Symbolic Regression for Interpretable Emulation of Proton Collective Flow in Intermediate-Energy Heavy-Ion Collisions

Abstract

Symbolic regression provides an interpretable machine-learning approach for constructing explicit analytic relations between physical inputs and observables. In this work, we develop symbolic-regression emulators for the isospin-dependent Boltzmann-Uehling-Uhlenbeck (IBUU) transport model and compare their performance with deep neural network (DNN) emulators. Using the same transport-model data employed in our previous emulator studies, we show that symbolic regression can reproduce the proton mid-rapidity slope $F_1$ of transverse flow $v_1$ and elliptic flow $v_2$ with accuracy comparable to that of DNNs, while providing explicit analytic expressions and substantially faster prediction once trained. We further demonstrate the use of symbolic regression in the reverse direction by constructing analytic relations that predict the in-medium nucleon-nucleon cross-section modification factor $X$ from the flow observables. Although the symbolic-regression models require substantially longer training times and exhibit greater run-to-run variation than DNNs, their analytic form and rapid evaluation make them promising tools for future transport-model sensitivity and uncertainty analyses.