A proposed model to estimate flow coefficients from charged-particle densities using Deep Learning

In the first microseconds of the Big Bang, all known matter in the Universe was in a state of Quark-Gluon Plasma (QGP). The primary purpose of heavy ion collisions is to investigate QGP's formation and properties. For this purpose, measuring the so-called flow coefficients is an essential observable. In this thesis, we approach the challenge of estimating the flow coefficients $v$$_{n}$ for $n$ = 1, 2, ..., 5 from charged-particle densities $_{dηdϕ}^{d^2Nch}$ using Deep Learning (DL). For this reason, we construct a controlled experiment by simulating nuclei collisions with a Glauber model. Next, we use the calculations from the Glauber to produce charged particles from various particle production models. Lastly, we distribute the final state particles in the relevant coordinate system (η, ϕ). We propose a model based on a Convolutional Neural Network (CNN) capable of analyzing patterns found in charged-particle densities to predict flow coefficients, with confidence intervals quantified by systematic uncertainties. We found that the proposed model can estimate the flow coefficient event by event, with high accuracy and precision for central and semi-central collisions. The performance was evaluated on a test set of 3.000 simulated Pb-Pb events, and yielded a Pearson product-moment correlation coefficient (PPMCC) of 0.42 for directed $v$$_{1}$ , 0.90 for elliptical $v$$_{2}$ , 0.86 for both triangular $v$$_{3}$ and quintuple $v$$_{5}$ , and 0.87 for the quadruple $v$$_{4}$ flow coefficients. The proposed model lays the groundwork for a novel technical approach towards estimating flow coefficients with the possibility of assisting already established methods.