Theoretical calculation and machine learning aided design of functional materials for energy conversion

Abstract

This thesis investigates the integration of machine learning (ML) and theoretical calculations to design and optimize functional materials for photocatalytic applications. By combining experimental techniques with theoretical calculations, that is, finite-difference time-domain (FDTD) simulations, and density functional theory (DFT) calculations, this work aims to accelerate the discovery of efficient, selective, and scalable photocatalytic systems for CO2 reduction and seawater splitting. The central focus is on leveraging ML and advanced simulations into experiments to provide new insights into plasmonic photocatalysts and microenvironmental perturbations in photoreaction. The first study explores the development of Ag-TiO2 core-shell photocatalysts for the selective reduction of CO2 to methane (CH4). A significant contribution of this work is the use of FDTD simulations to model and optimize microenvironmental perturbations, thereby enhancing the catalytic activity of the plasmonic core-shell nanoparticles. Additionally, DFT simulations demonstrate that localized surface plasmon resonance (LSPR)-induced electric field enhancements lower the energy barriers for CO2 activation and methanation. Experimentally, this system achieves 100% selectivity for CH4 with a production rate of 75 umol/g/h. This study emphasizes the advantages of microenvironmental engineering in optimizing photocatalytic activity and selectivity, with FDTD and DFT simulations further elucidating the mechanisms of microenvironmental perturbations. The second study focuses on the design of Co-NC@Cu core-shell photocatalysts for solar-driven hydrogen production from seawater. By dispersing single Co atoms on a nitrogen-doped carbon (NC) shell surrounding a Cu core, this novel catalyst achieves a hydrogen production rate of 9080 umol/g/h and a solar-to-hydrogen (STH) conversion efficiency of 4.78%. A key highlight of this work is the detailed investigation of the local coordination environment of the single Co atoms, as well as the thermodynamic and kinetic effects of electric field perturbations on the catalytic process. DFT calculations reveal that the single Co atoms act as highly active sites for hydrogen evolution, exhibiting low energy barriers for the reaction. Furthermore, the electric field's role in enhancing the reaction thermodynamics and kinetics was elucidated, providing insights for further optimization of catalytic performance. Integrating single atoms, photothermal effects, and localized surface plasmon resonance (LSPR) demonstrates a robust and efficient design for seawater splitting. The third study showcases a comprehensive workflow combining ML and DFT calculations to accelerate the discovery and optimization of single-atom-based (SA) 2D photocatalysts. Using a dataset of Janus-TMD materials as a case study, ML models were trained to identify high-activity catalytic sites and screen potential substrates for photocatalytic CO2 reduction. The ML-driven predictions successfully prioritized optimal single-atom catalysts, with experimental validation confirming the activity and selectivity of two synthesized Janus substrates MoOSe with single-atom Pt. Photocatalytic experiments demonstrated the potential of the ML-guided design in delivering efficient and selective catalysts, underscoring the synergy between computational and experimental approaches. The growing dataset of atomic structures, intermediates, Janus configurations, and adsorption models provides a robust foundation for refining ML models and driving innovations in SA-based 2D materials discovery. In conclusion, this thesis demonstrates the successful integration of ML, FDTD, and DFT techniques with experimental approaches for the design of advanced functional materials, which contribute to the development of sustainable energy solutions through CO2 reduction and hydrogen production. Show more