Machine-Learning-based Investigation of Atomic Packing Effects: Chemical Pressures at the Extremes of Intermetallic Complexity

Intermetallic phases represent a domain of emergent behavior, in which atoms can combine into complex geometrical arrangements with repeat patterns containing thousands of atoms or even long-range order incompatible with a 3D unit cell. The formation of such arrangements points to unexplained driving forces that, if understood, could be harnessed in materials design. DFT-Chemical Pressure (CP) analysis has emerged as an approach to visualizing how atomic packing tensions can drive complexity and create potential functionality. However, its applications have hitherto been limited in scope by the dependence on electronic structure calculations. In this Article, we develop a Machine Learning (ML)-based implementation of the CP approach, drawing on DFT-CP schemes collected in the Intermetallic Reactivity Database. To illustrate its capabilities, we explore one of the quintessential instances of intermetallic complexity, Mg2Al3. The structure’s ML-derived CP scheme points to an origin in simple matching rules for the assembly of Frank-Kasper polyhedra.