Emulators for stellar profiles in binary population modeling

TL;DR

Addresses the challenge of efficiently emulating detailed internal stellar profiles in binary population synthesis. The authors combine PCA-based dimensionality reduction with fully-connected neural networks trained on POSYDON v1 MESA grids to predict density-profile shapes from three binary initial parameters, followed by a post-processing step to ensure physical consistency. They demonstrate that the emulator achieves accuracy comparable to, and in some cases better than, a rescaled nearest-neighbor approach while offering major memory and storage savings. This framework enables scalable, high-fidelity population studies and provides a foundation for incorporating profiles into on-the-fly prescriptions and future grid expansions.

Abstract

Knowledge about the internal physical structure of stars is crucial to understanding their evolution. The novel binary population synthesis code POSYDON includes a module for interpolating the stellar and binary properties of any system at the end of binary MESA evolution based on a pre-computed set of models. In this work, we present a new emulation method for predicting stellar profiles, i.e., the internal stellar structure along the radial axis, using machine learning techniques. We use principal component analysis for dimensionality reduction and fully-connected feed-forward neural networks for making predictions. We find accuracy to be comparable to that of nearest neighbor approximation, with a strong advantage in terms of memory and storage efficiency. By providing a versatile framework for modeling stellar internal structure, the emulation method presented here will enable faster simulations of higher physical fidelity, offering a foundation for a wide range of large-scale population studies of stellar and binary evolution.

Figures (10)

• Figure 1: Log density profiles of all 1661 CO-HMS testing set binaries at the end of MESA evolution. The three main groups of central densities are hydrogen-burning stars at log(density)$\simeq$1, helium-burning stars at log(density)$\simeq$3, and carbon-burning stars at log(density)$\simeq$5. The highest density values are stars at carbon depletion, which is the latest stage at which we stop MESA simulations.
• Figure 2: Principal components of CO-HMS grid density profiles plus offsets for visual distinction. The first eight components are shown plotted in order of explained variance from top to bottom.
• Figure 3: Percentage of dataset variance explained by principal components versus number of principal components included.
• Figure 4: Two examples of PCA reconstructions (orange) of true density profiles (blue) shown for binaries in the HMS-HMS grid. The bottom panel shows a binary whose profile is in an area of the dataset's representation space not well covered by the principal component basis, and so the reconstructions in that area are less accurate than the example shown in the top panel.
• Figure 5: Neural network architecture for generating principal component weights of emulated density profiles.