Back to search

Georgia Institute of Technology

Advancing on-the-fly Machine Learned Force Fields and Leveraging Electronic Structure Theory for Computational Catalysis

Abstract

dc:description.abstract

Discovering better heterogeneous catalysts to reduce the energy footprint of the chemical industry requires searching a massive chemical space. Density functional theory calculations are too expensive to fully traverse this space, and existing machine learned potentials are not sufficiently general to provide a robust replacement. This thesis focuses on developing robust Bayesian models for uncertainty prediction to guide continuous and autonomous model development. First, neural network potentials built on Gaussian multipole descriptors were shown to fail badly when no example of a held-out element is present despite the universal descriptor design. A Cu/BaTiO\textsubscript{3} CO\textsubscript{2}-hydrogenation catalyst illustrates the tradeoff between electronic structure and machine learning for a complex material. An on-the-fly force field built with the Gaussian multipole descriptors was shown to capture complex potential energy surfaces efficiently, but the variance from Bayesian linear regression models was found not to be a robust estimate of uncertainty. This motivated a ground-up diagnosis on minimal systems which revealed that a misspecified homoscedastic noise model cannot represent deterministic, spatially structured residuals. Finally, Bayesian linear regression was shown to be unsuitable for active learning because its variance collapses away from the training data. In aggregate, the work provides a roadmap for autonomous and self-improving atomistic modeling workflows.

Degree

thesis:*
Name thesis:degree_name
Chemical Engineering, PhD
Department thesis:degree_department
Chemical and Biomolecular Engineering
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Timmerman, Lucas
Advisor dc:contributor.advisor
  • Medford, Andrew J.
Committee members dc:contributor.committeemember
  • Suryanarayana, Phanish
  • Thomas E. Gartner
  • Vassiliki-Alexandra Glezakou
  • Realff, Matthew
  • Yang, Julia
  • Fung, Victor

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/82212

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Timmerman, Lucas. Advancing on-the-fly Machine Learned Force Fields and Leveraging Electronic Structure Theory for Computational Catalysis. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/82212