We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

gr-qc

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

General Relativity and Quantum Cosmology

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

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 PSDs. 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 $0.986$ accuracy and true and false positive rates of $0.984$ and $1.6\times 10^{-3}$, respectively. On the classification problem, the best performing network classifies signals with $0.962$ accuracy. We also find that our trained classifier displays close to ideal behavior for the data we consider.
Comments: 18 pages, 15 figures, 2 tables
Subjects: General Relativity and Quantum Cosmology (gr-qc); High Energy Astrophysical Phenomena (astro-ph.HE)
Cite as: arXiv:2302.00666 [gr-qc]
  (or arXiv:2302.00666v1 [gr-qc] for this version)

Submission history

From: Adhrit Ravichandran [view email]
[v1] Wed, 1 Feb 2023 18:53:02 GMT (6829kb,D)

Link back to: arXiv, form interface, contact.