Speaker
Description
In this talk, I will discuss a coarse-grained, statistical (probabilistic) representation of the potential energy surface (PES) of solid-state systems we have recently developed. It is constructed using a combination of the first-principles random structure sampling and structure transformation modeling. The random structure sampling identifies statistically relevant (i.e., more probable) local minima on the PES, along with their energy distribution and their sizes on the PES (their “widths”). The structure transformation modeling then measures the “depths” of these minima, which allows to group the shallow ones with deeper minima into which they would rapidly transform effectively increasing the width of the deeper local minima. In our previous work, we demonstrated that the experimental realizability of crystalline phases – including those exhibiting lattice disorder – correlates with the total widths of their corresponding PES minima. This insight enables predictions of the likely outcomes of materials synthesis across different chemistries and synthesis methods. We also quantified specific PES features that make certain chemistries difficult to crystallize, making them more likely to form glassy or amorphous phases. I will conclude by discussing current challenges and opportunities for integrating machine learning into our methodology to further advance and accelerate predictive synthesis and materials discovery in general.
