Building an Interatomic Potential for FeNi: a potential rare earth-free, 'gap' magnet

In this work, we present a machine learned interatomic potential (MLIP) for ferromagnetic FeNi alloys, Fe_1−xNi_x, 0.4 ≤ x ≤ 0.6, trained on an extensive set of ab initio density functional theory data. The FeNi system has an L1₀ ordered phase which has been shown to be a promising ‘gap’ magnet [4]. The atomic ordering temperature is low, below FeNi’s Curie temperature, and the ordering is hampered by slow kinetics. Here we present results of kinetic Monte Carlo (kMC) simulations using our generated MLIP to investigate this feature. Empirical interatomic potentials (IPs), that are typically used for kMC simulations of intermetallic systems, have been recently shown to provide insufficiently accurate barrier heights for quantitative results [2] which motivates our use of IPs trained on ab initio calculations.

Using the atomic cluster expansion (ACE) [1] descriptors, we acquire the reference data with a modified version of the ACEHAL (ACE Hyperactive Learning) [3] package which allows multiple concurrent processes to explore the atomic configurational space. We use the resulting interatomic potential and kMC simulations to explore the atomic ordering and to estimate diffusion coefficients. We analyse the results to gain insight into the kinetics of the A₁/L1₀ disorder/order transformation process and possible strategies for speeding it up.

References

[1] R Drautz. Physical Review B, 99:014104, 1 2019.

[2] A Fisher, J B Staunton, H Wu, P Brommer. Modelling and Simulation in Materials Science and Engineering, 32(6):065024, July 2024.

[3] C van der Oord, M Sachs, D P Kovacs, C Ortner, G Csanyi. npj Computational Materials, 9(1):168, 2023.

[4] C D. Woodgate, C E. Patrick, L H. Lewis, and J B. Staunton. Journal of Applied Physics, 134(16):163905, 10 2023