| Abstract: |
Precise halo mass (Mh) measurements are crucial for cosmology and galaxy formation. HaloFlow (C. Hahn et al. 2024) introduced a simulation-based inference (SBI) framework that uses state-of-the-art simulated galaxy images to infer Mh with precision. However, for HaloFlow to be scientifically useful on real data, it must remain reliable even when the underlying galaxy-formation physics differ from those in the simulations on which it was trained. In practice, this means that HaloFlow must gener-Alice across different galaxy formation and feedback models. Without accounting for these differences,HaloFlow produces biased and overconfident Mh posteriors when applied to simulations with different physics. We introduce HaloFlowDA, an extension of HaloFlow that integrates domain adaptation (DA) with SBI to mitigate these cross-simulation shifts in galaxy photometry and morphology. Using synthetic galaxy images forward-modeled from the IllustrisTNG, EAGLE, and SIMBA simulations, we test two DA methods: Domain-Adversarial Neural Networks (DANN) and Maximum Mean Discrepancy (MMD). Incorporating DA significantly reduces bias and improves calibration, with MMD achieving the most stable performance, lowering the normalized residual metric, β, by an average of 34% and up to 64% when trained and tested on different simulations. HaloFlowDA enables consistent, simulation-trained inference models to generalize across domains, establishing a foundation for robust Mh inference from real HSC-SSP observations and future Rubin LSST observations.
|
