Learning Grain Boundary Thermodynamic Spectra in Polycrystals

The thermodynamics of alloy components determine material structures, properties and hence the processability and performance matrices of engineering alloys. Incorporating grain boundaries in thermodynamic models is not a simple task: grain boundary defects in polycrystals occupy a wide 5-dimensional configuration space, imposing a computational barrier in evaluating thermodynamic data across the chemical space. This complexity calls for a combination of recent advances in data science approaches and computational materials science. This talk will first discuss an isotherm approach in modeling equilibrium solute behavior at grain boundaries and show that relevant spectral thermodynamic quantities can be extracted from atomistic simulations. Then, data-science and machine-learning enabled accelerated models are presented to extend the work to many alloy systems. Finally, a full-spectral case study of a binary alloy system with the inclusion of rapid estimations of dilute segregation energies, solute-solute interactions at grain boundaries and excess vibrational entropy of segregation is presented. These accelerated models open opportunities to construct an embedded alloy grain boundary atlas, which ultimately can assist in alloy design.