(114g) Machine-Learned Interatomic Potentials for Modeling Catalytic Reactions Under Thermodynamic Conditions: Challenges and Opportunities

Despite significant advancements in computational hardware and simulation algorithms, modeling reactive systems under thermodynamic conditions that closely mimic experimental settings remains a considerable challenge in chemical sciences. The nature of chemical bond formation and breakage necessitates a quantum mechanical treatment, which often limits us to studying small system sizes due to the high computational expenses associated with density functional theory (DFT) calculations. The machine-learned interatomic potentials (ML-IAPs) trained on the molecular dynamics trajectories utilizing density functional theory (DFT) provide an attractive solution to this problem. The ML-IAP offers a considerable reduction in computational costs compared to direct DFT calculations while preserving accuracy comparable to DFT simulations. In this presentation, we share our recent work modeling complex reaction systems using ML-IAP under experimental thermodynamic constraints. We will highlight the potential of these IAPs to model large system sizes, allowing us to incorporate collective/cooperative events in catalysis. We will also note the areas that need further development.