Post on 31-Dec-2015
description
Electron identification capabilities of the CBM experiment at FAIR
Semen Lebedev GSI, Darmstadt, Germany and LIT JINR, Dubna, Russia
Claudia HöhneGSI, Darmstadt, Germany
Ivan KiselGSI, Darmstadt, Germany
Gennady OsoskovLIT JINR, Dubna, Russia
S.Lebedev@gsi.de
S. Lebedev et al, Electron Identification in CBM DPG 2010 2/13
The CBM experimentAim: Investigation of the QCD phase diagram -> measurement of hadronic and leptonic
probes in large acceptance
MVD+STS: tracking, momentum determination, vertex reconstruction
RICH & TRD:
electron identification pion suppression 104
for low mass vector mesons and J/ψ reconstruction
up to 10MHz interaction rate → rare probes
RICHTRD
TOF ECAL
PSDSTS
magnet
TOF: hadron identificaion
ECAL: ECAL: photons, photons, ππ00, , γγ
PSD: event characterization
S. Lebedev et al, Electron Identification in CBM DPG 2010 3/13
The CBM RICH detector
RICH hits (blue), found rings (red), track projections (green).
ElectronsElectrons
PionsPions
Radius versus momentum for Radius versus momentum for reconstructedreconstructed ringsrings..
RICH characteristics:• radiator: CO2 length 1.5 m; Pth,π =4.65 GeV/c
• glass mirror of 6 mm thickness:
3m radius; 11.8 m2 size• photodetector Hamamatsu H8500 MAPMT:
2.4 m2 -> 55k channels
Mean number of hits per electron ring is appr. 21
RICH: electron identification by Cherenkov radiation
All components are available on market.
S. Lebedev et al, Electron Identification in CBM DPG 2010 4/13
Reconstruction in the CBM RICH detector
Ring finding in RICHRing finding in RICH ring-track matchingring-track matchingExtrapolation of STS tracks onto the Extrapolation of STS tracks onto the photodetector planephotodetector plane
Main problems of ring recognition in CBM RICH:• high ring density (~80 rings per central Au-Au at 25 AGev, many secondary
electrons);• many overlapping rings;• distortions and elliptic shape of the rings;• measurement errors: multiple scattering, B-field, detector granularity• ring-track matching (high density of projected tracks)
S. Lebedev et al, Electron Identification in CBM DPG 2010 5/13
Hough Transform ring finding
Hough Transform:large combinatorics => slowLocalized Hough Transform: much less combinatorics => fast
S. Lebedev et al, Electron Identification in CBM DPG 2010 6/13
Ring finding resultsSimulation: central Au+Au collisions at 25 AGeV beam energy (UrQMD)
5e+ and 5e- embedded as signal
Accepted rings = rings with >= 5 hits
Ring reconstruction efficiency for embedded e+ and e-.
Nof fakes = 3.56 per eventNof clones = 1.03 per event
2x Intel Xeon X5550 processors at 2.67GHz (8 cores)
Time/ev., ms
Serial 5.8Parallel 3
Ring finder was systematically studied for different variations:• high ring density (>factor 2);• reduced number of hits per ring (70%);• additional errors due to mirror surface inhomogeneity;
appr. 80 rings per one event
S. Lebedev et al, Electron Identification in CBM DPG 2010 7/13
Electron Identification in RICH
ElectronsElectrons
PionsPions
Radius versus momentum for reconstructedRadius versus momentum for reconstructed rings rings in central Au+Au collisions at 25 AGeV beamin central Au+Au collisions at 25 AGeV beam energy energy
for UrQMD eventsfor UrQMD events (large RICH) (large RICH). .
Pions which were matched to secondary electron RICH rings
Pions – dashed lineElectrons – solid line
BB
AA
A
B
Ellipse fitter
Ring-track distance
S. Lebedev et al, Electron Identification in CBM DPG 2010 8/13
TRD detector
TRD characteristics:• Each station consists of several identical layers
• Each layer consists of a multilayer dielectric radiator, and of a gaseous detector (85%Xe+15%CO2).
• Pad readout with coordinate resolution 0.03-0.05 cm across and 0.27-3.3 cm along the pad (tracking!)
The CBM TRD is intended for tracking and improved electron identification for p > 1.5 GeV/c.
TR production by relativistic particles when crossing material boundaries with different dielectric constants ε.
TR production for γ > 1,000 -> electron/pion separation
S. Lebedev et al, Electron Identification in CBM DPG 2010 9/13
Event Reconstruction in TRD
TRD event reconstruction:• tracking -> collect hits in tracks (track propagation from STS)• electron identification -> using energy losses
electrons
Using only standard cuts is not enough -> advanced algorithms were implemented, which allow to reach pion suppression 200-700 at 90% electron efficiency.
Energy loss: compare simulation (red) and experimental data (black)
π
e-
Sum of energy loss in 12 layers of TRD.Long tail of Landau distribution for π
Methods:• Likelihood• Artificial Neural Network (ANN)• Ordered statistics (mediana)• Boosted Decision Tree (BDT)
S. Lebedev et al, Electron Identification in CBM DPG 2010 10/13
Results of electron Identification in TRD (I)
Electron identification algorithms were systematically studied for different variations:• TR parameters (different radiators);• TRD detector layout;• Error on energy loss resolution; 90% electron efficiency
electrons and pions with parameters θ = (2.5, 25), ϕ = (0, 360)for certain 1.5 GeV/c momentum
Method name π suppression
Boosted Decision Tree 660Artificial Neural Network 534Photon cluster counting 150Ordered statistics (mediana) 140Simple cut on energy loss sum 5
S. Lebedev et al, Electron Identification in CBM DPG 2010 11/13
Results of electron Identification in TRD (II)
Statistics: 1M electrons and 1M pions with parameters θ = (2.5, 25), ϕ = (0, 360) for certain momenta (1, 1.5, 2, 3, 4, 5, 7, 9, 11, 13 GeV/c).
90% electron efficiency
Pion suppression vs. momentum
Black: BDT method Blue: ANN method
S. Lebedev et al, Electron Identification in CBM DPG 2010 12/13
Results of electron Identification in CBM
The RICH detector alone yields a pion suppression factor of 500 at an electron identification efficiency of 83.3% while in combination with TRD
a factor 104 is reached at 69% efficiency.
S. Lebedev et al, Electron Identification in CBM DPG 2010 13/13
Summary
• Development of fast (parallel!) and efficient algorithm for ring recognition in CBM RICH (3 ms per event for parallel version).
• Study of different algorithms for electron identification in TRD.
• RICH and TRD provide sufficient good e-identification (efficiency, π suppression) to allow feasibility of low-mass vector meson and J/ψ measurements.
S. Lebedev et al, Electron Identification in CBM DPG 2010 14/13
• Backup
S. Lebedev et al, Electron Identification in CBM DPG 2010 15/13
Algorithm of electron Identification in TRD
Two steps of the algorithm:• energy loss transform• evaluate probability using BDT
Transform:• Prepare probability density function (PDF) for ordered energy losses (store them in file)• Sort energy losses• Calculate likelihood ratio for each energy loss:L = PDF_pi / PDF_pi + PDF_el
Task: distinguish electrons and pions using 12 energy loss measurements
BDT:•Boosted decision tree (BDT) classifier from TMVA package was used.• Before using BDT has to be trained
Transformation is very important step, without this step classifiers could not be trained properly.
S. Lebedev et al, Electron Identification in CBM DPG 2010 16/13
Robustness of el. id. method
• One should consider not only a pion rejection procedure, as it is, it is necessary to take into account its robustness to such experimental factorsexperimental factors as calibration of measurements, pile up of signals etc. • Most probable value of Eloss for pion is around 1.05-1.5 keV• Add error to the energy loss for each hit: Eloss=Eloss+Gauss(0, Sigma)• BDT method was used
90% electron efficiency