Abstract
Nanoparticles are collections of atoms in the size range of 1--500 nm that exhibit unique and size dependent properties, distinct from bulk materials. This lends nanoparticles to a wide variety of applications, such as in medicine and in industry, including applications in catalysis towards energy storage. Because the specific properties of a nanoparticle depends on its size and precise atomic structure, being able to predict the structure of a nanoparticle is prerequisite to predicting its properties. This thesis explores the computational modelling of nanoparticles, both towards structural elucidation with global optimisation algorithms, as well as modelling catalytic activity of nanoparticles.
First, two new methods are described that aid in the development of global optimisation techniques. Specifically, a scheme for classifying what type of structure a nanoparticle adopts is devised, based upon the expansion of an existing classification scheme to define new classes relating to icosahedral nanoparticles. Being able to classify nanoparticle structures aids in examining of the behaviour of global optimisation algorithms, and the types of structures the search discovers. The second method allows for the quicker estimation and benchmarking of global optimisation algorithms by applying a statistical model from survival analysis. In this way, the average performance of global optimisation algorithms for structural prediction can be estimated without needing 100% of replicate trials to complete, greatly reducing the computational cost of benchmarking novel structural prediction methods.
Next, a prominent example of a nanoparticle global optimisation algorithm, the basin-hopping algorithm, is augmented to incorporate memory into the search with a blacklist of recently visited structures, towards preventing the algorithm from revisiting recently visited areas of the potential energy surface. Two different modes are programmed that differ in response to proximity to the blacklist-one mode restarts the search from a random structure upon getting to close to the blacklist, where the other simply rejects the move that achieves close proximity, and attempts a new move from the current position instead. The latter mode is shown to outperform the former for certain benchmarking systems, but for other systems both modes failed to enhance performance, due to the specific topographies of the potential energy surfaces of these cases.
A novel global optimisation algorithm is developed that applies for the first time the divide-and-conquer paradigm to nanoparticle global optimisation in an a priori manner. This approach sees the potential energy surface divided into distinct regions to be searched separately and in isolation. This has the effect of preventing the search from focusing on one area of the search space for a disproportionate amount of time, and is shown to enhance the speed of structural elucidation for a variety of systems. For one specific system, algorithmic performance was not enhanced as the initial exploration of the potential energy surface was insufficient for describing the wider variety of structures present for this case, highlighting how the algorithm could be further improved.
Finally, a catalytic study of Cu nanoparticles with irregular structures obtained from global optimisation is performed. Modelling suggests these irregular Cu nanoparticles exhibit enhanced activity for the electrochemical CO2 reduction reaction, however selectivity for this reaction over the competing hydrogen evolution reaction must also be considered. The hydrogen evolution reaction is modelled on the irregular nanoparticles, and activity compared to more commonly studied regular structures, as well as bulk Cu surfaces. The irregular structures exhibited unique activity compared to the regular systems, highlighting the need to obtained detailed nanoparticle structures. However, significant barriers still exist to adsorption of H and desorption of H2, suggesting a high selectivity for the CO2 reduction reaction.