Machine Learning for the Electronic Structure of Bulk Crystalline Materials and Beyond

The design of new materials with desired properties is a crucial step in making many innova­tive technologies viable and thus is a precursor for the broader impacts on society brought by such innovation.

A typical challenge in materials design is to find materials that fulfill specific requirements on efficiency, cost, environmental impact, length of life, safety, and other prop­erties.

A standard theoretical tool for investigation of candidate materials is the prediction of electronic structure from numerical solutions of equations based on quantum mechanics, e.g., using the framework of density-functional theory. However, data-driven methods are increasingly adopted in many research fields.

These methods can extract desired relations purely out of data, and generate models capable of near-instant predictions.

The aim of this project is to replace some of the standard use of such numerical solutions of quantum mechanics-based equations by machine learning models.

We will develop and train these models to be capable of making as accurate predictions of electronic structure properties for crystalline bulk materials but at a small fraction of the computational cost.

In a set of planned tasks, we make a combined push forward in a set of key areas to enable this use of machine learning models.

These models will be transfor­mative for theoretical materials design since they enable instant solution of inverse­ design problems via inversion (properties to material).