An improved sampling strategy for global energy minimization of multi-component systems

Effective initial sampling plays an important role in capturing key details about the energy surfaces of multi-component, multi-sublattice phases for the purposes of accurate convergence toward the global minimum energy configuration of a given system. It is shown that, when using the appropriate statistical distribution, both quasi-random and pseudo-random sampling methods compare well with the standard uniform grid-based technique. Moreover, the combination of random sampling with uniform grid points, while maintaining sampling performance for equilibrium calculations in the Al-Co-Cr system, significantly increases performance for a fictive 10-component system.