High-throughput and data mining with <i>ab initio</i> methods

Accurate ab initio methods for performing quantum mechanical calculations have been available for many years, but their speed, complexity and instability have generally constrained researchers to studying only a few systems at a time. However, advances in computer speed and ab initio algorithms have now created fast and robust codes, where large numbers of calculations can be performed automatically, making it possible to do high-throughput ab initio computation. High-throughput computations can be used to efficiently screen and optimize for desired properties in broad classes of materials, as well as create large databases for data mining applications that can guide both experiments and further calculations. This paper discusses some of the challenges associated with preparing, running, collecting and assessing ab initio results in a high-throughput framework. An example application is given in the area of crystal structure prediction for binary alloys. The high-throughput results are in good agreement with known data, and suggest many possible new compounds not yet seen experimentally. Data mining techniques are used to find correlations among structural energies, and the correlations are then used to accelerate identification of stable crystal structures in new alloys.