Machine- and deep-learning regression for equations of state of compact objects
Phys. Rev. C 114, 035804 – Published 9 September, 2026
DOI: https://doi.org/10.1103/v98s-bh7z
Abstract
One of the outstanding challenges in nuclear physics is achieving a comprehensive understanding of the equation of state that governs dense nuclear matter, and, by extension, the internal composition and structure of compact astrophysical objects such as neutron stars and quark stars. The primary approach to addressing this problem has thus far been predominantly theoretical, involving the construction of equations of state based on plausible nuclear and particle physics theories. These theoretical predictions are subsequently evaluated against empirical results, which are obtained primarily from heavy-ion collision experiments and, increasingly, from astrophysical observations of compact objects. In recent years, several alternative analytical methodologies have been developed that integrate the aforementioned theoretical and empirical approaches. These methods incorporate advanced data analysis and processing techniques, including Bayesian inference, and machine-learning and deep-learning frameworks, to enhance the interpretation and constraining of the equation of state for dense nuclear matter. In this study, the aforementioned techniques, particularly machine learning and deep learning, are employed with the objective of enhancing our understanding of the equation of state by leveraging information derived from mass-radius diagrams of compact objects. More specifically, this study utilizes a large dataset of equations of state constructed using multimodal formulations of neutron stars, along with a corresponding dataset for quark stars. These equations of state are consistent with known physical constraints and have been developed to reflect the highest degree of physical plausibility. Through the application of machine-learning and deep-learning techniques, it becomes possible to infer or reconstruct the underlying equations of state based on input from mass-radius diagram data.
