Machine Learning Interatomic Potentials for Million-Atom Simulations of Multicomponent Alloys

Abstract

Machine learning interatomic potentials (MLIPs) with broad chemical flexibility are important for atomistic simulations of compositionally complex materials such as high-entropy alloys. Here, we study two state-of-the-art MLIP frameworks, the neuroevolution potential (NEP) and the graph atomic cluster expansion (GRACE), for 16 elemental metals and multicomponent alloys. GRACE potential with Finnis-Sinclair type shows substantially higher training efficiency and consistently, though only slightly, better accuracy for mechanical properties, thermal stability, and chemical extrapolation. In contrast, NEP achieves an approximately 60-fold higher inference speed, making it attractive for million-atom molecular dynamics simulations. We further examine uncertainty quantification strategies and find that ensemble-based uncertainty correlates robustly with model error, whereas D-optimality is less reliable for the systems considered here. Large-scale nonequilibrium molecular dynamics simulations of shock propagation further show that NEP, combined with ensemble-based uncertainty quantification, enables efficient and reliable simulations under extreme dynamic conditions.

Figures (10)

• Figure 1: Benchmarking the accuracy and efficiency of UNEP-v1 and GRACE-FS-M potentials. Error distributions for (a) energy, (b) forces, and (c) virial stress are compared alongside (d) the computational cost (wall time) required for training.
• Figure 2: Uncertainty quantification for UNEP-v1 and GRACE-FS-M potentials. (a, b) Uncertainty estimates for UNEP-v1 using ensemble learning at the atomic and structural levels, respectively. (c-f) Uncertainty estimates for GRACE-FS-M using (c, d) ensemble learning and (e, f) the D-optimality criterion, each shown at the atomic and structural level.
• Figure 3: Computational speed comparison between UNEP-v1 and GRACE-FS-M for (a) pure Cu and (b) the Al$_{31}$Cr$_{6}$Cu$_{22}$Ni$_{32}$V$_{9}$ HEA. UNEP-v1 is evaluated on A100 and H100 GPUs, while GRACE-FS-M is run on 192 CPU cores.
• Figure 4: Thermal stability of goldene assessed by MD simulations. (a) Maximum energy drift as a function of temperature. (b, c) Atomic snapshots of the structure after relaxation at 1,500 K, showing the different structures obtained by two MLIPs at extreme conditions.
• Figure 5: Assessing the thermal stability of HEAs via MD simulations. The maximum energy drift, a key metric for stability, is plotted for (a) the Al$_{31}$Cr$_{6}$Cu$_{22}$Ni$_{32}$V$_{9}$ alloy with chemical short-range order (CSRO) and (b) a random 16-element alloy across a range of temperatures.