Understanding the pressure-induced structural evolution and thermophysical response of Mg–Zn intermetallics via a neuroevolution potential

We developed a fourth-generation neuroevolution potential (NEP) for the Mg–Zn binary system, trained on a comprehensive DFT dataset covering pure elements, solid solutions, intermetallics, defects, and high-temperature/high-pressure configurations. The potential achieves DFT-level accuracy with an energy root-mean-square error of 3.43 meV/atom and faithfully reproduces the equations of state, phonon dispersions, and elastic constants of the C14/C15 Laves phases, Mg 4 Zn 7 , and Mg 2 Zn 11 . Molecular dynamics simulations of the Mg 4 Zn 7 Frank-Kasper phase from 0 to 1000 K and 0 to 50 GPa reveal a pressure-adaptation mechanism characterized by continuous topological distortions of icosahedral and Friauf polyhedra, accommodating approximately 30% volume collapse while preserving long-range order without melting. The NEP accelerates calculations by orders of magnitude relative to DFT, enabling atomistic studies of precipitation, interfaces, and defects in lightweight alloys. The training dataset and potential files are publicly available at https://github.com/jinshuaishuai0305/Mg-Zn-NEP-potential.git.