High-throughput data mined prediction of inorganic compounds and computational discovery of new lithium-ion battery cathode materials

Abstract

The ability to computationally predict the properties of new materials, even prior to their synthesis, has been made possible due to the current accuracy of modern ab initio techniques. In some cases, high-throughput computations can be used to create large data sets of potential compounds and their computed properties. However, regardless of the field of application, such a computational high-throughput approach faces a major problem: to be relevant, the properties need to be computed on compounds (i.e., stoichiometries and crystal structures) that will be stable enough to be synthesized. In this thesis, we address this compound prediction problem through a combination of data mining and high-throughput Density Functional Theory. We first describe a method based on correlations between crystal structure prototypes that can be used with a limited computational budget to search for new ternary oxides. In addition, for the treatment of sparser data regions such as quaternaries, a new algorithm based on the data mining of ionic substitutions is proposed and analyzed. The second part of this thesis demonstrates the application of this highthroughput ab initio computing technique to the lithium-ion battery field. Here, we describe a large-scale computational search for novel cathode materials with specific battery properties, which enables experimentalists to focus on only the most promising chemistries. Finally, to illustrate the potential of new compound computational discovery using this approach, a novel chemical class of cathode materials, the carbonophosphates, is presented along with synthesis and electrochemical results.