Machine learning methods to predict the formation of binary compact objects

Aug 3, 2026

Upload by Anda

Thesis work evaluates machine learning approaches, including Random forest and XGBoost, for predicting the formation of binary compact objects using only stellar-binary initial conditions (masses, metallicity, orbital parameters) plus key binary-evolution parameters such as common envelope efficiency. Models are trained on simulations generated with rapid binary population synthesis using the SEVN code. Results show limited ability to predict merging systems within the Hubble time, with performance near random chance, while distinguishing bound from non-bound compact-object binaries more effectively.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages46

Reading Time1 h 56 min

AccessFree to download

Introduction

Gravitational waves & GW detectors

Sources of GW: compact objects

Binaries of stellar blackhole

Integrating theory with SEVN to explore the parameter phase and taking assistance of ML

Machine learning

Types of ML algorithms

Supervised and unsupervised learning

Random forest

SEVN

Data, training and validation

Training the model

Challenges encountered while training with SEVN data

UNIVERSIT DEGLI STUDI DI PADOVA Dipartimento di Fisica e Astronomia “Galileo Galilei”Master Degree in Physics of Data Final Dissertation Machine learning methods to predict the formation of binary compact objects Thesis supervisor Prof. Michela Mapelli Thesis co-supervisor Dr. Giuliano Iorio Candidate Vivek Kashyap Janardhana Academic Year 2022/2023 Machine learning methods to predict the formation of binary compact objects Vivek Kashyap Janardhana Abstract In this thesis I tested different machine learning methods (Random forest, XGBoost) to predict the formation of binary compact objects solely based on initial condition of stellar binaries (masses, metallicity and orbital parameters) and on the parameters of binary evolution (i.e., common envelope efficiency) . I trained the machine learning methods on simulations performed with rapid binary population synthesis code (BPS) code SEVN. I found that the tested methods can predict the formation of merging (within the Hubble time) and non-merging binary compact objects. The performance of predicting merging compact objects is found to be 50%, which is equivalent to random chance. This suggests that the methods cannot accurately predict the formation of merging compact objects. However, the methods show good performance in distinguishing between systems that produce bound and non-bound systems. Contents 1 Introduction 1 1. 1 Gravitational waves & GW detectors . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Sources of GW: compact objects . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Binaries of stellar blackhole . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.4 Integrating theory with SEVN to explore the parameter phase and taking assistance of ML ................................. 8 2 Machine learning 9 2.0.1 Types of ML algorithms ......................... 9 2.0.2 Supervised and unsupervised learning ................. 10 2.0.3 Random forest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 3 SEVN 15 3. 1 SEVN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.1.1 Binary population synthesis code . . . . . . . . . . . . . . . . . . . . 15 3.1.2 SEVN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.2 Data, training and validation . . . . . . . . . . . . . . . . . . . . . . . . . . 20 3.2.1 Training the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 3.2.2 Challenges encountered while training with SEVN data . . . . . . . 22 Chapter 1 Introduction 1.1 Gravitational waves & GW detectors The concept of gravitational waves traces its origins back to the revolutionary theory of general relativity proposed by Albert Einstein in 1915 . In 1918, Einstein made a remarkable prediction: the existence of gravitational waves as a consequence of his theory. These waves, he proposed, are ripples in the fabric of spacetime itself, propagating through the universe at the speed of light. Einstein’s prediction of gravitational waves was a profound revelation that not only challenged our understanding of gravity but also presented a new perspective on the nature of the cosmos. It indicated that gravity, far from being merely a force acting at a distance, is an intrinsic property of the spacetime continuum. Gravitational waves became an essential aspect of Einstein’s general theory of relativity, expanding our understanding of the gravitational interaction and its influence on the behavior of massive objects in the universe. The significance of gravitational waves lies in their potential to unveil the secrets of the most violent and energetic events in the cosmos. These include the collision of massive black holes, the coalescence of neutron stars, and the explosive cataclysms of supernovae. By detecting and studying gravitational waves, scientists can gain unique insights into the dynamics, properties, and evolution of these

FAQs

Which inputs are used to predict the formation of binary compact objects?

The models use initial conditions of stellar binaries (masses, metallicity, orbital parameters) and binary-evolution parameters, specifically including common envelope efficiency.

How were the machine learning models trained?

They were trained on simulations produced with rapid binary population synthesis using the SEVN code.

How well do the methods predict merging compact objects?

Predicting merging systems within the Hubble time achieves about 50% performance, which is equivalent to random chance, indicating poor predictive accuracy for mergers.

Aug 3, 2026

Upload by Anda