High entropy alloys (HEAs) are a new class of materials incorporating four or more elements in similar concentrations. Due to their superior radiation resistance properties compared to pure elements and conventional alloys, HEAs are promising materials for applications in structural elements of future fusion and fission reactors [1].

In this work, we focus on the bcc-based HEAs Ta-Ti-V-W system, which has been shown from our combined density functional theory (DFT), cluster expansion and Monte-Carlo simulations to have the lowest solid solution temperature among five-component (Cr-Ta-Ti-V-W) HEAs [2]. The accurate and fast machine learning (ML) interatomic potentials has been developed based on thousands of DFT calculations for the representative structures of alloys obtained using a DFT-based MC database for various alloy compositions, structures with different short-range ordering and different classes of calculations: structure optimisations, calculations with applied strains, with a presence point defects, as well as ab initio molecular dynamics (MD) simulations both for bcc and liquid phases. Different ML approaches, namely linear ML, quadratic noise ML and kernel models, and various types of atomic descriptors, have been tested in order to achieve a good balance between accuracy, speed and predictive power [3]. The root-mean-square errors between the energies, forces and stresses computed using DFT and ML potential are below 2 meV/atom, 0.2 eV/Å and 0.01 eV/Å3, respectively. The developed ML potentials have been applied in MD simulations to study the chosen properties of HEAs as a function of composition and temperature, and the obtained results have been compared with the available DFT results.

References

[1] O. El-Atwani, N. Li, M. Li, et al. Science Advances, 2019, 5, eaav2002.

[2] D. Sobieraj, J.S. Wróbel, T. Rygier, et al. Physical Chemistry Chemical Physics, 2020, 22, 23929.

[3] A.M. Goryaeva, J. Dérès, C. Lapointe, et al. Physical Review Materials, 2021, 5, 103803.