Catalytic Large Atomic Model (CLAM): A Machine-Learning-Based Interatomic Potential Universal Model

Catalysis involves complex reactions with dynamic changes in catalyst morphology, challenging the capabilities of traditional Density Functional Theory (DFT) methods. To address this, we present the Catalytic Large Atomic Model (CLAM), a machine-learning-based interatomic potential designed for heterogeneous catalysis. Trained on a comprehensive dataset that includes metals, alloys, oxides, clusters, zeolites, 2D materials, and small molecules, CLAM ensures high accuracy across diverse catalytic systems. We also introduce a "local fine-tuning" algorithm that enhances the model’s applicability by accelerating structural optimizations and transition state searches while maintaining precision. Additionally, CLAM facilitates rapid reaction network construction and efficient kinetic analysis. This work advances computational catalysis by providing a universal and robust tool for catalyst design and mechanism exploration.