Muon Identification with Muon Telescope Detector at the STAR Experiment

0.23 MB·

by  T. C. Huang

Muon Identification with Muon Telescope Detector at the STAR
Experiment
T.C. Huanga, R. Mab, B. Huangd, X. Huangc, L. Ruanb, T. Todorokib, Z. Xub, C. Yange, S. Yange, Q.
Yange, Y. Yanga,∗, W. Zhae
aNational Cheng Kung University,Tainan 70101, Taiwan
bBrookhaven National Laboratory, Upton, New York 11973, USA
6 cTsinghua University,Beijing 100084, China
1 dUniversity of Illinois at Chicago, Chicago, Illinois 60607, USA
0 eUniversityof Science and Technology of China, Hefei 230026, China
2

l
u
J

5 Abstract
1
The MuonTelescope Detector (MTD) is a newly installed detectorin the STAR experiment. Itprovidesan

] excellent opportunity to study heavy quarkoniumphysics using the dimuon channel in heavy ion collisions.
t
e Inthis paper,we reportthe muonidentificationperformancefor the MTD using proton-protoncollisionsat
d
√s = 500 GeV with various methods. The result using the Likelihood Ratio method shows that the muon
-
s identification efficiency can reach up to 90% for muons with transverse momenta greater than 3 GeV/c
n ∼
and the significance of the J/ψ signal is improved by a factor of 2 compared to using the basic selection.
i
.
s Keywords: STAR, MTD, muon identification, muon, dimuon, quarkonium
c
i
s
y
1. Introduction the MuonTelescopeDetector (MTD), dedicatedto
h
p measuring muons was proposed in 2009 and was
The Solenoidal Tracker At RHIC (STAR) [1] is
[ installed from 2012 to 2014 [5, 6]. In this paper,
one of the two large high energy nuclear physics
we present the muon identification performance of
3
experiments at the Relativistic Heavy Ion Collider
v this new detector. There have been many studies
(RHIC) at Brookhaven National Laboratory. Af-
0 on muon identification from different experiments,
1 ter 15 years of operation, the STAR experiment and more details can be found in Refs. [7, 8, 9].
9 has provided many important results, which have
Thispaperisarrangedasfollows. InSection2,a
2 helped to improve our understanding of Quantum
briefdescriptionoftheSTARdetectorispresented.
0
Chromodynamics(QCD).Inparticular,evidenceof
. The data sets and event selection are described in
1 the existence of the Quark Gluon Plasma (QGP)
Section3. InSection4,threemethodsusedtoiden-
0 opened a new window to get a deeper insight into
tify muon candidates are described. We present in
6
QCD [2]. Heavy quarkonia states are ideal probes
1 section 5 the efficiency as well as the resulting sig-
to study the properties of QCD matter. For in-
: nal significance for J/ψ from these three methods.
v stance,quarkoniumsuppressioninthemediumdue
Finally, a summary is given in Section 6.
i
X to the color-screening of surrounding partons can
provide information about the partonic nature of
r
a the QGP and its temperature [3, 4]. Quarkonia 2. The STAR detector
areidentifiedbyreconstructingtheirinvariantmass
(M ) in the dilepton decay channel. Muons can The STAR detector is a general purpose particle
inv.
be reconstructed more precisely due to their re- detector optimized for highenergy nuclear physics.
duced bremsstrahlung radiation in material com- The main subsystems relevant to this analysis in-
pared to electrons. A new subdetector in STAR, clude the Time Projection Chamber (TPC), the
MagnetSystemandthe MTD.TheTPCisthe pri-
∗Correspondingauthor. mary tracking detector for charged particles and
Email address: [email protected] (Y.Yang) provides particle identification via measurements
Preprint submitted toNIMA July 18, 2016
of the energy loss (dE/dx) [10]. It covers full az- world-wide differential J/ψ cross section measure-
imuthal angles (0 < φ < 2π) and a large pseudo- ments [14]. The reconstructedmuonp inMC was
T
rapidity range (η < 1). The transverse momenta also slightly smeared by a Gaussian function, with
| |
(p ) and charge (q) of charged particles are mea- mean = 1.004 p and width = 0.022 p , to
T T T
× ×
sured by the curvature of their trajectories in the match the reconstructed J/ψ mass distribution in
0.5 Tesla solenoidal field generated by the Magnet data.
System. There are 30 bars, known as “backlegs”,
3.2. Track selection
outside the coil to provide the return flux path for
the magnetic field [11]. They are 61 cm thick at Tracksselectedfor the muonidentificationstudy
a radius of 363 cm corresponding to about 5 ab- have to meet the following requirements: pT is
sorption lengths. These backlegs play an essential greater than 1 GeV/c; the distance of closest ap-
roleinenhancingthemuonpuritybyabsorbingthe proachto the collision vertex should be less than 3
background hadrons from collisions. The MTD is cm to suppress secondary decays; number of TPC
a fast detector based on the Multi-gap Resistive clusters used in reconstruction should be greater
Plate Chamber technology to record signals, also than 15 (the maximum possible is 45) to have
referredto assignals(“hits”)generatedbycharged good momentum resolution; number of TPC clus-
particles traversing it. It provides single-muon and ters used for the dE/dx measurement is greater
dimuontriggersbasedonthenumberofhitswithin than 10 to ensure good dE/dx resolution; the ra-
apredefinedonlinetimingwindow. TheMTDmod- tio of the number of used TPC clusters over the
ules are installed at a radius of about 403 cm, and number of possible clusters is not less than 0.52 in
cover about 45% in azimuth within η < 0.5 [5]. ordertorejectsplittracks. Tracksarealsorequired
Installation of the full MTD was 10%| |, 63%, and to project to MTD hits that fire the triggers.
100% completed for the 2012, 2013, and 2014 run Figure 1 shows the invariant mass spectrum of
years respectively. As shown in cosmic ray data, opposite-sign dimuon pairs with the selection cri-
the timing resolution of the MTD is 100 ps and teria described above applied to both candidate
the spatial resolutions are 1-2 cm in∼both rφ and daughters. The J/ψ signal is clear around 3.1
z directions [12]. ∼ GeV/c2. More than 1500 J/ψ candidates are
present in the data sample used here. Lighter
mesons, like ω, φ and η particles particles are ob-
3. Dataset and event selection scured by large backgrounds at low M .
inv.
3.1. Data and Monte Carlo
4. Muon identification
Data for this study were collected by the STAR
4.1. Methods
detector during the RHIC proton-proton run at a
To distinguishmuoncandidatesfromthe hadron
center of mass energy of 500 GeV in 2013. Events
background,therearefourvariables,∆ToF,∆y q,
in the data sample were selected using the MTD ×
∆z and nσ used in this study. ∆ToF is the dif-
dimuon trigger which requires at least two MTD π
ference between the calculated time-of-flight value
hits in coincidence with the bunch crossing. The
from track extrapolation with a muon particle hy-
datasetrepresentsanintegratedluminosityof28.3
pb−1. pothesis and the measured one from the MTD de-
ThedetectorresponsetotheJ/ψ µ+µ− signal tector. ∆z and ∆y are the residuals between the
→ MTD hit position and extrapolated track position
was studied using a Monte Carlo (MC) simulation.
on the MTD, where z is along the beam pipe and
The MC sample was generated by a single-particle
y is perpendicular to z along the surface of each
generator with flat distribution in p , φ and η for
T MTDmodule(approximatelyrφ). ∆y ismultiplied
J/ψ. These simulated signals were then passed
by charge (∆y q) to eliminate the charge depen-
through the full GEANT3 [13] simulation of the ×
dence. nσ is the difference between the measured
STAR detector, and “embedded” into real events, π
dE/dxandthetheoreticalvalueassumingthetrack
followed by the standard reconstruction procedure
is a pion (for simplicity with pre-existing codes),
as used for real data. The kinematic distributions
± normalized to the dE/dx resolution of the TPC:
ofthe embedded J/ψ andµ wereweightedbythe
pT spectrumofJ/ψ inpp collisionsat500GeVde- nσ = (logddEx)measured−(logddEx)π,theory. (1)
termined via interpolation through a global fit of π σ(log dE)
dx measured
2
×106
Counts/Bin0.3 2]Events/0.04 [GeV/c 23×103 SσχMN2 J/ =//=nψB d0=3 f.= . 010= 5158 65586±3.±430.±006.1.0/±031010765.4.1 0 G4GeeVV//cc22 Probability Density00..42 MDBaaCct akS gSigrionguannla d(l e ((mOSSbSe-)SdS.)) Probability Density00..23 MDBaaCct akS gSigrionguannla d(l e ((mOSSbSe-)SdS.))
0.1
0.2
1 fit
signal
background
0
0 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 0
0.1 Mµµ [GeV/c2] −80−60−40−20 0 20 40 60 80 −50−40−30−20−10 0 10 20 30 40 50
Opposite-sign ∆z [cm] ∆y×q [cm]
Same-sign
00 1 2 3 4 5Mµµ [Ge6V/c2] Probability Density00..23 MDBaaCct akS gSigrionguannla d(l e ((mOSSbSe-)SdS.)) Probability Density00..46 DBaactak gSriogunnadl ((OSSS-)SS)
0.1 0.2
Figure1: The dimuon mass spectrum with basic selection
described in Section 3.2 applied to both muons. The black 0 0
solidandredopencirclesareforthemuonpairswithoppo- −5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
siteandsamesignsincharge, respectively. Theinsetshows nσπ ∆ToF [ns]
the dimuon mass fits for J/ψ peak. The solid red line is
a combined fit to the signal and background with a single
Gaussianplusafourth-orderpolynomial,andthetwoverti-
calblackdashedlinesindicatethemasswindow[2.92-3.25 Figure 2: The probability density function of ∆z, ∆y×q,
GeV/c2]usedtoselecttheJ/ψ candidates. ∆ToF, and nσπ variables. The black points are the signal
from data (OS-SS), the solid line histograms are the signal
fromMC,andthedashedlinehistogramsarethebackground
fromdata(SS).
The probability distribution functions (PDFs) of
each variable for same sign dimuon pairs (SS)
within the M window [2.92, 3.25] GeV/c2 ground rejection power can be obtained by com-
inv.
paring the yield and significance of the J/ψ signal
(shown by vertical dashed black lines in the inset
from before and after the selection. Several sig-
of Fig. 1) were used to characterize backgrounds.
nal (single Gaussian or Crystal-Ball function) and
The PDFs of pure muons were then obtained by a
background(third-orderorfourth-orderpolynomial
subtraction of the background PDFs from those of
function) models are used to fit the dimuon mass
opposite sign dimuon pairs (OS) within the same
spectra. The efficiency is calculated by using the
M window. Similar distributions are extracted
inv.
average fitted values for the number of J/ψ from
from MC as well except for the ∆ToF distribution
all combinations of signal and background models,
because the timing signal of MTD is not modeled
and the fit model uncertainty is determined by us-
in the simulation. Figure 2 shows the comparisons
ing the maximum deviation of any fit result from
for the PDF of each variable between signal (data
the average.
andMC)andbackground. Thesignaldistributions
in data and MC are in reasonable agreement. Straight cut method
•
Three methods, straight cut, N-1 iteration and The simplest way to reduce background is to
LikelihoodRatio,areutilizedinthis paperto iden- directly apply cuts onthese four variables. An
tify muon candidates. The performance of these about 2.5σ window cut on ∆z and ∆y q,
± ×
three methods is quantified using a tag-and-probe and an asymmetric window cut1, 1.5σ to
−
procedure. In the low muon p region (p < 3.5 +2.5σ, on nσ are used to retain high effi-
T T π
GeV/c), the tagged muon is the one with higher ciency while rejecting background, where σ is
p , while the probed muon is the one with lower the width of the signal distributions as shown
T
p . However,inthehighmuonp region(p >3.5 inFig.2. Anempiricalasymmetriccut isused
T T T
GeV/c), in contrast, the tagged (probed) muon is
the one with lower (higher) p to increase statis-
T 1Themeanvalueofnσπ formuonsisshiftedtotheright
tics. The muon identification cuts are applied on by∼0.5σcomparedtopions;therefore,amorestrictcuton
theprobedmuons,andthentheefficiencyandback- lownσ isappliedtoreducethepioncontamination.
3
for ∆ToF since the hadron background has a
longtailtotheright. Specifically,theselection atio50 atio14
F r40 F r12
criteria are 5 < ∆ToF < 0.2 ns, ∆z < 20 PD PD10
cm, ∆y q−<20 cm and 1<nσ| <|3. 30 8
π
| × | − 6
20
4
N-1 iteration method
• 10 2
An advanced way to select muon candidates, 0
0 −2
called N-1 iteration, is to vary one variable to
−4
optimize the J/ψ signal significance (S/√B) −10−80−60−40−20 0 20 40 60 80 −50−40−30−20−10 0 10 20 30 40 50
with the otherN-1 variables fixedat eachiter- ∆z [cm] ∆y×q [cm]
ation step. The values of the cuts determined o50
using this method are −4.8<∆ToF<0.7 ns, DF rati40 PlinDeFa rr aetxiotrapolation
−105.5<<∆nzσ<<193.c6m. ,−9<∆y×q <14cmand P30 lliinneeaarr ffiitt (±1σ)
π
− 20
Likelihood Ratio method 10
•
A more sophisticated way to reduce the back-
0
ground level and keep high purity simulta-
neously is using more powerful multivariate −10−5 −4 −3 −2 −1 0 1 2 3 4 5
methods,suchastheLikelihoodRatiomethod. nσπ
The basic idea is to create a discriminative
variable in the form of a likelihood ratio R = Figure3: The PDFratios of ∆z, ∆y×q, ∆ToF, and nσπ
(1 Y)/(1 + Y), where Y = Qyi and each variables. Thesolidredlineindicatesthebin-to-bininterpo-
−
y =PDFbkg/PDFsig isaratiobetweenback- lationandthedashedredlineisalinearfittoparameterize
i i i
theratio.
ground and signal PDFs. Due to the limited
statistics in data for the signal PDFs which
causes large fluctuations, the embedded MC Thesystematicuncertaintiesfromusingdifferentfit
sampleisusedtoconstructthesignalPDFs. In models, as described in Section 4.1, are about 5 -
this method, only three variables, ∆y q, ∆z 7% in data. The p smearing uncertainty in MC
T
×
and nσπ, are used to calculate the R value for is evaluated by varying the mean and width in the
theprobedmuon. Thecutvalueson∆ToFare smearingfunction to matchthe mean and width of
fixed from the N-1 iteration method. Figure 3 reconstructed J/ψ mass within 1σ. In addition,
showsthePDFratios(PDFbkgd/PDFsig)for, for the Likelihood Ratio method,±we compared the
∆y q,∆z,andnσπ,respectively. Abin-to-bin results from using sideband or same-sign data as
×
interpolation (solid red line) is used to obtain the background PDFs, from extracting the ratios
the ratios between points in the middle region in the middle regionvia bin-to-bin interpolationor
while a linear fit (dashed red line) is used for via fitting with a third-order polynomial function,
the side regions where statistics are low. Fig- and from varying the fit function by 1σ in the
±
ure4(a)showsthe discriminatingpowerofthe side region shown as the blue dot-dashed lines in
LikelihoodRatiomethod,andthecutvalueon Fig. 3. The maximum deviation from the average
R variable, R > 0.2, is chosen to maximize of these results is assigned as the uncertainty re-
−
the significance of the J/ψ signal as shown in lated to determining the PDF ratios. The total
Fig. 4(b). systematicuncertaintiesindifferentp binsare0.9
T
- 8.7% (0.6 - 2.6%), 0.9 - 10.5% (0.2 - 2.1%) and
4.2. Systematic uncertainties 0.6 - 15.4% (0.5 - 3.8%) for straight cuts, N-1 iter-
ation and Likelihood Ratio method in data (MC),
The following sources of systematic uncertainty
respectively.
onthe muonidentificationefficiencyareconsidered
forthesethreemethods: thebackgroundandsignal
fit models used in data, and the smearing on muon 5. Results
p inMC.Forthe LikelihoodRatiomethod,differ-
T
ent methods to build the PDF ratios and different Theperformancesofdifferentmuonidentification
procedurestoextracttheratiosarealsoconsidered. methods are evaluated by using J/ψ µ+µ− sig-
→
4
00.25 B1.4
1
0. S/
R)/ Signal (OS-SS) 1.2
y( 0.2
bilt Background (SS) 1
a
b
Pro0.15 0.8
0.6
0.1
0.4
0.05
0.2
0 −1 −0.5 0 0.5 1 0−1 −0.5 0 0.5 1
Likelihood Ratio (R) Likelihood Ratio cut
(a) (b)
Figure4: (a)DistributionsoftheLikelihoodRatio(R)forsignal(blue)andbackground(red). (b)ThecutvalueonRvariable
isoptimizedusingthesignalsignificance.
nals with the selection cuts applied on the probed
(subleading)muonsasshowninFig.5. Allofthem
have the capability to reduce the background level
by morethan65%while keeping the J/ψ efficiency ×103
3
relatively high. The muon identification efficien- 2V/c] µtagged & µprobed matched to MTD
cies are calculated relative to the basic selection Ge µprobed with Straight Cuts
describedinSection3.2,andshownasafunctionof 04 [ µprobed with N-1 Iteration
pT in Fig. 6. The plateau efficiency (pT > 3 GeV) nts/0. 2 µprobed with Likelihood Ratio
isabout90%for the LikelihoodRatio method, and e
v
E
about 80% for the other methods. For the J/ψ
signal, the Likelihood Ratio method provides an
overall efficiency of about 80% and improves the
1
significance by a factor of 1.38. Detailed compar-
isons betweenallthree methods are summarizedin
Table 1.
After applying the muonidentificationselections
0
determined using the Likelihood Ratio method on 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8
both muons, not only the significance of the J/ψ Mµµ [GeV/c2]
signalisenhancedbyafactorof2(S/√B =31.89),
but also the peaks of the light mesons, such as ρ,
ω and φ, become clearer as shown in Fig. 7. This Figure 5: The dimuon mass spectra with various muon
offers a good opportunity to study light mesons, identification methods. The solid curves are combined fits
to the mass distribution with a Gaussian distribution plus
heavy quarkonium, and the dimuon continuum at
a fourth-order polynomial function. The dashed lines show
the STAR experiment. the fitted background. The solid circles are for the basic
selection applied to both muons; the open boxes, open cir-
clesandsolidtrianglesareforthecasesthatthesubleading
6. Conclusion muonisselectedusingstraightcut,N-1iterationcutandthe
LikelihoodRatiomethods,respectively.
TheMTDisanewlyinstalleddetectordedicated
to triggering on and identifying muons in STAR
with low kinematic cutoff. In this paper, we evalu-
5
Basic selection Straight cut N-1 iteration Likelihood Ratio
Signal 1656 143 114 1006 67 20 891 58 12 1346 74 22
± ± ± ± ± ± ± ±
Background 10861 104 317 2682 52 80 1962 44 58 3743 61 79
± ± ± ± ± ± ± ±
S/B 0.15 0.01 0.01 0.38 0.03 0.01 0.45 0.03 0.01 0.36 0.02 0.01
± ± ± ± ± ± ± ±
S/√B 15.89 1.37 1.12 19.44 1.31 0.49 20.11 1.33 0.39 22.01 1.23 0.43
± ± ± ± ± ± ± ±
ε — 0.61 0.01 0.05 0.54 0.01 0.05 0.81 0.01 0.07
signal
± ± ± ± ± ±
1 ε — 0.75 0.01 0.02 0.82 0.01 0.01 0.66 0.01 0.01
bkgd.
− ± ± ± ± ± ±
εMC — 0.62 0.01 0.03 0.51 0.01 0.03 0.80 0.01 0.04
signal ± ± ± ± ± ±
Table1: Comparisonoftheperformanceforthreemuonidentificationmethods. εsignaland1−εbkgd.arethemuonidentification
efficiency and the background rejection rate relative to the basic selection, respectively. The first and second errors are the
statisticalandsystematicuncertainties,respectively.
D efficiency1.21 D efficiency1.21
Muon I00..68 Muon I00..68 Bin30×103
s/ ρ,ω Basic selection (×0.1)
0.4 0.4 nt
u
0.2 0.2 Co φ Likelihood Ratio
Straight Cuts N-1 Iteration
00 2 4 6 8 10 00 2 4 6 8 10 20
pµ [GeV/c] pµ [GeV/c]
T T J/ψ
D efficiency1.21 MC 10
Muon I00..68 DToattaal errors
0.4 Systematic errors 00 1 2 3 4 5 6
0.2 Mµµ [GeV/c2]
Likelihood Ratio
00 2 4 6 8 10
pµ [GeV/c]
T Figure 7: The dimuon mass spectrum with both muons se-
lected with Likelihood Ratio method compared with that
usingbasicselectionwhichisscaledby0.1. Thesignificance
Figure6: Themuonidentificationefficienciesasafunction of J/ψ signal is greatly enhanced and the light mesons are
ofpT withstraightcuts,N-1iterationandLikelihoodRatio clearlyvisible.
methods. Thebluesolidpoints arefromMC,whilethe red
open circles are from data. The error bars are the total
uncertainties (statistical plus systematic), whilethe shaded
boxesrepresentthesystematicuncertainties.
6
ated three different muon identification methods:
straight cut, N-1 iteration and Likelihood Ratio
method. Each of these can reduce the background
level by more than 65% and keep the J/ψ signals
withabout60%ofefficiency. Withthismuoniden-
tification capability, the MTD opens the door to
study heavy-ionphysics with muons,especially the
quarkonium states, in the STAR experiment.
7. Acknowledgments
We thank the STAR Collaboration, the RACF
atBNL andNationalChengKungUniversity(Tai-
wan) for their support. This work was supported
in part by the U.S. DOE Office of Science under
the contract No. de-sc0012704 and the Goldhaber
Fellowship program at Brookhaven National Lab-
oratory. L. Ruan acknowledges a DOE Office of
Science Early Career Award. We thank Dr. Gene
Van Buren (BNL) for the proofreading.
References
References
[1] K. H. Ackermann et al. (STAR Collaboration) , Nucl.
Instr.andMeth.A499(2003), 624.
[2] M.Gyulassy,arXiv:nucl-th/0403032 (2004).
[3] T.Matsui,H.Satz, Phys.Lett.B178(1986), 416.
[4] A.Mocsy,Eur.Phys.J.C 61(2009), 705.
[5] L.Ruanet al,J.PhysG36(2009), 095001.
[6] Z.Xu,BNLLDRDproject07-007; BeamTestExperi-
ment(T963) atFermiLab.
[7] ATLASCollaboration,Eur.Phys.J.C74(2014),3130.
[8] TheCMSCollaboration,J.ofInstr.7(2012), 10.
[9] LHCbCollaboration,J.ofInstr.10(2013), 10.
[10] M.Andersonet al.(STARCollaboration), Nucl.Instr.
andMeth.A499(2003), 659.
[11] F. Bergsma et al. (STAR Collaboration), Nucl. Instr.
andMeth.A499(2003), 633.
[12] C. Yang et al, Nucl. Instr. and Meth. A762 (2014),
1-6.
[13] CERNProgramLibraryLongWriteupW5013.
[14] W.Zhaet al.,arXiv:1506.08985(2015).
7