2024

Hirvonen, Henry

This thesis focuses on a phenomenological modeling of ultrarelativistic heavy-ion collisions. The primary objective is to investigate and constrain the properties of the quark-gluon plasma (QGP) by comparing fluid-dynamical simulation results with various flow observables measured at CERN-LHC and BNL-RHIC. To achieve this, the existing EKRT+fluid dynamics heavy-ion collision framework is further developed, and machine learning techniques are utilized to reduce the computational cost of complex simulations. These types of advancements are crucial for the improving understanding of the QGP properties. The introduction of the thesis provides a general description of the employed heavy-ion collision framework and discusses the novel features introduced in this thesis. The main contributions of this work can be categorized into three development areas: dynamical decoupling, neural networks, and the Monte-Carlo EKRT initial state model. Firstly, incorporating a dynamical decoupling into the fluid-dynamical framework improved the description of peripheral collision systems, resulting in a better agreement with the measured flow coefficients compared to constant temperature decoupling. Secondly, neural networks were trained to predict flow observables directly from the initial state, effectively replacing the computationally expensive hydrodynamic simulations and reducing the required computation time by several orders of magnitude. Finally, a new Monte-Carlo EKRT initial state model was introduced and successfully applied to the studies of rapidity distributions of charged particles and their flow coefficients, as well as midrapidity flow observables. Keywords: relativistic heavy-ion collisions, quark-gluon plasma, relativistic hydrodynamics, machine learning