Search for: All records

We present an approach to approximating static properties of glasses without experimental inputs rooted in the first-principles random structure sampling. In our approach, the glassy system is represented by a collection (composite) of periodic, small-cell (few 10 s of atoms) local minima on the potential energy surface. These are obtained by generating a set of periodic structures with random lattice parameters and random atomic positions, which are then relaxed to their closest local minima on the potential energy surface using the first-principles methods. Using vitreous SiO2 as an example, we illustrate and discuss how well various atomic and electronic structure properties calculated as averages over the set of such local minima reproduce experimental data. The practical benefit of our approach, which can be rigorously thought of as representing an infinitely quickly quenched liquid, is in that it transfers the computational burden to linear scaling and easy to converge averages of properties computed on small-cell structures, rather than simulation cells with 100 s if not 1000 s of atoms while retaining a good overall predictive accuracy. Because of this, it enables the future use of high-cost/high-accuracy electronic structure methods, thereby bringing the modeling of glasses and amorphous phases closer to the state of modeling of crystalline solids. 

Stability prediction is accelerated by treating the convex hull as a probabilistic object, allowing for an efficient active learning process that minimizes the number of thermodynamic calculations necessary to define the convex hull. 

Active learning is a valuable tool for efficiently exploring complex spaces, finding a variety of uses in materials science. However, the determination of convex hulls for phase diagrams does not neatly fit into traditional active learning approaches due to their global nature. Specifically, the thermodynamic stability of a material is not simply a function of its own energy, but rather requires energetic information from all other competing compositions and phases. Here we present Convex hull-aware Active Learning (CAL), a novel Bayesian algorithm that chooses experiments to minimize the uncertainty in the convex hull. CAL prioritizes compositions that are close to or on the hull, leaving significant uncertainty in other compositions that are quickly determined to be irrelevant to the convex hull. The convex hull can thus be predicted with significantly fewer observations than approaches that focus solely on energy. Intrinsic to this Bayesian approach is uncertainty quantification in both the convex hull and all subsequent predictions (e.g., stability and chemical potential). By providing increased search efficiency and uncertainty quantification, CAL can be readily incorporated into the emerging paradigm of uncertainty-based workflows for thermodynamic prediction. 

Local bonding environments can be characterized via ensemble averages of PDFs to provide insight into the relationship between synthetic temperature and structure.