Physics-informed machine learning for modelling defect-driven catalytic phenomena

Chaudhari, Amit 2025. Physics-informed machine learning for modelling defect-driven catalytic phenomena. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (81MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (211kB)

Abstract

The growing global demand for clean and sustainable energy requires rapid advances in the atomic-level understanding of critical materials, especially transition metals, rare-earth metals and their oxides. These materials form the foundation of industrial catalysts; however, it is very challenging to construct predictive models to optimise their performance. Density functional theory (DFT) is a core method in modern computational chemistry but is limited in terms of accuracy and scalability, which prohibits the simulation of complex electronic effects, length scales and disorder of real materials. This thesis aims to address these challenges through the application of computational methods and workflows that complement DFT and allow better understanding of experimentally observed metal oxide effects in catalysis. In Chapter 3, I investigate the simulation of electron polarons in Nb- and W-doped TiO2, which are important materials for photocatalysis and solar cells but poorly understood at the atomic level. I apply Hubbard corrected DFT+ to accurately simulate these materials and show that careful determination of both the Hubbard value and projector is required to match experimental observations. Refinement of the Hubbard projector mitigates numerical instabilities and enables robust simulations across a wide range of materials, which is achieved using supervised machine learning in Chapter 4. In Chapter 5, I examine the role of metal oxide supports in enhancing the sulfur tolerance of Ni-based methane steam reforming catalysts. Combining DFT+, grand canonical Monte Carlo sampling and machine learned interatomic potentials, I probe oxygen buffering and bulk phase transformations that govern sulfur oxidation and catalyst regeneration. The simulations rationalise experimental observations across different supports, linking atomic-scale defect chemistry with macroscopic catalytic performance. Overall, this work demonstrates how refined DFT+ methods, machine learning and multiscale modelling can provide a pathway to more comprehensive simulations of complex catalytic materials that are far beyond current capabilities.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Chemistry
Date of First Compliant Deposit:	3 March 2026
Last Modified:	03 Mar 2026 16:40
URI:	https://orca.cardiff.ac.uk/id/eprint/185423

Actions (repository staff only)

Edit Item