[論文レビュー] Hierarchical Crystal Structure Prediction of Zeolitic Imidazolate Frameworks Using DFT and Machine-Learned Interatomic Potentials

Yizhi Xu, Jordan Dorrell+5 more|arXiv (Cornell University)|Jan 8, 2026

Metal-Organic Frameworks: Synthesis and Applications被引用数 0

ひとこと要約

The paper performs hierarchical crystal structure prediction (CSP) for ZnIm2 ZIFs using DFT-guided data and machine-learned interatomic potentials to survey a vast crystal-energy landscape, identifying thousands of minima and many novel topologies.

ABSTRACT

Crystal structure prediction (CSP) is emerging as a powerful method for the computational design of metal-organic frameworks (MOFs). In this article we employ CSP to perform high-throughput exploration of the crystal energy landscape of zinc imidazolate (ZnIm2). As the most polymorphic member of the zeolitic imidazolate framework (ZIF) family, ZnIm2 has at least 24 reported structural and topological forms, and new polymorphs still being regularly discovered. With the aid of custom-trained machine-learned interatomic potentials (MLIPs) we have performed a high-throughput sampling of over 3 million randomly-generated crystal packing arrangements and identified 9626 energy minima characterized by 1493 network topologies, including 864 topologies that have not been reported before. Experimentally-reported structures of ZnIm2, falling within the search boundaries, were all matched with the predicted structures, demonstrating the power of the CSP method in sampling experimentally-relevant ZIF structures. Finally, through a combination of topological analysis, density and porosity considerations, we have identified a set of structures representing promising targets for future experimental screening. as well as demonstrated how structures of mechanochemically-synthesized MOFs can be identified via matching the experimental powder diffraction patterns with the simulated patterns from the predicted structures.

研究の動機と目的

• Motivate crystal structure prediction as a tool for designing metal-organic frameworks (MOFs).
• Demonstrate high-throughput CSP to map the energy landscape of ZnIm2 ZIFs.
• Show that MLIPs enable large-scale sampling with accuracy close to DFT.
• Identify promising structural targets via topology, density, and porosity analyses.
• Demonstrate matching of experimental powder patterns to predicted structures for mechanochemical MOFs.

提案手法

• Use CSP to generate and sample over 3 million randomly-generated crystal packing arrangements of ZnIm2.
• Apply custom-trained machine-learned interatomic potentials (MLIPs) to rapidly evaluate energies of structures.
• Identify energy minima and classify them into 1484 network topologies, including 855 novel topologies.
• Perform topological analysis, density, and porosity assessments to select promising targets.
• Compare predicted structures to experimentally-reported ZnIm2 forms, matching all but one within search boundaries.
• Show how mechanochemical MOF structures can be identified by matching simulated and experimental powder diffraction patterns.

実験結果

リサーチクエスチョン

• RQ1 Can CSP combined with MLIPs efficiently map the crystal-energy landscape of ZnIm2 ZIFs at scale?
• RQ2 How many unique network topologies and energy minima exist for ZnIm2 within the CSP search space?
• RQ3 To what extent do CSP-predicted structures align with experimentally observed ZnIm2 forms?
• RQ4 Can topological, density, and porosity metrics identify experimentally relevant or synthesizable targets?
• RQ5 Can predicted structures be identified in experimental patterns from mechanochemical synthesis via powder diffraction matching?

主な発見

• Identified 9609 energy minima characterized by 1484 network topologies.
• Among these, 855 topologies are not previously reported.
• All but one experimentally-reported ZnIm2 structures fell within the search boundaries and were matched by predicted structures.
• High-throughput CSP combined with MLIPs effectively samples experimentally-relevant ZIF structures.
• Analyses yielded a set of structures representing promising targets for future experimental screening.
• Demonstrated how predicted structures can be matched to experimental powder diffraction patterns to identify mechanochemically-synthesized MOFs.