Exotics Working group
CIEMAT group

The CDF Collaboration


Miguel Vidal (vidal@fnal.gov)

Oscar Gonzalez (oglez@fnal.gov)

P.0. Box 500, M.S. 318 (MADRID)
Batavia, Illinois 60510

Search for Scalar top decaying into Charm and Neutralino

Public Web Page

Using the CDF detector in the Run II of the Tevatron, we have analyzed events containing two jets and missing transverse energy in order to look for the presence of new physics. At least one of the jets was required to be tagged as originating from a heavy-flavor quark in order to enhance the sensitivity for events containing a scalar top decaying into charm and neutralino. The analysis was optimized via one Neural Network to reduce the heavy flavor multijets background plus a flavour separator to enhance the c jet contribution. Results were obtained with 2.6fb-1 of data and the achieved sensitivity allows to exclude stop masses up to 180 Gev/c2 at 95% C.L.

- Basic Event Selection -
- Control Regions -
- Signal Region -
- Systematics Uncertainties -
- Limits -
Download plots in EPS format by clicking on the plot

Public Note


In the present analysis we search for direct stop quark production. We look for direct stop pair production, where the stop decays to charm and neutralino. The neutralino is taken to be the Lightest Supersymetric particle (LSP) and R-parity conservation is assumed. Therefore, the stop signature is 2 c jets and large missing transverse energy (MET) from the LSP escaping detection.

For this analysis we use the MET+JETS trigger. We define three control regions, and one signal region. In the three control regions predicted distributions are compare with those measured in data requiring single and double tagged events (using the SecVtx tagging algorithm). We then perform a signal optimization, requiring one tag events only, by using a Neural Network (NN) in order to reduce the main background at this point, the heavy-flavour QCD multi-jet production. After that we apply a flavour separator (CHAOS) to enhance the c jet contribution in the final state. Finally we extract a limit based on the shapes of the final discriminant.


Basic Event Selection :

The present analysis is made using a three-level logic MET+jets trigger. A sequence of cuts on the MET is required at each level plus aditional cuts over the jets at level 2. All the event processed in the analysis are required to have:

  • At least 2 Jets
  • MET > 50 GeV
  • ET > 25 GeV and |η| < 2.4
  • Leading Jet ET > 35 GeV
  • 1 Central Jet |η| < 0.9
  • Jet EM Fraction < 0.9
Since it is expected a 2 c jets final state, we are requiring one of the jets as originating from a heavy-flavour quark using a Secondary Vertex tagging algorithm.


Control Regions :

The SM processes predicted with Monte Carlo or data-driven methods were tested in control regions defined such that they result in background-dominated samples in which signal contribution is negligible. Two regions were defined by reversing the selection requirements introduced to suppress specific background processes and a third one was defined in order to check the analysis tools in a signal-like environment, but avoiding the application of cuts that would enhance the signal contribution to a measurable level.

  • QCD Region Plots

    Lead Jet ET MET

  • Lepton Region Plots

    Lead Jet ET MET

  • Pre-Optimization Region Plots

    Lead Jet ET MET


    Signal Region :

    An optimization process using a Neural Network (NN) plus a flavour saparator (CHAOS), developed for this analysis, was made in order to reduce the background contribution.
    The process takes the Pre-optimization selection as benchmark. The first step in this optimization is to reduce the HF multijet background as much as possible. We apply a couple of cuts requiring only two jets in the final region and ΔΦ(MET,TrackMET) < 90 degrees where TrackMET is the MET calculated using the CDF tacking system. This two cuts remove easily a lot of HF multijet background due mainly to miss-measurements. After that we apply a NN trained with signal MC versus taggable jets (HF multijet-like) The optimized signal is:

    • Optimization Signal: m(stop) = 125 GeV/c2, m(neutralino) = 70 GeV/c2
    The set of variables used to train the NN is:
    • MET
    • TrackMET
    • HT
    • ET,j1
    • ET,j2
    • ηj1
    • ηj2
    • ΔΦ(j1,j2)
    • minΔΦ(MET,jets)
    • ΔΦ(MET,TrackMET)

    We apply a cut on the NN output selecting events in the region between 0 and 1. The region -1 to 0 is used as another control region in which we normalize the amount of HF multijet contribution to data.

    HF multijet-NN output

    As a final step in the optimization process we apply a flavour separator to enhance the c jet contribution in the final region. Charm Hadron Oriented Separator (CHAOS) is a 2D Neural Network developed for this analysis. Using the sum of the outputs in 1D we fund a since separation between c and other flavours. We apply a cut on the CHAOS sum of the outputs at 1.65 getting the final region in this way.

    CHAOS outputs

    We found the output of the HF multijet-NN after applying CHAOS as the best discriminant. No significant deviation of the SM was found in the analysis.

    Final discriminant

    HF multijet-NN output aplying CHAOS Final numbers

    Kinematics in final region


    Systematics Uncertainties :

    We address systematic uncertainties from different sources:

    • Jet Energy Scale (JES): A systematic error in the calorimeter energy scale affect the total transverse energy on the jets. The effect in the final region is negligible.
    • Tagging Scale Factor (SF): The difference between data and MC in c-tagging efficiency (~10%) is taken as systematic uncertainty. The resulting uncertainty in the final region is 3.6%.
    • CHAOS Scale Factor: The difference between data and MC is taken as systematic uncertainty. The resulting uncertainty in the final region is 9.2%.
    • Mistag Rate: The systematic error assigned to the tag rate matrix is 4.8%.
    • Luminosity: The systematic uncertainty in the luminosity is taken to be 6%, affecting to the normalization of all the MC estimated backgrounds.
    • ISR/FSR: The uncertainty associated with the initial and final state radiation was evaluated by generating sample with more/less ISR/FSR. The effect in the final region is 1.7%.
    • PDF: The PDF uncertainty has been determined to be 3.8% on the acceptance.
    • HF QCD Multijet Background: We assign a conservative 30% uncertainty in the prediction based on the variation observed when matrix definition is changed.
    • Top-Pair Production cross section: We quote the uncertainty in the CDF measured value (11%) of the top-pair production cross section.
    • Single Top Production cross section: We quote the theoretical uncertainty in the single-top cross section (13%)
    • Diboson Production cross section: We quote the theoretical uncertainty being 10% in the WW and WZ cross sections and 20% for the ZZ process.
    • Single EWK Boson Production cross section: Although the cross section for Z and W production are known to a high precision, we are using the heavy flavour processes in Alpgen to perform estimations of Z/W+multijet processes. Because of this, we estimate a 40% uncertainty in the predictions.
    • Top quark mass: In the current analysis, the top-pair production background is estimated using MC with a top quark mass of 175 GeV/c2. Since our signal optimization is based on a Neural Network trained with top-pair processes we include a systematic error due to the top pair NN output dependence on the top quark mass. We compute this error measuring the number of top-pair events in the final selection by using a top quark mass of 172.5 GeV/c2.
    • Differences in shape between Alpgen and Pythia: We include a shape systematic uncertainty in the final selections due to the differences between Alpgen and Pythia generators used to estimate the Z/W + jets processes.
    • HF QCD Multijet and mistag estimation after CHAOS: We quote the uncertainty in the final region due to this estimations of 3.6% and 8.2% respectively.


    Limits :

    In the final region the number of observed events events are in good agreement with the expected events from the SM processes.

    Since no significant deviation from the SM prediction was observed, the results were used to extract an exclusion limit for the cross section of the described process. Using shapes from the final discriminant, we set a 95% C.L. limit. For the assumed model, this analysis is able to exclude stop masses up to 180 Gev/c2 at 95% C.L.