Work

Accelerating Catalyst Discovery Using Quantum Chemistry and Machine Learning

Year: 2026
Type: dissertation
Abstract: This thesis presents a series of computational studies that combine quantum chemistry with machine learning (ML). The first focus is on using high-fidelity quantum chemistry to gain a deepened mechanistic understanding of experimentally reported catalysts. The second focus is on developing data-driven ML methods to accelerate the discovery of new materials. This work demonstrates the interplay between these two approaches: density functional theory (DFT) studies provide the physical insights needed to build robust ML models, while ML provides the speed to screen large chemical spaces that are beyond the reach of DFT alone.I use DFT studies to explore and seek to explain the performance of two experimentally proven, high-performing catalysts for the oxygen evolution reaction (OER). For Ba-doped Co3O4, DFT calculations showed that Ba doping stabilizes the active CoOOH surface and opens a more efficient reaction pathway. For the highly active and durable Mn0.75Ru0.25O2 catalyst, my DFT work showed that an asymmetric Mn-O-Ru dual-site center creates a low-barrier reaction pathway. Charge transfer from Mn to Ru electronically stabilizes the catalyst and prevents it from dissolving, tackling the widely held trade-off between activity and stability. The second part of the thesis uses these findings to build a platform for accelerated catalyst discovery. I created a novel ML workflow that uses a quantum-inspired similarity metric as a key feature in a crystal graph convolutional neural network (CGCNN). This similarity-informed approach doubled the data efficiency for predicting adsorption energies across the vast chemical space of high-entropy alloys. The model enabled me to identify a new, high-performance FeCoNiIrRu high-entropy alloy for the oxygen reduction reaction (ORR), a prediction that was then confirmed by experiments. By combining detailed mechanistic analysis with accelerated data-driven screening, this thesis demonstrates a powerful and practical method for the rational design of next-generation catalysts.
Author: Chang, Yuxin
Language: English
Open Access status: green
OpenAlex ID: W7203959065
Cites: 105
Cited by:
Related to: 10