Particle Identification with the Transition Radiation Detector

Transcription

1 Particle Identification with the Transition Radiation Detector Yvonne Pachmayer, University of Heidelberg for the TRD PID Group Particle identification algorithms Performance

2 Transition Radiation PID with ALICE TRD p > 1 GeV/c Test Beam Data TR creation for γ > 800 TR photons absorbed in the drift volume preferentially close to radiator Time evolution of the signal recorded improved e/ separation power Specific energy loss and + TR production measured

3 TRD PID Methods Different analysis methods Truncated mean method LQ1D = 1-dim likelihood (total integrated charge) LQ2D = 2-dim likelihood (charge + position) Multi-dimensional PID LQ3D/LQ7D = 3/7-dim likelihood (charge + position) (Neural network (NN)) Bayesian approach within ALICE Combines PID method from various detectors Priors for ITS+TPC Propagation factors

4 Onset of Transition Radiation Muons (cosmic rays) de/dx only Compilation of TRD de/dx +TR data from proton-proton collisions, cosmic rays and test beam data Fit with Aleph parameterisation + TR contribution X-G. Lu, doctoral thesis, Universität Heidelberg, Germany

5 Truncated Mean Method cpass0/cpass1 Time-bin calibration Cosmics/V0/... X-G. Lu, doctoral thesis, Universität Heidelberg, Germany

6 Truncated Mean Method Truncation of TRD signal to obtain (almost) Gaussian distribution N signal clusters above baseline (up to 6 number of time bins) Signal clusters not completely independent (ion drift) Truncation fraction f = 0.55 Truncated mean signal from the M smallest signal clusters:

7 Truncated Mean Method Performance Idea by Marian to combine TRD truncated mean with TPC measurement Advantage in case part of TPC off However no manpower for this at the moment

8 Truncated Mean Method Summary PID via truncated mean accessible in AliPIDResponse ESDs and AODs supported Cuts on signal and n σ supported (up to triton) (Automatic η, cluster and centrality correction) Parameters available for LHC13b-f : p-pb 13TeV (best resolution ~12%) LHC15n : pp 5 TeV LHC15o : Pb-Pb 5 TeV Florian Herrmann Lukas Altenkämper

9 Transition Radiation Likelihood Methods TR creation for γ > 800 TR photons absorbed in the drift volume preferentially close to radiator Time evolution of the signal recorded improved e/ separation power Specific energy loss and + TR production measured

10 TRD - Likelihood analysis method A. Andronic P ( Ei e ) Probability that the energy deposited Ei in (layer i) was produced by an electron Likelihood to be an electron: Pe P ( Ei e ) = Pe + Pπ or respectively other particle species N Pe = P ( E i e ) i=1 N Pπ = P ( E i π ) i=1 N = number of modules (e.g. ALICE TRD Nmax = 6)

11 Implementation in AliRoot Documentation The TRD PID can be applied in two ways: via a cut on the probability of a given particle species AliPIDResponse::EDetPidStatus AliPIDResponse::ComputeTRDProbability (const AliVTrack *track, Int_t nspecies, Double_t p, AliTRDPIDResponse::ETRDPIDMethod PIDmethod) Double_t p returns the probabilities of a given particle species (electron, muon, pion, kaon, proton). Values between 0 to 1 are returned (The default value is 0.2). With AliTRDPIDResponse::ETRDPIDMethod PIDmethod one can select if the LQ1D/LQ2D/LQ3D/LQ7D method should be used: AliTRDPIDResponse::kLQ1D or analogous for other methods electron identification at a fixed electron efficiency level Using the function Bool_t AliPIDResponse::IdentifiedAsElectronTRD(const AliVTrack *vtrack, Int_t &ntracklets,double_t efficiencylevel,double_t centrality, AliTRDPIDResponse::ETRDPIDMethod PIDmethod) Int_t &ntracklets returns the number of tracklets used for the PID calculation. Double_t efficiencylevel: available electron efficiencies are (0.05%, 0.1%, 0.15%..., 0.85%, 0.9%, 0.95%). Double_t centrality for pp and p-pb collisions the centrality is -1; for Pb-Pb the respective centrality is used. (The TRD signal has a centrality dependence in Pb-Pb) With AliTRDPIDResponse::ETRDPIDMethod PIDmethod one can select if the LQ1D/LQ2D/LQ3D/LQ7D method should be used: AliTRDPIDResponse::kLQ1D or analogous for other methods Assumes that a TOF epid cut of ±3σ was applied

12 References and Thresholds available Period Comment LHC10h Pb-Pb 2.76 TeV LHC11h Pb-Pb 2.76 TeV LHC12c-f pp (using data from LHC13 p-pb) LHC13b-f p-pb LHC15f pp 13 TeV LHC15n pp 5 TeV LHC15o Pb-Pb 5 TeV Florian Herrmann

13 Electron Identification Likelihood on total charge (LQ1D) Simplest method of TRD PID: Based on the charge measured Likelihood on total charge in a single tracklet Data sample: V 0 sample topological identification + TOF&TPC PID e from conversion, pions from K0-decays, protons from Λ-decays D. Lohner, doctoral thesis, Universität Heidelberg, Germany

14 Electron Identification Likelihood on total charge (LQ1D) Simplest method of TRD PID: Based on the charge measured Likelihood on total charge in a single tracklet Data sample: V 0 sample topological identification + TOF&TPC PID e from conversion, pions from K0-decays, protons from Λ-decays M. Fasel, doctoral thesis, TU Darmstadt, Germany, Nov 2012

15 Electron Identification Likelihood on total charge (LQ1D) Cross-Checks Cross-Check with electrons from π0 γγ and then γ e+e- Electron identification efficiency Electron likelihood Tracking efficiency in MC (realistic detector description) M. Fasel, doctoral thesis, TU Darmstadt, Germany, Nov 2012

16 PID Systematic Uncertainty

17 Electron Identification Usage of Temporal Evolution of the Signal (I) Time evolution of the signal recorded TR photons absorbed in the drift volume preferentially close to the radiator Characteristic peak at large drift times enables even better separation of electrons and pions Treat slices as independent J. Klein, NIMA706(2013)23

18 Single Electron Analysis p-pb mb

19 Electron Identification Usage of Temporal Evolution of the Signal (II) References and thresholds determined for Electron efficiencies: Number of tracklets LQ1D robust and simple method with reasonable pion suppression LQ2-7D and NN improvement in pion suppression (#chambers in a stack): 1-6

20 Electron Identification Usage of Temporal Evolution of the Signal (II) The more differential, the more statistics needed Likelihood methods Same data sample for performance study as for for reference/threshold creation No systematics on electron efficiency Ideal tracks: 6 TRD tracklets + missing slices cut Limited in p due to statistics Treat slices as independent NN Statistics NN structure

21 Electron Identification Usage of Temporal Evolution of the Signal Example Application using LQ2D Physics Case: J/ψ analysis Data sample Pb-Pb (centrality: 0 40%) Single track transverse momentum pt > 2 GeV/c PID: TPC only and TPC+TRD TPC cuts: nσe < 2, rejection: nσπ > 3.5, nσp > 4 cut on standard deviation of (de/dx - <de/dxe hyp>)/σ TRD PID LQ2D method, electron likelihood > tracklets (chambers in a stack) PID only used when available (10/18 TRD supermodules installed in 2011) Despite reduced coverage the TRD PID yields an improvement of S/B by ~30% TRD PID improves S/B ratio I. Arsene

22 Performance in Pb-Pb (2010) TRD signal has a centrality dependence D. Lohner, doctoral thesis, Universität Heidelberg, Germany

23 Performance in Pb-Pb (2010) TRD signal has a centrality dependence D. Lohner, doctoral thesis, Universität Heidelberg, Germany

24 Performance in Pb-Pb (2015)

25 Thresholds/Parameters Not created charge dependent For study input sample and sample a. u. PID Performance Charge Dependence used for evaluation with the respective charge selection Charge dependent effects LQ1D less effected LQ2D positive charge smoother p shape p > 1.5 GeV/c stat. fluctuations 25

26 Missing Slices Cut Pion Efficiency 6 tracklets Missing slices cut None of the contributing tracklets has a missing slice for LQ1D and LQ2D No missing slices cut Min 1 slice per layer not empty LQ1D: Qtot >0 LQ2D: Q0 && Q1 > 0 Checks via true quantiles Not calculated thresholds TRD Meeting 26

27 Missing Slices Cut Pion Efficiency for different Electron Efficiency No gap & no missing slices cut Recover performance via lower electron efficiency cut TRD Meeting 27

28 Missing Slices Cut LQ3D Similar effect for LQ3D Less fast rise with no cut, probably statistics TRD Meeting 28

29 Comparison Data vs MC Pion Efficiency 6 tracklets Parameter creation analogous to data MC samples LHC13b2_efix_p1 LHC13b2_efix_p2 LHC13b2_efix_p4 Performance evaluated using the V0 sample used to create the references and thresholds MC shows an even stronger pion rejection than data when applying the missing slices cut Is the method analogous to data not valid? Differences in PID Response Is the separation of the different particle species larger? However common scaling factor Differences in tracking 29

30 Likelihood Methods Summary PID via likelihood method accessible in AliPIDResponse ESD/AOD analysis LQ1D, LQ2D, LQ3D, LQ7D available References and thresholds (layers: 1-6; efficiency %; centrality dependent) Not considered: charge dependence Thresholds: assume that a TOF eid cut was applied before

31 TRD PID QA (I) PID QA 'framework' for general user to judge that PID worked independent of his/her analysis task (parameters loaded etc.) for experts to spot any problems the general user reports should run on the lego train (in GSI train framework: PIDqa.root) Data ESD MC ESD Data AOD MC AOD Basic x x x x Likelihood x x x x Truncated Mean x x x x V0 Basic x x V0 Likelihood x x V0 Truncated Mean x x MC truth Basic x x 31

32 Likelihood TRD PID QA (II) Likelihood LQ1D and LQ2D vs p for 5 particle species (electron to proton) TPC nsigma vs p w/o TRD PID (LQ1D and LQ2D 90% electron efficiency) 32

33 Monte Carlo PID Response M. Voelkl Present implementation GEANT3 and 'after burner' Comparison MC vs test beam Test beam: only total integrated charge available (w/o radiator TR) de/dx and de/dx +TR GEANT3 vs GEANT4 GEANT4 Tuning of parameters ongoing No perfect representation possible due to varying incident angle on fibres Test beam vs ALICE (no material in front of TRD - Bremsstrahlung)

34 Neural Network with TMVA (I) TRD PID with NN TMVA Martin Kroesen LQ2D NN training in momentum ranges ideal tracks ideal tracks

35 Neural Network with TMVA (II) Test with local AliRoot version TMVA w/o TRD PID Implementation needs tuning Slow & large weights file Next steps Performance studies Non ideal tracks Pb-Pb TRD PID 90% electron efficiency

36 Single Electron Analysis p-pb mb Tracking Systematics Martin Fleck, Master Thesis PID systematics 3% Tracking systematics ~10%

37 Back-up

38 Data Sample Data samples needed to reference/threshold creation Test beam (total integrated charge) Cosmics (track inclination) V0 sample topological identification + TOF&TPC PID e from conversion, pions from K0-decays, protons from Λ-decays D. Lohner, doctoral thesis, Universität Heidelberg, Germany

39 Truncated Mean Method nσ Approach (I) Lukas Altenkämper Input: V0 sample

40 Truncated Mean Method nσ Approach (II) Lukas Altenkämper Input: V0 sample Use βγ information to produce correction map Compare mean signal with expected signal in βγ - η bins

41 Truncated Mean Method nσ Approach (III) Lukas Altenkämper nσ rather constant in after correction Only shown for pions, also works for other particle species Pions identified via TOF/TPC (not V0 sample)

42 TRD - Likelihood analysis method Definitions A. Andronic electron efficiency is the rate of electrons, which are identified correctly pion efficiency is the rate of pions which are misidentified as electrons pion suppression is the reciprocal of pion efficiency Design goal: example ALICE TRD Pion efficiency <1% at 90% electron efficiency

43 Total Integrated Charge (Slice 0) Lennart Volz Fit Function XianGuo:

44 Charge dependence of total integrated charge Electrons 44

45 Charge dependence of total integrated charge Pions 45

46 6 tracklets 6 tracklets

47 Missing Slices Lennart Volz Data Data p > 1 GeV/c MC p > 1 GeV/c p-pb mb Small layer dependence (7% 9%) Minor charge dependence Minor p dependence Correlation with other missing slices Similar dependence in MC In analysis no missing slice cut reduces the statistics

48 Pion Efficiency Track Parameter Dependence