Cleaning up a particular mess: Creating a novel Physics-Informed Graph Neural Network for high-multiplicity TPC track reconstruction.

The Tagged Deep Inelastic Scattering (TDIS) experiment at Jefferson Lab aims to probe the mesonic content of nucleons by tagging low-momentum recoil hadrons with a multiple Time Projection Chamber (mTPC) in coincidence with scattered electrons. Traditional clustering methods for mTPC track finding will be inadequate for the experiment due to high expected multiplicity. Therefore, we turn to Graph Neural Networks (GNNs) to model a series of non-linear decision boundaries that will segment particle hits into trajectories. Our GNN introduces novel layers that, when combined with known physical properties of the experiment, transform the generally non-differentiable solution to a clustering problem into a differentiable solution compatible with the backpropagation algorithm.