Presentation is loading. Please wait.

Presentation is loading. Please wait.

Summary of the Hadronic Calibration Workshop 1) Test-beam and Monte Carlo Validation 2) Local Hadron Calibration 3) Jet calibration.

Similar presentations


Presentation on theme: "Summary of the Hadronic Calibration Workshop 1) Test-beam and Monte Carlo Validation 2) Local Hadron Calibration 3) Jet calibration."— Presentation transcript:

1 Summary of the Hadronic Calibration Workshop Tancredi.Carli@cern.ch 1) Test-beam and Monte Carlo Validation 2) Local Hadron Calibration 3) Jet calibration and Etmiss 4) In-situ calibration 5) Tools Agenda: 3 rd ATLAS Hadronic Calibration Workshop Milan, Italy, April 26-27, 2007 For new-comers: very nice pedagogical introduction by M. Lefebvre and P. Loch I try to give a simple overview of interesting points (personal) Impossible to summarize the many detailed discussions in 2 days workshop !

2 Electron and muon well described: G4.8 (new mult. scatt.) slightly better than G4.7 Pions: MC Validation Etot in TileCal ~ -25% ~ -45% Energy independent Long. Energy Profile dE/dx Default QGSP has showers too narrow and too short Adding nuclear cascade models (Bertini) make shower wider and longer in better agreement to data To be addressed: In HEC with Bertini: shower shapes ok, but pions response 5% too high and resolution ~15% too good With Bertini Without Bertini TileCal Without Bertini With Bertini

3 P. Speckmayer MC Validation in CTB CTB TileCal 90 o energy fraction in the highest Cells of each layer V. Kazanine Second moment shower length MC: QGSP+Bertini (best knowledge) mean shower length Second radial cluster moment Agreement Tile standalone and CTB Lar/Tile +20% -30%

4 Contributions to the jet signal: Try to address reconstruction and calibration through different levels of factorization physics reaction of interest (parton level) lost soft tracks due to magnetic field added tracks from underlying event jet reconstruction algorithm efficiency detector response characteristics (e/h ≠ 1) electronic noise dead material losses (front, cracks, transitions…) pile-up noise from (off-time) bunch crossings detector signal inefficiencies (dead channels, HV…) longitudinal energy leakage calo signal definition (clustering, noise suppression,…) jet reconstruction algorithm efficiency added tracks from in-time (same trigger) pile-up event The Problem P. Loch

5 Hadronic Calibration Models Model I: Physics object based (Global): –first reconstruct hadronic final state objects like jets and missing Et using calorimeter signals on fixed electromagnetic energy scale –then calibrate the jets in-situ using physics events feedback calibration to calorimeter signals for missing Et calculation –a priori use “MC Truth” in simulations for normalization uses full physics simulations to determine hadronic calorimeter calibration some direct bias due to choice of physics final state and jet reconstruction Model II: Detector-based objects (Local): –reconstruct calorimeter final state objects like cell clusters first calibrate and those using a local normalization and corrections (reference local deposited energy in calorimeter) –reconstruct physics objects in this space of calibrated calorimeter signals –apply higher level corrections for algorithm inefficiencies determined in-situ or a priori, as above no direct physics object bias, but strong dependence on simulations for determining local calibration functions P. Loch

6 Local Hadron Calibration in a Nut-shell Define calo signal using topo- cluster with noise cut 4/2/2 (seed/neighbors/perimeter) Classify as em, had or unknown Apply weights w=E tot /E rec from MC to hadronic cluster Apply out-of-cluster corrections Apply dead material corrections to had and em cluster separately Example: Weighting Table 2<  <2.2 Example: Classification Table 2<  <2.2 4<E cl <16 GeV New phase space method: Instead of calibration hits Count clusters: hadronic e.m. Dead material

7 Weighting in Local Hadron Calibration - Resolution unweighted weighted Improvement due to weighting Resolution with respect to true deposited energy in calorimeter S. Menke

8 G. Pospelov,S Menke 100 GeV charged  Linearity Resolution Slight over estimation of total energy (still to be understood) TOPO_EM: sum of energy in all topo clusters at em scale TOPO_DM: “topo_em” + reconstructed DM energy. Dead material correction recovers mean, but not fully resolution Local Hadron Calibration Performance in ATLAS

9 Large eta: problem with FCAL em scale Signal recovered at high energy, problem at low energy: E=20 GeV: 5% missing Resolution improves for E>50 GeV only Neutral pion (not shown): stability of signal under weights EM scale off for low energy and eta>3.2 Linearity Resolution Charged pions S. Menke, G. Pospelov

10 DM Corrections in Barrel CTB2004 Mean Resolution Problem at energies below 20 GeV. Poor correlation between DM loss and layer energies. Example: DM losses between LAr and Tile, 9 GeV pions T. Carli, K. J. Grahn  =0.45 E=50 GeV DM corrections work on MC and data.

11 Weighting in Barrel CTB 2004 P. Francavilla Apply ATLAS weights to CTB Look at Data vs MC: Need to extract weights/DM from CTB in future Pion mixed with protons Drop related to possible problem in TileCal (on-going task force) On Had-scale Data/MC as good as on em-scale for linearity (just shift), but resolution 30-40% worse in Data than in MC on had-scale (about ok on em.scale) Ongoing efforts to validate LocalHadronCalib on test-beam: best knowledge MC and standard method from Athena (first results soon)

12 Alternative Layer Weighting in Barrel CTB04 (Bias) :Reduced, small Improvement by 16% (20%) in RMS(  E) at 9 ( 50) GeV Works also at low energy ! 50 GeV pions at  =0.45 9 GeV pions  =0.45 (T. Carli, F Spano’, P. Speckmayer) Exploit information of shower fluctuations in longitudinal layers Build-up correlation matrix, map-out layer weights from calculated eigenvectors True E deposit Weighted E E on e.m. scale

13 Collect all electromagnetic energy cell signals into projective towers Cancel noise by re-summation of these towers –Towers with E 0 until the resulting protojet has E>0 (all cells are kept!) Run jet finding on the protojets –Results are “uncalibrated” electromagnetic energy scale calorimeter tower je Apply cell level calibration Additional corrections for residual Et and η dependencies of the reconstructed jet energy, and since recently also for jet algorithm dependencies, are applied More corrections determined from in-situ calibration Introduction Global Jet calibration Schemes P. Loch

14 Cone QCD jets with R = 0.7 For calorimeter tower jet (or recently topo-jets on em scale): 1) Find matching truth jet 2) Extract cells from tower jet 3) Fit cell signal weights w i with constraint Correct residual (Et,η)-dependent signal variations after cell weighting Introduction Global Jet calibration Schemes 1) “H1-style” Calibration Weights  Successfully applied in many physics analysis baseline for a long time, benchmark in the near future from MC Use weighted sampling energy sums in jets Weights are parameterized as function of E jet, E em, E had in a given jet Few numbers, does not need cells Not quite optimal but fast and a good candidate for HLT jet calibration 2) Sampling energy based (Chicago) 3) Pisa weighting use jet energy together with cell signal in weighting functions Fully parametrized weighting functions Final recommendation for jet calibration basis needs ATLAS collision data Parametrisation of w=f(E cell )

15 Global Jet calibration – Linearity Performance Cone R=0.7 tower isolated jets - At high energy fine (<1%), at low energy <100-200 GeV worse 2-4% - Some deviation beyond 1% for Pisa. uncalibrated calibrated uncalibrated calibrated

16 The energy density calibration methods have better resolution compared to sampling method at higher energies. - the difference decreases at higher eta. Global Jet calibration – Resolutions Performance Design goals  (E)/E = 50%/  E  3% for |  | < 3  (E)/E = 100%/  E  5% for 3 < |  | < 5 Typically achieved: 0<  <0.7 ??

17 Calorimeter Cluster Jets Use topo cluster with local hadronic calibration –Factorizes hadronic calibration, signal definition corrections, dead material corrections e/h corrected at the detector level, no jet context needed –Uses “3-d energy blobs” rather than towers Implied noise suppression → cluster provide signal of (constant) minimum significance over fluctuations Clusters are freely located in calorimeter Seed splitting due to fixed geometry grid like for tower jets less likely –Provides better calibrated input to jet finder Relative mis-calibration much smaller, ~5% at most Allows possible input selection to be more comparable with particle jets calorimeter domain jet reconstruction domain physics analysis domain “Towers” less optimal in noise suppression Present base-line: P. Loch

18 S.Menke/G. Pospelov K T -jets from TopoClusters +dead material +Out-of-cluster weighted em-scale Local hadron calibration – Linearity on Jets Apply local hadronic calibration to jets -Flat response in Et within +/- 2% ~50-400 GeV range -Rapidity dependence ok up to | η |≈2.7 (em scale calibration problem in FCal) -Still missing: calibrations for e.m. clusters additional jet correction for low energy particles (not reaching calorimeter) ? jet algorithm corrections ? …like out-of-cone showering

19 Missing Et Reconstruction: Options and Default Strategy highlighted boxes indicate the default configuration Refined MET calibration Based on cells with calibration from physics objects each cell belongs to one or no physics object Calibration is directly derived from physics object calibration Cells in : electrons, photons, taus, jets, muons unused TopoClusters, outside of TopoClusters Uses reconstructed high quality muons with Pt from external spectrometer Based on jet energy correction for dead material between Lar/Tile Old ATLAS standard

20 Ztautau Wmunu H1-Style 0.53 Wenu Zee RefinedLocHad 0.530.60 0.54 0.480.63 0.600.610.46 Ex(y)miss Resol = K *  SumET J1 J2 H1-Style 0.50 J3 RefinedLocHad 0.500.47 0.49 0.50 0.53 0.500.550.50 J4 J5 J6 0.52 0.54 0.52 0.54 0.56 0.57 Refined Calibration gives the best results on channels with electrons For jets all method about equal Missing Et Performance Example: Ex(y)miss Resol = K *  SumET

21 In-Situ Calibration using gamma-jet Events Gaussian likelihood: Calibrated jet energy: Idea: derive in-situ calibration from data only by requiring Et-balance between photon and Jet in gq   q M. Hurwitz A. Farbin 17<p T <35 GeV Before cuts: After cuts: Balanced not equal calibrated !

22 See good balance using binned fit Nonlinearity with respect to truth at low energies –Good balance doesn’t guarantee a calibrated jet Performance depends on sample being used to assess it –With nice balance cuts, calibration looks very good above ~200 GeV In-Situ Calibration using gamma-jet Events uncalibrated calibrated Apply calibration to all events M. Hurwitz A. Farbin

23 In-Situ Calibration using gamma-jet Events Doug Schouten Use Etmiss projection to derive jet calibration More inclusive: less sensitive to jet algorithm Less sensitive to photon/jet balance? Result: With pile-up Without pile-up Pile-up needs to be taken Into account Works for E>100 GeV

24 Physics Effects in Pt-balance in Z+jet Events A. Gupta et al. To assess performance study average pt-balance of leading jet with E T >10 GeV Pt-balances within +-5-10% only Better at high energy Problems below 100 GeV Sensitivity to physics modeling (Herwig vs Pythia)

25 Conclusion Hadron calibration in an collider experiment is rather complex and challenging Unlikely that it works at day 1 Need to put tools in place and study performance and limitations of methods In my opinion, important to factorise and study effect-by-effect  only possibility to understand something) Local Hadron Calibration (for me) most promising to give best long term performance Strategy to get all weights from MC, need to be validated on test-beam Likely to fail due to detector geometry etc. (need a lot of work at the beginning) Data drive in-situ approach will most likely have best performance on early data  need correct functional form fit calibration parameters from data


Download ppt "Summary of the Hadronic Calibration Workshop 1) Test-beam and Monte Carlo Validation 2) Local Hadron Calibration 3) Jet calibration."

Similar presentations


Ads by Google