Mr Jonathan Ting1, Professor Amanda Barnard1
1School of Computing, Australian National University, Canberra, Australia
A data-driven approach to materials design can accelerate the discovery and development of metallic nanoparticles, using synthetic (computational) databases that capture different structural, processing, and property features. In order to speed up chemical reactions, surface characteristics of nanoparticle catalysts are particularly important, but they are highly complex and typically include a variety of atomic configurations that will react differently. One of the greatest challenges in the rational design of metallic nanoparticles is to identify surface characteristics relevant to particular reactions that can be controlled during synthesis. Previous attempts to label the surface atoms are mostly based on simple domain-driven metrics that lack direct association with experimental observations.
We propose an entirely data-driven approach to label the surface atoms of metallic nanoparticles based on their similarities in high-dimensional space, which includes their local environment to identify more complex patterns. The method uses iterative label spreading, a semi-supervised machine learning method especially suited for clustering of small data sets with high dimensionality. This algorithm does not contain hyper-parameters or require input assumptions, and returns different numbers of clusters with varying sizes depending on the atomistic structure of the nanoparticles.
The model automatically clusters the surface atoms into groups that can be related to specific known reactions, allowing multiple reactions to be considered simultaneously without researcher intervention. The approach thus exhibits great utility as a way to label metallic nanoparticle surfaces. The method was demonstrated on synthetic palladium nanoparticles generated using molecular dynamics, an important catalyst with significant potential for chemical engineering.
Biography:
Mr Jonathan Ting is a PhD candidate at the Australian National University School of Computing, working on the application of machine learning techniques to the rational design of multimetallic nanoparticles as catalysts.