| Autor: |
Balyakin IA; Institute of Metallurgy of the Ural Branch of the Russian Academy of Sciences, Ekaterinburg, Russia. NANOTECH Centre, Ural Federal University, Ekaterinburg, Russia., Yuryev AA, Gelchinski BR, Rempel AA |
| Jazyk: |
angličtina |
| Zdroj: |
Journal of physics. Condensed matter : an Institute of Physics journal [J Phys Condens Matter] 2020 May 13; Vol. 32 (21), pp. 214006. |
| DOI: |
10.1088/1361-648X/ab6f87 |
| Abstrakt: |
The structural and dynamics properties of melts are directly related to their solidification processes, and consequently to the properties of as-cast solid alloys. Ab initio molecular dynamics (AIMD) is a powerful tool that can study both of these factors. However, the main disadvantage of this method is its low performance which is critical for simulation of the multicomponent liquids. At the same time the atomistic simulation of multicomponent liquids has found its application for prediction of the formation of high-entropy alloys-a novel class of materials with enhanced mechanical properties. An effective method to solve the problem of AIMD low performance may be the design of pair or many-body potentials for classical molecular dynamics. One of the promising approaches is high-dimensional neural networks-the method of constructing many-body potentials for classical molecular dynamics from ab initio data. Thus, in this work, the high-dimensional neural network potential for multicomponent liquid VZrNbHfTa melt was constructed. The structure of this melt obtained by AIMD and high-dimensional neural network potential was compared by analyzing partial radial distribution functions. Dynamics of the melt obtained by both methods was also compared analyzing velocity autocorrelation functions and mean-square displacement for each type of atom in multicomponent VZrNbHfTa melt. It was shown that structure and dynamics are reproduced well by high-dimensional neural network potential (HDNNP). Some differences between HDNNP- and AIMD-obtained structure and dynamics are explained by finite-size effect and lack of statistics in AIMD simulation along with inherent errors in energy and force estimations made by high-dimensional neural network potentials. Analysis of melt structure via partial radial distribution functions and chemical short range order parameters led to the conclusion that vanadium atoms are repulsed from all the atoms of another type in liquid VZrNbHfTa system, which lowers the probability of single phase disordered solid solution formation. Diffusivity in multicomponent melt was found to decrease with increasing mass and size of an atom. |
| Databáze: |
MEDLINE |
| Externí odkaz: |
|
