Off-campus SJSU users: To download campus access theses, please use the following link to log into our proxy server with your SJSU library user name and PIN.

Publication Date

Summer 2023

Degree Type

Thesis - Campus Access Only

Degree Name

Master of Science (MS)

Department

Chemistry

Advisor

Gianmarc Grazioli; Alberto A. Rascón; Annalise Van Wyngarden

Abstract

The photodissociation of acetaldehyde occurs through the absorption of UV light, which decomposes the molecule to form radical products and there are four major observed pathways during this process. It is not well understood how the photodissociation of acetaldehyde starts from seemingly the same initial conditions, but ultimately forms different reaction products. This motivates our work probing the dynamics of these pathways using machine learning-guided ab initio molecular simulations. Support vector machine (SVM) classifiers were first trained to predict the reaction products from phase space data over different ranges of timesteps from the simulations. Analysis of the highest accuracy models begins by extracting the support vectors from the classifiers. The nearest neighboring support vectors are obtained from trajectories of different products that approached opposite sides of the decision boundary, which are ideal regions to search for transition states. Next, the Euclidean distance of the support vectors between all pairs of products are calculated to determine the nearest out-of-class neighboring support vectors. The trajectories contributing the most support vectors are identified in search of bounding points of potential transition states. Thus, we aim to build an artificial intelligence for probing the dynamics of reactive trajectories in phase space and perform quantitative analysis on it to gain interpretable data about the regions of phase space where the product differentiation occurs.

Download

Share

COinS