This project focuses on the development of advanced machine learning-based interatomic potentials (forcefields) for simulating molten salts. Traditional forcefields often fail to accurately capture the complex interactions in these systems, especially when transition metals are present. By leveraging machine learning techniques, we can create highly accurate models that learn from high-fidelity quantum mechanical data. These models are capable of predicting the forces between atoms with unprecedented accuracy, enabling large-scale molecular dynamics simulations that were previously intractable.

The primary goal is to enable reliable simulations of molten salt reactors and thermal energy storage systems. The forcefields developed in this project allow for the detailed study of structural and dynamic properties of these materials, such as viscosity, thermal conductivity, and diffusion coefficients. This work is crucial for understanding the corrosion mechanisms in next-generation nuclear reactors and for designing more efficient and safer energy systems. The insights gained from these simulations provide direct support to experimental efforts and accelerate the materials discovery process.

This research explores the use of Graph Neural Networks (GNNs) to predict the redox potentials of iron-based metal complexes. Redox potentials are a critical property for a wide range of applications, including catalysis, batteries, and biological systems. However, experimentally measuring or computationally calculating these values for a large number of candidate molecules is a time-consuming and expensive process. GNNs offer a promising alternative by learning the relationship between the molecular structure and its electronic properties.

By representing molecules as graphs, GNNs can effectively capture the intricate local and global chemical environments that determine the redox potential. This project aims to build a robust and accurate GNN model that can rapidly screen vast chemical spaces for promising candidates with desired electrochemical properties. The successful development of such a model will significantly accelerate the discovery of new catalysts and materials for energy storage, contributing to the advancement of sustainable energy technologies.

Mechanochemistry, the study of chemical reactions induced by mechanical force, is a rapidly growing field with applications in materials synthesis, pharmacology, and green chemistry. This project utilized reactive molecular dynamics (ReaxFF) simulations to investigate the fundamental molecular mechanisms underlying shear-driven mechanochemical reactions. By applying shear and pressure to molecular systems, we observed bond-breaking and bond-forming events in real-time, providing insights that are often difficult to obtain from experiments alone.

The simulations focused on understanding the molecular mechanisms behind shear-activation of chemical species and how shear-driven reaction pathways compared against thermal pathways. This research had direct implications for the design of more efficient and environmentally friendly chemical processes. By collaborating closely with experimental chemists, we aimed to validate our simulation results and provide guidelines for designing new mechanochemical synthesis routes. Further, since shear-driven reactions play crucial roles at lubricating interfaces, the findings from this work will also facilitate the development of more durable and efficient lubricants and additives.