Exciton diffusion in amorphous organic semiconductors: Reducing simulation overheads with machine learning.

Simulations of exciton and charge hopping in amorphous organic materials involve numerous physical parameters. Each of these parameters must be computed from costly ab initio calculations before the simulation can commence, resulting in a significant computational overhead for studying exciton diffusion, especially in large and complex material datasets. While the idea of using machine learning to quickly predict these parameters has been explored previously, typical machine learning models require long training times, which ultimately contribute to simulation overheads. In this paper, we present a new machine learning architecture for building predictive models for intermolecular exciton coupling parameters. Our architecture is designed in such a way that the total training time is reduced compared to ordinary Gaussian process regression or kernel ridge regression models. Based on this architecture, we build a predictive model and use it to estimate the coupling parameters which enter into an exciton hopping simulation in amorphous pentacene. We show that this hopping simulation is able to achieve excellent predictions for exciton diffusion tensor elements and other properties as compared to a simulation using coupling parameters computed entirely from density functional theory. This result, along with the short training times afforded by our architecture, shows how machine learning can be used to reduce the high computational overheads associated with exciton and charge diffusion simulations in amorphous organic materials.

Development of a Structural Comparison Method to Promote Exploration of the Potential Energy Surface in the Global Optimization of Nanoclusters.

A structural comparison method (SCM) was created to quantify the structural diversity of nanoclusters and was implemented into a global optimization algorithm to evaluate structural diversity between generated clusters on the fly and promote exploration of the potential energy surface. The SCM evaluated topological differences between clusters using the common neighbor analysis and provided a numerical measure of similarity between the two clusters. The SCM was implemented into a genetic algorithm by integrating it into a new structure + energy fitness operator such that structural diversity of clusters in the population and their energies were used to assign fitness values to clusters. The efficiency of the genetic algorithm with this new fitness operator was benchmarked against several Lennard-Jones clusters (LJ38, LJ75, and LJ98) known to be difficult cases for global optimization algorithms. For LJ38 and LJ75, this new structure + energy fitness operator performed equally well or better than the energy fitness operator. However, the efficiency of locating the global minimum of LJ98 decreased using this new structure + energy fitness operator. Further analysis of the genetic algorithm with this fitness operator showed that the algorithm did indeed promote exploration of the PES of LJ98 as desired but hindered refinement of clusters, preventing it from locating the global minimum even if the energy funnel of the global minimum had been located.

A theoretical investigation of 38-atom CuPd clusters: the effect of potential parameterisation on structure and segregation.

Understanding the structure of bimetallic clusters is increasingly important due to their emerging practical applications. Herein we investigate the structure of 38-atom CuPd clusters using a genetic algorithm with cluster energies described by the semi-empirical Gupta potential. Selected clusters are then refined with density functional theory. Three different parameterisations of the Gupta potential are used and their performance assessed to understand what features of bulk and surfaces are necessary to capture for accurate description of small clusters. Three general regions of motif stability exist; for the Pd majority clusters (Pd38 to Cu4Pd34) the truncated octahedron is most stable, while for clusters of intermediate compositions (Cu5Pd33 to Cu25Pd13) a "pancake" icosahedron is most stable, and for the Cu majority clusters (Cu26Pd12 to Cu38) again the truncated octahedron is most stable. CuPd clusters tend to segregate to a Cu-core, Pd-shell structure if possible, and at higher Cu compositions, the Pd segregates to the faces of the cluster. Using multiple parameterisations of the Gupta potential ensures the full variety of possible structures is found, and improves the search for the most stable CuPd clusters.

Excited States of Triphenylamine-Substituted 2-Pyridyl-1,2,3-triazole Complexes.

A new 2-pyridyl-1,2,3-triazole (pytri) ligand, TPA-pytri, substituted with a triphenylamine (TPA) donor group on the 5 position of the pyridyl unit was synthesized and characterized. Dichloroplatinum(II), bis(phenylacetylide)platinum(II), bromotricarbonylrhenium(I), and bis(bipyridyl)ruthenium(II) complexes of this ligand were synthesized and compared to complexes of pytri ligands without the TPA substituent. The complexes of unsubstituted pytri ligands show metal-to-ligand charge-transfer (MLCT) absorption bands involving the pytri ligand in the near-UV region. These transitions are complemented by intraligand charge-transfer (ILCT) bands in the TPA-pytri complexes, resulting in greatly improved visible absorption (λmax = 421 nm and ϵ = 19800 M-1 cm-1 for [Pt(TPA-pytri)Cl2]). The resonance Raman enhancement patterns allow for assignment of these absorption bands. The [Re(TPA-pytri)(CO)3Br] and [Pt(TPA-pytri)(CCPh)2] complexes were examined with time-resolved infrared spectroscopy. Shifts in the C≡C and C≡O stretching bands revealed that the complexes form states with increased electron density about their metal centers. [Pt(TPA-pytri)Cl2] is unusual in that it is emissive despite the presence of deactivating d-d states, which prevents emission from the unsubstituted pytri complex.

Calculations of the effect of catalyst size and structure on the electrocatalytic reduction of CO2 on Cu nanoclusters.

The structure and catalytic properties of Cu nanoclusters of sizes between 55 and 147 atoms were examined to understand if small Cu clusters could provide enhancement over traditional catalysts for the electrocatalysis of CO2 to CO and carbon-based fuels, such as CH4 and CH3OH, compared to bulk Cu surfaces and large Cu nanoparticles. Clusters studied included Cu55, Cu78, Cu101, Cu124, and Cu147, the structures of which were determined using global optimisation. The majority of Cu clusters examined were icosahedral, including the perfect closed-shell, partial-shell, elongated and distorted icosahedral clusters. Free energy diagrams for the reduction of CO2 showed the potential required for the formation of CO is notably smaller for all cluster sizes considered, relative to Cu(111). Less variation is observed for the limiting potential for the formation of CH4 and CH3OH. However, it was found that clusters that are either a distorted motif or contain vacancy defects yielded the best activity and provide an interesting synthesis target for future experiments.

Contrasting motif preferences of platinum and gold nanoclusters between 55 and 309 atoms.

The atomic structure of size-selected Pt clusters in the range 10-600 atoms is investigated with aberration-corrected scanning transmission electron microscopy and reveals significantly different behaviour from the existing data for Au clusters. The Pt clusters show a dominance of the FCC motif from relatively small sizes, whereas traditionally for Au multiple motifs - the icosahedron, decahedron and FCC motifs (and related structures) compete. The new data motivates a comprehensive computational investigation to better understand similarities and differences in the structures and energetics of the two different metallic clusters. Low energy structures of Pt and Au clusters with 55, 101, 147, 228 and 309 atoms (±2%) are identified using a global optimisation algorithm, and the relative energies found by local minimisation using density functional theory. Our computational results support the experimental observations; for Au clusters all motifs are comparably stable over the whole size range, whereas for Pt, the motifs only compete at the smallest sizes, after which the FCC motif is the most stable. Structural analysis suggests the greater tendency of Au towards amorphisation enables the icosahedron and decahedron to remain competitive at larger sizes.