Speaker
Description
The impact of active galactic nucleus (AGN) radiation feedback on galaxy evolution remains an open question in astrophysics, particularly regarding its ability to regulate star formation and drive outflows. In this work, we present a detailed investigation of AGN radiation feedback using state-of-the-art cosmological simulations, analyzing its effects on gas thermodynamics, metal enrichment, and outflow kinematics. Our recent study demonstrates that AGN radiation significantly influences gas cooling rates and star formation histories, with varying efficiencies depending on host galaxy mass and environment.
To further refine feedback models, we integrate machine learning (ML) techniques and high-performance surrogate modeling. We introduce CODES, a benchmark for evaluating coupled ODE system surrogates, enabling systematic testing of radiation feedback prescriptions across multiple regimes, including extrapolation, uncertainty quantification, and gradient correlation. Additionally, we employ a Bayesian model comparison framework that uses a k-sparse variational autoencoder (VAE) to assess the realism of simulated galaxies compared to SDSS observations, revealing discrepancies in AGN-driven outflows across different simulation models.
Finally, we present RUBIX, a GPU-accelerated tool for forward modeling integral field unit (IFU) observations from simulations. RUBIX achieves a 600× speedup over traditional methods, allowing for efficient modeling of AGN-driven outflows and their observational signatures, paving the way for direct comparisons with IFU surveys.
By combining high-resolution simulations with ML-driven analysis, our work provides new insights into the role of AGN radiation feedback in galaxy evolution, helping to constrain theoretical models and improve the fidelity of cosmological simulations.
