Validating a Machine Learning Approach to Identify Quenched Jets in Heavy-Ion Collisions

TL;DR

This work presents a supervised LSTM model that learns jet-quenching signatures from jet substructure and parton-shower history to predict jet-by-jet energy loss in heavy-ion collisions. Trained on Jewel photon–jet samples and tested with Delphes CMS detector realism, the method correlates with true quenching and remains robust to detector effects. Cross-checks using observables not included in training—photon–jet imbalance, fragmentation, and jet shapes—validate that the model captures genuine quenching features rather than background fluctuations. The approach offers jet-by-jet quenching discrimination applicable to experimental data, aiding the disentangling of competing energy-loss mechanisms in the QGP.

Abstract

Jet quenching is a phenomenon in heavy-ion collisions arising from jet interactions with the quark-gluon plasma (QGP). Its study is complicated by the interplay of multiple physics processes that affect jet observables. In addition, detector effects may influence the results and must be accounted for when identifying quenched jets. We employ a Long Short-Term Memory (LSTM) neural network trained on jet substructure, incorporating parton shower history, to predict jet-by-jet quenching levels. Using photon-jet samples from the \textsc{Jewel} event generator, we show that the LSTM predictions strongly correlate with true jet energy loss. This validates that the model effectively learns the features of jet-QGP interaction. We simulate detector effects using \textsc{Delphes} simulation framework and demonstrate that the method identifies quenching effects in a realistic environment. We test the approach with photon-jet momentum imbalance, jet fragmentation function, and jet shape, which were not included in the training, confirming its ability to distinguish true quenching features.

Figures (9)

• Figure 1: Scatter plot of average momentum density $\langle \rho \rangle$ and multiplicity for the underlying event simulated by Angantyr within Pythia 8.3.
• Figure 2: JES of reconstructed Jewel-Med (PbPb) jets from Delphes Energy Flow candidates. The upper panel shows JES before jet energy calibration, while the lower panel shows JES after calibration. (a–c) JES as a function of jet $p_{\mathrm{T}}$, $\eta$, and $\phi$, respectively. (d–f) Corresponding distributions after jet energy corrections.
• Figure 3: JES of reconstructed Jewel-Vac (pp) jets from Delphes Energy Flow candidates. The upper panel shows JES before jet energy calibration, while the lower panel shows JES after calibration. (a–c) JES as a function of jet $p_{\mathrm{T}}$, $\eta$, and $\phi$, respectively. (d–f) Corresponding distributions after jet energy corrections.
• Figure 4: JER of reconstructed medium/vacuum jets from Delphes Energy Flow candidates. (a): JER for medium jets; (b): JER for vacuum jets.
• Figure 5: Binary classification performance of the LSTM neural network at the GEN level (left) and RECO level (right). (a, b) Classification output; (c, d) ROC curve.