Disentangling CO Chemistry in a Protoplanetary Disk using Machine Learning

Amina Diop, Ilse Cleeves, Dana Anderson, Jamila Pegues

Molecular abundances in protoplanetary disks are sensitive to the local physical conditions including the gas temperature, the radiation field, and cosmic rays. Physical conditions are often intertwined, which makes it challenging to understand the interplay between chemical and physical processes. We present an exploration of this complex interplay using machine learning techniques. This approach is tested on understanding the chemistry of CO, a commonly used gas-mass tracer in protoplanetary disks. We use a representative disk model and harness the diversity of chemical environments within the disk to train the machine learning algorithm. A regression model is built to understand how CO depends on the gas density, X-ray and cosmic ray ionization rate, gas temperature, and UV flux. The model suggests that there is a tendency toward CO depletion in the disk, with the cosmic ray ionization rate and gas temperature being the most important drivers of chemistry. We generate a model with a higher cosmic ray ionization rate and one with a higher initial C/O ratio and note that depletion appears to be more severe in these models. The former effect could be due to an increased production of ions that destroy CO, while the latter process creates an oxygen-poor environment that can enhance the lock up of carbon into C2H instead of CO. Finally, we study the behavior of our models at different time steps and observe that depletion exacerbates over time, potentially due to the buildup of species that destroy CO. Our results agree with previous work, which suggests that machine learning techniques can be used to understand complex chemical processes in a novel fashion.