EmulART: Emulating radiative transfer—a pilot study on autoencoder-based dimensionality reduction for radiative transfer models

Abstract Dust is a major component of the interstellar medium. Through scattering, absorption and thermal re-emission, it can profoundly alter astrophysical observations. Models for dust composition distribution are necessary to better understand curb their impact on A new approach serial computationally inexpensive production such models here presented. Traditionally these studied with help radiative transfer modelling, critical tool attenuation reddening observed properties galaxies active galactic nuclei. Such simulations present, however, an approximately linear computational cost increase desired information resolution. Our efficient model generator proposes denoising variational autoencoder (or alternatively PCA), spectral compression, combined approximate Bayesian method spatial inference, emulate high from low models. For simple spherical shell anisotropic illumination, our proposed successfully emulates reference simulation starting less than 1% information. emulations at different viewing angles present median residuals below 15% across dimension 48% dimensions. EmulART infers estimates $$\sim $$ ∼ 85% missing input, all within total running time around 20 minutes, estimated be 6 $$\times × faster target resolution simulations, up 50 when applied more complicated simulations.