Optimal Observables for the Chiral Magnetic Effect from Machine Learning
Yuji Hirono, Kazuki Ikeda, Dmitri E. Kharzeev, Ziyi Liu, Shuzhe Shi
TL;DR
We address the CME detection challenge by introducing a physics-informed ML framework that builds a parametric observable from flow harmonics and optimizes its coefficients to maximize CME sensitivity while suppressing backgrounds. The observable O combines linear and bilinear terms in P-even and P-odd harmonic components, with its coefficients optimized via gradient descent using AVFD simulations. Results show strong background suppression and CME sensitivity gains (up to 18.7 sigma for selected orders), outperforming conventional gamma and delta correlators. The approach yields interpretable observables and is extensible to additional experimental inputs, including isobar comparisons, offering a route to conclusive CME evidence.
Abstract
The detection of the Chiral Magnetic Effect (CME) in relativistic heavy-ion collisions remains challenging due to substantial background contributions that obscure the expected signal. In this Letter, we present a novel machine learning approach for constructing optimized observables that significantly enhance CME detection capabilities. By parameterizing generic observables constructed from flow harmonics and optimizing them to maximize the signal-to-background ratio, we systematically develop CME-sensitive measures that outperform conventional methods. Using simulated data from the Anomalous Viscous Fluid Dynamics framework, our machine learning observables demonstrate up to 90\% higher sensitivity to CME signals compared to traditional $γ$ and $δ$ correlators, while maintaining minimal background contamination. The constructed observables provide physical insight into optimal CME detection strategies, and offer a promising path forward for experimental searches of CME at RHIC and the LHC.