Machine-learning approach to the design of OSDAs for zeolite beta

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Zeolite beta synthesis relies on organic structure directing agents (OSDAs).
• Predicting OSDA effectiveness traditionally requires computationally intensive molecular dynamics simulations.
• A need exists for faster, more efficient methods to design novel OSDAs.

Purpose of the Study:

• To develop a machine learning strategy for designing effective OSDAs for zeolite beta.
• To replace computationally expensive molecular dynamics calculations with a neural network prediction model.
• To accelerate the discovery of new OSDAs for zeolite beta synthesis.

Main Methods:

• Trained a neural network on 4,781 candidate OSDAs to predict stabilization energies.
• Utilized an evolutionary design algorithm to explore the chemical space of feasible OSDAs.
• Validated predicted stabilization energies against molecular dynamics computations.

Main Results:

• The neural network predictions showed high correlation with molecular dynamics results.
• The evolutionary algorithm thoroughly sampled chemically feasible OSDAs.
• Identified 469 OSDAs with stabilization energies below -17 kJ/(mol Si), outperforming known agents.

Conclusions:

• The machine learning strategy is effective for designing OSDAs for zeolite beta.
• The identified OSDAs offer significant improvements and expand the pool of potential agents.
• This approach is expected to facilitate the synthesis of new zeolite beta materials.