Diphoton events are divided into four independent subsamples according the position and type of the photon candidate. In CC events (the most sensitive category), there are two photons in the central region of the detector. In CP events, one photon is in the central region and one is the plug region. For each of these categories, the two highest p_T photons in the sample are selected. If a CC or CP event is not identified, then two additional categories are considered. In C'C events, both photons are central but one has converted and is reconstructed from its e+e- decay products. Finally, in C'P events, one photon is in the plug region and the other is a central conversion photon.

For each channel, simulation of the diphoton mass spectrum for a Higgs boson with a mass of 115 GeV/c² decaying into two photons. The mass resolution of the diphoton channel is extremely good compared to dijet decay modes (Gaussian sigma / mean < 3%).

The invariant mass distribution is used as a discriminant for separating Higgs signal from the background. A fit is made to the diphoton invariant mass distribution of the data using the sideband region for each Higgs mass hypothesis and then interpolating to the signal region of that test mass. The fit in the signal region is used to establish the background expectation. Example fits below are made from a Higgs test mass window of 115 GeV/c² with the corresponding data vs. fit residual also shown.

Invariant mass distributions of the data compared to the background model showing Higgs signal scaled to the expected and observed limits obtained for each respective channel alone. Top plots are for CC, CP, C'C, and C'P respectively. Lower plots are the same distributions, but zoomed in on the signal mass hypothesis at 115 GeV.

Observed and expected limits are calclulated for each diphoton category alone, relative to the SM prediction.

95% C.L. upper limits are set on the production cross section times branching fraction relative to the standard model prediction using a Bayesian binned likelihood approach.