Algorithms and Methods for Particle Identification with ALICE TOF Detector at Very High Particle
Multiplicity
TOF simulation group
B.Zagreev
ACAT2002, 24 June 2002
ALICE Time-Of-Flight detector (TOF)R=3.7m, S=100m2, N=160000
Problems• Need of very high time resolution (60 ps - intrinsic, 120 ps -
overall)• High multiplicity dN/dY8000 primaries (12000 particles in
TOF angular acceptance)– 45(35)% of them rich TOF, but they produce a lot of
secondaries• High background
– total number of fired pads ~ 25000 => occupancy=25000/160000=16%
– but only 25% of them are fired by particles having track measured by TPC
• Big gap between tracking detector (TPC) and TOF– big track deviation due to multiple scattering– TRD tracking ???
Procedure
• Software framework for ALICE - Aliroot (ROOT based + GEANT3). Then we have the same environment for simulation and reconstruction.– Tracking (Kalman filtering)– Matching– Time measurements– Particle identification
Matching
• Probe tracks algorithm• Kalman filtering• Combined method (Kalman + probe tracks)
Probe tracks algorithm• All tracks are ordered according their transverse
momentum (the higher momentum the less track errors)
• Starting from the highest momentum track, for each track at the outer layer of TPC, a statistically significant sample of probe tracks is generated and tracked in Aliroot (GEANT geometry and medias, magnetic field etc.)
• So for a given track we have a set of TOF pads crossed by these probe tracks. We chose, roughly, the pad crossed by biggest number of probe tracks.
Probe tracks algorithm
The end of reconstructed track (r, p) in TPC or TRD
Fired pads
Kalman filtering + probe tracks algorithm
The ends of reconstructed track (r, p)
S2
3
S1
R1
R2
R1<R2 but S1<S2 !
TPC (TRD)
TOF
Time measurements
• Time-amplitude and other corrections• Time zero calculations
1. Consider a very small subset (n) of primary “gold” tracks. Let l1…ln, p1…pn, t1…tn - length, momentum and time of flight of corresponding tracks. Now we can calculate the velocity (vi) of particle i in assumption that particle is pion, kaon or proton.
2. Then we can calculate time zero:
3. We chose configuration C with minimal 2(C) ~ (ti
0(C) - <ti0>(C))2
Combinatorial algorithm for t0 calculation
ii
ii t
vlt
p)K,,( 0
Combinatorial t0 distribution (250 events)
Results for t0 combinatorial algorithm
Now 30sec (PIII)
t0 calculation, all tracks as pionstpmltclt 1// 22
0
T0 calculations with not matched hits
Particle identification
1 Simple contour cut2 Neural network3 Probability approach
Mass distribution, 50 HIJING events, B=0.4T
1/ 222 ltcpm
Mass-momentum distribution, HIJING1/ 222 ltcpm
TOF efficiency and contamination
Neural network PID• ROOT based network constructor (Anton
Fokin, http://www.smartquant.com/neural.html)
• 1 hidden layer perceptron (different number of neurons)
• output: 3 neurons for , K or p• input parameters: mass, momentum and matching
parameter• Good results for not overlapping clusters of
particles. For realistic distribution performance is not so good
Mass-momentum distribution, HIJING1/ 222 ltcpm
neurons
neurons
Fit with 2D function
Probabilities for PID, (1.5-2 GeV/c)
50%50%
70%
PID at STAR experiment
e Kp
12 March 2002 Karel Safarik: ALICE Performance 17
Particle IdentificationParticle Identification /K K/p/K K/pTPC and ITSTPC and ITS ( (dEdE//dxdx))
/K K/p/K K/pTOFTOF
/K K/p/K K/pHMPIDHMPID (RICH) (RICH)
e/ e/TRDTRD //
PHOSPHOS Muon Muon detectordetector
0 1 2 3 4 5 p (GeV/c)
1 10 100 p (GeV/c)
x
Combine PID
pions
kaons
y
gK(x,y)~gK(x)gK(y)1D cuts
2D cut
gK(x)
gK(y)
Conclusions & plans• A number of methods and algorithms were developed for
particle identification at high multiplicity and background
• Results obtained are reasonable and allow to fulfil physical tasks
• Plans:– Complete probability algorithm, combine several detectors– Kalman filtering for matching– Try to realize iterative algorithm for tracking, matching and
particle identification