Simulation of many-jet processes at the (HL-)LHC using GPU-accelerated matrix element calculations

• Applicant: Dr. Enrico Bothmann
• Subject Area: Nuclear and Elementary Particle Physics, Quantum Mechanics, Relativity, Fields
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 510810461

Monte-Carlo event generators are a backbone of the LHC physics programme. They translate quantum field theory models into fully differential simulated event samples, which can be directly compared to experimental collision data. In physics analyses, they are used to predict both Standard Model (SM) background and New Physics signal rates.However, projections by the ATLAS and CMS particle detector collaborations show that given the increasing collision rates and complexity of the analysis use cases, event generators must be accelerated by factors between two and ten to fit the computing budgets over the next decade. Moreover, they are not yet ready to deliver precise predictions for final states with six or more jets. Having faithful simulations of many-jet states would enable experimental New Physics searches to take inter-jet correlations into account for event selection to improve their sensitivity significantly.We plan to develop new algorithms for tree-level matrix-element calculations to address both challenges, and integrate them into the existing HEP toolchain to make their advantages in terms of speed and multiplicity range available for experimental and phenomenology applications. Our earlier results, for a GPU-accelerated calculator of gluon amplitudes, provided realistic speed-ups of up to an order of magnitude in a chip-to-chip comparison. In this project, this should now be extended into a full SM event generator. This will then be integrated with a many-jet-ready merging algorithm, allowing merged simulations in a multiplicity range that is currently not accessible. Finally, the new algorithms will be made accessible within the widely used Sherpa event generator framework, to make the improvements directly available for all tree-level calculations within the framework. At each step of the programme, we will deliver computational and phenomenological studies. The latter will include a study of inter-jet correlations for many-jet final states.To achieve these goals, we will develop the new algorithms based on Berends-Giele recursion. For treating arbitrary QCD amplitudes, we plan to employ for the first time the minimal Melia colour decomposition, which is expected to give significant speed-ups due to its optimal scaling behaviour. Also for the first time, we plan to implement a recursive GPU-accelerated phase-space sampling. For the multi-jet merging, our first choice will be to adapt the recent developments of a sector shower merging algorithm, which is especially apt for large multiplicities.For the future, it would be natural to focus on one-loop calculations next, thus aiming to make full next-to-leading order calculations efficient, and extending the multiplicity range for which they can be used. Another future avenue is to use the GPU-accelerated event generator to fuel the next generation of deep learning applications to collision events.

• DFG Programme: Research Grants
• International Connection: USA
• Cooperation Partner: Dr. Stefan Höche