Capabilities and limitations of MACE interatomic potentials for Fe-Ni alloys: a systematic benchmark

We present a systematic benchmark of MACE interatomic potentials for iron–nickel alloys, focusing on structural, elastic, and finite-temperature properties of chemically disordered body-centered-cubic (bcc) and face-centered-cubic (fcc) phases. The reference dataset comprises spin-polarized PBE density-functional-theory (DFT) calculations for special quasirandom structures (SQS) span- ning composition, volume, and shear deformation. A system-specific MACE–sqs model trained on this dataset achieves validation errors of 2.0 meV/atom for energies and 24.3 meV/Å for forces. We compare MACE–sqs with several MACE foundation models, including models trained using Hubbard corrections, through equations of state, equilibrium volumes, cubic elastic properties, and thermal lattice expansion. MACE–sqs gives the most consistent overall performance for several zero-temperature structural and elastic benchmarks, although its relative performance depends on property and composition and it does not reproduce all experimental elastic trends. In the finite-temperature comparison, no single model is uniformly superior, and MACE–matpes agrees more closely with experiment for some compositions and comparison metrics. Finite-temperature calculations yield reasonable thermal-expansion coefficients near 300 K, while deviations at elevated temperatures and for magnetovolume sensitive compositions highlight the limitation of representing magnetic effects only implicitly through the spin-polarized reference data. These results define the tested application domain in which targeted SQS-based training improves upon broad foundation models and identify explicit magnetic degrees of freedom and broader configurational sampling as priorities for future development.