SPICE: Differentiable Synthetic Spectra Modeling for Stars with Surface Inhomogeneities

Speaker: Maja Jabłońska (Australian National University) In Person and zoom

Zoom info: https://stanford.zoom.us/j/98604058568

Traditional stellar spectral modeling often assumes uniform and symmetric stellar surfaces. However, observational evidence from interferometry, photometric time series, and spectroscopic surveys highlights the prevalence of inhomogeneities, such as temperature spots, pulsations, rotational flattening, and binary interactions. To address these complexities, we introduce SPICE (SPectral Integration Compiled Engine), an open-source Python package (JAX based) that integrates ML-driven spectral emulation to synthesize stellar spectra. SPICE combines mesh-based surface models with a transformer-based neural intensity emulator, enabling high-accuracy and flexibility modeling of stellar physics.