Ma, Teng; (2025) Investigation of Structural and Thermodynamic Properties of Gold and Its Alloys Using State-of-the-Art Machine Learning Force Fields. Doctoral thesis (Ph.D), UCL (University College London).

Text Ma_10214851_Thesis.pdf Access restricted to UCL open access staff until 1 November 2026. Download (8MB)

Abstract

Gold, traditionally considered catalytically inert in its bulk form, exhibits high activity when structured as nanoparticles or alloys. The catalytic properties of these nanoparticles are strongly size-dependent, creating a demand for computational tools capable of accurately modelling systems of varying dimensions. While ab initio methods, such as Density Functional Theory (DFT), provide high accuracy, their computational cost restricts their application to small systems and short timescales. In contrast, empirical potentials can simulate larger systems but often sacrifice accuracy. Machine learning force fields (MLFFs) offer a promising compromise, delivering computational speed comparable to empirical methods while approaching the accuracy of ab initio calculations. However, developing robust MLFFs for metallic nanoparticles is challenging due to the vast structural diversity arising from defects, surface motifs, internal grain boundaries, and the existence of numerous isomers for a given size. In this doctoral research, a kernel-based MLFF approach was selected over neural network alternatives due to its superior utility in guiding the selection of critical training structures. The training strategy combined active learning with offline methods. One cluster ablation algorithm was developed to enhance training efficiency by extracting local atomic environments within a defined cutoff radius from large nanoparticles. The resulting MLFF was rigorously validated by accurately predicting key structural and thermodynamic properties of gold nanoparticles. Furthermore, the methodology was successfully extended to AuPd bimetallic systems. The force field demonstrated its predictive power by correctly modelling surface segregation phenomena in AuPd nanoparticle structures over 5 nm which is consistent with prior experimental observations.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item