Speaker
Description
"In $H\to\gamma^*\gamma\to ee\gamma$ decays, low-mass electron pairs with high transverse momentum can produce overlapping electromagnetic showers in the calorimeter. A single cluster matched to two opposite-charge tracks forms a merged dielectron candidate, requiring dedicated identification to retain signal and reject photon and jet backgrounds. This poster presents merged dielectron identification and performance at ATLAS and its application to $H\to\gamma^*\gamma$.
In Run 2, merged dielectrons were identified using a cut-based method. In Run 3, a boosted decision tree (BDT) is trained to distinguish signal merged dielectrons from photon and jet backgrounds. It combines standard electron-identification shower-shape variables with track-to-calorimeter pointing variables. Training uses simulated $H\to\gamma^*\gamma$ signal samples, Jet Filter background samples and $H\to\gamma\gamma$ background samples. To improve data-MC agreement, simulated shower-shape variables are corrected using converted-photon fudge factors. Background $p_{\mathrm{T}}$ and $\eta$ distributions are reweighted to match the signal, preventing the classifier from exploiting kinematic differences.
At the same background efficiency as the Run 2 identification, the simulated signal efficiency increases from approximately 62% to 90%, allowing more signal candidates to be retained. To evaluate the efficiency in data, converted photons from $Z\to\ell\ell\gamma$ decays are used as a proxy for merged dielectrons. A simultaneous fit in signal and control regions accounts for residual background. Comparing the extracted efficiency with simulation provides data-to-MC scale factors and uncertainties.
The identification is applied to the ATLAS $H\to\ell\ell\gamma$ measurement with $m_{\ell\ell}<30\,\mathrm{GeV}$, using $164\,\mathrm{fb}^{-1}$ of $pp$ collision data at $\sqrt{s}=13.6\,\mathrm{TeV}$. At the same cross-section and luminosity, it improves the overall expected significance by 5% relative to the Run 2 identification."