Grain boundary segregation spectra from a generalized machine-learning potential

Modeling solute segregation to grain boundaries at near first-principles accuracy is a daunting task, particularly at finite concentrations and temperatures that require accurate assessments of solute-solute interactions and excess vibrational entropy of segregation that are computationally intensive. Here, we apply a generalized machine learning potential for 16 elements, including Ag, Al, Au, Cr, Cu, Mg, Mo, Ni, Pb, Pd, Pt, Ta, Ti, V, W and Zr, to provide a self-consistent spectral database for all of these energetic components in 240 binary alloy polycrystals. The segregation spectra are validated against prior quantum-accurate simulations and show improved predictive ability with some existing atom probe tomography experimental data.