Generative Graph Neural Networks for Reconstructing Parton-Level Jet Showers after Hadronization

Recreating accurate parton-level event configurations from jets is a critical task for various physics analyses. However, hadronization processes cannot be computed using perturbative QCD. Hence, it is not tractable to analytically reconstruct parton-level events after hadronization.

We present a novel graph neural network approach to reconstruct jet showers at the parton-level from reconstructed particle jets. We utilize graph representations of particle and parton-level jets and state-of-the-art machine learning models i.e. a graph variational autoencoder (GVAE) and a deep autoregressive graph model (GraphRNN). Unlike traditional regression-based methods that focus on predicting individual particle properties, our method captures the entire parton-level event structure from jet data, offering a physically realistic reconstruction.

In this talk, we discuss jets originating from photon-tagged events to maximize partonic structure in a single reconstructed jet. However, we note that our method works for any jet multiplicity and process. We also discuss the performance of our method using the earth mover's distance metric, in addition to studying the impact of different sources of background such as underlying events, detector effects, and pileup events.