ABSTRACT:

Purpose: Finding and characterizing Earth-like exoplanets that could harbor life is a key decadal priority for astrophysics [1]. The composition and chemical makeup of planets is imprinted in their building material, the disks of gas, dust, and ices around young stars [2]. Recent observations from flagship observatories, like the James Webb Space Telescope, has shown diverse and rich organic chemistries around young stars [3]. This diversity is inherited from the galactic environment over millions of years, but the complexity of the process challenges simple chemical models, which cannot account for the dynamical nature of the process.

Methods: Embedding chemical histories in a digital twin of the Milky Way, I can investigate how chemical variability emerges in different galactic environments. Three outstanding challenges motivate the proposed methodology:
(i) chemical networks require high temporal resolution physical histories that are not available in the digital twin;
(ii) chemistry must be traced over millions of years from the sterile interstellar medium to the rich chemistry around young stars;
(iii) the factors in the galactic environment that condition the observed chemical complexity are unknown
To address (i) I will develop a physics-informed, learned super-resolution method to construct sub-grid histories from sparse data sets. A stochastic generative approach utilizing Diffusion Schrödinger Bridge models will ensure statistically consistent physical structure [4]. For (ii), I will use the largest model world-wide of a region in the Milky Way, encompassing hundreds of starforming complexes and tens of thousands of young stars evolving over 30 million years. To tackle (iii), I will use hierarchical Bayesian population modeling to determine which galactic conditions and evolutionary stages that sets the chemical composition of the solids which eventually assemble into planets.

Impact: This project will provide fundamental insights into the influence of galactic environments for the chemistry around young stars and the physical processes that drives the observed diversity, which can help inform future observational strategies for selecting young planetary nurseries. In addition, the novel machine-learning framework for reconstructing flow histories in fluid dynamic models from sparse data will be broadly applicable across scientific domains using fluid dynamics as a modelling framework.