Predicting Structural Properties of Pure Silica Zeolites Using Deep Neural Network Potentials

Abstract

Machine learning potentials (MLPs) capable of accurately describing complex ab initio potential energy surfaces (PESs) have revolutionized the field of multiscale atomistic modeling. In this work, using an extensive density functional theory (DFT) data set (denoted as Si-ZEO22) consisting of 219 unique zeolite topologies (350,000 unique DFT calculations) found in the International Zeolite Association (IZA) database, we have trained a DeePMD-kit MLP to model the dynamics of silica frameworks. The performance of our model is evaluated by calculating various properties that probe the accuracy of the energy and force predictions. This MLP demonstrates impressive agreement with DFT for predicting zeolite structural properties, energy–volume trends, and phonon density of states. Furthermore, our model achieves reasonable predictions for stress–strain relationships without including DFT stress data during training. These results highlight the ability of MLPs to capture the flexibility of zeolite frameworks and motivate further MLP development for nanoporous materials with near-ab initio accuracy.

Operando Modeling of Zeolite-Catalyzed Reactions Using First-Principles Molecular Dynamics Simulations

“…Although such force fields are specifically parametrized to reach high accuracy for a set of systems or conditions, they are often not transferable to other systems . Lastly, there are some notable examples of reactive force fields which allow simulation of reactive events within the field of catalysis, but still the overall outcome depends on the parametrization of the force field. − Interesting new directions are currently being explored to derive machine learning potentials (MLPs), where starting from underlying quantum-mechanical data sets a numerical potential is derived to describe the PES using a nonlinear regression method. − This approach is very promising, but the applications in the field of zeolite catalysis, which is characterized by an enormous complexity at various levels, is nearly nonexistent. ,− We elaborate in Outlook and Future Directions on new possibilities within the field of modeling zeolite catalysis when having access to cheaper methods to evaluate the PES.…”

Zeolite encapsulated organometallic complexes as model catalysts

“…[173][174][175][176] Machine learning based potentials (MLPs) have recently emerged as a promising method of accurately modelling the properties and dynamics of several systems and reactions. [177][178][179][180][181][182][183][184][185] As a result, MLPs are iteratively trained to 'learn' the potential energy surface of the system (based on limited DFT data) and can serve as a viable substitute for QM/MM and QM/QM schemes and apply to systems of arbitrary size at almost DFT level of accuracy. Consequently, MLPs can help to perform high-throughput screening of several catalyst configurations in multiple zeolites 179 if desired, in addition to being able to model system dynamics.…”

“…Consequently, MLPs can help to perform high-throughput screening of several catalyst configurations in multiple zeolites 179 if desired, in addition to being able to model system dynamics. 178 Dalton Transactions Frontier…”

Full‐scale modeling of chemical experiments

“…For instance, Bombarelli et al first constructed a complete set of high-throughput simulation frameworks, including the criteria for screening OSDAs, automatic acquisition of data from the experimental literature, automatic dynamics simulation, and construction of the database, and then new OSDAs were designed by combining the machine learning method; finally, the results of the experiments validated the efficiency of the design. [25][26][27][28][29]…”