Coupling Coagulation to Hydrodynamics with Machine Learning Model Surrogates

The state of the art in coupling dust coagulation to hydrodynamical simulations of protoplanetary disks are 2D simulations with ∼100 particle size bins. Many problems in planet formation, such as the streaming instability, are 3D in nature and require extremely high resolution in size space. Coupling 3D hydrodynamics with coagulation requires going beyond the current state of the art, by harnessing the power of machine learning (specifically neural differential equations) to obtain an efficient and accurate subgrid model for dust collisions. This will for the first time allow for a full treatment of the dust component, including coagulation, in simulations of the polydisperse streaming instability and pebble accretion.