Project 4: Creating Transferable Reactive Machine Learning Potentials at Unrestricted Coupled Cluster
Machine learning models of atomic and molecular interactions are increasingly employed in atomistic simulations. Training these models requires large datasets derived from quantum mechanical (QM) calculations, which provide information such as energies and forces for different molecular structures. The accuracy of these QM calculations is critical, as it directly impacts the performance of the trained machine learning models. While density functional theory is commonly used due to its efficiency, it often falls short for certain applications, such as modeling transition states in chemical reactions. To overcome these limitations, more accurate QM methods, such as coupled cluster calculations, can be utilized. This project aims to combine advanced machine learning techniques with high-fidelity QM data to develop transferable machine learning interatomic potentials. These potentials will be trained on datasets generated through unrestricted coupled cluster calculations, offering unprecedented accuracy and efficiency for predicting molecular properties. The project will involve generating coupled cluster datasets, training machine learning potentials, and rigorously benchmarking the resulting models to ensure they can accurately simulate complex reactive processes, including those in biological systems.
Team
powered by Borlabs Cookie