Probing Collectivity in String Models via Machine Learning

Abstract

This work is devoted to studying the potential of machine learning techniques in relativistic nuclear physics for distinguishing between various physical theories and, consequently, gaining a deeper comprehension of the underlying physical processes in ultra-relativistic nuclear collisions. Recent findings on the modeling of p + p and A + A interactions within the framework of the color string fusion model suggest that it is feasible to describe the experimentally observed event-by-event azimuthal asymmetry in a unified manner across various colliding systems. Such a description has become possible by considering two mechanisms of string interaction: (1) changes in the magnitude of the colour field in the region of string overlap in the transverse collision plane (2) Lorentz boosts applied to particles emerging as a result of string motion due to their mutual attraction. We demonstrate that it is feasible to train machine learning algorithms using \({{p}_{{\text{T}}}}\)–\(\phi \) distributions from event-by-event data to distinguish between the proposed sources of collective behaviour.