2024 workshop program
Obtaining Magnetic Fields in Molecular Clouds using Denoising Diffusion Probabilistic Models
The work
Abstract
Magnetic fields are important to the structure and processes of the interstellar medium, but are notoriously hard to estimate. While methods for predicting magnetic fields currently exist–such as the Davis-Chandrasekhar-Fermi (DCF) method–they may fail at times. Thus, we have built generative AI models, specifically Denoising Diffusion Probabilistic Models (DDPMs), to understand the relationship between polarization and magnetic fields in hopes of achieving higher efficacy than current estimation methods. We train the DDPMs on simulations of molecular clouds which include full pixel-by-pixel information of the strength and orientation of polarization and magnetic fields. Then, we apply the trained models to real polarized dust emission observations in order to make predictions about the magnetic field strength and orientation in actual molecular clouds. We compare these results to current methods for estimating magnetic fields such as the DCF method.