Efficient emulation of relativistic heavy ion collisions with transfer learning

TL;DR

This paper introduces a transfer learning approach to efficiently emulate heavy ion collision simulations, significantly reducing computational costs in uncertainty quantification across various collision systems.

Contribution

It presents a novel transfer learning method that maps parameter dependencies between emulators, minimizing the need for extensive simulations in heavy ion collision studies.

Findings

Transfer learning reduces emulator training costs.

Method enables efficient uncertainty quantification.

Applicable to multiple collision systems.

Abstract

Measurements from the Large Hadron Collider (LHC) and the Relativistic Heavy Ion Collider (RHIC) can be used to study the properties of quark-gluon plasma. Systematic constraints on these properties must combine measurements from different collision systems and methodically account for experimental and theoretical uncertainties. Such studies require a vast number of costly numerical simulations. While computationally inexpensive surrogate models ("emulators") can be used to efficiently approximate the predictions of heavy ion simulations across a broad range of model parameters, training a reliable emulator remains a computationally expensive task. We use transfer learning to map the parameter dependencies of one model emulator onto another, leveraging similarities between different simulations of heavy ion collisions. By limiting the need for large numbers of simulations to only one of the emulators, this technique reduces the numerical cost of comprehensive uncertainty quantification when studying multiple collision systems and exploring different models.