Strong and Electro-Weak Matter 2026

Neural network enhanced Bayesian global analysis of heavy-ion observables using EbyE-EKRT model and hydrodynamics with dynamical freeze-out

17 Aug 2026, 15:45

15m

F4050 (4th floor) (University of Helsinki Main Building)

talk (15min) Heavy-ion collisions and the quark gluon plasma Heavy Ion Contributions 2

Speakers

Harri Niemi (University of Jyväskylä) Henry Hirvonen (Vanderbilt University)Dr Jussi Auvinen (University of Wroclaw)Prof. Kari J. Eskola (University of Jyväskylä (FI))

Description

Bayesian global analysis of measured observables is nowadays a standard method for determining the properties and initial conditions of the hot QCD-matter produced in ultrarelativistic heavy-ion collisions. However, even with surrogate models like Gaussian process emulators (GPEs) reducing the amount of simulations, it can be computationally prohibitively expensive to produce sufficient training data for statistics-hungry rare observables. To overcome this, we introduce [1] a novel deep convolutional neural network (NN) -enhanced Bayesian global analysis of heavy-ion observables. We employ relativistic 2+1 D second-order viscous hydrodynamics with a dynamical freeze-out and initial conditions computed from the pQCD\&saturation -based event-by-event (EbyE) EKRT-model [2]. As constraints, we exploit data from $\sqrt{s_{NN}}=200$ GeV Au+Au collisions at RHIC, and from $2.76$ and $5.02$ TeV Pb+Pb and $5.44$ TeV Xe+Xe collisions at the LHC. We replace the slow hydrodynamical simulations by the fast NNs, which predict bulk observables directly from the initial energy density profiles EbyE [3], accounting for the QCD-matter properties [4]. With the NN output, we train the GPEs for obtaining the studied centrality-class averaged observables and their uncertainties. The NNs reduce the computing time by orders of magnitude. Our analysis results in a specific shear viscosity $\eta/s$ with a minimum-value plateau at temperatures $150\lesssim T \lesssim 230$ MeV with $0.12 \lesssim (\eta/s)_{\mathrm{min}} \lesssim 0.18$, and in a non-zero bulk viscous coefficient $\zeta/s$ at $200\lesssim T \lesssim 300$ MeV. At the freeze-out, the Knudsen number is $0.8-2.3$ and the ratio of the mean-free-path to the system size $0.3-1.2$, the data thus implying that the freeze-out indeed happens at the expected applicability limit of hydrodynamics.

[1] J. Auvinen, K. J. Eskola, H. Hirvonen and H. Niemi, arXiv:2603.26413 [hep-ph].
[2] H. Hirvonen, K. J. Eskola and H. Niemi, Phys. Rev. C 106, 044913 (2022).
[3] H. Hirvonen, K. J. Eskola and H. Niemi, Phys. Rev. C 108, 034905 (2023).
[4] H. Hirvonen, K. J. Eskola and H. Niemi, EPJ Web Conf. 296, 02002 (2024).