Search for hadronic decays of $W$ and $Z$ bosons in photon events in $p\overline{p}$ collisions at $\sqrt{s}=1.96\mathrm{TeV}$

We report on a search for the process $p\overline{p}\to \gamma +W/Z$ with $W/Z\to q\overline{q}$ in events containing two jets and a photon at the center-of-mass energy $\sqrt{s}=1.96\mathrm{TeV}$, using $184{\mathrm{pb}}^{-1}$ of data collected by the CDF II detector. A neural network event selection has been developed to optimize the rejection of the large QCD production background; it is shown that this method gives a significant improvement in both signal-to-noise ratio and signal sensitivity, as compared with an event selection based on conventional cuts. An upper limit is presented for the $\gamma +W/Z$ production cross section with the $W$ and $Z$ decaying hadronically.

Figure 1

(a,b,e) Feynman diagrams at the tree level for the process $q\overline{q}\to W(Z)\gamma \to q\overline{q}\gamma$. The s channel for the $Z\gamma$ production is forbidden in the standard model. (c,d) Radiative $W(Z)$ decay diagrams. A bremsstrahlung photon is emitted by one of the two quarks from the $W(Z)$ decay.

Figure 2

Prompt photon fraction (number of prompt photon/number of candidate photons) in the data sample as a function of the photon candidate $p_T$. Only statistical errors are reported.

Figure 3

Trigger efficiency for photon candidates as a function of $p_T$.

Figure 4

Normalized dijet invariant mass (solid line) distribution of the two leading jets of the selected $\gamma (W/Z)\to \gamma q\overline{q}$ MC events. The individual contributions from the $W$ and $Z$ bosons (dashed lines) are shown. The fit results are reported in the inset.

Figure 5

Comparison of a few distributions for signal (dashed lines) and background (solid lines) events. All distributions are normalized to 1. The observables reported are (from left to right, top to bottom): (1) $\Delta {\eta }_{jj}$, (2) $n_{trk}^{j_1+j_2}$, (3) $M_{j_2}/E_{j_2}$, (4) ${\eta }_{\max }^{\mathrm{jets}}$, (5) $\Omega$, (6) $d{E}_T^{j\gamma }$. A precise definition of these observables is provided in Table 3.

Figure 6

Best (full square) and worst (empty square) ANN input variable combination in terms of significance as a function of the number of input nodes for a signal efficiency of 75%. The fluctuations in the curves are due to small changes in the ANN internal parameters one has to introduce when the number of inputs increases.

Figure 7

The $N_{\mathrm{OUT}}$ normalized distributions for the signal (dashed line) and for the background (solid line) events.

Figure 8

Efficiency for the signal and the background as a function of the ANN output threshold $N_{\mathrm{CUT}}$.

Figure 9

Dijet mass distribution of the data after the ANN selection cut. An exponential function $e^{P_0+P_1·m_{jj}}$ is used to fit the two sidebands (solid line) and the result is interpolated inside the signal region (dashed line). The values $M_{\min }=52\mathrm{GeV}$ and $[M_{\mathrm{sig}}^L,M_{\mathrm{sig}}^H]=[68,116]\mathrm{GeV}$ are used to search for the $W/Z$ mass peak.

Figure 10

Excess of events in the data with respect to the background prediction deduced from the sideband fits (bin errors do not include the background prediction uncertainties). The turn-on effect can be noticed in the first two bins (they are not included in the background fit). No evidence of any excess from the $W/Z$ resonance production is found inside the signal region.