# Identifying quenched jets in heavy ion collisions with machine learning

**Authors:** Lihan Liu, Julia Velkovska, Marta Verweij  
**arXiv:** [2206.01628](https://arxiv.org/abs/2206.01628)  
**Date:** 2022-06-03  
**Summary page:** https://www.summarizepaper.com/en/paper/2206.01628v2/

## Key points

- Measurements of jet substructure in ultra-relativistic heavy ion collisions show influence of interactions with quark gluon plasma
- Modern data-driven techniques used to investigate modifications in jet showering process
- Machine learning approach developed for identifying quenched jets using Jewel and Pythia 8 models
- Neural network trained on sequential substructure variables extracted from jet clustering history
- Application of background subtraction and jet grooming techniques tested with experimental data to enhance effectiveness
- Feature engineering procedure involves reclustering jets using Cambridge/Aachen algorithm and soft drop groomer for identifying hard splittings
- Innovative method presented for identifying quenched jets amidst background of soft particles in heavy ion collisions
- Successful application of machine learning techniques to understand modifications to jet substructure induced by interactions with quark gluon plasma

## Summary

Measurements of jet substructure in ultra-relativistic heavy ion collisions have revealed that the jet showering process is influenced by interactions with quark gluon plasma. To further investigate these modifications, modern data-driven techniques are being utilized. This study presents a machine learning approach for identifying quenched jets. The simulation involves using the Jet Energy Loss Invariant (Jewel) model and a non-quenching model known as Pythia 8. Sequential substructure variables are extracted from the jet clustering history following an angular-ordered sequence and used to train a neural network constructed on top of a long short-term memory network. One key aspect of this study is the application of background subtraction and jet grooming techniques that have been previously tested with experimental data, demonstrating their effectiveness. This machine learning approach shows promise in exploring various jet classification topics using real experimental data. Furthermore, the feature engineering procedure begins with reclustering jets using the Cambridge/Aachen algorithm, resulting in a binary tree structure representing the clustering history. The soft drop groomer plays a crucial role in removing softer branches until identifying hard splittings that meet specific conditions. Each splitting is described by a feature vector containing substructure variables such as momentum fraction and angular distance. Overall, this study presents an innovative method for identifying quenched jets amidst a large background of soft particles generated in heavy ion collisions. The results showcase the successful application of machine learning techniques to understand and characterize modifications to jet substructure induced by interactions with quark gluon plasma. Special thanks are extended to collaborators and funding sources for their support throughout this research endeavor.