Understanding and controlling zeolite synthesis 

Toru Wakihara 

Institute of Engineering Innovation, The University of Tokyo, Tokyo, JAPAN 

Zeolites are artificially synthesized from silicon sources, aluminum sources, mineralizing agents and structure-directing agents in batch systems under hydrothermal conditions. To fulfill the wide industrial demands, efficient synthesis of zeolites with controllable crystallization kinetics is considered crucially important. It is believed that the amorphous aluminosilicate obtained by mixing raw materials first undergoes an induction stage, where its structure changes to a more stable amorphous state, and then a "crystal nucleus," the smallest entity that can be recognized as a crystalline phase, is formed. Thus, if the process of initial structural evolution of zeolites can be clarified and controlled, it will be possible to design and synthesize zeolites with controlled structure and composition in a rational manner. Furthermore, it is expected to make a significant contribution to the discovery of theoretical zeolites with unknown structure and composition. However, the multifaceted formation process of zeolites cannot be simply explained by the classical nucleation theory, in which monomers aggregate to reach a critical size. Furthermore, the aperiodic structure of the aluminosilicate precursor invalidates the conventional crystallographic characterization, and this amorphous-to-crystalline transition process is still a black box. This lecture will focus on the synthesis of zeolites, explaining how zeolites are formed in the hydrothermal synthesis process and how they can be controlled. 

Predicting zeolite polymorphism from experimental and computational data 

Rafael Gómez-Bombarelli 

Department of Materials Science and Engineering, MIT, Cambridge, USA 

Zeolites are nanoporous materials with many applications in separations and catalysis. Thanks to their topological diversity they are extremely versatile. Over 250 known polymorphs exist and thousands more are predicted to be possible. However, zeolite synthesis is a complex combinatorial problem with dozens of input variables, and it typically requires intensive trial and error in the lab. This results in challenges in controlling polymorphism (i.e. which phase will be formed under certain conditions) in a predictive fashion. 

Machine learning tools can leverage large datasets to address high-dimensional design challenges. We have recently shown that big-data tools, combining atomistic simulations, and automated literature extraction with natural-language process can produce insights and quantitative predictions for controlling these phases. For instance, solid-state interzeolite transformations can be rationalized by quantifying the similarity in chemical connectivity between frameworks. For templated zeolite synthesis, we have used over a million atomistic simulations, validated with exhaustive literature data extracted from the literature, to rationalize and design molecular templates.  

In this talk, we will provide a brief introduction to the discipline of machine learning, and delineate key concepts such as supervised vs. unsupervised learning, feature engineering vs. representation learning, training vs. test data, etc. Then, we will relate these foundational concepts in ML to successful applications in predicting zeolite synthesis.