Novel Approach for Clustering Zeolite Crystal Structures.

Informatics approaches play an increasingly important role in the design of new materials. In this work we apply unsupervised statistical learning for identifying four framework-type attractors of zeolite crystals in which several of the zeolite framework types are grouped together. Zeolites belonging to these super-classes manifest important topological, chemical and physical similarities. The zeolites form clusters located around four core framework types: LTA, FAU, MFI and the combination of EDI, HEU, LTL and LAU. Clustering is performed in a 9-dimensional space of attributes that reflect topological, chemical and physical properties for each individual zeolite crystalline structure. The implemented machine learning approach relies on hierarchical top-down clustering approach and the expectation maximization method. The model is trained and tested on ten partially independent data sets from the FIZ/NIST Inorganic Crystal Structure Database.