DOI
https://doi.org/10.25772/FECM-GQ07
Author ORCID Identifier
https://orcid.org/0000-0002-8398-8178
Defense Date
2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy
Department
Chemistry
First Advisor
Ka Un Lao
Abstract
The routine use of highly accurate computational chemistry methods is impeded by their inherent computational cost. Consequently, large-scale investigations or computations involving large systems often necessitate more computationally efficient methods, at the expense of accuracy. To achieve a more desirable trade-off between computational cost and accuracy, we leverage recent advancements in machine learning to integrate electronic structure theory methods with machine learning models. Long-range corrected (LRC) or range-separated (RS) density functionals with correct asymptotic behavior have a propensity to diminish both the delocalization and localization errors where the length of the range-separation is determined by the range-separation parameter, which is traditionally tuned non empirically. The tuning procedure typically necessitates a series of SCF calculations, preventing the applications of these methods to large systems. A machine learning model is developed to bypass the ab initio system-specific tuning procedure. The new method, LRC-wPBE(wgddML) is used to predict polarizabilities for a series of oligomers and surpasses the performance of all other popular density functionals, where polarizabilities are sensitive to the asymptotic density decay and are crucial in a variety of applications including the calculations of widely-used dispersion corrections and refractive index. In addition, $\Delta$-machine learning (ML) models are trained based on newly developed intermolecular features, which are derived from intermolecular histograms of distances for element/substructure pairs and rely solely on Cartesian coordinates, to simultaneously account for local environments as well as long-range correlations, are also developed to address deficiencies of the D3/MBD models, including the inflexibility of their functional forms, the absence of many-body dispersion contributions in D3, and the standard Hirshfeld partitioning scheme used in MBD. Then, we utilize our recently developed machine learning corrected ab initio dispersion (aiD) potential, coined D3-ML, to address the dispersion deficiencies in second-order Moller-Plesset perturbation theory (MP2) by replacing its problematic dispersion and exchange-dispersion terms with D3-ML. This leads to the development of a new dispersion-corrected MP2 method, MP2+aiD(CCD), which outperforms other spin-component-scaled and dispersion-corrected MP2 methods as well as popular ML models for predicting noncovalent interactions on diverse dataset and large systems.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
5-9-2024