Data-Driven Generation of Neutron Star Equations of State Using Variational Autoencoders
Student:
TBD
Mentors:
Sanjay Reddy (INSPIRE-HEP, email: sareddy@uw.edu)
Tianqi Zhao (INSPIRE-HEP, email: tianqi24@uw.edu)
Prerequisites:
Linear Algebra and Calculus are necessary. Thermodynamics, statistics, and scientific programming in Python are preferred.
What Students Will Learn:
The student will learn the basic knowledge of neutron stars and the equation of state (EOS) for the dense matter therein. They will also learn the basic knowledge of neural networks and how to implement them with TensorFlow or PyTorch. The ability to process large datasets and to take advantage of GPU in parallel computing will be developed.
Expected Project Length:
3 quarters
Project Description:
Neutron stars are among the most extreme astrophysical objects, with densities exceeding nuclear saturation. Their internal structure is governed by the equation of state (EOS), which relates pressure to energy density. While theoretical models such as Skyrme-type interactions and relativistic mean-field (RMF) calculations provide a wide range of possible EOSs, fully exploring the parameter space remains computationally challenging. To address this, machine-learning methods such as the Gaussian Process (GP) [1] and Variational Autoencoder (VAE) [2] have been proposed to learn from existing EOS models and generate new ones, facilitating a more efficient exploration of dense matter properties.
This project aims to train a VAE on a dataset of EOS models derived from Skyrme and RMF models. The VAE consists of an encoder that maps high-dimensional EOS data to a lower-dimensional latent space and a decoder that reconstructs the EOSs from this representation. By sampling from the learned latent space encoding EOS essentials, we can generate novel yet physically meaningful EOS models as existing models while exploring parameter regions beyond existing models. These generated EOSs will be analyzed for their astrophysical implications, including solving the Tolman-Oppenheimer-Volkoff (TOV) equations to determine neutron star mass-radius relations.
The project will result in a robust framework for EOS generation, providing new insights into the underlying properties of dense matter. Ultimately, this approach could enhance our understanding of dense matter physics and contribute to constraining neutron star properties through multi-messenger astrophysical observations.
References:
[1] Landry, P. and Essick, R., 2019. Nonparametric inference of the neutron star equation of state from gravitational wave observations. Physical Review D, 99(8), p.084049.
[2] Han, M.Z., Tang, S.P. and Fan, Y.Z., 2023. Nonparametric Representation of Neutron Star Equation of State Using Variational Autoencoder. The Astrophysical Journal, 950(2), p.77.