A deep learning approach for predicting multiple observables in Au+Au collisions at RHIC

TL;DR

Aimed at predicting multiple observables in Au+Au collisions at RHIC, the paper develops a physics-informed neural network trained solely on experimental data for $dN_{ch}/d\eta$, $v_2(p_T)$, and $dN_{ch}/(2\pi p_T dp_T d\eta)$ across energies and centralities. The method embeds the space–time stages of a heavy-ion collision into a six-layer architecture with locally connected layers and a dual-input design, enabling a data-driven surrogate that respects underlying physics. The model achieves low training and test losses and generalizes to energies not in the training set, with validations against CLVisc hydrodynamics using $TRENTo$ initial conditions and against global multiplicity systematics. This approach provides a practical tool to fill RHIC data gaps and to accelerate phenomenological studies of QGP properties, with future plans to extend observables, systems, and uncertainty quantification for global analyses.

Abstract

We develop a neural network model, based on the processes of high-energy heavy-ion collisions, to study and predict several experimental observables in Au+Au collisions. We present a data-driven deep learning framework for predicting multiple bulk observables in Au+Au collisions at RHIC energies. A single neural network is trained exclusively on experimental measurements of charged-particle pseudorapidity density distributions, transverse-momentum spectra and elliptic flow coefficients over a broad range of collision energies and centralities. The network architecture is inspired by the stages of a heavy-ion collision, from the quark-gluon plasma to chemical and kinetic freeze-out, and employs locally connected hidden layers and a structured input design that encodes basic geometric and kinematic features of the system. We demonstrate that these physics-motivated choices significantly improve test performance compared to purely fully connected baselines. The trained model is then used to predict the above observables at collision energies not yet explored experimentally at RHIC, and the results are validated using the energy dependence of the total charged-particle multiplicity per participant pair as well as comparisons to a CLVisc hydrodynamic calculation with TRENTo initial conditions. Our findings indicate that such physics-guided neural networks can serve as efficient surrogates to fill critical data gaps at RHIC and to support further phenomenological studies of QGP properties.

Figures (14)

• Figure 1: The scheme of the neural network model.
• Figure 2: The training and test loss of the neural network model as a function of the training epochs.
• Figure 3: The charged particle pseudorapidity density distributions produced in Au + Au collisions at $\sqrt{s_{NN}}=19.6, 62.4$ and $200$ GeV for different centralities. The symbols are experimental data taken from Refs. Back2003Back2006. The curves are the training results of the neural network.
• Figure 4: The charged particle pseudorapidity density distributions produced in Au + Au collisions at $\sqrt{s_{NN}}=130$ GeV for different centralities. The symbols are experimental data taken from Ref. Back2003. The curves are the test results of the neural network.
• Figure 5: The charged particle pseudorapidity density distributions produced in Au + Au collisions at $\sqrt{s_{NN}}=17.3, 39$ and $54.4$ GeV for different centralities. The curves are the predictions of the neural network. The dash-dotted lines are the results of CLVisc. The bands are the error bars.