|About this Abstract
||2017 TMS Annual Meeting & Exhibition
||Computational Materials Discovery and Optimization – From Bulk to Materials Interfaces and 2D Materials
||Predicting Novel Spinels Using Density Functional Theory Assisted Machine Learning
||Joshua Schiller, Elif Ertekin
|On-Site Speaker (Planned)
Predicting novel materials can be challenging due to the time-consuming effort necessitated by experimental searches or the resources and approximations utilized by theoretical ones. To overcome these obstacles, we have developed a framework that combines machine learning and density functional theory (DFT) in a unique way to achieve rapid materials discovery. We use experimental data to determine stable structures and DFT to identify unstable ones. This approach avoids relying on DFT to predict stability, since theoretically predicted, zero-temperature formation enthalpies of stable compounds are often distributed above the convex hull. Instead, we rely on simulations to predict instability, a less challenging task. Utilizing these classified structures and their constituent elemental attributes, we predict new stable compounds with several machine learning techniques. We have applied this methodology to the spinel phase space for the discovery of new materials and are working with experimental collaborators to synthesize some top candidates.
||Definite: None Selected