Neural Astrophysical Wind Models

TL;DR

This paper demonstrates that neural networks embedded in differential equations can discover complex astrophysical wind physics, such as mass-loading and flow geometry, without prior knowledge, offering a new approach for modeling non-linear inverse problems.

Contribution

The authors introduce a neural ODE framework that learns key physical functions in galactic wind models directly from data, without assuming their functional forms.

Findings

Neural networks successfully discover physics of wind mass-loading and geometry.

The method robustly handles singularities and unknown functional forms.

Neural ODEs show promise as interpretable tools for complex inverse problems.

Abstract

The bulk kinematics and thermodynamics of hot supernovae-driven galactic winds is critically dependent on both the amount of swept up cool clouds and non-spherical collimated flow geometry. However, accurately parameterizing these physics is difficult because their functional forms are often unknown, and because the coupled non-linear flow equations contain singularities. We show that deep neural networks embedded as individual terms in the governing coupled ordinary differential equations (ODEs) can robustly discover both of these physics, without any prior knowledge of the true function structure, as a supervised learning task. We optimize a loss function based on the Mach number, rather than the explicitly solved-for 3 conserved variables, and apply a penalty term towards near-diverging solutions. The same neural network architecture is used for learning both the hidden mass-loading and surface area expansion rates. This work further highlights the feasibility of neural ODEs as a promising discovery tool with mechanistic interpretability for non-linear inverse problems.