Jets in CMS Making jets from calorimeter or

Jets in CMS § Making jets from calorimeter or track information Ø key aspects of the CMS detector relevant for jet reconstruction Ø jet algorithms implemented in the CMS environment § Performance of algorithms Ø angular/energy resolutions, timing, mass resolutions, efficiencies § Working with jets in physics analyses Ø several topics relevant for a broad discussion Jorgen D’Hondt Vrije Universiteit Brussel – IIHE Jet Workshop – June 20, 2008 – Paris

The CMS detector in a nutshell J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 2

The CMS detector in reality! Installing the beam pipe J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 3

The CMS detector in reality! Installing the beam pipe J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 4

The CMS calorimeters HCAL: plastic scintillators brass absorber tower size Dh=0. 087 with Df = 5 -10 o ECAL: Pb. WO 4 crystals End. Cap 29 x 29 mm 2 Barrel 22 x 22 mm 2 ~1 o depth ~25 X 0 Crystals read-out by silicon avalanche photo-diodes (80% of the light collected in the first 25 ns) J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 5

CMS versus ATLAS Rick Cavanaugh Motivation to implement Particle Flow tools for jet reconstruction combining the calorimeter with the tracking system. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 6

Input for jet algorithms The general input for jet clustering algorithms are combined ECAL and HCAL “towers”: one or more HCAL cells and the corresponding ECAL cells. A tower is treated as a massless particle and its energy is the sum over all cells if they pass noise thresholds. Overall tower thresholds ET>0. 5 Ge. V ~4200 towers in total fraction of towers fired after thresholds in ttbar events. ttbar J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 7

Jet algorithms in CMS The CMS software framework supports 4 jet algorithms with several parameter settings: • Iterative Cone: used in the trigger algorithms because it has a short and predictable execution time per event, towers with ET>1 Ge. V are considered as seeds • Mid. Point Cone: same seeds requirements as for the Iterative Cone • Seedless-Infrared-Save or SISCone: external code equal to ATLAS (Salam/Soyez) • Fast k. T (Cacciari/Salam) E-scheme for all: the energy and momentum of a jet are defined as the sums of energies and momenta of its constituents J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 8

CMS PAS JME-07 -003 Jet algorithms in CMS To compare: total reconstruction time event ~10 s J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 9

Performance of jet algorithms Performance studies have been done using di-jet QCD and ttbar events. Matching efficiency between a Calo. Jet and a particle level jet at the vertex, with a matching criteria of DR<0. 5. The efficiencies of jets reconstructed with the Fast k. T and SISCone algorithms indicate better performance than jets reconstructed with the Midpoint Cone and Iterative Cone algorithms. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 10

Performance of jet algorithms The jet response Rjet = p. T/p. Tgen for the barrel region as a function of p. Tgen is shown for uncorrected jets. Very good agreement between the individual algorithms is found for all regions of the detector, indicating good correspondence between the values of D for the k. T algorithm and R for cone algorithms which are being compared. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 11

Performance of jet algorithms The h and f resolutions for jets in the barrel region are shown as a function of p. Tgen. Good agreement is found among all algorithms with comparable radius parameter, with marginal differences at low p. Tgen. Jets reconstructed with larger radius parameters yield slightly worse resolution both in h and f. Note that the position of the primary vertex is assumed to be at z = 0, which dilutes the h resolution w. r. t. taking the correct position measured with the tracking detectors into account. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 12

Performance of jet algorithms The h and f resolutions for jets in the barrel region are shown as a function of p. Tgen. Good agreement is found among all algorithms with comparable radius parameter, with marginal differences at low p. Tgen. Jets reconstructed with larger radius parameters yield slightly worse resolution both in h and f. Note that the position of the primary vertex is assumed to be at z = 0, which dilutes the h resolution w. r. t. taking the correct position measured with the tracking detectors into account. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 13

Performance of jet algorithms The jet energy resolutions derived from MC truth for jets in the barrel region. Jets reconstructed with Fast k. T show slightly worse resolution at low p. Tgen, while no significant impact of the radius parameter choice is observed. The resolutions are obtained additionally without using MC truth information by using the data-driven Asymmetry Method, which relates the jet p. T resolution to the resolution of the p. Timbalance between the two leading jets. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 14

Performance of jet algorithms The jet reconstruction performance in ttbar events is studied by selecting events with one (“lepton+jets”) or zero (“alljets”) electron or muon in the final state from a ttbar ALPGEN sample with no additional jets (“ttbar +0 jets”). Only events are considered for which all three decay products of one or both t(tbar) decay(s) can be uniquely matched to reconstructed calorimeter jets. hadronic decays J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 15

Performance of jet algorithms J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 16

Performance of jet algorithms Di-jet mass resolutions for Z’ decays are in good agreement (R=0. 5). J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 17

Jets for physics analyses Overview of jet ET cuts applied in Standard Model or related analyses: PTDR numbers ttbar (single-lepton) spin correlations : ttbar FCNC : ttbar (di-lepton) cross section : ttbar (single-lepton) cross section/mass : single-top Wt : single-top t- & s-channel : W+X (e/m) p. T>30 Ge. V [CMS Note 2006/111] p. T>30 -40 Ge. V [CMS Note 2006/93] p. Tunc>20 Ge. V [CMS Note 2006/77] p. T>30 Ge. V [CMS Note 2006/66 -64] p. T>35 -60 Ge. V [CMS Note 2006/86] veto p. T>20 Ge. V p. T>35 Ge. V [CMS Note 2006/84] veto p. T>30 Ge. V [CMS Note 2006/61] SUSY/top : p. T>30 Ge. V [CMS Note 2006/102] Usualy applied Iterative Cone algorithm (DR=0. 5). Pile-Up jets in some analyses reduced by longitudinal primary vertex matching, but no uniform applied procedure. To reduce QCD multi-jet background the p. T-cuts on jets have to be increased to about 40 Ge. V for top quark physics. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 18

Optimization of parameters Iter. Cone Optimize the matching between the parton and jet kinematics for several benchmark processes (here top quark processes: single-top, top pairs and tt. H). Need flexibility of the framework to allow optimization (eg. calibration for several parameters settings). Les houches hep-ph/0604120 Mid. Point J. D’Hondt Vrije Universiteit Brussel k. T Jet Workshop – Paris June 20, 2008 19

Comparing algorithms CMS Note 2006/066 Top quark mass analysis in lepton+jet final state. Run three jet algorithms (IC, MC and KT) and compare the directions of the four hard jets (after the event selection) → angular matching a<0. 3 85. 7% matched 87. 5% matched 78. 1% matched Requiring the three algorithms have to match in a to better than 0. 3 rad, the efficiency of this cut become 76. 1% (again after the normal event selection). . . unfortunatly not a strong reduction of the influence of jet reconstruction in the top mass systematic uncertainty. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 20

Pile-up jets lepton+jet (tt) after primary vertex constraint CMS Note 2006/066 #jets per event Using the primary vertex constraint reduces the amount of jets significant, although they remove generally low p. T jets. lepton+jet (tt) after primary vertex constraint Current Monte Carlo samples without pile-up collisions, hence few studies in this direction recently. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 21

Influence of pile-up jets Systematic effects on measurements are usually estimated by turning on/off the extra Pile-Up collisions. The effect on the number of selected events or on the measurant is quoted as an estimate of the systematic uncertainty (often scaled to represent a realistic knowledge of the expected value of Pile-Up collisions). E[X]=l Var[X]=l If Pile-Up was minor effect, it would not change the status of an event between selected or non-selected according to some event selection criteria. But the number of extra Pile-Up collisions is Poissionian distributed E[X]=3. 5 (low luminosity) and they can have a sever effect on the reconstruction of an event. Most of the event selections in proton-proton collisions are based on the presence of jets in the final state. A threshold is often applied on the ET of the reconstructed jets. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 22

Influence of pile-up jets Study the influence of this randomness ( event-by-event !! ) Simulate 1000 ttbar events (single-lepton) and add low-luminosity Pile-Up collisions according to a random Poisson distributed number (E[X]=3. 5) do this 100 times for each event. Apply an event selection on the reconstructed jets. As we expect 4 hard jets, an ET threshold is applied on the 4 th highest ET jet (before jet calibration) a fraction of the 100 ‘different’ events will be selected about the thresholds we apply in CMS The probability for each event to be selected can be determined, where the stochastic part is due to the randomness of the Pile-Up collisions in the hard event. High threshold low prob Low threshold high prob Medium thresholds : UNIFORM DISTRIBUTION !! J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 23

Influence of pile-up jets Define the variable F to quantify the uniformity of the distribution: F= (1 / #events). ∑ |Pi - ½| (mean deviation from ½) Determine this variable as a function of the ET-threshold on the 4 th jet. (ET>0. 5 Ge. V for Ecal. Plus. Hcal. Towers) just low-luminosity Pile-Up !! Usually jet thresholds are applied around ETcal>30 Ge. V which is about ETrec>15 Ge. V, hence we are very sensitive to the randomness of the Pile. Up collisions. Hence only about 50% of the events would remain selected if one applies a different random sequence of the Pile. Up collisions on the same hard events !! J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 24

Influence of pile-up jets Going to higher ET-thresholds of course reduces the sample of selected events. The mean probability <Pi> is plotted for an event to be selected as a function of the applied threshold. High threshold reduce the selected event sample, hence this is not the solution. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 25

Effect on a simple analysis A cross section can be determined from data events if one knows the efficiency of the event selection criteria, usually from simulation. The efficiency has a statistical uncertainty due to the limited size of the simulation sample, but also from the A robust analysis randomness of the Pile-Up. De/e~3% would have only one peak here Relative uncertainty on cross section (%) The effect of the randomness of the Pile-Up becomes very strong when the selection efficiency is small and only a small sample of events is selected (like in many of our analyses). J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 26

Leptons influencing jet reconstruction Most analyses have clustered all energy (above thresholds) in the combined Ecal. Plus. Hcal calorimeter. This includes also possible isolated and usually hard leptons, or does not include muons. average energy deposite around lepton in top decays (CMS Note 2006/024) hard muon hard electron Ttbar Pile-Up included (ORCA) region to neglect when clustering leptons propagated to calorimeter surface General discussion: should we remove the lepton information (tracker or calorimeter) before applying the jet clustering algorithms. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 27

Jet calibration ‘The’ benchmark for a long-term jet calibration effort. Estimated the effect Dmt of a global relative shift (%) on the jet energy scale. For the light quark jets an in-situ calibration has been applied by forcing the W boson mass constraint. heavy CMS Note 2006/066 light For precise top quark mass measurements the JES is very important. 1 Ge. V effect on mt : D(JES) ≈ 1. 5% (for b-jets) (effect of jet resolution to be checked) J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 28

Jet calibration strategy Factorized approach into natural pieces with additional optional corrections: Allows a thorough understanding of each individual part of a systematic uncertainty on the jet energy scale (factorized uncertainties). Most of the factors can be measured directly from collision data: • L 1: pile-up and effects of thresholds found in min-bias and zero-bias events. • L 2: jet response vs. η relative to barrel found using di-jet balance, etc. • L 3: jet response vs. p. T found in barrel using g/Z + jets, top, etc. Lots of work in progress and being put in place for first data later this year. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 29

Summary/Outlook § Jets are important in every aspect of the CMS experiment and therefore deserve our full attention § Close collaboration between theoretical and experimental studies is productive § Experiments should be ready to incorporate new algorithms whenever proposed by theoretical arguments § But be aware that before the new algorithm can be applied in physics analyses, lots of work need to be done by the experimentalists. J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 30

Back-up items J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 31

ECAL J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 32

HCAL J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 33

Jet constituents J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 34

Particle Flow: basic idea Associate hits with each sub-detector J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 35

Particle Flow: basic idea Make links across sub-detectors J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 36

Particle Flow: basic idea Apply particle identification and separate the particles J. D’Hondt Vrije Universiteit Brussel Jet Workshop – Paris June 20, 2008 37