Optimizing the throughput of the ATLAS Geant4 detector simulation

¹ Department of Physics, University of Warwick, Gibbet Hill Road, Coventry, CV4 7AL; United Kingdom.
² CERN, Geneva; Switzerland
³ Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh PA; United States of America.
⁴ Cavendish Laboratory, University of Cambridge, Cambridge; United Kingdom.
⁵ Physics Division, Lawrence Berkeley National Laboratory, Berkeley CA; United States of America.
⁶ High Energy Physics Division, Argonne National Laboratory, Argonne IL; United States of America.
⁷ University of California, Berkeley CA; United States of America.
⁸ Department of Physics, University of Washington, Seattle WA; United States of America.
⁹ INFN Sezione di Milano; Italy
¹⁰ Department of Particle Physics and Astrophysics, Weizmann Institute of Science, Rehovot; Israel.
¹¹ Physics Department, Ben-Gurion University of the Negev, Beer-Sheva; Israel.
¹² Fakultät für Mathematikund Naturwissenschaften, Fachgruppe Physik, Bergische Universität Wuppertal, Wuppertal; Germany.
¹³ Department of Physics, Nanjing University, Nanjing; China

^* e-mail: Ben.Morgan@warwick.ac.uk

Published online: 7 October 2025

Abstract

For Run 3 at the Large Hadron Collider the Monte Carlo Detector Simulation performed with Geant4 (Full Simulation) within the ATLAS experiment has undergone significant improvements to enhance its computational performance and overall efficiency. Here we provide an overview of the physics and technical optimizations implemented in the ATLAS Full Simulation for Run 3. Notable physics developments include use of Geant4’s range cuts and G4GammaGeneralProcess for gammas, and variance reduction techniques such as neutron/photon Russian Roulette, and Woodcock Tracking in the Electromagnetic Endcap Calorimeter (EMEC). Technical developments include optimizations of the geometric representation of the particularly complex EMEC detector to assist navigation, tailored switch-off of the magnetic field, and use of static linking and Link-Time Optimization within a single shared library. These enhancements collectively resulted in a major achievement, increasing the throughput of the simulation by a factor of 2 compared to the baseline simulation configuration used in Run 2. Further optimizations are under development, and we provide an overview of this continuing programme covering both near and longer term features.