Machine learning-accelerated chemistry modeling of protoplanetary disks

TL;DR

This paper introduces a machine learning approach using KNN regression to rapidly predict the chemical composition of protoplanetary disks, enabling efficient analysis of observational data.

Contribution

It presents a novel application of machine learning to accurately model disk chemistry using minimal physical parameters, improving speed and efficiency.

Findings

KNN can accurately predict disk chemistry from limited physical data

Correlations in physical conditions enable simplified modeling

Method can be integrated into Bayesian fitting procedures

Abstract

Aims. With the large amount of molecular emission data from (sub)millimeter observatories and incoming James Webb Space Telescope infrared spectroscopy, access to fast forward models of the chemical composition of protoplanetary disks is of paramount importance. Methods. We used a thermo-chemical modeling code to generate a diverse population of protoplanetary disk models. We trained a K-nearest neighbors (KNN) regressor to instantly predict the chemistry of other disk models. Results. We show that it is possible to accurately reproduce chemistry using just a small subset of physical conditions, thanks to correlations between the local physical conditions in adopted protoplanetary disk models. We discuss the uncertainties and limitations of this method. Conclusions. The proposed method can be used for Bayesian fitting of the line emission data to retrieve disk properties from observations. We present a pipeline for reproducing the same approach on other disk chemical model sets.