Solving Lane-Emden-Type Eigenvalue Problems with Physics-Informed Neural Networks

TL;DR

This paper introduces a novel physics-informed neural network approach to solve the Lane-Emden eigenvalue problems for various n values, accurately determining stellar radii with minimal errors, validated against theoretical solutions.

Contribution

The work demonstrates the first application of PINNs to solve Lane-Emden eigenvalue problems and accurately compute stellar radii, with hyperparameter optimization for improved performance.

Findings

PINNs achieve near-exact agreement with theoretical eigenvalues for n=0,1.

Errors below 0.0009% for n=2,3 and 0.05% for n=4.

Hyperparameter tuning enhances solution accuracy across different n values.

Abstract

The Lane-Emden equation, a nonlinear second-order ordinary differential equation, plays a fundamental role in theoretical physics and astrophysics, particularly in modeling the structure of stellar interiors. Also referred to as the polytropic differential equation, it describes the behavior of self-gravitating polytropic spheres. In this study, we present a novel approach to the solution of the eigenvalue problem which arises when considering the Lane-Emden equation for n = 0, 1, 2, 3, 4 using Physics-Informed Neural Networks (PINNs). The novelty of this work is that, we not only solve the Lane-Emden equation via PINNS but we also determine the eigenvalue, r, which is the stellar radius. Hyperparameter tuning was conducted using Bayesian optimization in the Optuna framework to identify optimal values for the number of hidden layers, number of neurons, activation function, optimizer, and learning rate for each value of n. The results show that, for n = 0, 1, PINNs achieve near-exact agreement with theoretical eigenvalues (errors < 0.000806%). While for more nonlinear cases, n = 2, 3 and n=4, PINNs yield errors below 0.0009% and 0.05% respectively, validating their robustness.