A Search For ZH→l⁺l^-bb
in 9.45/fb of CDF II Data

A brief outline of the search techniques is presented below. For more details please see the analysis documentation available here.

Online (trigger) Selection

Events are recorded by an online selection algorithm which requires the presence of 1 or 2 electron candidates, a muon candidate, or missing transverse energy (>30 GeV) and jets.

Lepton Identification

We select muon and electron candidates using neural network lepton identification algorithms.

Z Candidate Selection

We require a dilepton pair with a mass between 76 and 106 GeV/c². The dilepton mass for all candidate events is shown below.

Jet Selection

We require events to have 2 or 3 jets with a transverse energy of greater than 25 GeV and a pseudo-rapidity (η) of -2.0 < η < 2.0. The dijet mass of the two lead-E_T jets is required to be greater than 25 GeV/c². The dijet mass for all candidate events is shown below.

Jet Energy Corrections

We correct the energy of the two lead-E_T jets using an artificial neural network algorithm. The dijet mass for all candidate events after NN correction is shown below.

Data Model

Major backgrounds to the ZH search are Z+2 light flavor (lf) jets, Z+cc, Z+bb, tt, and diboson processes. We model the contributions from Z+lf,Z+cc, and Z+bb using an ALPGEN simulation. Contributions from WW,WZ,ZZ, and tt are modeled using a PYTHIA simulation. We use data-derived methods to model the contribution from misidentified Z→l⁺l^- decays.

PreTag Control Region

We validate our data model in the sample of reconstructed Z→l⁺l^- + 2 or 3 jet events before requiring the presence of one or more identified b-jets (PreTag). In general we find reasonable agreement between our model and the observed PreTag data. The distributions of various event quantities are shown here.

b-jet Identification

We use the new CDF HOBIT algorithm to identify (tag) b-jets. We define four exclusive analysis categories based on the b-quark content of an event : events with two 'Tight' Hobit b-tags (TT), events with one Tight and one 'Loose' Hobit b-tags (TL), events with a single Tight tag (Tx), and events with two Loose tags (LL).

3-stage Neural Network Signal Discrimination

We separate ZH from background processes by using a series of neural networks. The first neural network is trained to separate tt from ZH signal and can be seen here. The second neural network is trained to separate Z+jet and Z+charm backgrounds from ZH signal and can be seen here. The third neural network separates diboson backgrounds from ZH signal and can be seen here. Based on the values returned by the three expert networks, we assign candidates to one of 4 regions in our final discriminant. The 3-stage sorting logic is summarized in the flowchart below. In each of the 4 regions of the final discriminant, we plot the output of a neural network designed to simultaneously separate ZH from all backgrounds. We deploy 13 versions of this network, each optimized for a specific Higgs boson mass. The output of this network (as optimized for a 120 GeV Higgs boson) can be seen before and after sorting. The sorted discriminant is used in the extraction of upper limits on σ_ZH×BR(H→bb).

Systematics

The following table summarizes the systematic uncertainties affecting the shape and/or normalization of the components of our background model.

Event Totals

The predicted event totals are compared to the observed data for each of the 16 sub-channels in the following tables :

Z→e⁺e^- + 2 jets

Z→e⁺e^- + 3 jets

Z→μ ⁺μ ^- + 2 jets

Z→μ ⁺μ ^- + 3 jets

Results

We compute Bayesian 95% CL upper limits on the value of σ_ZH×BR(H→bb), and obtain expected upper limits from the median of the upper limits resulting from a series of background-only pseudo-experiments. The observed and expected limits are listed below and are shown in the figure here.

The dilepton mass for all candidate events:

The dijet mass for all candidate events:

The dijet mass for all candidate events after NN correction:

Output of the tt expert discriminant:

Output of the lf+charm expert discriminant:

Output of the diboson expert discriminant:

3-stage Neural Network Signal Discrimination:

120 GeV Optimized Network Output Before Sorting :

120 GeV Optimized Network Output After Sorting :

120 GeV Optimized Network Output for Total Model Under JES Variation :

95% CL Upper Limits :