Machine Learning-Based Detection of Non-Axisymmetric Fast Neutrino Flavor Instabilities in Core-Collapse Supernovae

Aug 5, 2026

Upload by Hazel

Dense neutrino environments in core-collapse supernovae and neutron star mergers can trigger fast flavor conversions when the neutrino electron lepton number (νELN) angular distribution crosses zero along certain directions. Prior studies mainly addressed axisymmetric νELN crossings, since non-axisymmetric cases lacked realistic angular-distribution models. This work presents a machine-learning approach trained on two core-collapse supernova simulations (with and without rotation), enabling detection of non-axisymmetric νELN crossings with accuracy above 90%.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Core-collapse supernovae and neutron star mergers

Collective neutrino oscillations and fast flavor conversions

νELN crossings and the need for angular distributions

Machine learning for fast flavor conversion detection

arXiv :2401 . 10915v1 [ astro-ph .HE] 12 Jan 2024 MPP-2023-271 Machine Learning-Based Detection of Non-Axisymmetric Fast Neutrino Flavor Instabilities in Core-Collapse Supernovae Sajad Abbar , 1 Akira Harada,2 and Hiroki Nagakura 3 1 Max-Planck-Institut f¨ur Physik (Werner-Heisenberg-Institut), F¨ohringer Ring 6, 80805 M¨unchen, Germany 2 RIKEN Interdisciplinary Theoretical and Mathematical Sciences Program (iTHEMS), 2-1 Hirosawa, Wako, Saitama 351-0198, Japan 3 Division of Science, National Astronomical Observatory of Japan, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan In dense neutrino environments like core-collapse supernovae (CCSNe) and neutron star mergers (NSMs), neutrinos can undergo fast flavor conversions (FFC) when their angular distribution of neu trino electron lepton number (νELN) crosses zero along some directions. While previous studies have demonstrated the detection of axisymmetric νELN crossings in these extreme environments, non axisymmetric crossings have remained elusive, mostly due to the absence of models for their angular distributions. In this study, we present a pioneering analysis of the detection of non-axisymmetric νELN crossings using machine learning (ML) techniques. Our ML models are trained on data from two CCSN simulations, one with rotation and one without, where non-axisymmetric features in neutrino angular distributions play a crucial role. We demonstrate that our ML models achieve detection accuracies exceeding 90% . This is an important improvement, especially considering that a significant portion of νELN crossings in these models eluded detection by earlier methods. I. INTRODUCTION Core-collapse supernovae (CCSNe) and neutron star mergers (NSMs) are among the most extreme astrophysical settings in the universe. These events, marking the dramatic finale of massive stars and the collision of densely packed remnants, reveal some of the universe’s most energetic and enigmatic events [1–4] . At the heart of these celestial dramas lies the emission of elusive neutrinos, which are generated in huge numbers during both CCSNe and NSMs. In the extreme and densely packed conditions within these events, neutrinos undergo an exciting journey marked by an intriguing phenomenon referred to as collective neutrino oscillations [5– 11] (for a recent review see Ref. [12]) . This fascinating behavior emerges from their coherent forward scatterings with the dense neutrino background. This nonlinear and collective dance gives rise to a rich tapestry of flavor transformations. One of the most interesting advancements in the field has been the discovery and the exploration of the fascinating phenomenon of fast flavor conversions (FFCs), which occurs on extraordinarily short scales (see, e.g. , Refs. [13–67]) . FFCs manifest on scales characterized by ∼ G−F1 n1 , potentially spanning just a few centimeters within the SN core. This is in contrast to the conventional slow modes, governed by the neutrino vacuum frequency and expected to extend over kilometer-scale distances within the SN environment. Here, GF represents the Fermi coupling constant, and nν denotes the neutrino number density. FFCs occur iff the angular distribution of the neutrino electron lepton number, νELN, G (v) = √2GF Z0 ∞ E(2~~ν~~2dπE)3ν [fνe (p) − fν¯e (p)], (1) crosses zero at some v = v (µ,ϕν ), with µ = cosθν [30] . Here, Eν , θν , and ϕν are the neutrino energy, the zenith, and azimuthal angles of the neutrino velocity, respectively, and fν ’s are the neutrino occupation numbers. Note that here and throughout this work we assume that νx and ν¯x (heavy-lepton neutrinos and antineutrinos) have the same angular distributions. Investigating νELN crossings requires access to comprehensive angular distributions of neutrinos. Yet, acquiring this detailed angular data presents a considerable challenge in modern CCSN and NSM simulations due to the extensive computational resources it demands. In

FAQs

What triggers fast flavor conversions in these simulations?

Fast flavor conversions occur when the angular distribution of the neutrino electron lepton number (νELN) crosses zero in some directions.

Why were non-axisymmetric νELN crossings difficult to study previously?

Non-axisymmetric cases had been elusive largely because there were no suitable models for their angular distributions.

How does the proposed machine-learning method improve detection?

The method trains on two CCSN simulations (rotating and non-rotating) and uses non-axisymmetric patterns in neutrino angular distributions to achieve detection accuracies exceeding 90%. It also recovers a portion of crossings that earlier methods missed.

Aug 5, 2026

Upload by Hazel