Predictor of crystalline structure inspired by nature

Russian scientists have reported a way to improve the prediction algorithms of the crystal structure, making the discovery of new compounds several times faster. The results of the study were published in Communications in Computer Physics.

Given the ever-increasing need for new technologies, chemists are looking for better materials with improved strength, weight, stability and other properties. The search for new materials is a difficult task and, if done experimentally, takes a lot of time and money, as it often requires testing a large number of compounds under different conditions. Computers can contribute, but they require good algorithms.

In 2005, Artem R. Oganov, now a professor at Skoltech and at the Institute of Physics and Technology of Moscow (MIPT), developed the evolutionary prediction algorithm of the USPEX crystal structure, which is perhaps the most successful algorithm in the field, used by many thousands of scientists around the world. USPEX only needs to know which atoms the crystal is made of. Then, it generates a small number of random structures whose stability is evaluated according to the interaction energy between the atoms. Then, an evolutionary mechanism explains the natural selection, the crossing and the mutations of the structures and their descendants, which gives particularly stable compounds.

In their recent study, scientists from Skoltech, MIPT and Samara State Technical University, led by Artem R. Oganov, have improved the first stage of the USPEX, which generates initial structures. Demonstrating that the purely random generation was not very efficient, the researchers took inspiration from nature to develop a generator of random structures based on a database of topological types of crystalline structures, merging the evolutionary approaches developed by Oganov and the topological approaches developed by Professor Vladislav Blatov of Samara. Knowing that almost all the 200 000 inorganic crystalline structures known to date belong to 3000 topological types, one can very quickly generate a set of structures similar to the desired structure. Tests have shown that, thanks to the new generator, the evolutionary search is carried out 3 times faster than the prediction tasks compared to the previous version.

"The 3,000 topological types are the result of an abstraction applied to real structures.Inversely, you can generate almost all known structures and an infinite number of unknown but reasonable structures from these 3,000 types. 39 is an excellent starting point for an evolutionary mechanism from the outset, you probably have a zone close to the optimal solution.You get the optimal solution from the beginning or you approach and you recover it by evolutionary improvement ", explains Pavel Bushlanov, the first author of the study and researcher at Oganov's laboratory in Skoltech.

Explore further:
Scientists solve the golden puzzle of calaveritis

More information:
Pavel V. Bushlanov et al, Topology-based crystalline structure generator, Communications in Computer Physics (2018). DOI: 10.1016 / j.cpc.2018.09.016