Deep Learning the Dark Universe with Gravitational Waves

Gravitational wave astronomy has opened an extraordinary new window to test the theory of gravity in the genuinely strong, highly dynamical and relativistic regime.

The LIGO-Virgo Collaboration has now detected over 50 mergers of compact binary systems and this number will considerably increase in the coming years.

There are currently two main issues related to the possibility of testing gravity with gravitational wave observations: the weakness of parametric tests of General Relativity to go beyond null tests and the very long inference time required by standard samplers which can take up to months.

Specific waveform models and new techniques to speed up statistical inference are therefore crucial to maximise the scientific return of already available and upcoming data.

In this project, we will construct an analytical model of the gravitational waves emitted during the late inspiral and merger of compact objects in theories of gravity that are cosmologically motivated, namely that have a chance to explain Dark Energy.

We will then leverage deep learning techniques to promptly produce the posterior for the corresponding parameters given the detector data.

To this aim, we will build up on two codes developed by one of the supervisors - ROMAN and PERCIVAL - which pioneered the use of machine learning in gravitational wave science.

We will then apply this new pipeline to the real LIGO-Virgo data and perform Bayesian inference of Dark Energy parameters.

All together this project will provide a new and complete framework to test the dark Universe with gravitational wave observations, exploiting state-of-the-art deep learning techniques.