Inclusive Jet Measurements in Pb-Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques
IInclusive Jet Measurements in Pb–Pb Collisions at5.02 TeV with ALICE using Machine Learning Techniques
Hannah Bossi for the ALICE Collaboration a , ∗ a Yale University,Wright Laboratory, New Haven, CT, USA
E-mail: [email protected]
These proceedings report on measurements of the jet spectrum and nuclear modification factor forinclusive full jets (containing both charged and neutral constituents) in Pb–Pb and pp collisionsat √ s NN = .
02 TeV recorded with the ALICE detector. These measurements use a machinelearning based background correction [1], which reduces residual fluctuations. This methodallows for measurements to lower transverse momenta and larger jet resolution parameter ( R ) thanpreviously possible in ALICE. In this method, machine learning techniques are used to correctthe jet transverse momentum on a jet-by-jet basis using jet parameters such as information aboutthe constituents of the jet. Studies that investigate the effect of the potential fragmentation biasintroduced by learning from constituents will also be discussed. With these studies in mind, theresults are compared to theoretical predictions. HardProbes20201-6 June 2020Austin, Texas ∗ Speaker © Copyright owned by the author(s) under the terms of the Creative CommonsAttribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). https://pos.sissa.it/ a r X i v : . [ nu c l - e x ] S e p nclusive Jet Measurements in Pb–Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques Hannah Bossi for the ALICE Collaboration
1. Introduction
Measurements of inclusive jet suppression (or R AA as defined in Equation 1 as the ratio ofthe per-event jet yield in Pb–Pb and the cross section in pp multiplied by T AA accounting for thecollision geometry) serve to search for signatures of jet quenching in heavy-ion (HI) collisions, R AA = (cid:104) T AA (cid:105) N events , PbPb d N AAjet d p T d η d σ pp d p T d η . (1)The nature and extent of this energy loss is expected to vary for different jet resolutionparameters R across p T scales. Jet quenching models describe the R -dependence of the R AA differently, with some models predicting this dependence to be stronger at lower jet p T [2–5].Recent experimental efforts have extended measurements at high p T to R = . R and low p T are also of interest.Reconstructing the jet transverse momentum ( p T ) in HI collisions is challenging due to thelarge fluctuating background from the underlying event (UE). For example, the fluctuations in thecharged particle momentum density in central (0–10%) Pb–Pb collisions at the LHC are ≈
18 GeV/ c [7]. Lower p T UE jets, commonly called fake jets, additionally contaminate the signal at low p T .One common treatment of this background (herein referred to as the area-based or AB method) isto subtract off the event-averaged momentum density (excluding the two leading jets) multiplied bythe jet area from the uncorrected jet p T [7].A leading track cut of 7 GeV/ c for R = 0.4 jets is typically applied in order to remove thefake jet contribution at the cost of an introduced bias. While the AB method effectively correctsfor the average background, it does not account for region-to-region fluctuations. These residualfluctuations are commonly handled in an unfolding procedure. Due to the limitations of the largebackground, lower jet p T and large R are less studied regions with inclusive jet probes.In these proceedings, inclusive jet measurements with ALICE utilizing a machine learning(ML) based method previously developed for charged particle jets [1] will be discussed. ThisML-based method allowed for the extension of charged particle jet measurements to lower p T for R = 0.6 jets, unprecedented in ALICE. These proceedings will present the extension of this methodto full jets, which is desirable as full jets are more aligned with the traditional definition of a jet.
2. Analysis Details
The results discussed in these proceedings utilize Pb–Pb data collected with the ALICE detector[8] in 2015 at √ s NN = .
02 TeV with an integrated luminosity of 0.4 n b − . Central events (0–10%)were analysed with a minimum bias trigger setup. The training of the ML estimator (see Section3) and the response matrix used for unfolding utilize detector level events and hybrid level events.These hybrid events were created using a PYTHIA8 [9] generated events propagated through aGEANT3 [10] simulation of the ALICE detector (herein referred to as the detector level ) whichwere embedded into real Pb–Pb data to mimic HI background effects (herein referred to as the hybrid level ). The ML approach will be compared with results from [11] which are based on theAB method. For the evaluation of R AA , the pp reference spectrum without a leading track bias wastaken from [11]. 1 nclusive Jet Measurements in Pb–Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques Hannah Bossi for the ALICE CollaborationThese proceedings focus on full jets that contain both the charged and neutral componentof the jet measured using charged particles registered by the Inner Tracking System [12] and theTime Projection Chamber [13] and electromagnetic clusters measured from the ElectromagneticCalorimeter [14]. The FastJet package [15] was used to cluster charged constituents above 150MeV/ c and neutral constituents above 300 MeV/ c into jets using the anti- k T algorithm [16] with a p T recombination scheme.
3. ML-estimator
The ML-based background estimator creates a mapping between the measured and recon-structed jet p T , correcting each jet for the background particles that overlay it. The ML estimatoritself employs a shallow neural network as implemented in scikit-learn [17] with three layers. Asinput the estimator utilized a minimal, discriminative set of features including jet properties and theproperties of the constituents of the jet.The ML estimator is a regression task where the regression target is the reconstruction ofthe detector level PYTHIA jet p T from the hybrid event used in training the estimator. Theperformance of the estimator is quantified with δ p T distributions measuring the difference betweenthe jet p T predicted by the ML estimator and the PYTHIA detector level jet p T . A narrowing ofthe width in δ p T corresponds to reduction of residual fluctuations after background subtraction.Figure 1 demonstrates that the ML estimators more effectively reduce the residual fluctuationsremaining after background subtraction than the AB method and leads to a reduced width of the δ p T distribution. Figure 1:
The δ p T = p T , rec − p T , det distributions for R = .
4. Results
The ML method was applied to full R = . R AA was then evaluated using this inclusive jet spectrumand the scaled pp reference spectrum from [11]. These results are shown in Figure 2 compared to theprevious result obtained using the AB method [11]. With the ML-based correction, measurementsof jet suppression are extended to 40 GeV/ c . The systematic uncertainties were also reduced.2 nclusive Jet Measurements in Pb–Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques Hannah Bossi for the ALICE CollaborationThe left panel of Figure 2 illustrates an investigation of the fragmentation bias of the ML-basedestimator. This bias is introduced by the inclusion of constituent information in training. Trainingon PYTHIA jets causes a bias towards a pp PYTHIA fragmentation which is suggested to differfrom the fragmentation pattern of jets in Pb–Pb by numerous studies (e.g.[18, 19]). For theseinvestigations, a toy model is used with three different modifications to the constituents of the jetand therefore the fragmentation function. The fractional methods refer to each constituent losing afraction of its energy at specified angles from the jet axis. The BDMPS method is a modificationwhere the emission angle and energy is selected by sampling the BDMPS gluon emission spectrum[20–22]. The potential fragmentation bias of such modifications is quantified by training the MLon the modified toy model and calculating the resulting R AA . Figure 2 demonstrates the relativerobustness of this method to the explored biases.Keeping these studies in mind, comparisons with theoretical predictions are shown in theright panel of Figure 2. Comparisons with JEWEL with recoils on/off [3], SCETg (with radiativeand collisional energy loss) [4], Hybrid model (with wake) [2], and LBT [5] are shown. Thismeasurement of inclusive jets to lower jet p T constrains models in a less studied region. Figure 2:
The R AA with the ML and AB method are show for R = . R AA are compared to the fragmentation bias curves for a toy model with three differentmodifications. The right panel compares the R AA to various theoretical predictions. The ML-based R AA andthe AB R AA [11] are compared to the same pp reference, which does not include a leading track bias.
5. Conclusion
In these proceedings, a ML based background correction is presented, which allows for theextension of inclusive jet suppression measurements down to 40 GeV/ c . The fragmentation biasintroduced by training the ML on the constituents from PYTHIA has been quantified, which alsodemonstrates the robustness of this method. Future analyses with this method will explore largerresolution parameters and extend tests of the fragmentation bias.3 nclusive Jet Measurements in Pb–Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques Hannah Bossi for the ALICE Collaboration
References [1] R. Haake, C. Loizides, "Machine-learning-based jet momentum reconstruction in heavy-ioncollisions,"
Phys. Rev. C
99, 064904 (2019).[2] D. Pablos, "Jet Suppression From a Small to Intermediate to Large Radius,"
Phys. Rev. Lett.
JHEP
JHEP
Phys. Rev. C.
99, 054911 (2019).[6] CMS Collaboration, "Measurement of Jet Nuclear Modification Factor inPb–Pb Collisions at √ s NN = 5.02 TeV with CMS," CMS-PAS-HIN-18-014, http://cds.cern.ch/record/2698506 .[7] ALICE Collaboration, "Measurement of Event Background Fluctuations For Charged ParticleJet Reconstruction in Pb–Pb Collisions at √ s NN = 2.76 TeV," JHEP √ s NN = . Phys. Rev. C.
JISNT , 5:P03003, 2010[13] J. Alme, et al., "The ALICE TPC, a large 3-dimensional tracking device with fast readout forultra-high multiplicity events,"
Nucl.Instrum.Meth. , A622:316–367, 2010.[14] ALICE Collaboration, "ALICE electromagnetic calorimeter technical design report," CERN-ALICE-TDR-014, CERN-LHCC-2008-014, 2008 [https://inspirehep.net/literature/794183][15] M. Cacciari, G.P. Salam and G. Soyez,
Eur.Phys.J.
C72 (2012) 1896.[16] M. Cacciari, G.P. Salam, and G. Soyez, "The anti-kT jet clustering algorithm",
JHEP √ s NN =2.76 TeV," Phys. Rev. C.
90, 024908 (2014).[19] ATLAS Collaboration, "Measurement of jet fragmentation in Pb+Pb and pp collisions at √ s NN = 5.02 TeV with the ATLAS Detector," Phys. Rev. C.
98, 024908 (2018).[20] R. Baier, Yu. L. Dokshitzer, A. H. Mueller, S. Peigne, D. Schiff, "Radiative energy loss ofhigh-energy quarks and gluons in a finite volume quark-gluon plasma",
Nucl. Phys.
B483(1997) 291–320.[21] R. Baier, Yu. L. Dokshitzer, A. H. Mueller, S. Peigne, D. Schiff,"Radiative energy loss andp(T) broadening of high-energy partons in nuclei,"
Nucl. Phys.
B484 (1997) 265–282.4 nclusive Jet Measurements in Pb–Pb Collisions at 5.02 TeV with ALICE using Machine Learning Techniques
Hannah Bossi for the ALICE Collaboration[22] R. Baier, Yu. L. Dokshitzer, A. H. Mueller, D. Schiff, "Quenching of hadron spectra in media,"