High Throughput Computational Materials Discovery

Griesemer, Sean Darius

doi:https://doi.org/10.21985/n2-amdd-bq82

Work

High Throughput Computational Materials Discovery

Public

Download PDF

Download All Files (.zip)

The field of materials discovery is undergoing an unprecedented transition from laboratory tocomputer. Behind this transition is the new ability to accurately compute material properties, especially energetic stability, from first principles with density functional theory (DFT). However, DFT remains computationally expensive, and DFT-based materials discovery is intractable, especially in high throughput, when the search space comprises a combinatorically explosive number of possible compositions and structures with many degrees of freedom. In this thesis, we develop ways to accelerate two legs of computational materials discovery: crystal structure solution and the search for new stable compounds, and employ our methods to conduct materials discovery in high throughput. For the first leg, crystal structure solution, we develop a novel method of rapidly solving crystal structures from experimental diffraction data by searching for candidate prototypes from materials databases and evaluating their DFT stabilities and diffraction pattern matches, and then deploy this method in high throughput to solve 521 structures of compounds with existing diffraction data. For the second leg, the search for stable compounds, we compare and improve the workflows of previously developed search methods based on data mining and machine-learned formation energy prediction, and then deploy the methods in high throughput to discover thousands of new compounds that DFT predicts to lie on the convex hull of stability. Finally, we provide a comprehensive literature review of recent efforts to develop artificial intelligence for the accelerated discovery of new materials that are stable at zero and finite temperature.

Creator