Full event interpretation with MLPF reconstruction in the CMS detector

About

The particle-flow (PF) algorithm reconstructs a global description of each collision by producing a comprehensive list of final-state particles. It is central to event reconstruction in the CMS experiment at the CERN LHC, and has been a focus of developments in light of planned high-luminosity running conditions with increased pileup and detector granularity. The existing PF implementation relies on several physics-motivated heuristics and assumptions that can be replaced by machine learning (ML) models trained directly on simulated data. A state-of-the-art, ML-based PF (MLPF) reconstruction algorithm implemented within the CMS software framework is presented. The MLPF algorithm performs a learnable, differentiable full-event reconstruction on GPUs, generalizes across detector conditions and collision energies, and replaces multiple traditional reconstruction steps with a single unified model. Physics performance comparable to standard PF reconstruction is achieved in both simulation and data, with improved jet energy resolution and inference time. In simulated top quark-antiquark events under LHC Run 3 (2023–2024) conditions, the jet energy resolution is improved by 10–20% for jets with transverse momentum between 30 and 100 GeV. Runtime performance is evaluated using simulated QCD multijet events with pileup corresponding to 55–75 interactions. For these events, a median runtime of 20 ms per event is achieved on an Nvidia L4 GPU, with better scaling with event size than the standard CMS particle-flow reconstruction, which processes the same events in approximately 110 ms per event.

0:00:00 Slide 1

0:00:07 Slide 2

0:00:44 Slide 3

0:01:12 Slide 4

0:02:01 Slide 5

0:03:35 Slide 6

0:04:14 Slide 7

0:05:29 Slide 8

0:06:36 Slide 9

0:07:15 Slide 10

0:07:50 Slide 11