Speaker
Description
We present a machine learning–based framework for modeling temperature-dependent non-perturbative quantities in the quark-gluon plasma (QGP), aimed at improving predictions for heavy quarkonia suppression in high-energy nuclear collisions. Deep neural networks are trained on lattice data to extract temperature profiles of the Debye screening mass $m_D(T)$ and the QCD running coupling $\alpha_s(T)$. These learned profiles are incorporated into a potential model to compute quarkonium thermal widths and binding energies by numerically solving the Schrödinger equation with a complex, medium-modified heavy-quark potential.
To determine dissociation temperatures $T_d $, we employ two complementary criteria: the upper bound criterion $ 2E_B = \Gamma_{\text{th}} $, [1], and the lower bound criterion $ E_B = 3T $, [2]. This unified ML-based approach enables a data-driven estimation of quarkonia dissociation across a broad temperature range, providing improved consistency with lattice QCD results and experimental suppression patterns observed in relativistic heavy-ion collisions. The framework offers a robust extension beyond perturbative techniques and can be adapted to model in-medium evolution in both isotropic and anisotropic backgrounds.
References:
1: A. Mocsy and P. Petreczky, Phys. Rev. Lett. 99, 211602 (2007)
2: S. Digal, P.Petreczky and H.~Satz, Phys. Lett. B 514, 57-62 (2001)
