AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-22
ML in Astrophysical Turbulence I: Predicting Prestellar Cores in Magnetized Molecular Clouds using eXtreme Gradient Boosting
In giant molecular clouds, where supersonic turbulence and magnetic fields make it hard to predict which gas parcels collapse into star-forming cores, the authors present an XGBoost framework that forecasts the 3D trajectories and gravitational collapse fate of prestellar gas, trained on about 2.1 million tracer particles from self-gravitating turbulent MHD simulations. The model predicts 3D spatial evolution over 0.25 freefall times (about 0.31 Myr) with R² > 0.99 and ranks core susceptibility with AUC 0.975, while SHAP analysis shows local density, kinetic energy, and magnetic field magnitude dominate and the model re-derives a sharp nonlinear collapse threshold at log₁₀ρ ≈ 0.8 above which magnetic fields amplify collapse predictions. The study also shows that point-wise ML models, though strong at kinematic advection, mathematically lack the spatial context needed to evaluate the local differential operators that physically stop collapse.