Neural Unfolding of the Chiral Magnetic Effect in Heavy-Ion Collisions

Shuang Guo (郭爽), Lingxiao Wang (王凌霄), Kai Zhou (周凯), Guo-Liang Ma (马国亮) · 2025

Read Neural Unfolding of the Chiral Magnetic Effect in Heavy-Ion Collisions by Shuang Guo (郭爽), Lingxiao Wang (王凌霄), Kai Zhou (周凯), Guo-Liang Ma (马国亮) (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05808.txt.

Introduction

Shuang Guo, Lingxiao Wang, Kai Zhou, and Guo-Liang Ma hunt a subtle effect in atom smashes with deep learning in a 2025 paper. The effect, called the chiral magnetic effect, would split electric charges in the hot soup made when heavy ions collide. Heavy background noise has kept proof out of reach. The team trains a U-Net model to run the collision backward in time.

The model learns from simulated events with varied amounts of the effect mixed in. Starting from the final spread of particle momenta, it rebuilds the charge split across the whole event history. The rebuild works at two freeze-out stages of the soup. The authors offer the tool as a fresh route to test for the effect.

Catalog record

CREATORShuang Guo (郭爽), Lingxiao Wang (王凌霄), Kai Zhou (周凯), Guo-Liang Ma (马国亮)
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-05808

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Neural Unfolding of the Chiral Magnetic Effect in Heavy-Ion Collisions Preprint: RIKEN-iTHEMS-Report-25 Shuang Guo (郭爽), Lingxiao Wang (王凌霄), Kai Zhou (周凯), Guo-Liang Ma (马国亮) Abstract The search for the chiral magnetic effect (CME) in relativistic heavy-ion collisions (HICs) is challenged by significant background contamination. We present a novel deep learning approach based on a U-Net architecture to time-reversely unfold CME dynamics, enabling the reconstruction of the CME signal across the entire evolution of HICs. Trained on the events simulated by a multi-phase transport model with different cases of CME settings, our model learns to recover the CME-induced charge separation based on final-state transverse momentum distributions at either the quark-gloun plasma freeze-out or hadronic freeze-out. This underscores the promise of deep learning approaches in forthcoming experimental searches for the CME and related other physics in HICs. I INTRODUCTION Topological fluctuations of gluon fields in the quantum-chromodynamic (QCD) vacuum can induce a chirality imbalance, which implies local parity violation in strong interactions Lee and Wick 1974 ; Kharzeev et al. 1998 ; Kharzeev and Pisarski 2000 . This imbalance, quantified by a chiral chemical potential ( \mu_{5} ), generates an electric current \mathbf{J} parallel to an external magnetic field \mathbf{B} , resulting in a dipole-like charge separation (CS) along \mathbf{B} —a phenomenon known as the chiral magnetic effect (CME) Kharzeev et al. 2016 ; Fukushima et al. 2008 ; Kharzeev 2006 ; Kharzeev and Zhitnitsky 2007 ; Hattori and Huang 2017 . Relativistic heavy-ion collisions provide a unique experimental platform for studying the CME, as they simultaneously create the high-temperature QCD environment required for chirality imbalance and generate extremely strong magnetic fields in non-central collisions Skokov et al. 2009 ; Bzdak and Skokov 2012 ; Deng and Huang 2012 ; Chen et al. 2021 ; Xu et al. 2015 ; Tang et al. 2013 ; Zhao et al. 2018 ; Zhao et al. 2024 ; Gao et al. 2020 ; Wu et al. 2021 ; Shou et al. 2024 . Consequently, extensive experimental searches for CME-induced charge separation have been carried out at both the Relativistic Heavy Ion Collider (RHIC) at Brookhaven National Laboratory (BNL) Abelev et al. 2009 ; Abelev et al. 2010 ; Adamczyk et al. 2014a and the Large Hadron Collider (LHC) at CERN Abelev et al. 2013 ; Acharya et al. 2018 . However, because the magnetic field exists only transiently during the early stages of the collision Skokov et al. 2009 ; Bzdak and Skokov 2012 ; Deng and Huang 2012 ; Chen et al. 2021 , the late stage evolution of heavy-ion collisions produces numerous significant background sources, such as elliptic flow, posing substantial challenges for CME detection Adamczyk et al. 2014b ; Wang and Zhao 2017 ; Zhao 2019 ; Feng et al. 2025 . To extract potential CME signals from final-state hadrons, various analysis techniques and observables have been developed Sirunyan et al. 2018 ; Xu et al. 2018 ; Tang 2020 . A notable example is the CME-motivated isobar collision program conducted at RHIC in 2018 Koch et al. 2017 ; Abdallah et al. 2022a . However, another major obstacle arises from substantial background contributions, particularly due to the difference in nuclear structure between two isobaric nuclei Li et al. 2018 ; Zhao et al. 2019 ; Xu et al. 2021 . As a result, none of the current methods can unambiguously identify the CME or its induced charge separation (CS) in the quark-gluon plasma (QGP) along the direction of the magnetic field in isobar collisions. Given that event statistics in Au+Au collisions at 200 GeV will increase by a factor of 10 or more by the end of 2025, more advanced analysis methods are required to search for the potential CME signal. Promising techniques include event shape engineering Schukraft et al. 2013 , event shape selection Collaboration et al. 2025a ; Collaboration et al. 2025b and spectator/participant plane methods Abdallah et al. 2022b ; Feng et al. 2022 ; Chen et al. 2024 ; Chen et al. 2025 . At the same time, the rapid advancement of computational hardware and machine learning (ML) algorithms has revolutionized the way to extract information from large-scale datasets. Deep learning (DL), as a specialized ML paradigm, demonstrates exceptional capability in identifying nonlinear patterns with complex correlations Schmidhuber 2015 ; LeCun et al. 2015 . Its hierarchical neural architectures and automated feature extraction mechanisms are particularly effective in modeling complex physical systems, finding widespread applications in nuclear physics Pang et al. 2018 ; Zhou et al. 2019 ; Liu et al. 2019 ; Huang et al. 2021 ; Du et al. 2020 ; Kuttan et al. 2020a ; Jiang et al. 2021 ; Shi et al. 2022 ; Wang et al. 2022 ; Aarts et al. 2025 ; Zhou et al. 2024 , particle physics Baldi et al. 2014 ; Baldi et al. 2015 ; Barnard et al. 2017 ; Broecker et al. 2017 ; Radovic et al. 2018 , and condensed matter physics Carrasquilla and Melko 2017 ; Van Nieuwenburg et al. 2017 ; Han et al. 2018 ; Carleo et al. 2019 . In the DL study of the CME, supervised learning approach with convolutional neural networks has demonstrated success in identifying charge separation signals within final-state pion spectra Zhao et al. 2022 . However, this approach faced two significant challenges: first, the CME signal undergoes substantial suppression during the dynamical evolution of heavy-ion collisions Deng and Huang 2012 ; second, flow-related backgrounds progressively dominate as the heavy-ion collision system evolves Wang and Zhao 2017 . Developing DL techniques to reconstruct the evolution of CME signals in the QGP based on the measured final-state hadrons could provide a powerful solution to address these challenges, potentially overcoming current limitations in CME detection. In this Letter, we propose a novel deep learning framework based on a multi-phase transport (AMPT) model to unfold the final-state CME signal and reconstruct its complete evolution throughout the entire HICs via time reversal. This paper is organized as follows: Section II.1 introduces the AMPT model which provides the training and test datasets, while discussing its key model parameters, initial conditions, and their relevance to the CME. Section II.2 presents the time-embedded U-Net architecture, detailing its encoder-decoder structure with skip connections that enables accurate reconstruction of CME signal evolutions. In Sec. III , we present three cases with distinct inputs, demonstrating the predicted event-average charge separation (CS) along the reverse time evolution and the CS distribution at each time step comparing with the ground truth values. Finally, we summarize and discuss the implications of our results in Sec. IV . II MODEL AND METHOD II.1 A multi-phase transport model The AMPT model used in this work incorporates the string melting mechanism Lin et al. 2005 , simulating the complete evolution of high-energy heavy-ion collisions through four important stages. The initial conditions are provided by the HIJING (Heavy Ion Jet INteraction Generator) model Wang and Gyulassy 1991 ; Gyulassy and Wang 1994 , which generates the spatial and momentum distributions of minijet partons from hard processes and soft string excitations from soft processes. In the string melting implementation, all excited strings of hadrons are converted to their constituent partons while preserving the valence quarks’

Read the complete text