Computer Generation of Hypothetical Zeolites

The basic principles used to generate hypothetical zeolites by different research groups will be described, including two global optimization techniques: i) techniques based on the Monte Carlo method, such as ZEFSAII, FraGen, and SCIBS; and ii) techniques based on genetic algorithms, such as EZs, P-GHAZ, and zeoGAsolver. ZEFSAII starts with a space group and a number of unique tetrahedral atoms, and adjusts the atom locations and cell dimensions according to a cost function, which is reduced by the Monte Carlo method. A powerful feature is that the diffraction pattern of an unknown zeolite can be provided as a reference, and the difference between the computed pattern and the reference contributes to the cost function. The SCIBS method also starts with a space group and the number of unique tetrahedral atoms. The set of all possible 4-valent directed graphs, consistent with the space group symmetry, is generated. Low-energy topologies are found by simulated annealing of a barycentric-coordinate embedding of the graph. Databases of zeolite frameworks are available. EZs simulates the pore, cage, and even small void space structures in zeolites, by imposing exclusion zones into the crystalline space in which atoms are forbidden to enter. P-GHAZ uses evolutionary algorithms combined with general purpose computation on graphic processing units to find hypothetical zeolite structures close to a thermodynamic feasibility criterion. zeoGAsolver employs genetic algorithms to generate hypothetical zeolites by randomly locating atoms in a target asymmetric unit cell and evaluating a score function that defines structural aspects of zeolites described only as a collection of tetrahedral atoms (hence without oxygen). FraGen and zeoGAsolver include a chemical property introduced in the genetic algorithm that allows to classify the new generated points according to their Wyckoff multiplicity and forces the generated set to satisfy a condition related to the resulting material density. Data selection and reduction strategies will be described, and the accuracy of the fitness function will be critically evaluated. The combination of GPU cards with the genetic algorithms for zeolite structures determination and prediction problems will also be discussed. Examples of methods that successfully generated zeolitic structures will be given. In the cases discussed, GPUs allowed a broader sampling of the search space, on feasible timescales, enabling the generation of frameworks reported in IZA database and also hypothetical.