ARTICLE IN PRESS

0168-9002/$ - se

doi:10.1016/j.ni

�CorrespondChina, Hefei, A

fax:+86 551 360

E-mail addr

Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429

www.elsevier.com/locate/nima

Extensive particle identification with TPC and TOF at theSTAR experiment

Ming Shaoa,b,�, Olga Barannikovac, Xin Donga,d, Yuri Fisyakb, Lijuan Ruana,b,Paul Sorensend, Zhangbu Xub

aUniversity of Science and Technology of China, Hefei, Anhui 230026, ChinabBrookhaven National Laboratory, Upton, NY 11973, USA

cPhysics Department, Purdue University, West Lafayette, IN 47907, USAdLawrence Berkeley National Laboratory, Berkeley, CA 94720, USA

Received 13 June 2005; received in revised form 7 October 2005; accepted 19 November 2005

Available online 6 January 2006

Abstract

Particle identification (PID) capabilities are studied by using the Time Projection Chamber (TPC) and a Time-Of-Flight (TOF)

detector together at STAR. The identification capability of charged hadrons is greatly extended compared with that achieved by TPC

and TOF separately. Particle spectra from pþ p, dþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 200GeV and AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV are

used to develop the methods. The transverse momentum (pT) ranges of p, and pðp̄Þ identification are from �0:3GeV=c to �10GeV=c.

The high pT reach is limited by statistics in current data sets. An important conceptual advance was developed to identify electrons by

using a combination of dE=dx in TPC and velocity information from the TOF detectors, which is important for future low-mass dilepton

program at STAR.

r 2006 Elsevier B.V. All rights reserved.

PACS: 29.40.Cs; 29.40.Gx; 29.85.þc

Keywords: Particle identification; TOF; TPC; dE=dx; STAR

1. Introduction

One of the goals of the relativistic heavy-ion program atRHIC is to study quantum chromodynamics at extremecondition [1]. A unique strength of the solenoidal tracker atRHIC (STAR) [2] is its large, uniform acceptance capableof measuring and identifying a substantial fraction of theparticles produced in heavy-ion collisions. Detectorsrelevant to the study presented in this article are the TimeProjection Chamber (TPC) [3], and a proposed barrelTime-Of-Flight (TOF) [4]. For stable charged hadrons,the TPC provides p=KðpþK=pÞ identification to pT ’

0:7ð1:1ÞGeV=c by the ionization energy loss ðdE=dxÞ as

e front matter r 2006 Elsevier B.V. All rights reserved.

ma.2005.11.251

ing author. University of Science and Technology of

nhui 230026, China. Tel.: +86 551 3607940;

1164.

ess: swing@ustc.edu.cn (M. Shao).

usually been quoted and presented in the physics analyses[3]. However, direct particle identification (PID) capabilityfor stable hadrons at intermediate/high pT is important forthe study of collective flow and strong early-stage interac-tion in the dense medium formed in relativistic heavy-ioncollisions [1]. STAR PID capability can be furtherenhanced by the proposed TOF. A TOF system with atime resolution of t100 ps at STAR is able to identifyp=KðpþK=pÞ to pT ’ 1:6ð3:0ÞGeV=c, as demonstrated inFig. 1 (see also Refs. [5,6]). In addition, with relativistic riseof dE=dx from charged hadrons traversing the TPC atintermediate/high pTð\3GeV=cÞ and diminished yields ofelectrons and kaons at this pT range, we can identify pionsand protons up to very high pTð’ 10GeV=cÞ in pþ p, pþA and AþA collisions at RHIC. An important conceptualadvance was developed to identify electrons by using acombination of dE=dx in the TPC and velocity informa-tion from the TOF detectors. This has been used to

ARTICLE IN PRESS

p (GeV/c)

0.5 1 1.5 2 2.5 3 3.5 4

1/β

0.8

1

1.2

1.4

1.6

1.8

2

2.2

2.4

102

103

104

TOFr PID (62 GeV Au+Au run)

Fig. 1. 1=b vs. momentum for pions, kaons and (anti-)protons from TOFr

at 62.4GeV AuþAu collisions. The separation between pions and kaons

((anti-)protons) is achieved to pT�1:6ð3:0ÞGeV=c.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429420

measure charm yield via its semileptonic decay [7]. Electronidentification and hadron rejection power will bediscussed indetail. This provides a basic tool for thefuture dilepton measurements with the azimuthal 2pcoverage of a barrel TOF system. The proposed dileptonmeasurements will provide a penetrating probe into thenew state of dense matter produced in central heavy-ioncollisions at RHIC, since leptons do not participate instrong interactions occurring during hadronization andfreeze-out.

Fig. 2. Distribution of log10ðdE=dxÞ as a function of log10ðpÞ for electrons,

pions, kaons and (anti-)protons. The units of dE=dx and momentum (p)

are keV=cm and GeV=c, respectively. The color bands denote within �1sthe dE=dx resolution. I70 means Bichsel’s prediction for 30% truncated

dE=dx mean.

2. Experiment setup

At STAR [2], the main tracking device is a TPC,covering full azimuthal angle and �1:5 units of pseudo-rapidity. A dE=dx resolution of �8% can be achieved byrequiring the tracks of charged particles to have at least 20out of a maximum of 45 hits in the TPC. Detaileddescriptions of the TPC and its electronics system havebeen presented in Refs. [3,8]. One tray of prototype TOFdetector based on multi-gap resistive plate chambers [9](TOFr) was installed in STAR in 2003. It covers 1=60 inazimuth and �1pZp0 in pseudo-rapidity at the outerradius of the TPC, 220 cm from the interaction point. Twoidentical pseudo-vertex position detectors (pVPD) wereinstalled to record the start time for the TOFr, each 5.4maway from the TPC center along the beam line. EachpVPD covers �19% of the total solid angle in4:4pjZjp4:9. More information of the TOFr and thepVPD can be found in Refs. [10,11]. The data used in thisstudy were collected from Au+Au collisions at

ffiffiffiffiffiffiffiffi

sNNp

¼

62:4GeV in 2004, and pþ p and dþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 200GeV in 2003 at RHIC. The time resolution ofthe TOF system is �105 ps ðAuþAuÞ and �120 ps ðdþAuÞ respectively, which include pVPD’s contribution of�55 ps ðAuþAuÞ and �85 ps ðdþAuÞ.

3. Stable Hadron identification

3.1. PID at intermediate/high pT by TPC

At 3opTt10GeV=c, there is a difference of about 15%in the dE=dx between pions and kaons due to the pionrelativistic rise of the ionization energy loss. The differencebetween that of pions and (anti-)protons is even larger.This allows us to identify pions from other hadrons atthis pT range by the TPC alone at 2s level. The dE=dx

resolution is �8%, as demonstrated in Fig. 2. Shown inFig. 3 is the nsp distribution for charged hadrons at4ppTp4:5GeV=c and jZjo0:5, where nsp is the normal-ized dE=dx of pions. The normalized dE=dx is definedby nsY

X ¼ logððdE=dxÞY=BX Þ=sX , in which X ;Y can bee�;p�;K� or pðp̄Þ. BX is the expected mean dE=dx of aparticle X, and sX is the dE=dx resolution of TPC. For anideal calibration, the nspp distribution is a normal Gaussiandistribution. The nsp of positive and negative chargedhadrons are displaced by þ5 and �5, respectively in Fig. 3.Fig. 4 shows the pT dependence of nsKp and nspðp̄Þ

p relative tonspp, as predicted by the Bichsel function for the energy lossin thin layers of P10 [3]. From Fig. 3 and measurementfrom TOF [6], the yield difference between positive andnegative inclusive charged hadrons is approximately that ofproton and anti-proton (hþ � h� ¼ ðp� p̄Þ þ ðKþ �K�Þþðpþ � p�Þ ’ ðp� p̄Þ, kaons’ contribution is small). There-fore, the peak positions of dE=dx distribution from hþ �h� should represent well that of protons (sincensh

þ�h�

p ’ nsðp�p̄Þp ¼ nspðp̄Þp ). Indeed, the calibrated Bichsel

function for protons matches the dE=dx peak position of(hþ � h�), as shown in Fig. 4. In addition, as shown later,the dE=dx difference of protons and pions in themomentum range where PID selection is possible byTOF is also found to be consistent with the Bichselfunction. These cross-checks confirm that the Bichselfunction can be used to constrain the relative dE=dx

ARTICLE IN PRESS

pT (GeV/c)

nσx π-

nσx π

nσπk-nσπ

π(|η|<0.5)

nσπp-nσπ

π(|η|<0.5)

nσπp-nσπ

π(|η|<0.5 h+-h-)

-3

-2.5

-2

-1.5

-1

-0.5

0

2 3 4 5 6 7 8

Fig. 4. The relative dE=dx peak position of K–p (squares) and p–p(circles) in unit of standard resolution width (sp) of pion dE=dx as

function of pT. The triangles are the peak positions of the dE=dx

distribution of hþ � h� ¼ ðp� p̄Þ þ ðKþ �K�Þ þ ðpþ � p�Þ ’ ðp� p̄Þ.

Fig. 3. dE=dx distribution normalized by pion dE=dx at

4opTo4:5GeV=c and jZjo0:5, and shifted by �5 for positive and

negative charged particles, respectively. The distribution is from mini-

mum-bias AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV.

nσπ-5 -4 -3 -2 -1 0 1 2 3

h- /h+

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Fig. 5. h�=hþ vs. dE=dx distribution normalized by pion dE=dx at

3:5opTo4:0GeV=c and jZjo0:5. The distribution is from minimum-bias

AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429 421

position between kaons, (anti-)protons and pions. Toextract pion yield, we performed a six Gaussian fit to thedE=dx distributions of positive and negative hadronssimultaneously as shown in Fig. 3. The nsKp � nspp andnspðp̄Þ

p � nspp are fixed in a six-Gaussian fit, where the sixGaussians are for p�, K� and pðp̄Þ at a given pT bin. Thesigma of the six Gaussians are chosen to be the same. Thepion yields extracted from the fit can be found in Ref. [12].As shown by Fig. 2 and discussed in Ref. [12], the PID ofpions at �3opTo�7GeV=c is obtained with this method.The pT reach is limited by statistics and not by PIDcapability in this study.

Fig. 4 shows that the dE=dx separation between kaonand proton is less than one s at pT between 3 and 5GeV=c

and larger for larger pT. However, there are a few methodswe can use to identify protons and cross-check thecontaminations from kaons. Fig. 5 shows the ratio ofnegative hadrons over positive hadrons as a function ofdE=dx in unit of nsp at 3:5opTo4:0GeV=c. Since theenergy loss of particles in TPC is independent of its charge

sign, the dependence of h�=hþ on nsp is due to differentparticle composition and due to the dE=dx separationbetween pion, kaon and proton. Fig. 5 shows two plateaus.One from p̄=p and the other from p�=pþ. We will be ableto select protons with reasonable purity by requiringnspo� 2:5 in this particular pT bin. We can then use anindependent measurement of K0

S from V0 method toconstrain the charged kaon yields in the six-Gaussian fitor to estimate the charged kaon contaminations describedabove to obtain reliable proton yield. Detailed studies andmeasurements are underway.

3.2. Hadron PID at intermediate pT by TPC+TOF

Figs. 1 and 2 show that TOF alone is not able to separatebetween pions and kaons and TPC alone is not able toseparate pions, and protons at the pT range roughlybetween 2.0 and 4:0GeV=c. However, the dependence of1=b on pT from TOF and that of dE=dx on pT are differentin this pT range and therefore, by combining these two, wewill be able to extend our PID capability. Fig. 6 shows thehadron distribution as a function of nsp and mass square(m2), with m2 ¼ p2ððtTOF � c=lÞ2 � 1Þ, where p is themomentum, tTOF is the TOF, c is the speed of light invacuum, and l is the flight path length of the particle. Thepion, kaon and (anti-)proton peak can be clearly seen inFig. 6. The solid line in Fig. 7 is the projection of Fig. 6 tothe m2 direction at 3opTo4GeV=c. At this pT range, thepion and kaon bands are merged together, and cannot beclearly separated from the (anti-)proton band with theTOF. However, if we require nsp40 and then plot the m2

distribution again (dashed line in Fig. 7), the kaon and(anti-)proton bands are greatly suppressed and a clear pionsignal is observed. This is due to the rather large differencebetween nspp and nsKK or nspðp̄Þ

pðp̄Þ in this pT range, as shown inFig. 4. Similarly, the pion and kaon bands are suppressedsignificantly with respect to the (anti-)proton bandwhen nspðp̄Þo0 is applied. This helps us get a cleaner

ARTICLE IN PRESS

-1 -0.5 0 0.5 1 1.5 20

5

10

15

20

25

30

Cou

nts

35

PID (TOF+TPC)

nσπ>0nσp<0

m2 [(GeV/c2)2]

Fig. 7. m2 distribution without dE=dx cut (black solid line), with nsp40

cut (red dashed line), and nspo0 cut (blue dotted line).

-5 -4 -3 -2 -1 0 1 2 3 4 5

m2 [(

GeV

/c2)2

]

-1

-0.5

0

0.5

1

1.5

2

0

20

40

60

80

100

120

(2<pT<4 GeV/c)

nσπ

Fig. 6. Particle distribution as a function of nsp and m2 for

2opTo4GeV=c. The pion, kaon and (anti-)proton peaks can be clearly

seen.

HistoPt_10Entries Mean 0.3185RMS 0.3995

-0.5 0 0.5 1

Cou

nts

1.5

1

10

102

10326112

pT range: (1.400000,1.600000)

m2 [(GeV/c2)2]

Fig. 8. m2 distribution from TOFr for 1:4opTo1:6GeV=c. Arrow shows

the cut at the pion mass in order to obtain a clean pion sample for dE=dx

studies.

Entries 9582Mean -0.1394RMS 1.039

54.97 / 57Constant 396.9 ± 5.1Mean -0.1802 ± 0.0099Sigma 0.9457 ± 0.0073

-5 -4 -3 -2 -1 0 1 2 3 4 5

Cou

nts

0

50

100

150

200

250

300

350

400

h1

χ2/ndf

nσ distribution

1.4<pT<1.6

nσπ

Fig. 9. nsp distribution using the pion sample resulting from the cut

described in previous figure. The solid lines are fits to a Gaussian function.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429422

(anti-)proton signal. Therefore, the combination of theTPC and TOF enhances the PID at 2opTo4GeV=c whereneither TPC nor TOF can well separate the hadrons.

The invariant yields of pions, kaons and (anti-)protonswere calculated for 62.4GeV AuþAu collisions. Theresults were shown elsewhere in Ref. [6]. The pions and(anti-)protons were identified at 0:2opTo�5GeV=c, andthe kaons were identified at 0:2opTo�3GeV=c. Theupper PID pT reach is due to low statistics with the smallacceptance of TOFr.

3.3. Cross-check at low/intermediate pT

Since dE=dx plays a key role in our methods discussedabove, dE=dx calibration has to be carefully studied.Again, with PID capability at overlapping pT range, we canuse identified particles (pions, protons, etc.) from TOFr tocross-check the characteristics of nsðp;K; pðp̄ÞÞ distribution

at low/intermediate pT. Fig. 8 demonstrates an example ofchoosing a pion sample by cutting on m2. We can then plotthe nsp distribution of this sample and fit it with aGaussian function as in Fig. 9. The mean and sigma of theGaussian are listed in Table 1 as a function of pT. Theparameters (mean and sigma) obtained from protonsamples are also listed in Table 1. We note that the widthof nspp is systematically larger than that of pions, partly dueto the different resolution from different track lengthresulting in dispersion of nspp. The values of theseparameters for pions and (anti-)protons are almostindependent of pT and consistent with that obtained bypure dE=dx method (see Fig. 3). The initial calibration ofdE=dx is found to be about �0:2 sigma off zero. This isconsistent with the findings from fitting to pion peaks atrelativistic rise. The last two columns in Table 1 show themeasured hnspðp̄Þ

p � nsppi compared to Bichsel functioncalculation.

ARTICLE IN PRESS

Efficiency0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(K+

p) C

onta

min

atio

ns

10-2

10-1

PositiveNegative

rdE/dx π± Identification, 4.0 <pT<4.5 GeV/c

Fig. 10. Contamination from kaonsþ(anti-)protons to pions with a

dE=dx selection window vs. pion efficiency from the same selection at

4opTo4:5GeV=c. Dashed line is for pþ and solid line is from p�.

Table 1

nsp Gaussian fit parameters of the pion and (anti-)proton sample obtained using m2 cut

pT nspp nspðp̄Þp onspðp̄Þ

p � nspp4

ðGeV=cÞ Mean Sigma Mean Sigma Measurement Expectation

0.9–1.0 �0:185� 0:006 0:980� 0:004 4.727 70.014 1:110� 0:010 4.912 70.020 4.936

1.0–1.2 �0:171� 0:005 0:977� 0:004 3.176 70.021 1:171� 0:015 3.347 70.026 3.421

1.2–1.4 �0:150� 0:007 0:958� 0:005 1.562 70.013 1:112� 0:011 1.712 70.020 1.872

1.4–1.6 �0:180� 0:010 0:946� 0:007 0.464 70.015 1:073� 0:012 0.645 70.025 0.768

1.6–1.8 �0:195� 0:013 0:938� 0:010 �0.322 70.019 1:021� 0:014 �0.127 70.032 �0.036

1.8–2.0 �0:228� 0:021 0:923� 0:015 �0.834 70.023 0:989� 0:016 �0.606 70.044 �0.616

2.0–2.5 �0:171� 0:029 0:972� 0:023 �1.458 70.025 1:034� 0:019 �1.287 70.054 �1.174

2.5–3.0 �1.912 70.056 1:161� 0:048 �1.91 70.01 �1.72

3.0–3.5 �2.387 70.113 1:111� 0:121 �2.40 70.01 �2.06

3.5–4.0 �2.50 70.02 �2.27

4.0–4.5 �2.52 70.04 �2.35

4.5–5.0 �2.47 70.08 �2.44

5.0–6.0 �2.47 70.15 �2.47

6.0–7.0 �2.60 70.28 �2.47

The differences between nspðp̄Þp and nspp are also listed and compared with Bichsel calculations. At pTX3GeV=c, onspðp̄Þ

p � nspp4 are calculated via

hþ � h�.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429 423

3.4. Contamination in pion identification

Good particle-by-particle identification is crucial foranalyses such as collective flow and fluctuation whichprovide important physics at RHIC. Not only can pionyields be obtained by statistically fitting of the histogram,pion identification can also be obtained track by trackusing a dE=dx cut on the nsp. Fig. 10 illustrates thecontamination to pions in percentage from kaons and(anti-)protons when we select nsp greater than a givenvalue (threshold). We can achieve 495% pion purity at50% pion efficiency when requiring nsp40. As have beenshown previously, nspp is actually not centered at zero dueto insufficient calibration. However, this can be easilycorrected for. If one can tolerate a pion contamination of

10%, then the efficiency can be as high as �80%. Theserepresent the worst-case scenario since at higher pT thebaryon enhancement is less and the contamination shouldbe smaller.

3.5. Contamination in proton identification

The contamination in (anti-)proton identification is alittle more complicated. As shown in Fig. 2, the separationof dE=dx between protons and kaons is �1s at inter-mediate/high pT. Therefore the purity of protons atpTo4GeV=c is mainly driven by the m2 resolution ofTOFr. The m2 resolution is formulated by

dm2 ¼dp2

b2g2�

p2

b22dt

t�

p2

b22dL

L

where p is the momentum, t is the TOF and L is the tracklength of the particle. Fig. 11 depicts the m2 resolution forall charged particles as a function of pT, at �0:5pZp0.The momentum resolution is taken from Ref. [3]. dt and dL

are chosen to be 110 ps and 0.5 cm, respectively. Atintermediate/high pT, the m2 resolution increases approxi-mately quadratic with momentum. If we require m2

X0:88(mass square of pðp̄Þ) and nspðp̄Þo0, which means an (anti-)proton efficiency of 25%, the contamination from pionsand kaons is �10% at 4opTo5GeV=c. We have used theinvariant yields of p=K=p from Refs. [6,13] in thisestimation.At higher pTð45GeV=cÞ, TOFr is not effective in

identifying (anti-)protons anymore. The contamination to(anti-)proton identification from kaons is mainly deter-mined by the \1s separation between them. Therefore, at50% efficiency (nspðp̄Þo0), kaon contamination is about20% assuming kaon and proton yields are equal. Again we

ARTICLE IN PRESS

pT (GeV/c)

1 1.5 2 2.5 3 3.5 4 4.5 5

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

10

102

103

δ(m

2 ) [(

GeV

/c2)2

]

Fig. 11. m2 resolution as a function of pT.

-10 -8 -6 -4 -2 0 2 4 6 8 10nσπ±5

-10 -8 -6 -4 -2 0 2 4 6 8 100

20

40

60

80

100

120

140

160

180

200C

ount

s

220

dE/dx distribution in minbias 0.5<|η|<0.9, 4.0<p_T<4.5 GeV/c

Fig. 12. dE=dx distribution normalized by pion dE=dx and offset by �5

for positive and negative charge at 4opTo4:5GeV=c and 0:5ojZjo0:9,respectively. The distribution is from minimum-bias AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV.

80

100

120

dE/dx distribution in minbias d+Au 4.0<pT<4.5 GeV/c

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429424

can use different methods to evaluate the contaminationand efficiency as described in Section 3.4.

At STAR, the resolution of dE=dx measurementdepends on many factors, including magnetic field setting,event multiplicity, beam luminosity, track length and driftdistance. It also depends on the number of hits in TPC usedto calculate the ionization energy loss for a given track.Excellent readout electronics system (pulse shape control,fine linearity and large dynamic range, etc.) and carefuloffline calibration are important to achieve high-qualitydE=dx measurement. PID may have different efficiencyand purity, depending on the experimental setting and cutsused in analysis. Basically, dE=dx resolution is improvedwith longer track length, shorter drift distance, strongermagnetic field, lower multiplicity and beam luminosity, aswell as more hits in TPC used for track fitting. Figs. 12 and13 shows pion identification at 0:5ojZjo0:9 in AuþAucollisions at

ffiffiffiffiffiffiffiffi

sNNp

¼ 200GeV and 1:04jZj40:5 in dþAucollisions at 4ppTp4:5GeV=c, respectively. Due to longertrack and shorter drift distance for particles produced athigher jZj, the dE=dx resolution gets better. Thus theseparation between pions and kaons or(anti-)protons inFigs. 12 and 13 are larger than that in Fig. 3, and betteridentification can be achieved. From the relative heights ofpion and proton peaks, it is also obvious that the particlecompositions in AuþAu at

ffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV and dþAuat

ffiffiffiffiffiffiffiffi

sNNp

¼ 200GeV are different.

-10 -8 -6 -4 -2 0 2 4 6 8 10-10 -8 -6 -4 -2 0 2 4 6 8 100

20

40

60

Cou

nts

nσπ±5

Fig. 13. dE=dx distribution normalized by pion dE=dx

(4opTo4:5GeV=c) and offset by �5 for inclusive hadrons at

0:5oZo1:0 and �1:0oZo� 0:5, respectively. The distribution is from

minimum-bias dþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 200GeV.

4. Electron identification at low/intermediate pT

In addition to its hadron PID capability, TOFr detectorcan be used to identify electrons by combining it withdE=dx from TPC. We were not able to measure dileptonwith the prototype MRPC TOF tray installed in STAR dueto small TOF acceptance. However, single electron spectraare sensitive to charm production and future azimuthal 2pcoverage of TOF will enable us to do dilepton physics [7].In this section, we will discuss the electron identificationmethod, the hadron rejection power at low pT and rejection

ARTICLE IN PRESSM. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429 425

of the photonic background from g conversions and Dalitzdecays [14].

The top panel of Fig. 14 shows the 2-D scatter plot ofdE=dx as a function of the particle momentum (p) forcharged particles with TOFr matched hits from dþAucollisions. The bottom panel shows that slow hadrons areeliminated and the electron band is well separated from thehadron band with a particle velocity (b) requirement atj1=b� 1jo0:03. Electrons can then be selected with thefollowing cut (1):

dE=dxðpÞ42:4þ 0:65� ð1� e�ðp�0:15Þ=0:1Þ þ 0:1� p (1)

where p is in GeV=c and dE=dx is in keV/cm. With thecombination of dE=dx from TPC and b from TOFr,

2

3

4

5

6

dE/d

x (k

eV/c

m)

(a)

Particle momentum p (GeV/c)

0 0.5 1 1.5 2 2.5 3

2

3

4

5

6

Ene

rgy

loss

(b)TOFr d+Au @ 200 GeV

|1/ β-1| < 0.03

Fig. 14. dE=dx in the TPC vs. particle momentum (p) without (upper

panel) and with (lower panel) TOFr velocity cut of j1=b� 1jo0:03.

dE/dx (keV/cm)

0 1 2 3 4 5

Cou

nts

1

102

1030.4 < p

T/(GeV/c) < 0.5

χ2/ndf = 80.5/60

Fig. 15. dE=dx distribution with TOFr velocity cut in two pT bins. The arrow

shows a Gaussian plus exponential fit to the dE=dx distribution, while that in

electrons can be identified above p�0:15GeV=c, while thehigh pT reach is limited by the statistics in this analysis.

4.1. Contamination for electron identification

Hadron contamination to electron identification withTOFr velocity cut was studied from the dE=dx distributionin each pT bin. A Gaussian function with an exponentialtail is used in the fit. At pT ’ 223GeV=c, 2-Gaussian fit isalso performed and shows little difference from theGaussian plus exponential fit. Fig. 15 shows the results intwo pT bins from dþAu collisions. The arrows denote thecut from Eq. (1). Hadron contamination ratio is estimatedfrom these fits, and shown in Fig. 16. The electronefficiency is obtained by varying the cut in Eq. (1). Atlow pT, high electron efficiency and low hadron con-tamination can be achieved. Even at intermediate pT,the contamination can be reduced to a low level (�1%)if a lower electron efficiency (�50%) is chosen. Fig. 17

dE/dx (keV/cm)

0 1 2 3 4 5

Cou

nts

1

102 2.0 < pT

/(GeV/c) < 2.5

χ2/ndf = 53.1/62

s denote the cut from Eq. (1). The curve in left plot (0:4opTo0:5GeV=c)

right plot (2:0opTo2:5GeV=c) is a 2-Gaussian fit.

Hadron contamination

10-3 10-2 10-1 1

Ele

ctro

n E

ffici

ency

0.5

0.6

0.7

0.8

0.9

1

1.1

0.4<pT<0.5

0.8<pT<0.9

1.2<pT<1.4

2.0<pT<2.5

Fig. 16. The contamination from hadrons to electron identification as a

function of electron efficiency in different pT range, with TOFr velocity

cut.

ARTICLE IN PRESS

Transverse momentum pT (GeV/c)

0 0.5 1 1.5 2 2.5 3

e/h

10-2

10-1 All hadrons

After TOFr PID cut

Fig. 17. Electron/hadron ratio vs. transverse momentum (pT), with and

without TOFr velocity cut.

Table 2

Electron selection criteria in AuþAu

Method TOFþ dE=dx

jVertexZjo 30 cm

Primary track ? Yes

nFitPts 4 25

ndEdxPts 4 15

Rapidity ð�1:0; 0Þ

w2=ndf (0., 3.0)

b from TOF j1=b� 1jo0:03TOFr hit quality 30oADCo300

�2:7ozlocal=cmo3:4jylocal � yCjo1:9 cm

TOFp hit quality th1oADCoth22:0ozlocal=cmo18:00:4oylocal=cmo3:2

nFitPnts is the number of fit points of a track in the TPC, ndEdxPts is the

number of hits of a track for dE=dx calculation, ylocal and zlocal are

corresponding local coordinates on a TOF module, yC is the ylocal of a

readout cell center, and th1;2 are cell-by-cell lower/upper ADC cuts for

TOFp.

Particle momentum p (GeV/c)

0 0.5 1 1.5 2 2.5 3

dE/d

x (k

eV/c

m)

2

4

6 TOF Au+Au @ 62.4 GeV

|1/ β-1| < 0.03

Fig. 18. dE=dx vs. particle momentum after a TOF b cut

(j1=b� 1jo0:03).

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429426

shows electron to hadron ratio vs. pT before and afterTOFr velocity selection. The hadron rejection powercan be evaluated by (hadron contamination)�ðe=hÞ=ðelectron efficiencyÞ for a specific condition. It wasestimated to be at 10�5 level at pTo1:0GeV=c. At higherpT, additional hadron rejection from electromagneticcalorimeter in similar fiducial coverage can help reach thesame rejection power [15].

4.2. Single electrons from Au+Au collisions atffiffiffiffiffiffiffiffi

sNNp

¼

62:4 GeV

The purpose of the data analysis of dþAu and pþ pcollisions is to set up the baseline for heavy-ion collisions.Since the large data sample of AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 200 GeV is not available yet, the smaller sample ofAuþAu collisions at

ffiffiffiffiffiffiffiffi

sNNp

¼ 62:4 GeV has been analyzedto understand the necessary techniques for AuþAucollisions. There is a significant increase of multiplicitiesin AuþAu collisions as compared to that in pþ p anddþAu collisions, and a consequent decrease of trackquality. In this section, we discuss the electron measure-ments from TOF and dE=dx method and the photonicbackground estimation. Since the charm yield is lowcompared to photonic contributions at

ffiffiffiffiffiffiffiffi

sNNp

¼

62:4 GeV, any excess of electrons above the photonicbackground with acceptable errors is not expected. Thisprovides a testing ground to evaluate the quality ofbackground reconstruction.

STAR has accumulated �15 million events in a relativelyshort run with AuþAu collisions at

ffiffiffiffiffiffiffiffi

sNNp

¼ 62:4 GeV.With the minimum bias trigger (0280%) and vertex z

position selection, the events used in this analysis are �6:4million. The resolution for TOF detectors is �110 ps forTOF system, with �55 ps start timing resolution included.The hadron PID capability was reported in Ref. [6]. As indþAu and pþ p collisions, electrons can be identified bycombining TOF and dE=dx in the TPC. Electrons wereselected to originate from the primary interaction vertex

according to the criteria shown in Table 2. TOFp in thistable is another TOF tray based on traditional techniquewith fast scintillator plus photomultiplier. It has similarcoverage as TOFr. More details can be found in Ref. [11].Fig. 18 shows the dE=dx vs. particle momentum after a b

cut from TOF. The electron dE=dx band can be separatedfrom that of hadrons. The dE=dx resolution in AuþAucollisions decreases when compared to that in dþAu andpþ p due to much higher multiplicities and a multiplicitydependence of dE=dx calibration which is not currentlyimplemented. We fit the dE=dx distribution aroundelectron peak with both two-Gaussian function andexponentialþGaussian function to extract the electronraw yields in each pT bin.Hadron contamination becomes larger at pT41:5GeV=c

if we select electrons with the same efficiency as in pþ pand dþAu collisions. Fig. 19 shows the hadron contam-ination fractions under different electron selections. The

ARTICLE IN PRESS

Transverse momentum pT (GeV/c)

0 0.5 1 1.5 2 2.5 3

Had

ron

cont

amin

atio

n (%

)

10-3

10-1

1TOF Au+Au @ 62.4 GeV

(-1.→3.) σe

(-0.5→3.) σe

(0.→3.) σe

Fig. 19. Hadron contamination fractions for different electron selections

in AuþAu 62.4GeV collisions.

Invariant Mass Me+e- (GeV/c2)

0 0.5 1 1.5 2

Cou

nts

0

5000

0.4<pT<0.6

Fig. 20. Main photonic background reconstruction in AuþAu collisions.

The crosses depict the real electron pair candidate invariant mass

distributions and the histograms represent the combinatorial back-

grounds. The photon conversion peak is clear in �0 mass region, while

the p0 Dalitz is hard to see.

Table 3

Partner candidate selection criteria in AuþAu

Charge Opposite to tagged track

Primary/global ? Global

nFitPts 4 15

nFitPts/nMax 4 0.52

w2=ndf (0., 3.0)

se ð�1:; 3:0Þa

dca of eþ,e� (0.0, 3.0) cm

nMax is the number of maximum possible points of a track in the TPC,

and dca is the distance of closest approach to the primary interaction

point.aSeveral different se cuts have been studied.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429 427

selection se40 with electron efficiency of about 50% isnecessary to have the contaminations less than 10% atpT41:5GeV=c. At lower pT , the identification is notaffected as much by the high multiplicity since most of thehadrons are rejected by TOF selection.

4.3. Estimate of electrons from photonic source in AuþAu

Photon conversions g! eþe� and p0! geþe� Dalitzdecays are the dominant photonic sources of electronbackground. Since only one TOF tray (120th of theproposed TOF) is available, electron identification of botheþe� daughters from conversion or Dalitz decays is notpossible at this momentum. To measure the backgroundphotonic electron spectra, the invariant mass and openingangle of the eþe� pairs were constructed from an electron(positron) in TOFr and every other positron (electron)candidate reconstructed in the TPC [16]. A secondaryvertex at the conversion point was not required. In AuþAu collisions, due to large multiplicity, the partner trackreconstruction will lead to a large combinatorial back-ground due to hadron contamination without an addi-tional PID from TOF. Even with a stringent dE=dx

selection, this combinatorial background is still significant.In the future, dilepton studies would be possible with 2pcoverage of TOF for selecting clean eþe� pairs. Fig. 20shows the electron pair candidate invariant mass dis-tribution. The tagged electron was selected from TOFwith the cuts shown in Table 2 and additional 0onseo3.The partner track candidate was selected according toTable 3.

The combinatorial background distribution was pro-vided by rotating the partner track momentum p

!!� p!

,and normalized to the distribution of the real electron paircandidates in the region 0:8oMeþe�=o2:0 ðGeV=c2) in theinvariant mass spectrum. Fig. 20 shows the results in0:4opTo0:6GeV=c where pT is the pT of tagged electrons.The plot shows that the combinatorial background was

very well reproduced. The peak from photon conversion indetector material is clearly visible near zero mass region,and its offset from zero is due to the opening angleresolution [16] in TPC tracking. The p0 Dalitz contributionis not visible in this case, possibly because the Dalitzcontribution is much smaller and distribution is muchbroader compared to conversion processes. We subtractedthe combinatorial background from the real distribution,and integrated the remaining distribution from 0 to0:15GeV=c2 to get the reconstructed main photonicbackground raw yield. In this case, we assumed that bothphoton conversion and p0 Dalitz decays were recon-structed.The background raw yield need to be corrected for the

reconstruction efficiency, as we did it for dþAu and pþ pcollisions [7]. This efficiency was calculated from AuþAu62.4GeV HIJING events with full detector MC simula-tions. After jV Zjo30 cm cut, �53K events were used in thecalculation. We took all TPC electron tracks without anydE=dx and TOF hit cut to improve the statistics. Theprocedure is the same as we did in pþ p and dþAu

ARTICLE IN PRESS

Transverse momentum pT (GeV/c)

0 0.5 1 1.5 2 2.5 3

Par

tner

Fin

ding

Effi

cien

cy

0.4

0.6

0.8

1

d+Au 200 GeV HIJING

Au+Au 62.4 GeV HIJING

Fig. 21. Photonic background reconstruction efficiency from AuþAu

62.4GeV HIJING simulations. Also shown on the plot is that from dþ

Au 200GeV HIJING simulations.

Transverse momentum pT (GeV/c)

0 0.5 1 1.5 2 2.5 3

Bkg

d/T

otal

0.5

1

1.5

2

TOF Au+Au @ 62.4 GeV

Photonic Bkgd (-1,3),(-1,3)

Photonic Bkgd (0,3),(-1,3)

Photonic Bkgd (0,3),(0,3)

Fig. 22. The reconstruction efficiency-corrected photonic background

over the raw inclusive electron yield under different electron/positron

track selections. The numbers in the brackets on the plot show the se cutfor the tagged track and the partner track, respectively.

M. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429428

collisions [7]. Fig. 21 shows the background reconstructionefficiency in AuþAu 62.4GeV compared with dþAuresults. Due to the higher multiplicities, the reconstructionefficiency is relatively lower in AuþAu than that indþAu.

This efficiency was used to correct for the photonicbackground raw yield obtained from above. In addition, a�5% fraction of other photonic background (from dþAuresults) and the nse selection efficiency were also included.The corrected background spectrum is compared with theinclusive spectrum as shown Fig. 22. The reconstructedbackground matches the total inclusive electron spectrum.This is expected since the charm yield is more than an orderof magnitude lower than the inclusive electron yield at62.4GeV [14]. This also demonstrates that we canreconstruct fully the photonic background from g conver-sions and Dalitz decays.

The performance of the electron identification andbackground studies for AuþAu system makes us con-fident on extracting the charm signal at 223GeV=c fromthe coming 30M minimum bias 200GeV AuþAu data.With the STAR upgrade of inner tracker–Heavy FlavorTracker (HFT), we will be able to reject photon conver-sions by requiring additional hits in HFT, to rejectelectrons from charm semileptonic decays by its displacedsecondary vertex, and to further reconstruct/reject the lowpT partner of the electrons from the Dalitz decays usinginner tracker only. Electron identification at pT41GeV=c

can be improved by combining dE=dx from TPC, velocityfrom TOF and energy from electromagnetic calorimeter.Detailed studies are underway and are beyond the scope ofthis article.

5. Conclusion

In summary, we have developed a technique to extendthe particle identification up to high pT at STAR. Bycombining information from the TPC and TOF, we canmeasure pion, kaon and (anti-)proton in the intermediate/high pT range and electron in low/intermediate pT (high pT

limited by statistics). Preliminary spectra obtained in AuþAu collisions at

ffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV shows reliable PID topT�7GeV=c for pions and (anti-)protons, and �3GeV=c

for kaons, respectively. The electrons were identified at0:15opTo4GeV=c, in dþAu, pþ p collisions at

ffiffiffiffiffiffiffiffi

sNNp

¼

200GeV and AuþAu collisions atffiffiffiffiffiffiffiffi

sNNp

¼ 62:4GeV. Apurity of over 95% for pion identification and �70290%for proton identification was obtained with this method atintermediate/high pT. We achieved a hadron rejectionpower at 10�5 level for the electron identification at low pT.

Acknowledgements

We thank the STAR Collaboration, the RHIC Opera-tions Group and RCF at BNL, and the NERSC Center atLBNL for their support. This work was supported in partby the HENP Divisions of the Office of Science of the USDOE; the Ministry of Education and the NNSFC ofChina.

References

[1] STAR whitepaper, Nucl. Phys. A 757 (2005) 102.

[2] K.H. Ackermann, et al., Nucl. Instr. and Meth. A 499 (2003)

624.

[3] M. Anderson, et al., Nucl. Instr. and Meth. A 499 (2003) 659.

[4] The STAR TOF Collaboration, proposal for a large area time of

flight system for STAR, http://wjllope.rice.edu/TOF/TOF/Docu-

ments/TOF_20040524.pdf.

[5] J. Adams, et al., (STAR), Phys. Lett. B 616 (2005) 8;

L. Ruan, Ph.D. Thesis, University of Science and Technology of

China, nucl-ex/0503018.

[6] M. Shao, (STAR), J. Phys. G 31 (2005) S85;

L. Ruan, (STAR), J. Phys. G 31 (2005) S1029.

[7] J. Adams, et al., (STAR), Phys. Rev. Lett. 94 (2005) 062301.

[8] M. Anderson, et al., Nucl. Instr. and Meth. A 499 (2003) 679.

ARTICLE IN PRESSM. Shao et al. / Nuclear Instruments and Methods in Physics Research A 558 (2006) 419–429 429

[9] E. Cerron Zeballos, et al., Nucl. Instr. and Meth. A 374 (1996) 132.

[10] B. Bonner, et al., Nucl. Instr. Meth. A 508 (2003) 181;

M. Shaom, et al., Nucl. Instr. and Meth. A 492 (2002) 344.

[11] W.J. Llope, et al., Nucl. Instr. and Meth. A 522 (2004) 252.

[12] Z. Xu, (STAR), Int. J. Mod. Phys. A 20 (2005) 3783.

[13] J. Takahashi, (STAR), J. Phys. G 31 (2005) S1061.

[14] Xin Dong, Ph.D. Thesis, University of Science and Technology of

China, 2005, nucl-ex/0509011.

[15] M. Beddo, et al., Nucl. Instr. Meth. A 499 (2003) 725.

[16] J. Adams, et al., (STAR), Phys. Rev. C 70 (2004) 044902;

I. Johnson, Ph.D. Thesis, U.C. Davis, 2002, http://www.star.bnl.gov/

STAR/central/theses/2002/johnson_ian.pdf.