Determination of impact parameter for CEE with digi-input neural networks

TL;DR

This paper explores neural network models that directly analyze digitized signals from detectors to accurately determine the impact parameter in nucleus-nucleus collisions, especially at low energies where traditional methods falter.

Contribution

It introduces neural network approaches that utilize raw detector signals for impact parameter estimation, demonstrating improved accuracy over traditional methods at low beam energies.

Findings

Mean absolute error of 0.479 fm in impact parameter prediction

Neural networks effectively handle raw digitized signals

Models outperform traditional phase space input methods

Abstract

The impact parameter characterizes the centrality in nucleus-nucleus collision geometry. The determination of impact parameters in real experiments is usually based on the reconstructed particle attributes or the derived event-level observables. For the scheduled Cooler-storage-ring External-target Experiment (CEE), the low beam energy reduces correlation between the impact parameter and charged particle multiplicity, which decreases the validity of the explicit determination methods. This work investigates a few neural network-based models that directly take the digitized signals from the external Time-of-flight detectors as input. The model with the best performance shows a mean absolute error of 0.479 fm with simulated U-U collisions at 0.5 AGeV. The performances of the models implemented with digi inputs are compared with reference models with phase space inputs, showing the capability of neural networks to handle the original but potentially interrelated digitized signal information.