Predicting crystal structures with evolution

Even for simple solids, calculating the arrangement of the constituent atoms from first principles is exceedingly difficult, partly because of the need to sort through an astronomical number of possible ways that atoms can compose a basic repeatable unit cell—roughly 10³⁹ different arrangements for 30 identical atoms. Enter Artem Oganov, a materials scientist at ETH Zürich, and Colin Glass, a PhD student, who approached the problem by combining electronic structure calculations and a specifically developed evolutionary algorithm that requires only the chemical composition; no additional input from experiment is needed. In exploring the myriad possibilities, the algorithm proceeds in a step-by-step, continual-optimization fashion that avoids configurations less likely to succeed. The algorithm is efficient and robust: In their first tens of tests, over a range of extreme pressure and temperature conditions and with up to 40 atoms per unit cell, Oganov and Glass have reproduced experimentally known structures with nearly perfect success and predicted several new structures. For example, calcium carbonate (CaCO₃) is known to be stable at pressures above 40 GPa and is thought to be a major host of carbon in Earth’s lower mantle. But the structure of that high-pressure phase of CaCO₃, called postaragonite, was completely unknown until the new algorithm predicted it, along with a higher-pressure form with unusual tetrahedral carbonate ions. Both structures (shown here) have since been experimentally confirmed by Japanese colleagues. The ETH researchers have also found stable and metastable phases of carbon, oxygen, hydrogen, and other elements and compounds. (A. R. Oganov, C. W. Glass, J. Chem. Phys. 124 , 244704, 2006 http://dx.doi.org/10.1063/1.2210932 .)