Radiative Transfer Modeling of a Shadowed Protoplanetary Disk Assisted by a Neural Network

We present observations and detailed modeling of a protoplanetary disk around the T Tauri star V1098 Sco. Millimeter-wavelength data from the Atacama Large Millimeter/submillimeter Array (ALMA) show a ring of large dust grains with a central cavity that is filled with molecular gas. Near-infrared data from the Very Large Telescope (VLT) detect the scattered starlight from the disk surface and reveal a large shadow that extends over its entire southern half. We model the ALMA continuum and line data to determine the outer-disk geometry and the central stellar mass. Using radiative transfer models, we demonstrate that a misaligned inner disk, tilted in both inclination and position angle with respect to the outer disk, can reproduce the salient scattered-light features seen with VLT. Applying an image threshold algorithm to compare disk morphologies and training a neural network on a set of high-signal-to-noise-ratio models, we forward model the data and determine the inner-disk geometry. We find that the rotation axes of the inner and outer disks are misaligned by 38° and constrain the mass and location of a perturbing planetary or substellar companion. The technique of simulation-based inference that is illustrated here is broadly applicable for radiative transfer modeling of other objects.