From Code to Cloud: Machine Learning Approaches to Complex Interstellar Chemistry 92 views

By gathering a complete picture of the chemistry within astronomical source, we can infer a wide array of chemical and physical properties, such as source structure and chemical morphologies. Molecules act as tracers to highlight processes occurring in astrochemically relevant sources, linking our understanding of these environments to the completeness of their chemical inventories through the use of observed molecules and their abundance upper limits. Disentangling interstellar complexity is reliant on the detection of new molecules, as each new discovery develops our understanding of chemical and physical evolution. However, such detections are becoming increasingly challenging as chemical complexity grows combinatorially, often revealing gaps in our fundamental understanding of these processes. Rather than replace traditional experimental, modeling, and observational methods, machine learning has demonstrated promise as a complement to existing efforts. In this thesis, I will demonstrate the utility of machine learning methods in addressing current challenges in astrochemistry through a series of proof-of-concept applications. In Chapter 2, I leverage a pre-trained model to provide abundance predictions of a recently discovered interstellar anion and radical counterpart. I directly compare these results with predictions derived from a chemical modeling code to highlight the efficiency of machine learning approaches. Chapter 3 extends a modified pipeline previously used in the characterization of the quiescent dark cloud TMC-1, to the chemically diverse star-forming region Orion KL to reproduce abundances and provide counterfactual base predictions. In Chapter 4, I explore various dimensionality reduction techniques across several hyper-dimensional ALMA data sets, assessing their effectivess in compressing data into more manageable data products for astrochemical and astronomical use case. I conclude this thesis by outlining future directions in latent space exploration for astrochemical morphology studies and potential implications in light of upcoming observatory upgrades.

PHD (Doctor of Philosophy)

astrochemistry; machine learning; radio astronomy

English

2025-08-06