Computationally driven design of materials involves identifying key material properties that drive performance, calculating those properties, e.g., with quantum mechanics, determining correlations between material properties and performance, and then screening materials to identify those that are expected to optimize performance. A grand challenge is reducing the amount of time required to screen the materials, which would enable a more widespread search. Machine learning has been advantageous in this regard by identifying correlations between tabulated features of materials with their performance-driving properties. Specifically, using graph theory, a material's performance-driving properties can be quantified if one knows the material's structure along with the elemental properties of its constiuents. In this presentation, we discuss how we have used a combination of quantum mechanics and machine learning to link performance-driving properties of substituted metal ferrites to their performances for energy delivery for use in catalysis, biomedicine, and electronic devices. Specifically, we predict the saturation magnetizations of the ferrite materials, as these help define the potential for energy delivery. Recent literature suggests that the magnetic properties of substituted ferrites can be tuned via their compositions. Saturation magnetizations are calculated for ~1,000 substituted ferrite bulk compositions using density functional theory and input to machine-learning. The machine learning model predicts the saturation magnetization based on tabulated electronic and thermodynamic properties of the elemental constituents. The developed ML model enables the screening of vast chemical spaces and helps inform synthesis.

Computational screening of substituted metal ferrites for catalysis and other applications: Density functional theory and machine learning study

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.