High energy physics experiments such as those operated at the Large Hadron Collider (LHC) fundamentally rely on detailed and realistic simulations of particle interactions with the detector. The state-of-the-art Geant4 toolkit provides a means of conducting these simulations with Monte Carlo procedures. However, the simulation of particle showers in the calorimeter systems of collider detectors with such tools is a computationally intensive task. For this reason, alternative fast simulation approaches based on generative models have received significant attention, with these models now being deployed in production by current experiments at the LHC. In order to develop the next generation of fast simulation tools, approaches are being explored that would be able to handle larger data dimensionalities stemming from the higher granularity present in future detectors, while also being efficient enough to provide a sizable simulation speed-up for low energy showers.
A shower representation which has the potential to meet these criteria is a point cloud, which can be constructed from the position, energy and time of hits in the calorimeter. Since Geant4 provides access to the (very numerous) individual physical interactions simulated in the calorimeter, it also provides a means to create a representation independent of the physical readout geometry of the detector. This project will explore different approaches to clustering these individual simulated hits into a point cloud, seeking to minimise the number of points while preserving key calorimetric observables.
(As we typically receive a large number of responses and we are not able to reply to all initial messages, please only contact us after completing the test)