Qian Hu, Jessica Irwin, Qi Sun, Christopher Messenger, Lami Suleiman, Ik Siong Heng and John Veitch
{"title":"Decoding Long-duration Gravitational Waves from Binary Neutron Stars with Machine Learning: Parameter Estimation and Equations of State","authors":"Qian Hu, Jessica Irwin, Qi Sun, Christopher Messenger, Lami Suleiman, Ik Siong Heng and John Veitch","doi":"10.3847/2041-8213/ade42f","DOIUrl":null,"url":null,"abstract":"Gravitational waves (GWs) from binary neutron stars (BNSs) offer a valuable understanding of the nature of compact objects and hadronic matter, and the science potential will be greatly enhanced by the third-generation (3G) GW detectors, which are expected to detect BNS signals with order-of-magnitude improvements in duration, detection rates, and signal strength. However, the resulting computational demands for analyzing such prolonged signals pose a critical challenge that existing Bayesian methods cannot feasibly address in the 3G era. To bridge this critical gap, we demonstrate a machine learning–based workflow capable of producing source parameter estimation and constraints on equations of state (EOSs) for hours-long BNS signals in seconds with minimal hardware costs. We employ efficient compression of the GW data and EOS using neural networks, based on which we build normalizing flows for inference that can deliver results in seconds. The optimized computational cost of BNS signal analysis with our framework shows that machine learning has the potential to be an indispensable tool for future catalog-level BNS analyses, paving the way for large-scale investigations of BNS-related physics across the 3G observational landscape.","PeriodicalId":501814,"journal":{"name":"The Astrophysical Journal Letters","volume":"633 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2025-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"The Astrophysical Journal Letters","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.3847/2041-8213/ade42f","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Gravitational waves (GWs) from binary neutron stars (BNSs) offer a valuable understanding of the nature of compact objects and hadronic matter, and the science potential will be greatly enhanced by the third-generation (3G) GW detectors, which are expected to detect BNS signals with order-of-magnitude improvements in duration, detection rates, and signal strength. However, the resulting computational demands for analyzing such prolonged signals pose a critical challenge that existing Bayesian methods cannot feasibly address in the 3G era. To bridge this critical gap, we demonstrate a machine learning–based workflow capable of producing source parameter estimation and constraints on equations of state (EOSs) for hours-long BNS signals in seconds with minimal hardware costs. We employ efficient compression of the GW data and EOS using neural networks, based on which we build normalizing flows for inference that can deliver results in seconds. The optimized computational cost of BNS signal analysis with our framework shows that machine learning has the potential to be an indispensable tool for future catalog-level BNS analyses, paving the way for large-scale investigations of BNS-related physics across the 3G observational landscape.