Experimental Aspects of Jet Reconstruction (Lecture 2) Peter Loch University of Arizona Tucson, Arizona, USA e-mail: loch-at-physics.arizona.edu
Outline Jets at LHC Jet reconstruction in ATLAS More from jets at LHC Brief overview on jet features at LHC Focus QCD! Jet reconstruction in ATLAS Strategy I calibration implementation (jet context based) Tower and cluster jets In-situ calibration Forward jets More from jets at LHC Mass reconstruction Jet shapes Summary & Outlook
Jets at LHC New kinematic regime for jet physics W. Stirling, LHCC Workshop “Theory of LHC Processes” (1998) *annotation from J. Huston, Talk @ ATLAS Standard Model WG Meeting (Feb. 2004) New kinematic regime for jet physics Jets can be much harder Jets get more narrow in general (kinematic effect ~αs) Higher energies to be contained in calorimeters Jet reconstruction challenging Physics requirements typically 1% jet energy scale uncertainty top mass measurement in ttbar LHC is a top factory! hadronic final states in at the end of long decay chains in SUSY Quality takes time Previous experiments needed up to 10 years of data taking to go from ~4% down to ~1% Can often no be achieved for all kinds of jets and in all physics environments
Inclusive Jet Cross-section at LHC Changing jet flavours in QCD Jets at low pT most likely produced by gluon fusion Large phase space for radiation Expectation are multi-jet final states even for 2→2 processes More likely quark jets at higher pT Less radiation (Sudakov suppression) Less jets in events Narrower jets (again) Large kinematic range pT range 10-5000 GeV/c Di-jet mass reach several TeV/c2 Multitude of “jet flavours” generated in pp collisions at LHC → expect corresponding variety of jet shapes with (possibly) specific calibrations!
QCD Jet Multiplicities @ LHC Jet multiplicities in 2→2 scattering Di-jet mass range
Jet Production @ LHC Expectations: Jet energy scale error very quickly systematically dominated Large statistics in unexplored kinematic range already at low luminosity Calibration channels quickly accessible Especially for photon+jet(s) Dominant direct photon production gives access to gluon structure at high x (~0.0001-0.2)
QCD Measurements at LHC just from cross-section, can be improved by 3/2 jet ratio, but no competition for LEP/HERA! Strong Coupling test of QCD at very small scale ( ) PDFs di-jet cross section and properties (Et,η1,η2) constrain parton distribution function Compositeness Deviation from SM sensitivity to compositeness scale Λ up to 40 TeV @ 300 fb-1 (all quarks are composites)
Physics Environments @ LHC Physics environment different from Tevatron Increased underlying event activity (more phase space) Already at lowest (initial) luminosities ~1031-1032 cm-2 s-1 Additional activity from pile-up Proportional to instantaneous luminosity A.Moraes, HERA-LHC Workshop, DESY, March 2007 LHC prediction: x2.5 the activity measured at Tevatron! Number charged tracks in transverse region CDF data (√s=1.8 TeV) CDF data: Phys.Rev, D, 65 (2002) Rick Field’s (CDF) view on di-jet events pT leading jet (GeV)
Physics Environment @ LHC (2) Section: Jets at LHC Et ~ 58 GeV Et ~ 81 GeV no pile-up added LHC design luminosity pile-up added Physics Environment @ LHC (2) Additional challenges: Pile-up High bunch crossing rate at LHC (40 MHz) and large pp cross-section (~75 mb) About 23 (soft) collisions in addition to the triggered hard scattering at design luminosity 1034 cm-2 s-1 Poisson distributed Statistically independent Low pT scattering mostly (“minimum bias”) Generate lots of additional particles in addition to the underlying event ( ~370 particles/unit rapidity per bunch crossing (3700 within ATLAS) ~1,800 charged tracks in ATLAS/bunch crossing Similar dynamics but no correlation with hard scattering Calorimeter signals typically to slow Short signal shaping, bi-polar shaping function (ATLAS) Long signal history (~500-600 ns), Out of time pile-up Keep as much signal as possible High magnetic field (4 T) and high signal threshold (CMS) Suppress small signal contribution P. Savard et al., ATLAS-CAL-NO 084/1996 R = 0.7
Physics Environments @ LHC (3) Discovery physics at LHC Expect extreme busy final states Lots of leptons, missing Et and jets O(10) in SUSY Many x 10 jets in black hole production Good spatial resolution power to find the jets Good energy resolution for reliable missing Et calculation Need large rapidity coverage Tag vector boson fusion production of Higgs and exotics WW, WZ, ZZ with associated quark jets These “tag jets” often go forward Jets are uncorrelated with each other, but balance the central system (Higgs) Tagjet direction η 100 GeV 600 GeV 300 GeV Direction of tag jets in Higgs VBF production for mH = 100, 300, 600 GeV this is a very old plot!
Jet Finders in ATLAS Defaults are seeded cone and kT Also implementations of CDF mid-point, Cambridge/Aachen recursive recombination (0th order kT), “optimal jet finder” (event shape fit) More options: move to FastJet libraries CMS, theory No universal configuration Narrow jets W->jj in ttbar, some SUSY Wider jets Inclusive jet cross-section, QCD mW N.Godbhane, JetRec June 2006 Algorithm Cone Size R Distance D Clients Seeded Cone EtSeed = 1 GeV fS/M = 0.5 0.4 W mass spectroscopy, top physics Kt (FastKt) 0.7 QCD, jet cross-sections 0.6 P.-A. Delsart, June 2006
Tower Jets in ATLAS Sum up electromagnetic scale calorimeter cell signals into towers Fixed grid of Δη x Δφ = 0.1 x 0.1 Non-discriminatory, no cell suppression Works well with pointing readout geometries Larger cells split their signal between towers according to the overlap area fraction Tower noise suppression Some towers have net negative signals Apply “nearest neighbour tower recombination” Combine negative signal tower(s) with nearby positive signal towers until sum of signals > 0 Remove towers with no nearby neighbours Towers are “massless” pseudo-particles Find jets Note: towers have signal on electromagnetic energy scale Calibrate jets Retrieve calorimeter cell signals in jet Apply signal weighting functions to these signals Recalculate jet kinematics using these cell signals Note: there are cells with negative signals! Apply final corrections
Determination of Tower Jet Calibration Sample of fully simulated QCD di-jet events from hard scatter pT>17 GeV/c to kinematic limit Electronic noise included in simulation Match reconstructed calorimeter jet with close-by particle jet Both jets reconstructed with seeded cone R=0.7 pTseed>1 GeV/c Overlap threshold 50% Match exclusive: only accepted if only one jet close by Calorimeter jets are based on tower signals in a grid of ΔηxΔη = 0.1x0.1 Access cell signals in jet H1 motivated cell signal weighting strategy, see lecture 1 Determine cell signal weights in resolution optimization fit using truth particle jet energy as normalization Weights are function of cell location and cell signal density Dense signals – em, less dense signals hadronic Re-calculate jet four-momentum using cell weights Jet energy and direction change
“H1” Style Cell Signal Weighting in ATLAS Fit constraint: Jet four-momentum calculation after fit Final corrections for residual signal non-linearities Algorithm dependencies Available for seeded cone R=0.4, kT D=0.4, D=0.6 Signal dependencies (cluster/tower) “massless pseudo-particles”
Tower Jet Performance Signal linearity Energy resolution Relative to matched MC truth jet! Energy resolution Relative to truth S. Padhi, ATLAS Physics Workshop 07/2005 S. Padhi, ATLAS Physics Workshop 07/2005
ATLAS Tower Jet Features Algorithm and physics environment Remember: no universal calibration Fa,s recovers some signal linearity cannot redo cell weight fits for all possible scenarios Calorimeter signal choice Loss of signal if taking clusters instead of towers Noise cut! Width ~unchanged Weights still work old but educational! F. Paige, ATLAS Software Workshop 09/2004 jets/0.02 Reconstructed Et/True Et tower jets cluster jets F. Paige, ATLAS Jet/EtMiss Working Meeting 02/02/2005
Input Selection in Tower kT Jets Early attempt to deal with “vacuum” effect in kT Remove low energetic particles and signals from final state before jet finding by ET cut Problematic: ET uncalibrated for towers → cut effectively higher (~30%) for calorimeter jets! jet transverse energy ok! input bias very good agreement!!
Effect on Tower Jet Shapes Experimentally accessible Need to look at it again with data Again biggest problem uncalibrated constituents of jets We could feed jet calibration back to tower signals Distance from jet axis Distance from jet axis em scale!
Cluster Jets in ATLAS Attempt to factorize Noise suppression Noisy cells are removed Hadronic calibration Signal weighting in cluster context, no jet bias Dead material corrections Limited to vicinity of clusters Cannot correct if no signal at all nearby Out-of cluster corrections Efficiency correction for clustering algorithm Provides calibrated input to jet finding Relative mis-calibration O(5%) Instead of O(30%) Clusters can be interpreted as massless pseudo-particles ATLAS convention, see later! Cluster Jets in ATLAS
It’s All In The Pictures…
Why Cluster Jets At All? Reduce noise contribution Fixed cone tower jet Fixed cone cluster jet
Cluster Jet Signal Linearity Flat response in Et and rapidity Forward region problems under study Missing ~8% jet energy ~3% cluster mis-classification Cluster classified as em, but really is hadronic ~3% signal efficiency Signal of low energetic particles below cluster threshold ~2% electromagnetic calibration Em clusters need own calibration Basic scale insufficient Remember: calibration so far comes from single particle! No jet context whatsoever! Studies under way… S.Menke/G. Pospelov March 2007 T&P S.Menke/G. Pospelov March 2007 T&P
Jet Energy Scale Corrections Use of kinematic constraints in pp Photon/Z+jet pT balance Central value model dependent! Topology Hard cuts on back-to-back no problem at LHC! Kinematic limit ~400 GeV/c pT(photon) for 1% First shot at jet energy scale! W mass Powerful but very special jets W color-disconnected Narrow jets Di-jet balance Extrapolation tool to high pT Also detector uniformity Topology dependence Normalization strategy Match reconstructed pT balance with particle level balance Unfold topology dependences
Photon+Jet Resolution bias Choice of selection variables Average photon+jet pT Problematic in jet pT correction Change of event selection?? pT projection as function of pT(photon) more appropriate? photon Pt cut average Pt cut (pTγ+pTparton)/2
Jet Energy Scale From W Challenge: pile-up and W boost Pile-up can “improve” jet energy scale! P. Savard, P. Loch, CALOR97 η(W) ~1.8
Very High pT Jets Reconstruction concern: leakage Find indicators in jet signal to tag leakage Late showering in calorimeter Use muon spectrometer hits behind jet No energy measurement, but good tag Frank Paige, ATLAS T&P Week February 2006
Jet Finder Efficiencies in ATLAS Efficiency Only free parameter: matching radius Rm No kinematics matching! Purity Relates to fake rate Jets in VBF! Jets in VBF!
Forward Jets Pile-up Noise in Calorimeter Cells Complex region S. Menke, ATLAS Physics Workshop 07/2005 Forward Jets Complex region Low pt jets Especially in VBF! Same fluctuations (in Et) from pile-up as in central region! Signal significance a concern What do we need from it High tagging efficiency Need to find VBF Kinematics reconstruction Limited by fluctuations Limits “invisible Higgs” reconstruction e/jet separation Forward/backward asymmetry in Z decays Access to sinθw beyond LEP precision due to large rapidity coverage very detector dependent performance! ATLAS only here! very old! jet energy fluctuations jet cone size
Forward Jets: Kinematic Reconstruction In presence of pile-up Design luminosity covered by endcap RMS Pile-Up in R=1.0 Jet Cone @ 1034 s-1cm-2 RMS Pile-Up in R=0.4 Jet Cone @ 1034 s-1cm-2 Jet Cone Size R Signal Significance 1034 cm-2s-1 P.Loch & P.Savard, CALOR97 Forward tag jets in Higgs production have average Et of 30-50 GeV only (descreasing with increasing η) acceptance limit ATLAS Physics and Detector Performance TDR, Vol 1., LHCC-99-14/15, Fig.9-24
Jet Masses Gained interest at LHC Mass measurement challenging Tower Jet Mass Jet Masses Gained interest at LHC Heavily boosted top decays All decay products reconstructed in one jet Jet mass only observable indicating top decay Mass measurement challenging Particle jet level mass is reference Simulations only! Mass of calorimeter jet is affected by shower spreads Enters: signal definition dependence, cluster shapes, noise,… No significant attempt at Tevatron or elsewhere So how about LHC Really ATLAS for this lecture! Truth Jet Mass Cluster Jet Mass Truth Jet Mass Tower Jet Mass Truth Jet Mass Cluster Jet Mass Truth Jet Mass
Jet Mass Reconstruction cluster jets tower jets Signal definition dependence Expected from four-vector recombination in jet finding E.g. tower jets have about constant number of constituents in cone jets: Different noise contribution in clusters and towers Tower split shower signals Energy sharing between towers depends on shower size wrt readout granularity and internal Et flow in jet → direction dependence Cluster collects shower signal Non-resolvable shower overlap limits mass resolution → direction dependence cluster jets tower jets cluster jets tower jets relative mass difference
Mass Reconstruction Sensitivities Contribution from low energetic particles lost Overall effect depends on signal definition How about effect on mass? Exercise: remove particles below pT threshold from jet and re-calculate mass Remember: towers are not calibrated More severe effect of cut in tower jets Clusters are calibrated More similar to particle selection in jets
Mass Sensitivity (2) change of mass QCD kT jets, D = 0.6 log10(least biased reconstructed mass/GeV)
Jet Shapes (1) 1st question: any relation between number of particles, towers, clusters in jets? Most interesting for kT D = 0.6 here Look at matching callorimeter/truth jets Note: not the most important variable! We already expect change of “jet picture” by detector signal definition Hints on resolution power for jet shape variables and mass
Jet Shapes (2) We expected clusters to represent indivdual particles Cannot be perfect in busy jet environment! Shower overlap in finite calorimeter granularity Some resolution power, though Much better than for tower jets! ~1.6:1 particles:clusters in central region ~1:1 in endcap region Best match of readout granularity, shower size and jet particle energy flow Happy coincidence, not a design feature of the ATLAS calorimeter!
Jet Shapes (3) Jet density Example: kT jets in QCD cluster jets Jet Shapes (3) tower jets hadron jets Jet density Calculate energy outside of cone R = 0.3 as function of pT and direction Classic Tevatron measurement Experimental indication of transition from (low pT) gluon to (high pT) quark jets Example: kT jets in QCD D = 0.6 Fraction of energy outside cone around jet axis (Rcone=0.3) log10(pTjet/GeV)
Jet Shapes (4) Integrated Et flow in jets Average projection of constituent pT vector on jet pT vector as function of distance from jet axis Differential spectrum Integrate up to R Can be done with real data! Perpendicular component indication of jet dynamics ? Jet by jet?
Jet Shapes (5)
Summary General Jet calibration What we can get from jets Everything you have seen here from ATLAS is based on simulations Real data can bring us surprises (good and bad) Jet calibration Local calibration based on clusters seems to be the way to go for a first order hadronic calibration Cluster signal (~200/event) good basis for jet finding in physics analysis context Final jet energy scale corrections depend on analysis choices for jet finders, -configurations, selected event topologies… What we can get from jets Strong interest to go beyond Tevatron jets Jet masses of great interest Jet shapes can test basic jet formation dynamics ?? New dimension added: jet signal shapes in calorimeters
Lots is left to do… Refined jet calibration Use calorimeter jet signal shapes We used cluster shapes already, now look at cluster distribution in jet Use inner detector tracks Large momentum fraction of jet in tracks indicates a very hadronic jet, i.e. more corrections All this has not been used much in the past, but we have to address a 1% systematic error requirement somehow! We are more than ready for data!