Efficient screening of nanoclusters as catalysts for the hydrogen evolution reaction

Heterogeneous catalysis is a key component in modern industry as catalyst breakthroughs improve existing or accommodate the emergence of new technologies. For instance, catalyzing the splitting of water for energy storage purposes efficiently and cheaply is a potentially disrupting innovation. Nanoclusters have the potential to replace existing catalysts due to their catalytic behaviour at the nanoscale. However, experimental testing is often slow and expensive, and focuses on gradual improvements of known catalysts, prohibiting the discovery of novel materials. Computer simulations offer a method to design a new catalyst from scratch, allowing nanoclusters to be screened efficiently for their catalytic activity. Existing screening methods are often designed for simple infinite surfaces, neglecting the shape and size effects of nanoclusters. The large search space of catalyst screening at the nanoscale also poses a challenge to computational screening methods. This dissertation deals with the development of new methods for efficient screening of nanoclusters, explicitly capturing size and shape effects. In particular, machine learning (ML) approaches were used to reduce computational cost and a large part of the work was devoted to the benchmarking of descriptors as a key step in ML. New nanocluster-adsorbate tools were developed, these are the efficient exploration of nanocluster configurations, the exclusion of redundant adsorption sites and a DFT-ML loop. The screening workflow was automated and connected to a database allowing for screening and management of large sets of data. The workflow was verified on the simple hydrogen evolution reaction, a key reaction to electrolytic water splitting, and a bimetallic dataset containing several compositions of Ti, Co, Fe, Ni, Cu and Pt was screened. The implementation of new tools was kept modular and the programming aspect of the work is captured in three packages which are all publicly available and benefit other computational materials science researchers. The developed tools are not restricted to the model reaction; they are kept general so that they can be applied to other catalytic reactions on nanoclusters of arbitrary shapes and sizes.