In the last few decades, with the increased water pollution and energy scarcity, there have been attempts to solve these issues by studying chemical reactions and designing new materials. For example, understanding the structural and chemical properties of heterogeneous interfaces is critical in many applications, including water treatment, gas storage, energy storage, or water splitting. In this aspect, ab-initio simulations have been a powerful tool for investigating structural and chemical properties such as charge transfer, chemical stability, hole/electron conduction, and reaction energetics within chemical materials. However, due to the substantial computational cost, large, complex chemical and material systems are challenging to calculate with current density functional theory (DFT) based quantum calculation tools. Therefore, with the help of machine learning (ML) techniques, we can accelerate the prediction of chemical properties and reaction dynamics of intricate systems by providing a large dataset obtained from DFT calculations to train the ML models. My dissertation is composed of two parts. In the first part, we utilize the first principles calculations to explore structure-property relationships, electronic structure, and reaction energetics of various systems. For instance, we study the hydrogen storage performance of metal-organic framework, the conductivity of DNA strands, and the electrooxidation of biomass on Cu,Co-spinel oxides. In the second part of my dissertation, we address machine-learning-assisted methods to accelerate the investigation of the structure-property relationship of materials for many extended systems. For example, we explore the bioactivities of perfluoroalkyl substances and X-ray absorption spectroscopy of disordered systems such as amorphous carbon systems. We propose the advantages of applying ML in computational chemistry and materials science research from these examples. First, ML accelerates exploring structure-property relationships. Second, ML can also be used to interpret complex systems with a large supercell which would be computationally expensive if we only rely on DFT calculations. In short, ML combined with first-principles calculations enables more efficient and effective investigation of larger systems.
