Search for Rare Z Decays into Two Reconstructed Photons at CDF

Karen Bland¹, Ray Culbertson², Jay Dittmann¹, Craig Group^2,3, and Costas Vellidis²

¹Baylor University, ²Fermilab, ³University of Virginia

[Contact]

Public note

We report the most sensitive search to date for forbidden and exotic decays of the Z boson to a pair of photons, a pair of neutral mesons, or a neutral meson and a photon. The search is using the full CDF dataset corresponding to 10.0 fb^-1 of integrated luminosity. The two decay products are reconstructed in the central electromagnetic calorimeter, spanning a range of |η| < 1.1, and their momenta are used to reconstruct the invariant mass of the pair that is used as a proxy to identify Z bosons. The event selection is identical to the CDF H→γγ search, employing an ANN for the central photon identification to maximize the prompt photon purity of the selected sample. The background from continuum diphoton production is estimated by fitting the sidebands of the mass spectrum outside of the Z-mass window. The Z→ee (Drell-Yan) background is estimated from Pythia Monte Carlo (MC). The signal acceptance is determined with a Z→γγ MC sample derived from an original Pythia Z→νν sample with the neutrinos changed to photons for the detector and trigger simulation and the reconstruction; the angular distribution is reweighted from the intrinsic Pythia Z→νν distribution to the expected Z→γγ, Z→π⁰γ or Z→π⁰π⁰ distribution. In the case of decays into π⁰γ and π⁰π⁰, the acceptance is corrected for the relative efficiency of π⁰ vs. γ identification, which is derived from MC samples produced with the "particle gun" generator. Bayesian 95% C.L. limits on the signal branching ratio are derived by fitting the signal and total background expectations to the data in the Z-mass window of [80,102] GeV/c² using a binned likelihood based on the m_γγ distribution. The observed branching ratio limits presented here are 1.66×10^-5 for Z→γγ, 2.28×10^-5 for Z→π⁰γ, and 1.73×10^-5 for Z→π⁰π⁰. The Z→γγ and Z→π⁰γ limits are more sensitive by factors of 3.1 and 2.3, respectively, than the most stringent Particle Data Group limits reported by experiments at the LEP collider. The Z→π⁰π⁰ branching ratio limit is the first limit reported in this decay mode.

Fig. 1: Signal Simulation

.eps    .gif

Fig. 2: Modeling of Drell-Yan (DY) Background
The MC prediction is scaled to the expected number of events in the data sample.

.eps    .gif

Fig. 3: Fit to Drell-Yan-Subtracted Data (linear scale)
The Drell-Yan shape is first subtracted from the data sample. In order to model
the non-resonant diphoton background, a fit is then made to the sideband
control regions of the data. The histogram with a red curve shows this fit. The fit is
interpolated into the signal region to model the non-resonant background in this region
(not shown). The resulting (data - fit)/(statistical uncertainty) shape is provided
to show the comparison of the non-resonant background model to the data, for both
the signal region and the sideband control region.

.eps    .gif

Fig. 4: Fit to Drell-Yan-Subtracted Data (log scale)
The same as the previous figure, but now in a log scale.

.eps    .gif

Fig. 5: Comparison of Background Model to Data
A comparison of the background modeling to the data is provided for
(left) the sideband and signal regions of the diphoton mass (m_γγ) distribution and
(right) the signal region of the m_γγ distribution alone. The 11 bins of the plot in the
signal region (right) are among the inputs provided when constructing the binned
likelihood used to calculate branching ratio limits.

.eps    .gif

.eps    .gif

Table 1: Event Yields
Event yields for the m_γγ signal region (Fig. 5, right), for the background
and data yields.

.tex    .gif

Table 2: Systematic Uncertainties
Summary of the systematic uncertainties included in the analysis. Along with
bin-by-bin statistical uncertainties on the Drell-Yan background, these systematic uncertaintes
are considered when calculating limits on the signal branching ratios.

.tex    .gif

Table 3: 95% C.L. Upper Limits on the Z→γγ, Z→π⁰γ, and Z→π⁰π⁰ Branching Ratios
We calculate a Bayesian C.L. limit based on a Posson binned likelihood constructed from
each bin in the signal region (2 GeV/c² bin width) of the background, data, and
and signal mass distributions. We assume a positve flat prior in the signal
branching ratios and integrate over the Gaussian priors for the systematic
uncertainties. A 95% C.L. upper limit is determined such that 95% of the
posterior density for the branching ratio falls below the limit. The observed 95% C.L. on the
branching ratio is calculated from the data. For comparison, thousands of simulated experiments
are used to calculate a set of expected 95% C.L. limits that are obtained without data, based
on expected backgrounds only. The medain of these simulated pseudoexperiments is referred to as
the expected limit. The region where 68% (95%) of these pseudoexperiments lie about the median is
the (±1σ) ±2σ expected region.

.tex    .gif

Fig. 6: Diphoton Mass Shapes with Signal Scaled to the Observed Limit
The background and data are shown with an upper limit on the number of signal
events observed to exclude at 95% C.L. The signal limit is obtained
using the branching ratio limit values in the above table.

.eps    .gif

.eps    .gif