Measurement of the top quark mass in topologies enhanced with single top quarks produced in the t -channel at s √ =8TeV using the ATLAS experiment
MMeasurement of the top quark mass in topologiesenhanced with single top quarks produced in the t -channel at √ s = 8 TeV using the ATLASexperiment Hendrik Esch On behalf of the ATLAS Collaboration. TU Dortmund, Exp. Physik IV, Otto-Hahn-Str. 4, 44227 Dortmund, GermanyE-mail: [email protected]
Abstract.
This article presents a measurement of the top quark mass in topologies enhancedwith single top quarks produced in the t -channel produced via weak interactions. Thedataset was collected at a centre-of-mass energy of √ s = 8 TeV with the ATLAS detectorat the LHC and corresponds to an integrated luminosity of 20 . − . To determine thetop quark mass a template method is used based on the distribution of the invariant massof the lepton and the b -tagged jet as estimator. The result of the measurement is m top =172 . ± . . ) ± . . ) GeV.
1. Introduction
In addition to the t ¯ t pair production via the strong interaction, in proton-proton ( pp ) collisionsat the LHC, top quarks can also be produced singly via the weak charged-current interactions,giving another possibility for measuring the top quark mass. The dominant process for singletop quark production is the t -channel exchange of a virtual W -boson. Important differencesfrom the production mode compared to t ¯ t , resulting in different sizes of certain systematicuncertainties, and the fact that the measurement of the top quark mass is obtained from astatistically independent sample, provide an excellent motivation for such a measurement andfor including it in future combinations with other measurements.In this article, the first measurement of m top in topologies enhanced with t -channel singletop quark production with the ATLAS experiment [1] is presented. Production of top quarkpairs also give a significant contribution to the sample, while W t production and s -channelproduction only give minor contributions. Events are characterised by an isolated high- p T charged lepton (electron or muon), missing transverse momentum from the neutrino and exactlytwo jets produced by the hadronisation of the b -quark and the light quark in the t -channel.The main backgrounds are W/Z +jets production, especially in association with heavy quarks,diboson production, and multijet production via QCD processes. Events from all single topproduction processes and t ¯ t production are treated as signal in the analysis. a r X i v : . [ h e p - e x ] N ov . Event selection Based on the expected signature of the signal, events are selected with exactly one isolatedelectron or muon, missing transverse momentum and exactly two jets, out of which one isrequired to be identified as a b -quark jet.Offline electron and muon candidates are required to be isolated and satisfy p T >
25 GeVand | η | < . k t algorithm with a distance parameter of 0.4. Thereconstruction is based on locally calibrated clusters with simulation-based as well as in-situcalibrations based on data. They are required to satisfy p T >
30 GeV and | η | < .
5. Jets within2 . < | η | < .
5, which have significant energy deposited in the endcap-forward calorimetertransition region, must have p T >
35 GeV. Exactly one of the selected jets is required to beidentified as a b -quark jet by the MV1c b -tagging algorithm. The algorithm is applied at anefficiency of 50% for b -jets in simulated t ¯ t events.In order to reduce the number of multijet background events, which are characterised by low E missT and low transverse W -boson mass m T ( W ), the event selection requires E missT >
30 GeVand m T ( W ) >
50 GeV. Another class of multijet background events are further reduced byapplying an additional cut, which is realised by the following condition between the lepton p T and the ∆ φ ( j , (cid:96) ): p T ( (cid:96) ) >
40 GeV (cid:18) − π − | ∆ φ ( j , (cid:96) ) | π − (cid:19) , (1)where (cid:96) denotes the identified charged lepton and j the reconstructed jet with the highest p T [2].
3. Background Estimation
To determine the normalisation of the multijets background, a binned maximum likelihood fitis performed to the E missT distribution in data after applying all selection criteria, with the cuton E missT removed. Template distributions for the multijet backgroud are obtained by differentmethods in the electron and muon channel, correspondingly.In the electron channel a jet-lepton model [3] is obtained by selecting simulated multijetevents with jets that have similar properties to selected electrons. In the muon channel ananti-muon method [3] is used, which builds a multijet model derived from collision data.The multijet template is fitted together with templates derived from MC simulation for allother processes whose rate uncertainties are accounted for in the fitting process in the formof additional constrained nuisance parameters. The corresponding E missT distributions afterrescaling the different backgrounds and the multijet template to their respective fit results areshown in [4] for both the electron and muon channel.
4. Neural network selection
Following the event selection described in Section 2, the selected sample is still dominatedby background processes. Multivariate analysis techniques are used to separate signal frombackground candidates. A neural network classifier [5] that combines a three-layer feed-forwardneural network with a preprocessing of the input variables is used to enhance the separationpower.The network infrastructure consists of one input node for each input variables plus one biasnode, 15 nodes in the hidden layer, and one output node which gives a continuous output inthe interval [0 , t -channel events as signal during thetraining and W +jets, Z +jets, and diboson processes are considered as background. Extensivestudies were done to ensure that using a signal sample with a fixed top quark mass does notbias the result of the measurement.The input variables to the neural network are selected so that for a minimal number ofvariables the best possible separation between the signal and background processes is achieved.ach variable is initially tested for agreement between the MC background model and observeddata events in control regions and, taking into account potential signal contributions, is alsotested in the signal region. NN output r e l . d i ff e r en c e -0.4-0.200.20.4 E v en t s / . DATAuncertaintyMultijetsZ+jets, dibosonW+jets, Wt, s-channelttt-channel
ATLAS =8TeVs -1 L dt = 20.3 fb ∫ Figure 1.
Neural network outputdistribution in the signal region normalisedto the number of expected events [2]. This leads to 12 variables remaining for thenetwork including variables obtained from thereconstructed W -boson and the top quark.The resulting neural network output distri-butions for the various processes in the signal re-gion is shown in Figure 1. Signal-like events haveoutput values close to one, whereas background-like events accumulate near zero.To enhance the signal sample with singletop and t ¯ t events a cut on the neural networkoutput variable at 0 .
75 is chosen. In the signalregion 19833 events that fulfill this cut areobserved in data while the expectation from SMbackgrounds amounts to 19470 ±
5. Measurement of m top with a template method In order to measure the top quark mass in the signal region after the cut on the neural network,a template method is used. Simulated distributions are constructed for m ( (cid:96)b ), which is sensitiveto the top quark mass, using a number of discrete values of m top . This m ( (cid:96)b ) estimator is definedas the invariant mass of the charged lepton plus the b -jet system.The resulting distribution in the signal region after the cut on the neural network output indata together with the prediction assuming m top = 172 . E v en t s / G e V DATAuncertaintyMultijetsZ+jets, dibosonW+jets, Wt, s-channelttt-channel -1 L dt = 20.3 fb ∫ Preliminary
ATLAS m(lb) [GeV]
60 80 100 120 140 160 r e l . d i ff e r en c e -0.4-0.200.20.4 Figure 2.
Distributions of m ( (cid:96)b ) forevents with an output value of the neuralnetwork larger than 0 .
75. The signal MCprocesses assume m top = 172 . m(l b) [GeV]
60 80 100 120 140 160 E v en t s / G e V ATLAS
Simulation Preliminary =165.0 GeV top m =172.5 GeV top m =180.0 GeV top m Figure 3.
Dependence of the m ( (cid:96)b )distribution of all top quark processes onmtop for the signal MC samples generatedwith different input top quark masses,together with the signal probability densityfunctions [2].The templates are parametrised and the parameters are then interpolated between differentvalues of m top . In Figure 3 the sensitivity of the m ( (cid:96)b ) observable to the input value of the topquark mass is shown by the m ( (cid:96)b ) distributions for three different mass points together withheir respective fitted parametrisations. All single-top and t ¯ t processes are treated as signaland the signal templates for m ( (cid:96)b ) are fitted using an analytic expression including Landauand Gauss parametrisations. The same parametrisation is used for the mass-independent m ( (cid:96)b )distribution of the background, which is dominated by W+jets and QCD-multijet production.In the final step a likelihood fit to the observed data distribution is used to obtain the valueof m top that best describes the data. The likelihood has three parameters: the top quark mass m top , the relative background fraction f , and the overall normalisation N . The fraction f isconstrained by a Gaussian distribution centred around the prediction from simulation f bkg . Thewidth of the Gaussian σ f bkg reflects the theoretical uncertainty on the background fraction.
6. Results
The result of the fit to 2012 ATLAS data in topologies enhanced with t -channel single top quarksevents is: m top = 172 . ± . . ) ± . . ) GeV . (2)The distribution of m ( (cid:96)b ) in the full dataset together with the corresponding fitted probabilitydensity functions for the signal and background is shown in Figure 4. m(lb) [GeV]
60 80 100 120 140 E v en t s / . G e V [GeV] top m170 171.33 172.67 174 - l n ( L ) = 8TeV datasBackground signalttSingle-top t-channel signal Preliminary
ATLAS -1 Ldt = 20.3 fb ∫ ± = 172.2 top Best fit: m
Figure 4.
Fitted m ( (cid:96)b ) distribution in data withthe normalisation and m top being the best fitvalues. The relative mixture for the dominantsingle top t -channel production process and theother top processes, dominated by t ¯ t , are shown inlight and dark blue, respectively, and correspondto the values determined in [4]. The inset showsthe corresponding − L profile as a function ofthe top quark mass [2].The result has a total uncertainty of about 2 GeV which is dominated by systematicuncertainties. The largest contribution comes from JES uncertainties and the modelling ofthe t -channel process. Due to the (cid:96) + 2-jet channel selection there is no statistical correlationbetween the dataset used in this analysis and any other analysis performed using the t ¯ t finalstate.The selection with exactly one tagged plus one untagged jet present in the final state leadsto a reduced combinatorial background and better mass resolution compared to the t ¯ t → lepton+jets or the t ¯ t all hadronic decay channels. The presence of only one neutrino is anadvantage with respect to the t ¯ t → dilepton decay channel where the assignment of the missingtransverse momentum to the neutrinos is ambiguous. These advantages in terms of systematicsare complementary to the advantages of other channels, e.g. the smaller contributions frombackgrounds, indicating good prospects for combined measurements in the future. Acknowledgements
The author thankfully acknowledges the financial support of the BMBF (FSP101-ATLAS).
References [1] ATLAS Collaboration, 2008
JINST S08003[2] ATLAS Collaboration, ATLAS-CONF-2014-055 (09/2014) https://cds.cern.ch/record/1951323[3] ATLAS Collaboration, ATLAS-CONF-2014-058 (09/2014) https://cds.cern.ch/record/1951336[4] ATLAS Collaboration, ATLAS-CONF-2014-007 (03/2014) https://cds.cern.ch/record/1668960[5] Feindt M, Kerzel U 2006
Nucl. Instrum. Meth. A559