The Augmented Potential Method: Multiscale Modeling Toward a Spectral Defect Genome

Abstract

The modeling of solute chemistry at low-symmetry defects in materials is historically challenging, due to the computation cost required to evaluate thermodynamic properties from first principles. Here, we offer a hybrid multiscale approach called the augmented potential method that connects the chemical flexibility and near-quantum accuracy of a universal machine learning potential at the site of the defect, with the computational speed of a long-range classical potential implemented away from the defect site in a buffer zone. The method allows us to rapidly compute distributions of grain boundary segregation energy for 1,050 binary alloy pairs (including Ag, Al, Au, Cr, Cu, Fe, Mo, Nb, Ni, Pd, Pt, Ta and V, W solvent), creating a database for polycrystalline grain boundary segregation. This database is ~5x larger than previously published spectral compilations, and yet has improved accuracy. The approach can also address problems far beyond the reach of any other method, such as handling bcc Fe-based alloys, or the complex solute-solute interactions in random polycrystals. The approach thus paves a pathway toward a complete defect genome in crystalline materials.