Machine learning (ML) is a powerful, fast, and low-cost tool being implemented in many fields, including catalyst invention and materials discovery, to accelerate these processes and lower the time and cost for simulation using traditional approaches such as the Finite Element Method (FEM). In addition, density functional theory (DFT) and time-dependent DFT (TD-DFT) calculations are powerful tools for rationalizing photocatalyst design. In this thesis, we first demonstrate the use of ML tools to predict the electric field, which is needed to rapidly and accurately determine the plasmonic enhancement from silica-coated gold nanoparticles, and then use the predictions as a guide for designing core-shell gold nanoparticles. Directed by the feature analysis, we successfully synthesized the...[ Read more ]

Machine learning (ML) is a powerful, fast, and low-cost tool being implemented in many fields, including catalyst invention and materials discovery, to accelerate these processes and lower the time and cost for simulation using traditional approaches such as the Finite Element Method (FEM). In addition, density functional theory (DFT) and time-dependent DFT (TD-DFT) calculations are powerful tools for rationalizing photocatalyst design. In this thesis, we first demonstrate the use of ML tools to predict the electric field, which is needed to rapidly and accurately determine the plasmonic enhancement from silica-coated gold nanoparticles, and then use the predictions as a guide for designing core-shell gold nanoparticles. Directed by the feature analysis, we successfully synthesized the predicted core-shell structures and confirmed the ML predictions by directly mapping the localized surface plasmon resonance (LSPR) using atomic resolution scanning transmission electron microscopy-electron energy loss spectroscopy (STEM-EELS). We found that nanoparticles with 14 nm silica shell on the 16±5 nm gold core, immobilized with sensitizers, demonstrate a ~3-fold increase in conversion for the photocatalytic oxygenation of anthracene concomitant with a 15% increase in selectivity to anthraquinone. Similarly, for photooxygenation of dihydroartemisinic acid (DHAA), we observed a 4% increase in the selectivity of the desired drug. The combination of ML, experiment, and NSET theory shows the synergetic effect of plasmonic enhancement and fluorescence quenching, leading to a universal equation for TOF enhancement for the generation of ¹O₂. Our results from TD-DFT calculations suggest that the presence of an electric field (obtained from ML) can favor the intersystem crossing of methylene blue to enhance ¹O₂ generation.

We then demonstrate the use of DFT and TD-DFT calculations to rationally design single and dual atom catalysts (SACs and DACs) for photocatalytic applications. From DFT and TD-DFT calculations, we found that the synergetic effect of triplet sensitization (through intersystem crossing (ISC) energy transfer) and triplet-triplet energy transfer (through Dexter energy transfer) plays a significant role in the photocatalytic activity of both SACs and DACs, leading to a universal equation towards ¹O₂ generation. FeN4 SAC and FeNiN8 DAC exhibit a low ISC energy gap (ΔE_ISC) of 0.039 eV and 0.108 eV while possessing a high Bader charge transfer of 0.366 e^─ and 0.405 e^─, respectively, indicating their promising application for photocatalytic reactions. Inspired by the DFT and TD-DFT calculations, N-doped, FeN4, CoN4, NiN4, and FeNiN8 SACs were synthesized, and their photocatalytic activity was measured towards ¹O₂ generation. XANES was used to confirm the electronic and atomic structure of SACs, while LEIPS and UV-Vis were used to measure the conduction band (CB) and valance band (VB) of samples, indicating that introducing metal atoms can enlarge their optical bandgap. Subsequently, synthesized SACs were used for the photooxygenation of anthracene and two-electron oxygen reduction reaction (ORR) to measure their photocatalytic activity. FeN4 SAC demonstrates a high reaction conversion of 86% and a high ¹O₂ quantum yield of 1.04, obtained from ESR spectroscopy. In contrast, NiN4 SAC exhibits a high H₂O₂ production because of its high OOH* adsorption energy, obtained from machine learning (ML). This thesis provides a ML- and DFT-guided strategy for designing SACs, DACs, and plasmonic catalysts for photocatalytic applications.