Starduster: A Multiwavelength SED Model Based on Radiative Transfer Simulations and Deep Learning
Yisheng Qiu, Xi Kang
Abstract
Abstract We present starduster , a supervised deep-learning model that predicts the multiwavelength spectral energy distribution (SED) from galaxy geometry parameters and star formation history by emulating dust radiative transfer simulations. The model is composed of three specifically designed neural networks, which take into account the features of dust attenuation and emission. We utilize the skirt radiative transfer simulation to produce data for the training data of neural networks. Each neural network can be trained using ∼4000–5000 samples. Compared with the direct results of the skirt simulation, our deep-learning model produces ∼0.005 mag and ∼0.1–0.2 mag errors for dust attenuation and emission, respectively. As an application, we fit our model to the observed SEDs of IC 4225 and NGC 5166. Our model can reproduce the observations and provide reasonable measurements of the inclination angle and stellar mass. However, some predicted geometry parameters are different from an image-fitting study. Our analysis implies that including a constraint at (rest-frame) ∼40 μ m could alleviate the degeneracy in the parameter space for both IC 4225 and NGC 5166, leading to broadly consistent results with the image-fitting predictions. Our SED code is publicly available and can be applied to both SED fitting and SED modeling of galaxies from semianalytic models.