Efficient Deep Learning Rhodium Potential and Feasibility Validation in Large-Scale Molecular Dynamics Simulations

JY Ye and ZJ He and L Ma and KY Chen and J Rong and YD Sui and XJ Fu and XH Yu and J Feng, JOURNAL OF ELECTRONIC MATERIALS (2025).

DOI: 10.1007/s11664-025-12337-0

Deep Potential (DP) technology integrates deep learning with quantum mechanical computations, enabling the efficient handling of complex data from density functional theory (DFT) while demonstrating excellent computational accuracy and data analysis capabilities. Rhodium (Rh), one of the rarest and most valuable platinum group metals, plays a crucial role due to its strategic importance in the automotive and electronics industries. However, the computational process for Rh is hindered by a lack of appropriate potential models, resulting in time-consuming and resource-intensive calculations that limit its research applications. To fill this gap, we developed a high-precision interatomic potential using the DP method, successfully applying it to classical molecular dynamics (MD) simulations, thereby offering a new computational tool. We systematically compared the predictions of the constructed DP potential function with results from DFT across various physical properties, including lattice parameters, stability, and defects, confirming that the constructed DP potential function exhibits excellent accuracy consistent with DFT in physical property predictions. Notably, in terms of thermal transport properties, the phonons dispersion and thermal conductivity results obtained from the developed DP model still remain in high consistency with those from the DFT method. Additionally, MD simulations based on the DP framework indicate that the crystal melts at a temperature of 2283 K, which is remarkably consistent with the experimentally measured melting point of 2237 K. With rising temperature, the transport of Rh atoms significantly enhances, with a self-diffusion coefficient of 7.54 x 10-11 m2/s at the melting point, exhibiting diffusion behavior similar to that of typical face-centered cubic metals. This study serves as a foundational step in the application of deep learning to potential energy modeling of single- element Rh systems, ensuring the accuracy and reliability of the model. By extending this approach to multi-component systems in future work, it aims to provide theoretical support for the efficient and precise design of advanced materials.

Return to Publications page