Diversity-driven, efficient exploration of a MOF design space to optimize MOF properties

Area of Science:

• Materials Science and Engineering
• Computational Chemistry
• Chemical Engineering

Background:

• Metal-organic frameworks (MOFs) offer promising properties for various applications.
• The vast design space of MOFs hinders efficient exploration and optimization.
• Discovering optimal MOFs for specific applications like ammonia (NH3) adsorption is challenging.

Purpose of the Study:

• To develop a novel algorithm, Vendi Bayesian optimization (VBO), for efficient MOF design space exploration.
• To introduce a new similarity function that integrates chemical and structural features for MOF comparison.
• To identify optimal MOF candidates for ammonia storage, removal, and capture applications.

Main Methods:

• Development of Vendi Bayesian optimization (VBO), combining Bayesian optimization with the Vendi score for diversity.
• Introduction of a novel kernel similarity function accounting for MOF chemical and structural properties.
• Statistical assessment of VBO using simulated campaigns for NH3 adsorption performance metrics.

Main Results:

• VBO consistently outperformed random search in identifying high-performing MOF designs for NH3 storage, removal, and capture.
• Identification of twelve extant and eight hypothesized MOFs with potentially record-breaking NH3 working capacity (ΔN_NH) between 23.6 and 29.3 mmol g⁻¹.
• Predicted MOFs exhibit thermal stability and low regeneration energy (approx. 10% of NH3 energy content).

Conclusions:

• VBO is an effective algorithm for navigating large MOF design spaces to find optimal materials.
• The developed similarity metric enhances VBO's ability to find MOFs with chemistry- and structure-dependent properties.
• Key design features for optimizing NH3 working capacity include a pore size of ~10 Å, heat of adsorption ~33 kJ mol⁻¹, and presence of Ca.