Recent advances in atomic-scale simulations for supported metal catalysts

• Published in: Molecular catalysis 2024-02, Vol.554, p.113862, Article 113862
• Main Authors: Yoon, Yeongjun, You, Hyo Min, Oh, Jinho, Lee, Jung-Joon, Han, Jeong Woo, Kim, Kyeounghak, Kwon, Hyunguk
• Format: Article
• Language: English
• Subjects: Computational catalysis; Heterogeneous catalysts; Machine learning potential; Supported metal catalysts
• ISSN: 2468-8231; 2468-8231
• DOI: 10.1016/j.mcat.2024.113862

•Standard DFT, which describes conditions at 0 K in ultra-high vacuum with no interaction with the reaction environment, has limitations in modeling supported catalysts.•Advanced simulation methods for structure search and dynamic structural evolution under reaction environments are introduced.•The advanced methods combined with machine learning models have huge potential in modeling supported catalysts. Over the years, there has been a continuous drive to precisely determine the structure and active sites of supported metal catalysts using atomistic-scale simulations. Computational simulations for catalysis inherently involve theoretical assumptions and may not entirely capture the complexity of experimental environments. With ongoing advances in both theory and computational resources, there is an endeavor to model “realistic” supported catalysts close to the experimental environment. This review article provides a brief overview of recent efforts in this direction. Density functional theory (DFT)-based global optimization is the most common method to model a supported catalyst, however, it has many limitations. It is not feasible with standard DFT to find multiple meta-stable structures of small nanoparticles (NPs) placed on a support and capture their dynamic nature. The DFT is also limited in its ability to simulate supported catalysts with nano-sized large particles due to high computational costs. Genetic algorithm (GA), grand-canonical Monte-Carlo (GCMC), kinetic Monte-Carlo (KMC), ensemble-average approach, and ab-initio molecular dynamics (AIMD) have been used to overcome the limitations. We also discuss the encouraging prospects of machine learning (ML) capabilities. The ML models combined with computational chemistry assist in determining the structure of supported NPs with dynamic nature under realistic conditions. The ML potential has the capability to deal with supported NPs containing large number of atoms more accurately than force fields. We anticipate this review will provide direction for further investigation into computational methods for modeling supported catalysts. [Display omitted]