Determination of the impact parameter distribution of low-intermediate energy heavy ion collisions using machine learning

Abstract

<p indent="0mm">The impact parameter is an important quantity for reasonably simulating heavy ion collisions (HIC) using the transport model, because many HIC observables obviously depend on it in the simulation. However, the impact parameter cannot be directly measured or determined from HIC observables, so reconstructing impact parameter distributions from HIC observables is highly desired. For intermediate energy HICs, impact parameters distribute in a wide range for the selected events due to the strong fluctuation mechanism. In this work, three methods for reconstructing impact parameter distributions are discussed, namely sharp cut-off approximation, Bayesian estimation, and neural network estimation of impact parameters. The dataset used for reconstructing the impact parameter distribution by machine learning methods are generated from an improved quantum molecular dynamics model (ImQMD). Finally, we propose to use an unsupervised machine learning method, namely the K-means algorithm, to sort the centrality of heavy ion collisions and its validity is also proved in theory.