Atlas describes structural regularities between proposed zeolites

Zeolites comprise a wide array of nanoporous crystalline materials that are used as catalysts in shape-selective transformations of hydrocarbons and as molecular sieves for separating mixtures. Scientists have theorized the existence of millions of zeolites, but, despite their chemical simplicity, creating these new structures has proved challenging. Artificial intelligence provides a way to comb through these numerous possibilities and leads to a better understanding of how to make them.

Helfrecht et al. present a new resource that describes regularities between zeolites. Applying a machine learning technique using the smooth overlap of atomic positions (SOAP) method on the Deem database of hypothetical zeolites, the group analyzed structural motifs of hypothetical zeolites and constructed an atlas of local structures.

“All too often machine learning studies are tweaked to reproduce human intuition,” said author Michele Ceriotti. “We compared approaches based on hard numbers, and demonstrated that some of the commonly-used descriptors are incapable of predicting stability or density of a given framework.”

Machine learning analysis of the molar energy and volume for each structure in the dataset allowed the group to assess the quality of several standard descriptors such as distances, angles and ring sizes, for head-to-head comparison with the SOAP method.

They used these assessments to produce a 3D point “cloud atlas” of local environments that showed good correlations with the contribution of a given motif to the density and stability of its parent framework.

Ceriotti hopes the findings lead to further application of machine learning in atomistic modeling. Next, they will investigate where in the space of millions of potential structures lie those that are most likely to yield successful syntheses.

Source: “A new kind of atlas of zeolite building blocks,” by Benjamin A. Helfrecht, Rocio Semino, Giovanni Pireddu, Scott M. Auerbach, and Michele Ceriotti, Journal of Chemical Physics (2019). The article can be accessed at https://doi.org/10.1063/1.5119751 .