Abstract:In collider experiments, the kinematic reconstruction of heavy, short-lived particles is vital for precision tests of the Standard Model and in searches for physics beyond it. Performing kinematic reconstruction in collider events with many final-state jets, such as the all-hadronic decay of top-antitop quark pairs, is challenging. We present HyPER: Hypergraph for Particle Event Reconstruction, a novel architecture based on graph neural networks that uses hypergraph representation learning to build more powerful and efficient representations of collider events. HyPER is used to reconstruct parent particles from sets of final-state objects. Trained and tested on simulation, the HyPER model is shown to perform favorably when compared to existing state-of-the-art reconstruction techniques, while demonstrating superior parameter efficiency. The novel hypergraph approach allows the method to be applied to particle reconstruction in a multitude of different physics processes.

What problem does this paper attempt to address?

The paper attempts to address the problem of efficiently and accurately reconstructing the kinematic information of short-lived heavy particles (such as top quarks) in high-energy physics collider experiments. Specifically, the paper proposes a new method based on hypergraph representation learning—HyPER (Hypergraph for Particle Event Reconstruction)—to tackle the challenges of particle reconstruction in multi-final-state jet events. ### Background and Motivation In high-energy physics experiments such as the Large Hadron Collider (LHC), the reconstruction of short-lived heavy particles is crucial for precise testing of the Standard Model and searching for new physics phenomena. These particles typically decay rapidly into relatively stable final-state particles, so their kinematic information can only be inferred by analyzing the momentum and energy of the final-state particles. However, in complex events involving multiple final-state jets, this reconstruction task becomes very challenging, especially in fully hadronic decay channels, such as the fully hadronic decay of top quark pairs (t¯t). ### Main Challenges 1. **High Multiplicity Final States**: Many heavy particle decay processes produce a large number of final-state jets, leading to a combinatorial explosion problem. 2. **Jet Assignment Complexity**: Final-state jets not only come from the decay of short-lived particles but may also originate from parton splitting or concurrent proton-proton interactions, increasing the difficulty of jet assignment. 3. **Limitations of Traditional Methods**: Existing methods based on χ² minimization have limited performance in handling complex events. ### Solution The paper proposes a new machine learning model—HyPER, which leverages hypergraph representation learning to construct a more robust and efficient representation of collider events. Specifically: 1. **Hypergraph Structure**: HyPER represents events as a hypergraph, where each node represents a final-state object (such as a jet), and hyperedges represent relationships among multiple nodes. This structure can more effectively capture the high-order correlations among multiple final-state jets. 2. **Message Passing Mechanism**: Feature embedding is performed on the graph structure using a Graph Neural Network (GNN), and then the hypergraph structure is used to further extract and classify hyperedges, thereby reconstructing short-lived particles. 3. **Efficient Parameter Utilization**: HyPER maintains high performance while having a model size much smaller than existing advanced methods such as SPANet and Topograph. ### Experimental Validation The paper validates the performance of HyPER on the task of reconstructing top quark pairs in the fully hadronic decay channel and compares it with existing advanced methods. The results show that HyPER exhibits higher reconstruction efficiency under various jet multiplicity conditions, with a particularly significant advantage in high multiplicity regions. ### Conclusion HyPER successfully addresses the challenge of reconstructing short-lived heavy particles in complex collider events, demonstrating the potential of hypergraph representation learning in high-energy physics. This method not only improves reconstruction performance but also significantly reduces the number of model parameters, providing a new tool for future experimental data analysis.

Reconstructing short-lived particles using hypergraph representation learning

HGPflow: Extending Hypergraph Particle Flow to Collider Event Reconstruction

Improved particle-flow event reconstruction with scalable neural networks for current and future particle detectors

Reconstructing particles in jets using set transformer and hypergraph prediction networks

Graph Neural Networks for Particle Reconstruction in High Energy Physics detectors

Distance-Weighted Graph Neural Networks on FPGAs for Real-Time Particle Reconstruction in High Energy Physics

Calorimetry with Deep Learning: Particle Simulation and Reconstruction for Collider Physics

Time-based Reconstruction of Hyperons at PANDA at FAIR

Reconstruction of Unstable Heavy Particles Using Deep Symmetry-Preserving Attention Networks

Graph Generative Models for Fast Detector Simulations in High Energy Physics

HyperTrack: Neural Combinatorics for High Energy Physics

Topological Reconstruction of Particle Physics Processes using Graph Neural Networks

Reconstruction of electromagnetic showers in calorimeters using Deep Learning

Using graph neural networks to reconstruct charged pion showers in the CMS High Granularity Calorimeter

NuGraph2: A Graph Neural Network for Neutrino Physics Event Reconstruction

Boosted Higgs Boson Jet Reconstruction Via a Graph Neural Network

PASCL: Supervised Contrastive Learning with Perturbative Augmentation for Particle Decay Reconstruction

Reconstruction of Particle Flow Energy Distribution Using Deep Learning Algorithms