Deep operator neural network applied to efficient computation of asteroid surface temperature and the Yarkovsky effect
Zhao, Shunjing et al. (2024)
- Published
- Nov 4, 2024
- Journal
- arXiv (Cornell University)
- DOI
- 10.48550/arXiv.2411.02653
At a GlanceAI
DeepONet predicts asteroid surface temperatures ~100,000× faster than simulations while preserving ~1% accuracy for Yarkovsky studies.
SummaryAI
Accurate asteroid surface temperatures are central to modeling thermal forces such as the Yarkovsky effect, but repeated high-fidelity simulations are often too slow for broad parameter studies. This work applies a deep operator neural network to learn asteroid thermophysical behavior, achieving about 1% average temperature error at five orders of magnitude lower computational cost. That speedup makes multidimensional thermal analyses much more practical and enables N-body orbital simulations that incorporate near-instantaneous, ML-based Yarkovsky estimates. The examples of Phaethon and 2001 WM41 illustrate how AI methods could accelerate studies of asteroid evolution and hazard-relevant dynamics.
Method SnapshotAI
The study uses a deep operator neural network surrogate model trained to emulate asteroid thermophysical simulations and embedded in N-body dynamics.
BackgroundAI
General background in asteroid thermophysics, orbital dynamics, and machine learning surrogate modeling is helpful.