Rapid Identification and Classification of Eccentric Gravitational Wave Inspirals with Machine Learning

Adhrit Ravichandran,Aditya Vijaykumar,Shasvath J. Kapadia,Prayush Kumar

2024-07-02

Abstract:Current templated searches for gravitational waves (GWs) emanated from compact binary coalescences (CBCs) assume that the binaries have circularized by the time they enter the sensitivity band of the LIGO-Virgo-KAGRA (LVK) network. However, certain formation channels predict that in future observing runs (O4 and beyond), a fraction of detectable binaries could enter the sensitivity band with a measurable eccentricity $e$. Constraining $e$ for each GW event with Bayesian parameter estimation methods is computationally expensive and time-consuming. This motivates the need for a machine learning based identification and classification scheme, which could weed out the majority of GW events as non-eccentric and drastically reduce the set of candidate eccentric GWs. As a proof of principle, we train a separable-convolutional neural network (SCNN) with spectrograms of synthetic GWs added to Gaussian noise characterized by O4 representative \texttt{PSD}s. We use the trained network to (i) segregate candidates as either eccentric or non-eccentric (henceforth called the detection problem) and (ii) classify the events as non-eccentric $(e = 0)$, moderately eccentric $(e \in (0, 0.2])$, and highly eccentric $(e \in (0.2, 0.5])$. On the detection problem, our best performing network detects eccentricity with $0.914$ accuracy and true and false positive rates of $0.862$ and $0.138$, respectively. On the classification problem, the best performing network classifies signals with $0.853$ accuracy. We find that our trained detector displays close to ideal behavior for the data we consider.

General Relativity and Quantum Cosmology,High Energy Astrophysical Phenomena

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the issue of how to quickly identify and classify gravitational wave (GW) signals with eccentric orbits. Specifically:

1. **Current Issues**:
- The current LIGO-Virgo-KAGRA (LVK) network assumes that compact binary coalescences (CBCs) enter the sensitive band of the LVK network already in circular orbits when detecting GWs. However, some formation channels predict that in future observation runs (such as O4 and beyond), a portion of detectable binaries may enter the sensitive band with observable eccentricities.
- Using Bayesian parameter estimation methods to constrain the eccentricity (e) of each GW event is a computationally intensive and time-consuming task.

2. **Research Objectives**:
- Propose a machine learning-based method that can quickly identify most non-eccentric GW events and significantly reduce the number of candidate eccentric GW events, thereby reducing the workload for subsequent Bayesian parameter estimation.

3. **Specific Implementation**:
- The authors trained a Separable Convolutional Neural Network (SCNN) using spectrograms of synthetic GW data superimposed with Gaussian noise as input.
- The network is used to (i) distinguish whether a candidate signal is eccentric or non-eccentric (referred to as the detection problem), and (ii) classify events as non-eccentric (e=0), moderately eccentric (e∈(0,0.2]), and highly eccentric (e∈(0.2,0.5]) (referred to as the classification problem).

Through this method, the authors hope to effectively handle a large number of GW signals in future observation runs and reduce the demand on computational resources.