Abstract

Magnetic fields in molecular clouds are dynamically significant throughout the star formation process: they provide support against gravitational contraction, regulate turbulent flows, and influence the development of cloud substructure. Observations of magnetic field morphology and magnitude can be used to compare different models and theoretical predictions of star formation. Unfortunately, measurements of both the line-of-sight (LOS) and plane-of-sky (POS) components of the magnetic field are challenging to obtain and current methods suffer from a number of limitations, making the development of new techniques for magnetic field mapping desirable. Recently, the Velocity Gradient Technique (VGT), based on the anisotropic nature of MHD turbulence, has been developed for tracing magnetic field orientation using readily-available spectroscopic data. However, this method involves a sub-block pixel averaging procedure, which limits the attainable resolution of the POS map. In this talk, I will present a deep learning method motivated by the success of the VGT, which uses a convolutional neural network (CNN) to produce pixel-by-pixel magnetic field maps from spectroscopic observations across a range of ISM conditions.