Screening out unfeasible hypothetical zeolite structures via the closest non-adjacent O⋯O pairs

Abstract

To boost function-led discovery of new zeolites with desired pores and properties, millions of hypothetical zeolite structures have been predicted via various computational approaches. It is now well accepted that most of these predicted structures are experimentally unrealisable under conventional synthetic conditions. Many structure evaluation criteria have been proposed to screen out unfeasible structures, among which the framework density-framework energy correlation criterion and the local interatomic distances criteria are the most frequently used. However, many hypothetical structures passing these criteria have been found unfeasible because of the existence of highly distorted framework rings. Here, we propose a new set of structure evaluation criteria to screen out such unfeasible structures. After optimising all existing zeolite structures as silica polymorphs, we find that the closest non-adjacent OO distances in existing zeolite rings generally show a normal distribution. By comparing the closest non-adjacent OO distances between existing and hypothetical zeolite structures, we are able to screen out many unfeasible hypothetical zeolite structures with distorted rings that are deemed feasible according to previous structure evaluation methods.

Necessity of Heteroatoms for Realizing Hypothetical Aluminophosphate Zeolites: A High-Throughput Computational Approach

“…To accelerate the discovery of new zeolitic materials, millions of hypothetical zeolite structures have been predicted via various computational methods. − Furthermore, many structure evaluation criteria have been proposed to predict the feasibility of the hypothetical structures. − However, their experimental realization still requires a priori knowledge on whether heteroatoms in addition to Al and P sources are necessary to be introduced into the synthetic system. Previous studies concerning the prediction of whether heteroatoms are necessary for specific AlPO structures are based on the substitution of Al atoms in pure AlPOs by heteroatoms.…”

Ranking the Synthesizability of Hypothetical Zeolites with the Sorting Hat

“…Restricting the range of correlations or discarding all angular information leads to degradation of classification performance, indicating that the structural features that distinguish real and hypothetical zeolites involve angular correlations and patterns in the relative positions of second and third neighbor atoms -i.e., at length scales beyond the typical indicators that have been hypothesized in previous studies [11][12][13][14][15][16][17][18].…”

“…Which hypothetical zeolites are most likely synthesizable, and in which chemical composition? Previous attempts to answer these questions [11][12][13][14][15][16][17][18] have relied on intuitive guesses for structural descriptors such as rings and angles, which provide incomplete [19] and thus biased results. In the present work, we answer all these questions via rigorous data science methods combining unsupervised and supervised machine learning, [20] along with the generalized convex hull (GCH) description of thermodynamic stability, [21] yielding a new and powerful approach for sorting real [22] and hypothetical [2][3][4][5] zeolites, as well as finding promising zeolite candidates and suggesting likely chemical compositions for them.…”

Toward a New Era of Designed Synthesis of Nanoporous Zeolitic Materials

“…Zeolite structure prediction generates a large number of theoretical zeolite structures, from which synthetic targets can be selected according to functional need. − Traditional zeolite prediction involves two major steps. The first step is to generate hypothetical zeolite structure models based on thermodynamic, topologic, or geometric regulations; − the second step is to evaluate the predicted structures and screen out unfeasible ones that would be difficult to realize experimentally. − Such two-step procedures lack efficiency. The model-generating step is computationally expensive, and the majority of the generated models are synthetically inaccessible and need to be screened out in the structure evaluation step.…”