Jet characterization in Heavy Ion Collisions by QCD-Aware Graph Neural Networks

TL;DR

This paper introduces GraphRed, a physics-aware graph neural network that improves jet particle identification in heavy ion collisions by effectively filtering background noise and integrating with existing jet algorithms.

Contribution

The paper presents a novel graph neural network architecture, GraphRed, tailored for jet identification in heavy ion collisions, demonstrating superior efficiency over previous methods.

Findings

GraphRed effectively identifies jet-induced particles in heavy ion events.

The method shows robustness and compatibility with existing jet algorithms like FastJet.

GraphRed outperforms previous frameworks in efficiency and accuracy.

Abstract

The identification of jets and their constituents is one of the key problems and challenging task in heavy ion experiments such as experiments at RHIC and LHC. The presence of huge background of soft particles pose a curse for jet finding techniques. The inabilities or lack of efficient techniques to filter out the background lead to a fake or combinatorial jet formation which may have an errorneous interpretation. In this article, we present Graph Reduction technique (GraphRed), a novel class of physics-aware and topology-based attention graph neural network built upon jet physics in heavy ion collisions. This approach directly works with the physical observables of variable-length set of final state particles on an event-by-event basis to find most likely jet-induced particles in an event. This technique demonstrate the robustness and applicability of this method for finding jet-induced particles and show that graph architectures are more efficient than previous frameworks. This technique exhibit foremost time a classifier working on particle-level in each heavy ion event produced at the LHC. We present the applicability and integration of the model with current jet finding algorithms such as FastJet.