Automated Approach to Accurate, Precise, and Fast Detector Simulation and Reconstruction.

Area of Science:

• High-energy particle physics
• Computational physics
• Machine learning applications in science

Background:

• Detector simulation and reconstruction pose significant computational challenges in particle physics research.
• Current methods require substantial computational resources, limiting the speed of analysis and model development.

Purpose of the Study:

• To develop a novel deep learning approach, parnassus, to address the computational bottleneck in particle physics detector simulation and reconstruction.
• To create fast surrogate models that minimize resource utilization by combining simulation and reconstruction into a single step.

Main Methods:

• Developed a deep learning model, parnassus, that processes detector input as a point cloud and outputs reconstructed particles as a point cloud.
• Utilized a publicly available dataset of jets from the Compact Muon Solenoid (CMS) experiment for training and validation.

Main Results:

• The parnassus model accurately replicates the CMS particle flow algorithm's performance on trained events.
• Demonstrated the model's ability to generalize to jet momentum and types outside the initial training distribution.

Conclusions:

• Particle-flow neural-assisted simulations (parnassus) offer an efficient solution for detector simulation and reconstruction.
• The developed deep learning approach enables fast surrogate models applicable within and beyond large scientific collaborations.