Using Machine Learning to infer planet mass from observed gap in protoplanetary disks

The observed gaps in the protoplanetary disk are often considered an imprint of planets orbiting around the central star. The width and the depth of the gaps depend on the mass of the planet along with the disk properties, for example, aspect ratio, viscosity, and particle size. An estimate of the masses and the location of the planets forming in the disk is crucial to distinguish between different planet formation scenarios like “core accretion” and “gravitational instability” models. We run a large number of two-dimensional hydrodynamic simulations in FARGO3D using GPU clusters to compute the gaps induced by planets in a dusty disk for a wide parameter range. The dataset is then used to train a Deep Neural Network to predict the planet mass from an observed disk gap in a protoplanetary disk. This machine learning technique provides a significant advantage over the existing empirical relations as our model can be trained for any number of relevant parameters and complex disk system. Our trained neural network provides an accurate prediction of the planet mass for an observed gap in a (for example, HL Tau) disk in much-reduced computing time.