Search for Standard Model Higgs Boson Production in Association with a W Boson using Neural Networks with 4.3fb-1 of CDF data

Yoshikazu Nagai, Shinhong Kim (University of Tsukuba)
Jay Dittmann, Martin Frank, Nils Krumnack (Baylor University)
Benjamin Kilminster (FNAL)
Anyes Taffard (UC Irvine)
Timo Aaltonen (University of Helsinki)
Weiming Yao (LBNL)
Adrian Buzatu, Andreas Warburton (McGill University)
Richard Hughes, Kevin Lannon, Jason Slaunwhite, Brian Winer, Homer Wolfe (Ohio State University)
Justin Keung, Evelyn Thomson (University of Pennsylvania)
Christopher Neu (University of Virginia)
 


- Abstract -
- Event Selection -
- NN b-jet energy correction -
- Bayesian Neural Network Inputs -
- Bayesian Neural Network Output -
- Systematics -
- Results -




Abstract:

We present a search for Standard Model Higgs boson decaying to two b-quarks and produced in association with a W boson. This search uses data corresponding to an integrated luminosity of 4.3/fb. We select events with a high-pT lepton, a neutrino, and two jets. We require at least one of the jets to be identified as a b-quark jet (tagged) using three different tagging algorithms (SECVTX, JetProb and NN b-tagging). The discrimination between the Higgs signal and the large backgrounds in the W + 2 jets bin is increased through the use of an bayesian neural network. We see no evidence for an excess of Higgs signal in the NN output distribution. We set a 95% confidence level upper limit on the WH cross section times the branching ratio of the Higgs to decay to a bbbar pair, expressed as a ratio to the SM cross section:
σ(ppbar -> WH)*BR(H->bbbar) < 5.3 x SM observed (4.0 expected) at M(H) = 115 GeV/c2
σ(ppbar -> WH)*BR(H->bbbar) < 2.8 to 38 x SM for M(H) = 100 to 150 GeV/c2





Event Selection:

We analyze 4.3/fb of events recorded triggers for high pT electrons, muons, large missing transverse energy plus 2 jets (MET + 2 Jets Trigger), and large missing transverse energy (MET trigger). We require events to have a high-pT lepton candidate, large missing transverse energy, and two jets with at least one b-tag with SECVTX. We classify our events into 3 lepton categories: central triggered electrons and muons, forward (plug) triggered electrons, and non-triggered leptons (isoalted tracks). Each lepton sample has distinct backgrounds and triggers requirements, and so has a different set of event selection cuts. The following tables summarize the cuts:
Event Selection for Lepton Triggered Events
CategoryDouble SECVTXOne SECVTX + One JetProbOne SECVTX + One NN b-tagOne SECVTX
Lepton Central isolated electron or muon and Plug isolated electron (Pt>20 GeV)
Missing Et > 20 GeV(Central), > 25 GeV(Plug)
Two Jets > 20 GeV, |&eta| < 2.0
b-tagging (one jet)tight SecVtx b-tag
b-tagging (another jet)tight SecVtx b-tagJetProb b-tagNN b-tagNo b-tag
QCD vetoPlug electron onlyQCD Veto


Event Selection for (MET + 2 Jets / MET) triggered Events
CategoryDouble SECVTX tagOne SECVTX + One JetProbOne SECVTX + One NN b-tagOne SECVTX tag
Lepton Isolated track (Pt > 20 GeV)
Missing Et > 20 GeV
Two Jets > 25 GeV (MET + 2 Jets), > 20 GeV (MET), |η| < 2.0
b-tagging (one jet)tight SecVtx b-tag
b-tagging (another jet)tight SecVtx b-tagJetProb b-tagNN b-tagNo b-tag
QCD Veto No QCD VetoQCD Veto

We estimate our expected background contribution to the sample in each lepton type and tag category. The following plots show our expected and observed number of background events as a function of jet multiplicity. Our search region is the two jet bin. The other jet bins are control regions.

2 SECVTX Tag Central Triggered Lepton Events 2 SECVTX Tag Plug Triggered Electron Events 2 SECVTX Tag Isolated Track Events
More Background Plots and Tables




NN b-jet energy correction

The most sensitive variable for WH analysis is the dijet invariant mass. Improvement on dijet mass resolution directly results in improvement of the final sensitivity. To further improve the dijet mass resolution, we develop a neural network based b-jet energy correction method.
Next plot is the example of the dijet invariant mass after NN b-jet correction.
Dijet invariant mass after NN b-jet energy correction
(2 SECVTX Tag, all lepton combined)




Bayesian Neural Network Inputs

We employ distinct Bayesian Neural Network discriminant functions which were optimized for one of three tagging categories: double SECVTX tag, one SECVTX tag + one JetProb/NN tag, and one SECVTX tag. We use BNN with 7 inputs, 8 hidden nodes, and one output node.
For each BNN function, we use next input variables.
    double SECVTX tag
  • Dijet invariant mass: The invariant mass reconstructed from the two jets. For this variable, we apply NN b-jet energy correction.
  • PT Imbalance: The scalar sum of the lepton and jet transverse momenta minus the MET.
  • M max (lep + ν + jet): The invariant mass of the lepton, MET and one of the two jets, where the jet is chosen to give the maximum invariant mass.
  • Q x &etalep: The charge of the lepton times the &eta of the lepton.
  • Sum ET (loose jets): The scalar sum of the loose jet transverse energy.
  • PT(W): The transverse momentum of the reconstructed W.
  • HT: The scalar sum of the transverse energies of the jets, the lepton, and the MET.
    one SECVTX tag + one JetProb/NN tag
  • Dijet invariant mass: Same variable as double SECVTX input.
  • Sum ET (loose jets): Same variable as double SECVTX input.
  • Q x &etalep: Same variable as double SECVTX input.
  • M min (lep + ν + jet): The invariant mass of the lepton, MET and one of the two jets, where the jet is chosen to give the minimum invariant mass.
  • HT: Same variable as double SECVTX input.
  • PT(W): Same variable as double SECVTX input.
  • MET: Missing transverse energy.
    one SECVTX tag
  • Dijet invariant mass: Same variable as double SECVTX input.
  • Sum ET (loose jets): Same variable as double SECVTX input.
  • Q x &etalep: Same variable as double SECVTX input.
  • PT(W): Same variable as double SECVTX input.
  • HT: Same variable as double SECVTX input.
  • MET: Same variable as one SECVTX tag + one JetProb/NN tag input.
  • PT Imbalance: Same variable as double SECVTX input.
The following plots show the NN Inputs for pretag (control region) central leptons.





More NN Input Plots





Bayesian Neural Network Output

The bayesian neural network output is a value between 0 and 1. Values close to 1 correspond to "more signal-like", values close to zero correspond to "less signal-like". We train a separate neural network for each Higgs signal mass. The following plots show the BNN output in central lepton channels. Left plot shows all double-tag/lepton categories combined distribution, and right plot shows single-tag distribution.

All Leptons, All double-tags (Linear scale) All Leptons, One SECVTX Tags (Linear scale)
All Leptons, All double-tags (Log scale) All Leptons, One SECVTX Tags (Log scale)
More NN Output Plots





Systematics:

We address systematic uncertainty on the signal acceptance from several different sources:
  • Lepton reconstruction
  • Trigger
  • Initial and final state radiation, and Parton distribution functions
  • Jet energy scale
  • b-tagging scale factor
  • Luminosity
The following link shows the systematic uncertainty for each lepton/b-tag categories due to each effect on the signal acceptance.

Systematics




Results:

We perform a binned likelihood fit of the neural network distribution where we constrain the backgrounds to their estimated rates within uncertainties. For optimal sensitivity, we perform a separate simultaneous search in each tag channel and lepton category. We set a limit on WH production cross section times branching ratio as a function of Higgs mass. We express our limits as a ratio to the Standard Model Cross section.


CDF Preliminary 4.3 fb-1
Limits for Combined Tag and Lepton Channels
Rates relative to Standard Model Expectation
Mass Observe Expect
100 4.0 2.8
105 4.5 3.1
110 5.0 3.5
115 5.3 4.0
120 4.9 4.6
125 7.0 6.0
130 7.5 7.4
135 11.8 10.1
140 15.7 14.1
145 25.0 21.8
150 37.6 33.7


Last modified: Mon Aug 31 14:39:35 CDT 2009