Explainable autoencoder for neutron star dense matter parameter estimation

In a recent study, we present a physics-informed autoencoder designed to encode the equation of state (EoS) of neutron stars into an interpretable latent space. This approach allows us to efficiently estimate the parameters of dense matter from observational data, such as mass-radius measurements and tidal deformabilities, as in the case of gravitational wave events like GW170817, and ‘‘invert’’ the NS observations to obtain the EoS.

With Francesco Di Clemente and Matteo Scialpi we have experimented with additions to the autoencoder loss function, which exploit redundant algebraic information between the ‘‘ground truth’’ input and resulting output. The article is available at arXiv:2501.15222.