Computational Materials Discovery using Evolutionary Algorithms

Artem R. Oganov

School of Materials Science and Engineering

Northwestern Polytechnical University. Xi’an, Shaan Xi, 710072, P. R. China.

Abstract

In the last decade, a number of approaches made it possible to discover new materials on the computer–prior to experimental verification. This holds a transformative potential for the development of new technologies. A special role in this development is played by our evolutionary algorithm USPEX, developed by me and my students since 2006. The methodology will be briefly reviewed and its predictive power will be illustrated by several examples of unusual chemistry at high pressures:

Discovery of new allotropes of the elements, including a new superhard phase of boron, γ-B, and an insulating and optically transparent hP4 form of sodium, an elemental electride.

Discovery of novel chemistry of well-studied elements – e.g., the new stable compound of helium, Na₂He.

Discovery of novel compounds, forbidden by classical chemistry – e.g. Na₃Cl, NaCl₃, NaCl₇, Mg₃O₂, etc.

To enable discovery of technologically useful materials, we implement multiobjective (Pareto) optimization and apply it to a variety of problems:

1. Search for novel thermoelectric materials, where we show the possibility of achieving the figure of merit ZT>>1.

2. Discovery of novel magnetic materials and their experimental synthesis.

3. Search for novel superhard materials.

4. Prediction of novel chemistry of nanoparticles and possible explanation of carcinogenicity of oxide dust.

Current limitations and future prospects of these methods will be discussed.

DOI: 10.12110/firstfmge.20171121.401