PARTICLE IDENTIFICATION IN THE HADES SPECTROMETER AND PARTICLE PRODUCTION IN C+C AT 2 A GeV

The HADES spectrometer installed at GSI Darmstadt is devoted to study production of di-electron pairs from proton- and pion-induced reactions and nucleus-nucleus collisions. Extraction of rare lepton pairs in high hadron multiplicity events requires efficient particle identification (PID). In HADES electrons are identified by a RICH as well as a Pre-Shower and a TimeOfFlight (TOF) detector. For all charged particles momentum is measured by a tracking system combined with a toroidal superconducting magnet, and the TOF detector provides velocity and energy loss.The particle identification method has been implemented, allowing efficient identification of particles, using full experimental information from all subdetectors. The basis of the method is test of hypothesis, that the reconstructed track can be identified as certain particle specie. Several measured variables associated to each identified track from various subdetectors are used to provide a set of probabilities in individual PID algorithms, which are then merged assuming their statistical independence. For the resulted PID probability calculation the Bayes method taking into account the prior abundance of individual particle types, as well as the known detector response, is implemented. The performance of the method - in terms of efficiency and purity - is then evaluated in detailed simulations.To demonstrate the method performance, single particle spectra of charged hadrons and electrons from C+C at 2 A GeV are presented and compared with results of corresponding simulations. In order to verify the method the proton and pion yields and transverse mass and rapidity distributions are compared with existing data.

P.Tlustý17, G.Agakichiev5, C.Agodi2, H.Alvarez-Pol19, E.Atkin13, A.Balanda4, G.Bellia2,3, D.Belver19, J.Bielcik5, M.Böhmer14, H.Bokemeyer5, J.Boyard16, P.Braun-Munzinger5, V.Chepurnov6, S.Chernenko6, T.Christ14, R.Coniglione2, H.Daues5, J.Diaz20, R.Djeridi8, F.Dohrmann18, I.Duran19, T.Eberl14, V.Emeljanov13, L.Fabbietti14, O.Fateev6, C.Fernandez19, P.Finocchiaro2, J.Friese14, I.Fröhlich8, B.Fuentes19, J.Garzon19, R.Gernhäuser14, M.Golubeva11, D.Gonzalez19, E.Grosse18, F.Guber11, J.Hehner5,

T.Heinz5, T.Hennino16, S.Hlavac1, J.Hoffmann5, R.Holzmann5, A.Ierusalimov6, I.Iori9,10, M.Jaskula4, M.Jurkovic14, B.Kämpfer18, K.Kanaki18, T.Karavicheva11, I.Koenig5, W.Koenig5, B.Kolb5, U.Kopf5, R.Kotte18, J.Kotulic-Bunta1, R.Krücken14, A.Kugler17, W.Kühn8, R.Kulessa4, A.Kurepin11, T.Kurtukian-Nieto19, S.Lang5, J.Lehnert8, C.Maiolino2, J.Marín19, J.Markert7, Y.Mishin13, N.Montes19, J.Mousa15, M.Münch14, C.Müntz7, L.Naumann18, J.Novotný17, W.Ott5, J.Otwinowski4, Y.Pachmayer7, Y.Panebratsev6, V.Pechenov6, T.Perez8, J.Pietraszko5, R.Pleskač17, V.Pospíšil17, W.Przygoda4, N.Rabin12, B.Ramstein16, A.Reshetin11, J.Ritman8, G.Rodriguez Prieto19, M.Roy-Stephan16, A.Rustamov5, J.Sabin Fernandez19, A.Sadovsky18, B.Sailer14, P.Salabura4, M.Sanchez19,

P.Sapienza2, A.Schmah5, C.Schroeder5, E.Schwab5, P.Senger5, R.Simon5, V.Smolyankin12, L.Smykov6, S.Spataro2, H.Stelzer5, H.Stroebele7, J.Stroth7,5, C.Sturm5, M.Sudol7,5, A.Titov6, A.Toia8, M.Traxler5, H.Tsertos15, A.Vazquez19, Y.Volkov13, V.Wagner17, W.Walus4, Y.Wang7, S.Winkler14, M.Wisniowski4, T.Wojcik4, J.Wüstenfeld7, Y.Zanevsky6, D. Žovinec1, P.Zumbruch5

HADES@GSIHADES@GSI

e-,e+,p, identification

real-time lepton triggering

M=1-2%@ /

operation with p, , HI beams; B : 0-30, T: 0-60 MeV

beam

RICH

Tracking (MDC)

TOF(eID)

SHOWER(eID)

Magnet

PRINCIPLE:• for each track a probability that it is of a particle type h is calculated, for all possible particle types • Bayes theorem implemented• cut on the resulted probability set to decide on PID

INPUT:• for each track (track candidate) with a given momentum we have a set of independent measured variables• in HADES: TOF/TOFino velocity, energy loss, RICH response, MDC hit, SHOWER response

OUTPUT:• a probability, that a given track corresponds to the particle type h• efficiency and purity for a selected cut

PID MethodPID Method

1)Institute of Physics, Slovak Academy of Sciences, Bratislava, Slovakia 2)Istituto Nazionale di Fisica Nucleare- Laboratori Nazionali del Sud, Catania, Italy 3)Dipartimento di Fisica, Universita di Catania, Catania, Italy 4)Smoluchowski Institute of Physics, Jagiellonian University of Cracow, Cracow, Poland 5)Gesellschaft für Schwerionenforschung , Darmstadt, Germany 6)Joint Institute of Nuclear Research, Dubna, Russia 7)Institut für Kernphysik, Johann Wolfgang Goethe-Universität, Frankfurt, Germany 8)II.Physikalisches Institut, Justus Liebig Universität Giessen, Giessen, Germany 9)Istituto Nazionale di Fisica Nucleare, Sezione di Milano, Milano, Italy 10)Dipartimento di Fisica, Universita di Milano, Milano, Italy 11)Institute for Nuclear Research, Russian Academy of Science, Moscow, Moscow, Russia 12)Institute of Theoretical and Experimental Physics, Moscow, Russia 13)Moscow Engineering Physics Institute (State University), Moscow, Russia 14)Physik Department E12, Technische Universität München, Garching, Germany 15)Department of Physics, University of Cyprus, Nicosia, Cyprus 16)Institut de Physique Nucléaire d'Orsay, CNRS/IN2P3, Orsay Cedex, France 17)Nuclear Physics Institute, Czech Academy of Sciences, Řež, Czech Republic 18)Institut für Kern- und Hadronenphysik, Forschungszentrum Rossendorf, Germany 19)Departamento de Física de Partículas. University of Santiago de Compostela, Santiago de Compostela, Spain 20)Instituto de Física Corpuscular, Universidad de Valencia-CSIC, Valencia, Spain

Normalized probability density distributions

of each measured variable xk determined for each particle type h

• from exp data when possible (good separation of particles)• interpolation and extrapolation of „difficult“ regions“ with overlap from different particles• from simulations if necessary (e.g. RICH response)

STEP I - p.d.f.‘s

- merging of information from several INDEPENDENT measurements, e.g. more subdetectors. If the probability density functions in each measurement are known,then the likelihood to observe track with measured for the particle type h is

- if correlated variables are used, different approach has to be applied, e.g. principal component analysis used in RICH p.d.f. calculation. The number of variables is reduced to number of statisticaly independent „eigenvalues“ by diagonalizing the correlation matrix.

STEP II - Combination of measurements

If probabilities of occurences of individual hypotheses == relative incident rates for each particle type in the PID case P(h) are known (or can be estimated from both experimental data and simulations), then

STEP III - application of Bayes' theorem

where is probability that a track with measured is of a type h.

There is clearly a need to take this into an account, as it changes the decision on the hypothesis test,compare Fig.1 with Fig.2

STEP IV - PID decision

1) A cut on the resulted probability is set to get PID, cut value is usually 0.5 (probability of given particle type > 50%)

> cut

2) For each particle type two „quality“ factors determined (from simulations):

• Efficiency defined as ratio of number of correctly identified particles and number of all particles of given particle type in the input sample

• Purity defined for a given particle type as ratio of the number of correctly identified particles to the number of all identifications of a given type

References:

BABAR Barlow et al. www.slac.stanford.edu/BFROOT/www/Statistics/Report/report.pdfFOCUS hep-ex/0108011STAR Fisyak www.usatlas.bnl.gov/~fisyak/d0/photons/pure.ps HERMES www.phys.ualberta.ca/~mvincter/hermes/documents/joe.ps.gzhypothesis testing: e.g. in A.G. Frodesen, O. Skjeggestad, and H. Tofte,Probability and Statistics in Particle Physics Columbia University Press, 1979, ISBN 8200019063

Hadron IDHadron ID Lepton IDLepton IDHadrons are identified mainly using velocity and momentum measurements.

-

+

C+C, 2AGeV

p*q [MeV/c]

v/c

d

p

e- e+ TOFino+PreSHOWER TOF

Pu

rity

Pu

rity

Eff

icie

ncy

Eff

icie

ncy

Efficiency and purity of hadron identification

Pion yields per reaction• Estimate of averaged number of participants Apart - URQMD events after 1st level trigger: Apart= 7.91, in EXP number of detected charged baryons (p+d) higher by 1.083 than in SIM (p only) - different centrality selection! For EXP Apart=7.91*1.083=8.57

• result: N / Apart = 0.148 ± 0.015 N = (N + N )/2

• TAPS for 0 shows N / Apart = 0.138 ± 0.014

Momentum distributions

simulationexperiment

Enhancement of low mom. pions

Rapidity distributions

emission 4 simulationexperiment

In EXP baryons bounded in clusters (no deuterons from URQMD)

mT distributions

+; experimenttwo exponents

; experimenttwo exponents

particles from decay N

21 /2

/12

1 TmTm

TT

TT eCeCdm

d

m

T from fits [MeV]

•In simulation inverse slopes from UrQMD input correctly reconstructed - no bias made by acceptance and momentum resolution• EXP data show two slopes shape•Different results for + and –: acceptance and correction?•Higher values of T of pions agree with simulation•Agreement with KaOS results: 40 40 ±± 3; 3; 86 86 ±± 3 3 MeV MeV

e- e+

q*p [MeV/c]

v/c

Velocity vs momentum of lepton candidates (RICH-track) hadron contamination <2%

• only tracks associated with the RICH ring are considered as lepton candidates

• RICH variables, SHOWER charge ratio and velocity used for ID

• leptons mainly from 0 Dalitz decay, analysis done simultaneously for EXP and URQMD data

Pattern quality: Pad matrix, signal heightPattern density: Number of pads/maximaPattern geometry: Ring-like shape

RICH observables

Principal Component Analysis:

Leptons identified by producing an electromagnetic shower

● An increase in charge on post1 or post2 is an electron signature. Charge multiplication factors

Pre

Post1Post2

0totalQ

1totalQ

2totalQ

0

1

1total

total

Q

Qf

0

2

2total

total

Q

Qf

SHOWER observables

Multiplicity distributions of leptons

Momentum distributions of leptons

e+

counts/event (kick track>0) counts/event (kick track>0)

e+

Experiment and AnalysisExperiment and AnalysisC+C 2AGeV commissioning run - lower momentum resolution data5% interaction target LVL1 triggered events (Mch.>3) : 36*106 events tracking and momentum reconstruction using inner chambers and outer TOF/TOFino+SHOWER detectorsp10% at 0.7 GeV/c)

SimulationLVL1-Trigger: 19.5 x 106 (C+C @ 2 AGeV) UrQMD events, processed through GEANT3

Analysis - HADES data analysis package HYDRA based on C++ and ROOT

good ID up to p<1000 MeV/c, above it +/p not separated due to low momentum resolution (incomplete tracking setup)

typical ring in RICH pad plane

This method diagonalizes the covariance matrix of the observables and yields the eigenvectors of the distribution, which have the maximum information content. This

allows to reduce the number of dimensions without significant loss of information.

most of RICH observables are stromgly correlated

e-

The method developement and analysis are under progress. Presented results on leptons are obtained by combining the probabilistic approach with a „standard“ cut method.

Shown results for both hadrons and leptons suffer from low momentum resolution and imperfect track reconstruction due to uncomplete detector setup.

e-e+

At present we are analyzing the data from more recent experiment with setup incorporating also outer MDC chambers, i.e. data with better momentum resolution.

+ -

| ( | , )k kk

L x h f x p hx

x |P h x

, , , ,

||

|h e K p d

L x h P hP h x

L x h P h

|P h x

( | , )k kf x p h

Fig.1 Probability density functions of velocities for protons and deuterons at particle momentum 750 MeV/c

Fig.2 Distribution of velocities for protons and deuterons at particle momentum 750 MeV/c