The detailed 3D distributions of dust density and extinction in the Milky Way have long been sought after. However, such 3D reconstruction from sparse data is non-trivial, but is essential to understanding the properties of star-formation, large-scale dynamics and structure of our Galaxy. In this work I will introduce our new fast and scalable algorithm for 3D dust modeling. Using advanced ML methods such as sparse Gaussian Processes and Variational Inference, our algorithm maps Star Formation Regions (SFRs) with millions of input sources in parsec scales within an hour on a single GPU. Our approach allows us to identify large-scale structures in the Milky Way while simultaneously peering into individual molecular clouds, providing insights into multi-scale processes such as fragmentation in molecular clouds.
In Dharmawardena et al., (2022 a, b), we model the 3D dust density distribution of 15 SFRs, exploiting distances and extinctions derived from Gaia DR2 and IR data (from Fouesneau et al., 2022). From these maps, we extract 3D boundaries, volumes, precise dust masses (12%
statistical uncertainty) and filling factors to study fragmentation within our regions. We recover a wider range of substructures such as new interconnecting and free standing filaments and star-formation feedback and supernovae cavities. In Dharmawardena et al., in prep we present a first look at our new 3D dust density maps of the Milky Way out to 2 kpc from the sun simultaneously showing both large scale structure at 100s of pc scale and smaller scale structure at 10s of pc. The maps’ comparison to the Gaia DR3 DIBs, YSO and evolved star samples will shed light on stellar processes taking place in the Milky Way in 3D.