Molecular dynamics simulations of liquid gallium alloy Ga–X (X = Pt, Pd, Rh) via machine learning potentials

• Published in: Inorganic chemistry frontiers Vol. 11; no. 5; pp. 1573 - 1582
• Main Authors: Fang, Fang, Lin, Jie, Li, Jiajia, Zhang, Yu, Fu, Qiuyi, Zhou, Quanquan, Li, Wei, Zhou, Guobing, Yang, Zhen
• Format: Journal Article
• Language: English
• Published: London Royal Society of Chemistry 27.02.2024
• Subjects: Atomic properties; Coordination numbers; Diffusion barriers; Diffusion coefficient; Gallium base alloys; Low temperature; Machine learning; Molecular dynamics; Palladium; Platinum; Rhodium; Simulation; Tensors; Transition metals
• ISSN: 2052-1553; 2052-1545; 2052-1553
• DOI: 10.1039/D3QI02410E

Liquid gallium (Ga) has achieved significant attention across numerous fields in recent decades due to its distinctive physicochemical properties. In particular, the exceptional fluidic nature of liquid Ga makes it an excellent solvent to dissolve transition metals to prepare liquid Ga alloy (LGA) catalytic systems (M. A. Rahim, J. Tang, A. J. Christofferson, P. V. Kumar, N. Meftahi, F. Centurion, Z. Cao, J. Tang, M. Baharfar, M. Mayyas, F.-M. Allioux, P. Koshy, T. Daeneke, C. F. McConville, R. B. Kaner, S. P. Russo and K. Kalantar-Zadeh, Low-Temperature Liquid Platinum Catalyst, Nat. Chem. , 2022, 14 , 935–941). Thus, it is an important scientific quest to understand the microscopic structures and properties of transition metal atoms in LGA. Here, we employed a newly developed machine learning-based moment tensor potential (MTP), combined with molecular dynamics simulations, to explore the coordination and diffusion behaviors of transition metal atoms in three LGA systems of Ga–Pt, Ga–Pd, and Ga–Rh. It is observed that the trained MTP can provide accurate descriptions of energies and forces, as well as local structures, for each LGA system. Besides, our simulation results reveal that the average coordination number of the transition metal atom with surrounding Ga atoms follows an order of Ga–Rh > Ga–Pt > Ga–Pd, while the diffusion coefficient of the transition metal atom in liquid Ga has an inverse order of Ga–Rh < Ga–Pt < Ga–Pd. This is primarily because the diffusion barrier of Rh in liquid Ga is maximum, yet that of Pd in liquid Ga is minimum. Furthermore, the results of mean square displacement and the van Hove function suggest a normal diffusion mechanism for all three studied transition metal atoms in liquid Ga. Overall, this work shows the feasibility of MTP-based machine learning in modeling LGA systems and offers a theoretical understanding of the mechanism of interaction between transition metals and Ga atoms.