The prediction of close packed and porous inorganic crystal structures

Abstract

There are an immense number of possible atomic arrangements that can be generated from the knowledge of the unit cell lattice parameters and constituents. Therefore finding the crystal structure that produces the 'observed' powder diffraction pattern can still be intractable, in some cases, without further information. Hence, techniques for structure prediction have a role to play. Our method (Phys. Chem. Chem. Phys. 1991 1 2535-2542) assumes that the required structure has a lattice energy at either the global minimum or a local minimum of a similar depth. By implementing a genetic algorithm(GA) to generate plausible structures followed by a local optimizer, in order to increase the accuracy of the atomic coordinates, we have efficiently found the structures of a wide range of known close packed oxides, including perovskites, pyrochlores and spinels. Our method, which is capable of reproducing different phases of a compound, is based upon that developed by Bush et al (1995) who successfully predicted the previously unknown structure of Li3RuO4. The method has been implemented within the General Utility Lattice Program (GULP), thus facilitating the calculation of the physical properties (for example, elastic constants) of the phase. Now this method has been extended to enable the prediction of porous materials. To accelerate and guide the GA when trying to generate such materials, several techniques have been investigated, including a simple additive penalty to the cost function when an ion is within a defined 'exclusion zone'(EZ) and the use of a nonlinear grid that has no points within EZ. Further refinements currently being investigated include the use of polyatomic units and symmetry information.