Machine Learning for Monte Carlo Simulation in the Fermilab Muon g-2 experiment

The Muon g −2 experiment at Fermilab aims to measure the muon magnetic anomaly 4 times more precise than the BNL E821 experiment. Geant4-based simulation package gm2ringsim has been developed to study the systematic errors arising from beam dynamics and detector acceptance. Gm2ringsim, while providing high-fidelity simulation needed for the experiment, is time-consuming and has limited the amount of dataset that can be produced for a systematic study. To provide an alternative solution, a "divide and conquer" approach is proposed where the typical Monte Carlo simulation is divided into beam and spin dynamics, muon decay, and positron detection. The last part which involves positron tracking and electromagnetic shower development in the calorimeter, is modeled using machine learning algorithms. In this presentation, I will present the status of this fast simulation together with a benchmark of the performance.