Neural networks unveiling the properties of gravitational wave background from supermassive black hole binaries

Supermassive black hole binaries (SMBHBs) are binary systems formed by black holes with masses exceeding millions of solar masses, and are expected to form and evolve in the nuclei of galaxies. The extremely compact nature of these objects leads to the intense and efficient emission of gravitational waves (GWs), which can be detected by the Pulsar Timing Array (PTA) experiment in the form of a gravitational wave background (GWB); that is, a superposition of GW signals coming from different sources. The modelling of the GWB requires some assumptions as to the binary population, and exploration of the whole parameter space involved is hindered by the great computational cost involved. We trained two neural networks (NN) on a semi-analytical modelling of the GWB generated by an eccentric population of MBHBs that interact with the stellar environment. We then used the NN to predict the characteristics of the GW signal in regions of the parameter space that we did not sample analytically. The developed framework allows us to quickly predict the amplitude, shape, and variance of the GWB signals produced in different realisations of the universe.