Rational Design of Efficient Catalysts using First-Principles and Machine Learning
Rising concerns about the impending energy crisis and environmental impact have garnered tremendous research interest in developing catalysts for sustainable fuel production and pollution abatement. However, the traditional catalysts such as metals, alloys, or metal oxides usually lack in one of the figures of merit of catalysis, i.e., activity, selectivity, stability, or cost, limiting their wide-scale applications. Even after extensive research into improving the figures of merit through several trial-and-error approaches such as nanostructuring, alloying, and using promoters and co-catalysts, the inherent catalytic performance has not reached the desired levels. Hence, there is a great urge for the rational search of entirely new class of catalysts. Using first-principles and machine learning (ML), we explored new efficient catalysts by devising design principles based on the electronic structure of the materials. The current thesis primarily investigates two-dimensional (2D) materials, which have gained immense attention for a wide variety of applications, led by the discovery of graphene. However, graphene has a zero-band gap, which can be opened by heterostructuring with other 2D semiconductors. The sublattice symmetry breaking is quantified through a simple counting scheme to evaluate a universal criterion for the band gap opening in graphene-based heterostructures. Apart from Dirac cone splitting, the Dirac cone shift is also assessed to check if the semiconducting nature of the graphene heterostructures is preserved. A tight-binding analysis further supports the proposed scheme. The concepts developed in this study are extended towards finding a new photocatalyst composed of a heterostructure of 2D semiconductors (C2N and WS2) for solar hydrogen generation application. The C2N/WS2 heterostructure has optimum band gap and band alignments suitable for the photocatalytic water splitting reaction. Moreover, it is a type-II heterostructure, making it benign for photocatalysis through electron-hole separation onto the two layers. It also exhibits other essential properties, such as high charge carrier mobilities, enhanced visible-light harvesting, and low HER overpotential, comparable to other state-of-the-art photocatalysts. Building on this work, we searched for highly efficient 2D photocatalysts from a large chemical space of the newly generated 3099 2D materials database. The screening is carried out in a high-throughput (HT) manner, where the first tier consists of finding highly stable 2D materials. The interpretable ML approach is employed to extract non-trivial physical insights from thermodynamic (formation energy and convex hull distance) and the overall stabilities of these 2D materials. The second and third levels of HT screening consist of selecting the most stable 2D materials that satisfy the fundamental requirements for a photocatalyst. The photocatalytic efficiencies of the selected 21 2D materials are also evaluated, with HfSe2 and ZrSe2 reaching the theoretical limit of the solar-to-hydrogen conversion efficiency. Further, hydrogen fuel can also be generated from formic acid oxidation, which is a much easier process. Moreover, methane and ethylene (products of CO2 reduction reaction) and ammonia (product of N2 reduction reaction) are also highly sought-after fuels and chemicals, usually generated through electrocatalysis. In this regard, the AgAuCuPdPt high-entropy alloy (HEA) was synthesized recently and exhibited unprecedented activity and selectivity for both formic acid oxidation and CO2 mitigation. In order to establish the observed trends in activity and selectivity compared to the standard catalysts for both reactions, the free energy profiles are generated. Here, the active sites are determined based on d-band center calculations. For nitrogen fixation, we explored single-atom catalysts (SACs) as possible electrocatalysts. Among 13 transition metal atoms anchored at the Mo-top position in MoS2, Co, Fe, and Ru SACs are found to be highly active and selective for nitrogen reduction reaction (NRR), outperforming the popular Ru(0001) catalyst. Further, new simple descriptors for the NRR activity are proposed in the form of charge transferred from the transition metal atoms to the dinitrogen intermediate and group number of the transition metal. The established key structure-activity relationships in this work are expected to significantly improve the rational search of stable and cost-effective catalysts with enhanced activity and selectivity.