Research

Sub-grid Modeling of Super-Eddington Accretion

We propose a physically motivated sub-grid prescription for modeling both mass and spin growth of black holes in the super-Eddington accretion regime within hydrodynamical simulations (using the GIZMO code). The model includes an inner photon-trapping region with a simulation-based fitting structure and an outer standard thin α-disk, seamlessly transitioning between them. We implement a self-consistent spin evolution mechanism that accounts for effects like Bardeen–Petterson alignment and inner disk precession, conditioned on accretion rate. This work lays the groundwork for more realistic black hole evolution modeling in cosmological simulations. (Kao et al. 2025)

Machine Learning for Quasar Classification

We present a novel approach to classifying Broad Absorption Line Quasars (BALQSOs) using large-scale spectroscopic data from SDSS DR16. By applying multiple dimensionality reduction techniques—such as PCA, t-SNE, LLE, and ISOMAP—we transform high-dimensional quasar spectra into compact representations. These compressed features are then fed into machine learning classifiers (including XGBoost and Random Forest), achieving outstanding performance. This study demonstrates the synergy of statistical learning and astronomy, enabling efficient, reliable identification of rare quasar subtypes. (Kao et al. 2024)