Speaker
Description
The Tile Calorimeter (TileCal) is the central hadronic calorimeter of the ATLAS detector at the LHC. As a sampling calorimeter composed of alternating steel absorbers and scintillating tiles. It measures hadronic shower energy through light signals that are digitized and reconstructed into physics objects such as jets and missing transverse momentum. A precise understanding of the TileCal response to hadrons is essential for accurate energy reconstruction and for controlling systematic uncertainties in physics analyses.
Dedicated test-beam campaigns are done in a controlled environment in which TileCal modules are exposed to different beams with known energy, impact position and incident angle. The dataset recorded during these campaigns is used for detailed studies of hadronic response under reproducible conditions, free from the complexity of full collision events.
This contribution presents a study of isolated hadrons in TileCal test-beam data recorded at the beam energies above 30 GeV and proposes a machine-learning approach for response modeling.
A physics-driven reference model is first constructed by characterizing the calorimeter response as a function of beam energy, detector region, and impact position, including linearity, resolution, and the presence of non-Gaussian tails. Based on this reference, a gradient-boosted decision-tree model is trained using directly measurable observables that describe the spatial and longitudinal energy deposition across calorimeter layers, shower shape variables, cluster topology, impact coordinates, and timing information when available. The model is used to predict and parameterize the calorimeter response on an event-by-event basis, with the goal of improving response modeling and supporting future calibration and performance studies.