[SY-D3] High-entropy alloys investigation using machine-learned potentials
High-entropy alloys (HEAs) are a class of materials promising for their potential durability and high heat-resistance. HEAs are defined as alloys consisting of at least four different components that form single-phase solid solutions due to high entropy of mixing.
It is rather difficult to find which components would form a HEA since solid solutions alloys tend to split into binary or mono-atomic alloys at low temperatures and can form metallic glasses at high temperatures. This makes it difficult to experimentally investigate HEAs. Hence there is a need in computational approaches to the design of HEAs.
In this work we propose a computational framework for predicting the temperature of the order-disorder (i.e., intermetallic-solid solution) transition for HEAs. We first construct a very computationally efficient machine-learning on-lattice model of interatomic interaction (an alternative to the cluster expansion model) [Shapeev A., 2017]. The model parameters are fitted to quantum-mechanical (DFT) data with accuracy of about 1meV/atom. Then we perform canonical Monte-Carlo simulations for b.c.c. (body-centered cubic), h.c.p. (hexagonal close-packed) and f.c.c. (face-centered cubic) lattices. We validate our results against the published works [Körmann F. et al, 2017][Huhn W.P. et al, 2013][Fernández-Caballero A. et al, 2017]. Particular approach in atomistic simulations significantly reduces calculation time, enables to increase explored atomic configurations range and preserves the accuracy level in comparison with ab initio calculations. In the case of successful development of the method of constructing phase diagrams using machine learning, human labor can be excluded from the routine process of studying phase diagrams. This will also reduce the computational costs consumed by conventional calculations from the first principles.
It is rather difficult to find which components would form a HEA since solid solutions alloys tend to split into binary or mono-atomic alloys at low temperatures and can form metallic glasses at high temperatures. This makes it difficult to experimentally investigate HEAs. Hence there is a need in computational approaches to the design of HEAs.
In this work we propose a computational framework for predicting the temperature of the order-disorder (i.e., intermetallic-solid solution) transition for HEAs. We first construct a very computationally efficient machine-learning on-lattice model of interatomic interaction (an alternative to the cluster expansion model) [Shapeev A., 2017]. The model parameters are fitted to quantum-mechanical (DFT) data with accuracy of about 1meV/atom. Then we perform canonical Monte-Carlo simulations for b.c.c. (body-centered cubic), h.c.p. (hexagonal close-packed) and f.c.c. (face-centered cubic) lattices. We validate our results against the published works [Körmann F. et al, 2017][Huhn W.P. et al, 2013][Fernández-Caballero A. et al, 2017]. Particular approach in atomistic simulations significantly reduces calculation time, enables to increase explored atomic configurations range and preserves the accuracy level in comparison with ab initio calculations. In the case of successful development of the method of constructing phase diagrams using machine learning, human labor can be excluded from the routine process of studying phase diagrams. This will also reduce the computational costs consumed by conventional calculations from the first principles.