Thursday 10 April 2025
The quest for accurate energy reconstruction has long been a challenge in particle physics, particularly when dealing with complex shower topologies in calorimeters. Now, researchers have turned to machine learning to develop a novel approach that significantly improves upon traditional methods.
The Dark Matter Particle Explorer (DAMPE) mission is designed to detect and study high-energy cosmic rays, including electrons and positrons. However, the complex topology of non-fiducial events – those where particles interact with the detector outside its central region – makes energy reconstruction particularly tricky. Classical methods struggle to accurately recover the initial energy, leading to poor resolutions and biased results.
To address this issue, researchers have developed a Convolutional Neural Network (CNN) that learns to predict the relative error between reconstructed and true energies from images of the BGO calorimeter. The training process involves splitting Monte Carlo simulations into training and validation sets, with input images normalised by the maximum deposited energy in the calorimeter.
The results are impressive: the CNN method achieves a mean value close to 0 across all energies, indicating an accurate reconstruction of initial energies. Moreover, the 68% containment values show significant improvement over classical methods, demonstrating tighter bounds on energy resolution.
One notable advantage of this approach is its ability to eliminate the need for unfolding, a step often required in flux analysis but prone to edge effects and biases. By directly estimating the true energy from calorimeter images, the CNN method avoids these issues altogether.
To further validate these results, researchers applied the CNN classifier to real data, selecting clean samples of electrons and dividing them into five energy ranges. Comparing these distributions with corresponding Monte Carlo simulations shows a strong agreement, with the ratio between real and simulated data indicating a good qualitative match.
This study marks an important step towards more accurate energy reconstruction in particle physics, particularly for non-fiducial events. By leveraging machine learning techniques, researchers can develop more sophisticated methods that better capture the complexities of high-energy interactions. As the field continues to push the boundaries of what is possible, it will be exciting to see how this technology evolves and its applications expand beyond DAMPE.
Cite this article: “Revolutionizing Cosmic Ray Energy Reconstruction with Deep Learning”, The Science Archive, 2025.
Machine Learning, Particle Physics, Energy Reconstruction, Calorimeters, Dark Matter Particle Explorer, Convolutional Neural Network, Monte Carlo Simulations, Energy Resolution, Unfolding, Flux Analysis
