintonearbycalorimetercells 2.HadronsdeposittheirenergiesinECALcells Unlike the situation in this cartoon, hadrons have O(1) chance to leave their energies (e.g. via Nuclear interaction) in ECAL before reaching HCAL. O(20%) of jet energy becomes absorbed in the ECAL in this manner Two crucial detector effects added to be more realistic 1.Energy-smearingintonearbycalorimetercells 2.HadronsdeposittheirenergiesinECALcells GEANT4 • ECALsmearing pattern/hadron-energy-deposit-in-ECAL willbesimulated withGEANT4whereasHCALsmearingpatternwillbedonebysimpleansatz UpgradedversionofMS,Spethmann,Tweedie2012(SeeAppendix of1204.0525) EnergysmearingintonearbyECALcells ü Themostimportant ingredient inourdetectormodel Aparticlehitting singleECALcell ± 𝜋 5x5ECALcellsunderneath oneHCALcell 𝟐. 𝟐𝐜𝐦 𝟐𝟐𝐜𝐦 𝐏𝐛𝐖𝐎𝟒(𝐂𝐌𝐒) EnergysmearingintonearbyECALcells Idealsituation NOsmearing ± 𝜋 :allenergyisdeposited inasingleECALcell 5x5ECALarrays EnergysmearingintonearbyECALcells Inreality Smearing ± 𝜋 :energyissmearedinto nearbyECALcells ü Sensitivetojetsubstructure variables Lightercolorrepresentsless energydeposit 5x5ECALarrays Smearing effectbecomesextremelyimportant injetsubstructureanalysisof thehyper-boosted heavyparticles(e.g.top/H/Z/W) WesimulateECALsmearingbyGEANT 1. 2. 3. Prepare9x9ECALcellswithsamedimensionasCMSECAL Shootsingle𝑒 ± , 𝜋 ± beamsonto ECALrepeatedly Buildupalibraryofshoweringprofilesfor𝑒, 𝜋 beams ü e-inducedshowersasproxiesfor e andγ ü 𝜋-inducedshowersasproxyfor allhadrons Energyisfixedtobe100GeV 𝑒 ±, 𝜋 ± q ParticlehittingaECALcellisreplacedwitha randomlychosensmearingprofilefromthe library *Correlationbetweencellsareautomaticallyfoldedin 5x5ECALarrays WesimulateECALsmearingbyGEANT singleECALcell 𝐈𝐦𝐩𝐚𝐜𝐭𝐩𝐨𝐢𝐧𝐭 𝑒 ±, 𝜋 ± ü Impactpointisnotalwaysatthecenterof ECALcell.Itleadstoasymmetricpattern. Wesub-divide ECALinto9x9=81sub-cells totakeintoaccountasymmetricshowering pattern 5x5ECALarrays - Wedonotsimulateasymmetricdetectorgeometry,e.g.particlecanhitacellwithanangle Electron-inducedECALshoweringpatternbyGEANT 𝟏𝟎𝐆𝐞𝐕𝒆¥ beam 𝟏𝟎𝟎𝐆𝐞𝐕𝒆¥ beam Eâ«ãã/E¤¢â¤¬«¢£«ã«â£©-¢,notw.r.tE£-£°ã¬«³-§¤£ - NearlypT-independent.Itjustifiesourproxiessimulatedat100GeV Pion-inducedECALshoweringpatternbyGEANT 𝟏𝟎𝟎𝐆𝐞𝐕𝝅± beam 3𝐓𝐞𝐕𝝅± beam Eâ«ãã/E¤¢â¤¬«¢£³¤-¢,notw.r.tE£-£°ã¬«³-§¤£ - NearlypT-independent.Itjustifiesourproxiessimulatedat100GeV ProfileansatzforHCAL 𝐇𝐂𝐀𝐋𝐜𝐞𝐥𝐥 𝟓×𝟓𝐄𝐂𝐀𝐋𝐜𝐞𝐥𝐥𝐬 2𝑟 𝑓 𝑟 ∝ ¿ 𝑟 + 𝑅¿ 𝑒 ±, 𝜋 ± 𝐸¤¢ 𝐸-¨£ ¿ 75%containmentinacentrallystruckcell 95%within 3x3arrayaboutit Grindhammer,Rudowicz,Peters1990 CMSNOTE2006-138 ReplaceallparticlesflowingoutthebackofanECALcellwitha continuousangularenergydistributionaccordingtotheaboveansatz Spuriousstructureduetosmearing Smearing intonearbycellscanintroduce spurious structurewhena rescaling isdone withineachHCALcell Mini-jetclustering ü dealswithHCALenergyspreading: e.g.inEM-flow,theentirecollectionofECALandHCALcellsareclustered intomini-jetswiththeanti-𝒌𝑻 algorithm withthesizecomparabletothe HCALsize.Rescalingiscarriedoutwithineachmini-jet 𝐄𝐂𝐀𝐋 𝐇𝐂𝐀𝐋 Validation of our approach against CMS high pT W-jet CMSPASJME-14-002 With no trackers With perfect trackers Threebenchmarkscenarios Model Tracking: twoextremes LHC ECAL material ECALcell HCAL cell CMS-type (PbWOú ) 0.02× 0.02 0.1× 0.1 FCC1 Perfect/absent PbWOú (Leadtungstate) 0.01× 0.01 0.05× 0.05 FCC2 Perfect/absent PureW (Tungsten) 0.005× 0.005 0.05× 0.05 q RawECAL&HCAL We will see how these detector models perform in three benchmark LHC/FCC detectors q EM-flow q Track-flow q Particle-flow q Particle-level - EffectiveMoliereradiusofpureWisbiggerthanwhatisassumed.ConsiderPureWasaplace-holder foranynewmaterialwithahalf-sizedeffectiveMoliereradius Filteredtop-jetmass & 𝝉𝟑𝟐 of10TeVtop/gluonatFCC1 : equivalentsituation to5TeVtop/gluon attheLHC - pile-upandmagneticfieldarenotincluded inthisstudy N-subjettiness is doing great whenever tracks are available 𝜏¾¿ seems to probe a property within JHU/CMS subjets, rather than in-between them 5TeVtop/gluondiscriminationatFCC1 :equivalentto2.5TeV top/gluon-jetsattheLHC • Particle-flow is universally the best option (as it should be) • Track-flow works better with N-subjettiness, and EM-flow is less effective at capitalizing on N-subjettiness gluon mistag rate Gluon, at 50% top-tag rate (detector level) τ32+mJ JHU/CMS combined 0.1 raw ECAL & HCAL EM-flow → FCC2 track-flow particle-flow → FCC2 particle-level 0.01 1 10 1 FCC1 FCC2 𝐄𝐂𝐀𝐋𝟐𝐱, 𝐇𝐂𝐀𝐋𝟐𝐱 𝐄𝐂𝐀𝐋𝟒𝐱, 𝐇𝐂𝐀𝐋𝟐𝐱 comparedtoCMS-typeECAL,HCAL 10 1 10 p (TeV) T • • JHU/CMS tagger never fully competitive with N-subjettiness (except for EM-flow at 10TeV) Combined tagger is universally better ü FCC2 brings EM-flow, particle-flow to the similar level of half-𝑝6 jets at FCC1 Comparisontoanexistingstudyusingtrack-basedvariables Larkoski,Maltoni,Selvaggi 2015 *WefirstvalidatedourprocedurebyreproducingLarkoski etal. When tracks are not available, JHU/CMS tagger does most job When tracks are available, 𝜏¾¿ does most job Comparisontoanexistingstudyusingtrack-basedvariables Larkoski,Maltoni,Selvaggi 2015 *WefirstvalidatedourprocedurebyreproducingLarkoski etal. Perfect (solid red) vs. imperfect (solid black) tracking Note that Larkoski et al. (dotted black) scaned over 𝜏¾¿ with the fixed jet mass window ECAL 2x EM-flow can cover up to 20 TeV at FCC2 ∼ 10TeV at FCC1 StrongMagnetsatFCC ü Beneficialtohigh-𝑝6 physics.Ithurtslow-𝑝6 physics 𝑝 6ÿ…Œq = 0.15× B r × ÿ#$ T 𝑚 CMS:4T,1.5m FCC:6T,6m ~0.9GeV ~5.4GeV • Thisimpliesthat𝑂(100GeV) processsuchasHiggs physicsbecomeslow-𝑝6 physicsat100TeV! E.g.𝐻 → 𝑏𝑏' withlow𝑝6 willbesignificantly under-reconstructed duetolost tracks(Weneedtomakesurethatwearecapableofrestoringthelosttracksbacktoour jetsviatrackreconstruction,e.g.particle-flow) Inasituationthatstrong magneticfield becomesproblematic, ithurtshigh-𝑝6 tracking efficiency,but EM-flowisinsensitivetothisissue Toconclude q TheperformanceofouroptimizationofJHUTopTagger combinedwithNsubjettiness 1. 2. 3. 4. Quark- andgluon-jetscanbesimultaneouslyoptimizedwithinJHUTopTagger AddingN-subjettiness toe.g.JHUTopTagger,canmake𝑂(1) improvementof top/gluondiscrimination N-subjettiness iseffectivewhentracksareavailable JHUismorerobustthanN-subjettiness undermorepessimisticdetectorassumptions q EM-flowlooksverypromising.Itcansolelycoverupto20TeVtopsassuming FCC2configuration(ECAL4x,HCAL2x) 1. Trackersbecomecrucialtotagtopsbeyond it 2. UnlesstheFCCdetectorsareconstructedwithnear-perfecttrackers,someadditional investmentinECALgranularitywouldbebeneficial
© Copyright 2026 Paperzz