Particle tracking is a challenging pattern recognition task in experimental particle physics. Traditionally, algorithms based on the Kalman filter are used for such tasks and show desirable performance in finding tracks originating from the interaction point. However, many Beyond Standard Model (BSM) theories predict the existence of long-lived particles (LLP). They have a longer lifetime and travel a distance before decaying to Standard Model particles, resulting in large radius tracks. For such displaced tracks, dedicated tunings are often required to reach sensible performance since the quality of seeds for the Kalman filter has a direct impact on its performance.
Recent studies show machine learning-based particle track finding algorithms using graph neural networks (GNN) achieve competitive physics and computing performance for tracks originating from the interaction point. In this work, we developed a GNN-based end-to-end particle track finding algorithm for the High Luminosity LHC and apply such an algorithm to displaced track datasets to study the performance of reconstructing displaced tracks. The algorithm is designed to be agnostic about global track position. The datasets are generated under the ACTS framework and simulated for a generic detector. As the result, we reconstruct prompt and displaced tracks simultaneously with high track efficiency and no significant drop for displaced tracks.
|In-person or Virtual?||Virtual|