A comprehensive investigation on the accuracy and efficiency of methods for melting temperature calculation using molecular dynamics simulations

Machine learning approaches have been extensively applied to improve the accuracy and reliability of potentials, addressing inherent limitations in molecular dynamics (MD) simulations. Notably, the precise determination of the melting temperature (Tm) relies on MD simulations, necessitating a comprehensive review of available melting models. This study aims to present a thorough comparison of various melting methods, including the single–phase method, hysteresis method, Z method, modified Z method, void method, modified void method, two–phase method, and sandwich method, considering both accuracy and efficiency. These melting models are systematically categorized into three groups: perfect crystals, crystal defects, and solid–liquid interfaces. The impact of factors such as atomic numbers, heating and cooling rates, crystal defects, and nucleation processes on Tm is discussed. Additionally, the temperatures corresponding to the superheating limit and the amorphous transition induced by heating and cooling rates during melting and crystallization processes are analyzed. Furthermore, the study explores the impact of different proportions of solid and the number of solid–liquid interfaces on melting behaviors. To further enhance the accuracy of Tm calculations, we propose a new approach termed the modified two–phase method, integrating the void method and two–phase method to account for crystal defects and solid–liquid coexistence, respectively.