Machine Learning Estimation of Reaction Energy Barriers and its Applications in Astrochemistry
| dc.contributor.advisor | Fournier, Rene Andre | |
| dc.contributor.author | Ji, Hongchen | |
| dc.date.accessioned | 2024-03-18T18:15:41Z | |
| dc.date.available | 2024-03-18T18:15:41Z | |
| dc.date.issued | 2024-03-16 | |
| dc.date.updated | 2024-03-16T10:43:21Z | |
| dc.degree.discipline | Chemistry | |
| dc.degree.level | Doctoral | |
| dc.degree.name | PhD - Doctor of Philosophy | |
| dc.description.abstract | We developed a machine learning model for fast estimating reaction energy barriers. The model was trained on data for 11,730 elementary reactions and barriers computed with an estimated accuracy of 2.3 kcal/mol by Grambow and coworkers. Although it was trained, and then applied, only for reactions involving atoms of H, C, N and O, our model can readily be generalized to molecules and reactions involving other elements. We designed and tested many molecular representations. Our best model has 363 features calculated from the chemical composition, structure, and energy of products and reactants that can all be obtained at a small computational cost. A Kernel Ridge Regression with Laplacian kernel was found to give the best fit to the data. It makes predictions with a mean absolute error of 4.1 kcal/mol for barriers smaller than 40 kcal/mol. We used this machine learning model to estimate the barriers of 136,081 hypothetical association reactions between molecules known to exist in the circumstellar envelopes and interstellar medium, where temperatures range between 10 and 150 K. A screening procedure identified reactions likely to occur (those with near zero barriers) that could play a role in the formation of relatively complex organic molecules (those with at least one double bond and containing at least one atom each of the elements H, C, N and O). Reactions identified as the most promising were investigated with density functional theory and coupled cluster quantum chemical methods to obtain reaction pathways and energies of reactants, intermediates, transition states and products with an accuracy of roughly 1 kcal/mol. We found no barrierless reaction but found two reactions with low barriers leading to the formation of N-methyleneformamide (CH2NCHO) and imine acetaldehyde (NHCHCHO). These molecules have not yet been observed in space. The low barriers that we calculated, and the possibility that these reactions would be facilitated by adsorption on ice covered dust grains in the interstellar medium, suggest that these two molecules may exist in space and could be detected by determinations of rotational constants from spectroscopic lines in the microwave region. | |
| dc.identifier.uri | https://hdl.handle.net/10315/41959 | |
| dc.language | en | |
| dc.rights | Author owns copyright, except where explicitly noted. Please contact the author directly with licensing requests. | |
| dc.subject | Chemistry | |
| dc.subject | Physical chemistry | |
| dc.subject.keywords | Theoretical chemistry | |
| dc.subject.keywords | Computational chemistry | |
| dc.subject.keywords | Reaction energy barrier estimation | |
| dc.subject.keywords | Machine learning | |
| dc.subject.keywords | Interstellar medium (ISM) and circumstellar envelope (CSE) | |
| dc.subject.keywords | Kernel ridge regression | |
| dc.title | Machine Learning Estimation of Reaction Energy Barriers and its Applications in Astrochemistry | |
| dc.type | Electronic Thesis or Dissertation |
Files
Original bundle
1 - 1 of 1