ľ¹Ï¸£ÀûÓ°ÊÓ

Multi-Messenger Astrophysics aspires to make use of different radiation channels or messengers to provide information about astrophysical events. Certain phenomena emit gravitational waves as well as electromagnetic counterparts. For a binary neutron star coalescence, these different messengers have different times of arrivals and one can use the one arriving the earliest to predict the arrival of the others. Scientists have now been able to introduce a new method for early-warning alerts. 

A gravitational wave produced by the coalescence of two compact objects, such as neutron stars, is divided in three stages: the inspiral - when the orbital motion of the two objects radiates away energy and the orbit shrinks -, the merger - when they touch and join - and the ring-down - when the newly formed body returns to its stable state -. An early detection of the inspiral of these events would produce an alert for other detectors, increasing the possibility of observing other types of radiation. This would enable us to improve our understanding of these physical processes. 

In recent years, machine-learning-based approaches have sparked the interest of scientists, due to their rapid identification of gravitational-wave transients. In this line of thought, researchers from the Institute for Gravitational and Subatomic Physics (GRASP)/Nikhef in collaboration with colleagues from the ľ¹Ï¸£ÀûÓ°ÊÓ of Liège have introduced a new method, based on convolutional neural networks (CNNs), in order to produce early-warning alerts for an inspiraling binary neutron star system.

This setup was tested with gravitational waves embedded in simulated detector noise.  Its detection rate is equivalent to the standard techniques while being faster. This deep learning pipeline can produce an early alert up to 100 seconds before the merger for the best-case scenario. 
 

Publication: 

Convolutional neural networks for the detection of the early inspiral of a gravitational-wave signal
Grégory Baltus, Justin Janquart, Melissa Lopez, Amit Reza, Sarah Caudill, and Jean-René Cudell
18 May 2021, Phys. Rev. D 103, 102003,  and .