Collaborative Research: CDS&E: Theory-infused Neural Network (TinNet) for Nonadiabatic Molecular Simulations

Photoinduced electron-stimulated reactions are ubiquitous in nature, as exemplified by photosynthesis, which is driven by hot electrons excited by solar radiation. Solar-driven reactions also hold great potential for transforming critical chemical reactions related to sustainable energy, environmental pollution control, and low-carbon chemicals and fuels manufacturing, particularly when aided by catalysis. To this end, the project harnesses the combined power of quantum chemistry, machine learning, and ultrafast pump-probe experimental techniques to reveal insights that will improve the efficiency of photocatalyzed reactions, enabling the development of technologies that support clean energy and a cleaner environment. The project includes educational and outreach activities, such as STEM-related activities for K-12 students to encourage their curiosity, critical thinking, and synthesis of clues with domain knowledge of surface physics, materials chemistry, and mathematics. Additionally, the data from this project will be integrated into college-level courses that provide next-generation training to students in the fundamentals of computational and data-enabled science and engineering. The project explores a computational and data science approach to modeling nonadiabatic reaction dynamics on metal surfaces co-driven by electronic excitations. The study advances knowledge of selective bond activation for rational design of heterogeneous catalytic systems beyond Sabatier volcano limitations. The approach builds on recent advances in molecular simulations enhanced by machine learning (ML) potentials for rapid sampling of adiabatic ground states. In many adsorbate-substrate systems, however, electronic transitions play an important role in channeling energy from excited charge carriers to chemical bonds of key reaction intermediates, the process of which is nonadiabatic in nature. A data-efficient and interpretable modeling framework with explicit considerations of electron-phonon interactions will be developed for understanding nonadiabatic surface reactions, aiming to maximize quantum efficiency and bond selectivity of energy transfer processes. With the development of ML-enabled nonadiabatic molecular simulations, fundamental questions regarding how the structure/composition of electronically-excited nanoparticles, e.g., plasmonic metals, and physical characteristics of light stimuli that impact the efficiency and selectivity of bond activation can be tackled. The project is co-funded by the Catalysis program in the Chemical, Bioengineering, Environmental and Transport Processes (CBET) Division and the Chemical Theory, Models, and Computational Methods (CTMC) program in the Chemistry (CHE) Division. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.