Catalysis is critical in our sustainable development by aiding synthesis of various essential compounds. However, the advancement of this field is hindered by the un-elucidated atomic events under operating condition. Despite modernization of the characterization techniques, observations are usually limited to the equilibrium state (i.e., the condition before and after the catalysis occurs) due to the intractable behavior of the system at the atomic level. Such limitations hinder the identification of the promoting (or demoting) factors of the catalysis that eventually play their roles during the non-equilibrium states.

In the raise of computing power, atomic simulation is highly capable to elucidate the non-equilibrium states of catalysis by providing explicit ‘molecular movie’ of catalytic events. This approach works by constructing the interatomic potential, from which the energy and forces of atoms that govern the dynamic of the system can be derived. Given the rapid progress in machine-learning (ML), first principles calculations such as Density Functional Theory (DFT) can be accurately predicted by ML model after learning from adequate database. This framework results in faster and more efficient method called machine-learning molecular dynamics (MLMD). However, the MLMD relies on high-throughput and automated calculations to build reliable database used to train ML model, therefore, the large-scale computer system is inevitably necessary. In this project, we aim to elucidate the non-equilibrium states of the catalytic system by means of MLMD simulations. Some of our preliminary results on the application of MLMD in the elucidation of dynamics of catalytic surface have been published in the reference [1] and [2]. 

This research holds significance in catalysis field which currently challenged by intractable non-equilibrium states in experiment. This research thus acts as link between interdisciplinary of experimental studies, computational studies, and data science, which lead to more confident understanding of catalysis and accelerating transfer of idea from simulation to actual synthesis. Further, the dynamics property of catalysts can be included in the catalyst database and improve the performance of AI-based material screening. Since the typical database include only the equilibrium (or as-prepared) states of catalyst, this project also promotes quality and usage of data in material science.

[1] Halim, H. H.; Morikawa, Y. Elucidation of Cu–Zn Surface Alloying on Cu(997) by Machine-Learning Molecular Dynamics. ACS Phys. Chem. Au 2022, 2 (5), 430–447. 

[2 ] Halim, H. H.; Ueda, R.; Morikawa, Y. Machine Learning Molecular Dynamics Simulation of CO-Driven Formation of Cu Clusters on the Cu(111) Surface. J. Phys. Condens. Matter 2023, 35 (49), 495001. 

• Elucidation of Cu–Zn Surface Alloying on Cu(997) by Machine-Learning Molecular Dynamics
• Machine Learning Molecular Dynamics Simulation of CO-Driven Formation of Cu Clusters on the Cu(111) Surface