Modeling the Neutron Star Equation of State with a Structured Variational Autoencoder

The equation of state (EOS) of dense matter determines the structure and observable properties of neutron stars, yet remains uncertain due to the breakdown of nuclear theories and limited experimental access at supra-nuclear densities. We present a new inference framework for the neutron star EOS based on a Structured Variational Autoencoder (VAE). Once trained, the encoder helps solve the Tolman–Oppenheimer–Volkoff (TOV) equations for key observables such as the maximum mass and canonical radius, while the decoder reconstructs the EOS from these observables and a range of variational latent parameters. The model is trained on EOSs derived from Skyrme and relativistic mean-field interactions, ensuring broad theoretical coverage. The latent space captures physically meaningful correlations between microscopic properties and macroscopic observables, allowing efficient exploration of the dense-matter parameter space. The framework is flexible enough to reconstruct other quantities related to the EOS, providing a fast, data-driven approach to connect nuclear models, neutron-star observations, and multimessenger constraints.