Search for Scalar Top Quark Production in p-pbar Collisions at sqrt(s)=1.96 TeV

We report on a search for the supersymmetric partner of the top quark (scalar top) decaying into a charm quark and a neutralino in p-pbar collisions at sqrt(s)=1.96 TeV. The data sample, collected by the CDFII detector at the Fermilab Tevatron, corresponds to an integrated luminosity of 2.6 fb-1. Candidate events are selected by requiring two or more jets and a large imbalance in the transverse momentum. To enhance the analysis sensitivity, at least one of the jets is required to be identified as originating from a charm quark using an algorithm specifically designed for this analysis. The selected events are in good agreement with standard model predictions. In the case of large mass splitting between the scalar top quark and the neutralino we exclude a scalar top quark mass below 180 GeV at 95% confidence level.

V. Saveliev w , 15 A. Savoy-Navarro aa , 15 P. Schlabach, 15 A. Schmidt, 24 E.E. Schmidt, 15 T. Schwarz, 15 L. Scodellaro, 9 A. The standard model (SM) of elementary particles and fundamental interactions, however successful, is still incomplete since it does not explain the origin of electroweak symmetry breaking and does not give an answer to the gauge hierarchy problem [1]. A possible extension of the SM, supersymmetry (SUSY) [2], solves these problems by introducing a symmetry that relates particles of different spin. R-parity [2] conserving SUSY models also provide a prime candidate for the dark matter in the universe [3], namely the stable lightest supersymmetric particle (LSP). In these models, the left-handed and right-handed quarks have scalar partners, respectively denoted asq L andq R , which can mix to form scalar quarks with mass eigenstatesq 1,2 . Several models [4] predict that this mixing can be substantial for the scalar top (stop), yielding a stop mass eigenstate (t 1 ) significantly lighter than other scalar quarks. If thẽ t 1 is sufficiently light, it can be pair-produced copiously in proton-antiproton collisions at center-of-mass energy of √ s = 1.96 TeV at the Tevatron. A stop in the mass range accessible at the Tevatron is expected to decay, among other possibilities, into a charm quark and the lightest neutralino (χ 0 1 ), which is often assumed to be the LSP. Assuming R-parity conservation, stops are produced in pairs. We consider a scenario where mt 1 < m b + mχ+ (whereχ + is a chargino), and mt 1 < m W + m b + mχ0 1 . Under these conditions the decayt 1 → cχ 0 1 is dominant. Results from searches using this final state at the Tevatron have been previously reported in [5,6].
In this Letter, we report the search fort 1 → cχ 0 1 decays in pp collision data from 2.6 fb −1 of integrated luminosity collected by the upgraded Collider Detector at Fermilab (CDF II) at the Tevatron. The final state contains two c jets from the hadronization of the c quarks and features an imbalance in transverse momentum ("missing transverse energy" or E T [7]) from the two LSPs escaping detection.

II. ANALYSIS DESCRIPTION
CDF II is a multipurpose detector, described in detail elsewhere [8]. The charged-particle tracking system consists of silicon microstrip detectors and a cylindrical open-cell drift chamber, both of which are immersed in a 1.4 T solenoidal magnetic field coaxial with the beam line. The silicon detectors provide coverage in the pseudorapidity [7] range |η| ≤ 2 and are used to identify events with long-lived particle decays. The drift chamber surrounds the silicon detectors, has maximum tracking efficiency up to |η| = 1, and is used for charged particle momentum measurements. Segmented sampling calorimeters, which surround the tracking system, are arranged in a projective tower geometry and are used to measure the energy of interacting particles in the region |η| < 3.6. Muon candidates are identified by drift chambers, which extend up to |η| = 1.5, and are located outside the calorimeter volume. Jets are reconstructed from the energy depositions in the calorimeter cells using an iterative cone jet-clustering algorithm [9], with a cone size of radius R = (∆φ) 2 + (∆η) 2 = 0.7 [7]. Energy corrections [10] are applied to account for effects such as nonlinear calorimeter response, underlying event, and the position of the interaction point, that influence the measured transverse jet energy.
Candidate events used for this search are selected by an online event selection system (trigger) that requires E T ≥ 35 GeV and two jets. Further selections are applied off-line to remove accelerator-produced and detectorrelated backgrounds as well as cosmic-ray events. After off-line event reconstruction, the events are required to have E T ≥ 50 GeV, and at least two jets with |η| ≤ 2.4 and E T ≥ 25 GeV. The highest-E T jet is required to have E T ≥ 35 GeV and at least one of the selected jets is required to be in the region |η| ≤ 0.9.
The hadrons in jets coming from b or c quark fragmentation, heavy-flavor (HF) jets, have a measurable flight path, yielding secondary vertices relative to the pp interaction point (primary vertex). We require the events to have exactly one jet identified as a HF jet by the secondary-vertex tagging algorithm [11], since requiring two HF-identified jets enhances the sample with events containing two b quarks and reduces the signal acceptance. This criteria defines a preselection which is used as the basis of background studies and further optimization for the signal sample, as described below.
Dominant SM backgrounds are pair and single top-quark production, electroweak single and di-boson production, HF multijet production, and light-flavor multijet events where one of the jets is falsely tagged as a HF jet (mistag). The latter two background contributions are estimated from data. The alpgen [12] event generator interfaced with the parton-shower model from the pythia [13] event generator is used to estimate the electroweak boson production, the madgraph [14] generator is used to model the single top events, while the pythia event generator is used to model the top-quark pair and diboson backgrounds. For the event generation the CTEQ5L [15] parton distribution functions (PDFs) are used. Simulated events are passed through the geant3-based [16] CDF II detector simulation [17] and are weighted by the probability that they would pass the trigger selection. The single top-quark and diboson contributions are normalized to the theoretical cross sections [18][19][20]. The contributions for the electroweak boson samples are normalized to the next-to-leading order cross sections calculated with mcfm [21]. We use the measured top-quark pair production cross section of σ tt = 7.02 ± 0.63 pb [22]. Mistags are estimated based on the mistag rate [11] observed in inclusive jet data. The mistag rate is parametrized as a function of jet E T , |η|, secondary-vertex track-multiplicity, the number of primary vertices in the event, primary vertex z-position, and the scalar sum of E T of all jets in the event. To estimate the HF multijet background from data, we use a multijet tag rate estimator (MUTARE) described elsewhere [23,24]. The HF multijet prediction from MUTARE is scaled by a multiplicative factor that is obtained in a signal-free region.
To avoid potential biases when searching for new physics, we test the various background contributions in distinct control regions that are defined a priori. The three control regions used to validate the SM prediction are denoted as multijet, lepton, and pre-optimization regions. The multijet control region is defined to have the second leading E T jet I: Comparison of the total number of expected and observed events in the control regions. The total uncertainty is computed by taking into account the (anti)correlations between the partial uncertainties. In the lepton region the HF multijet prediction is scaled to match the observed events in order to perform shape comparisons only.  direction ( j 2 ) aligned with the − → E T [7], where aligned means ∆φ( − → E T , j 2 ) ≤ 0.4 rad. This HF multijet enriched region is used to obtain the MUTARE parameterization to predict the HF multijet background in the other control and signal regions. The lepton control region is defined to have j 2 direction not aligned with the − → E T (∆φ( − → E T , j 2 ) ≥ 0.7 rad) and at least one isolated charged lepton (e or µ) with p T ≥ 10 GeV/c . This lepton region is used to validate the modeling of the top quark and electroweak W/Z boson backgrounds. The pre-optimization control region is defined to have the leading and second leading E T jets not aligned with the − → E T and to have no identified leptons, where the lepton definition is the same as in the lepton control region. We studied kinematic distributions such as jet E T , E T , as well as the total numbers of events predicted. All these quantities are in agreement with observations in the three control regions. As an example, the E T distributions for the pre-optimization region are shown in Fig. 1. Table I presents the expected event yields for various background processes in the control regions under study.
We optimize the sensitivity to stop production by applying additional event selection criteria using a set of cuts suppressing the HF multijet background. We select events with exactly two jets, as expected from the signal, and fulfilling the condition ∆φ( − → E T , − → E trk T ) < π/2. This variable is the angular difference between the calorimetry-based − → E T and the same quantity calculated using tracks ( − → E trk T ) [7]. When the E T in the event is real, these two quantities are usually aligned in φ. However, when the E T comes from calorimetry mis-measurements, as in HF multijet events with no real E T , the angular difference between the two quantities is randomly distributed.
To further improve the sensitivity we apply a neural network (NN) trained with the tmva package [25], to reduce the remaining HF multijet background. We train the NN using jet E T , jet η, the minimum ∆φ between the − → E T and any of the selected jets, ∆φ( − → E T , − → E trk T ), ∆φ( j 1 , j 2 ), E T , E trk T , and the summed E T of all the jets in the event. We choose a reference signal point with m(t 1 ) = 125 GeV/c 2 and m(χ 0 1 ) = 70 GeV/c 2 to perform the optimization. The signal acceptance is obtained using the pythia event generator and CTEQ5L PDFs. Total signal yields are normalized to the NLO production cross section determined with the prospino event generator [26] and the CTEQ6M [27,28] PDFs. The uncertainty of the NLO production cross section is estimated to be 20%, arising from the scale dependence and the uncertainties on the PDFs. The NN output lies within −1 and 1, where the background peaks at −1 and the signal peaks at 1. We define our signal region as events with NN output scores > 0.
The final stage in the optimization is the application of a charm hadron analysis oriented separator (CHAOS) technique [24], explicitly designed for this analysis to obtain a sample enriched in c jets. CHAOS is a NN producing a two-dimensional output and trained with the snns v4.3 package [29] to determine whether a jet identified as HF has been produced from the hadronization process of a light quark falsely tagged as a HF jet, a b quark, or a c quark. The two-dimensional output structure allows the separation of the three different targets during the same training process. Depending on the flavor of the original parton, the jet identified as HF and its secondary vertex have different characteristics, mainly related to the tracking. Using properties of the tracks forming the secondary vertex and the tracks of the jets within a neural network, CHAOS allows enhancement of the jet selection with a desired flavor, in particular c jets. We apply CHAOS to the jet identified as HF and we find that the optimal cut has a selection efficiency [24] of 34% for c jets, 7.3% for b jets, and 4.9% for light jets.
The systematic uncertainties on the signal and the background predictions, taking into account correlated and uncorrelated uncertainties, are studied. Correlated uncertainties, affecting both the background prediction and signal acceptance, are dominated by the uncertainties on the performance of the b-tagging algorithm and CHAOS, which are 4.4% [11] and 9.2% respectively, and the luminosity (6%) [8]. Uncorrelated systematic uncertainties on the background predictions are dominated by uncertainties on the MUTARE parameterization (30%), the mistag rate (16% [11] for light-flavor multijets), the top-quark pair-production cross section (11%), the single top-quark production cross section (13%), and the diboson production cross section (10% for W W /W Z and 20% for ZZ). The uncertainty on the normalization of the boson plus HF jets to the total inclusive boson production cross section translates into a 10% uncertainty in the SM predictions. Correlated and uncorrelated uncertainties are evaluated separately and combined in quadrature.

III. RESULTS
The signal region is analyzed after the background predictions are determined. We observe 115 events, where 132.0 ± 24.4 are expected from background, as summarized in Table II where yields before selecting events based on CHAOS output are also shown. Since no significant deviation from the SM prediction is observed, the results are used to calculate a 95% C.L. exclusion limit for thet 1 pair production cross section.
We have used the differences in shape of the NN output to set the limits. These limits are computed using a Bayesian likelihood method [31] with a flat prior probability for the signal cross section and Gaussian priors for the uncertainties on acceptance and backgrounds. Figure 2 shows the expected and observed limits as a function of m(t 1 ) for a neutralino mass of 80 GeV/c 2 .
We exclude, assuming BR(t 1 → cχ 0 1 ) = 100%,t 1 masses up to 180 GeV/c 2 at 95% C.L. In addition, a 95% C.L. limit is obtained in the mass parameter plane of the model. Figure 3 shows the excluded region in the stop-neutralino mass plane of the analysis, compared with results from previous analyses [5,6,30]. The limit obtained with the  The result is compared to the previous results from CDF [5], from D0 [6], and from LEP [30] experiments at CERN. present analysis improves the results of previous searches using a similar topology, and represents the world's best limit in the region of large mass splitting.

IV. SUMMARY
To summarize, we have searched for the pair production of stop decaying into a charm quark and a neutralino, in 2.6 fb −1 of CDF Run II data. We observe 115 candidate events, which are in agreement with SM background expectations of 132.0 ± 24.4 events. No evidence for stop is observed, and we exclude a region in the stop and neutralino mass plane at 95% C.L. as shown in Fig. 3. Assuming BR(t 1 → cχ 0 1 ) = 100%, we exclude stop masses up to 180 GeV/c 2 at 95% C.L. for a neutralino mass of 90 GeV/c 2 .