Abstract: Identifying the active sites driving DeNOx catalysis in iron-modified zeolites remains a significant challenge due to their structural complexity. This study focuses on dicationic iron species (monomers and dimers) hosted at cation-exchange positions associated with framework aluminum pairs to establish structure–activity relationships. We developed a comprehensive workflow integrating genetic algorithms (GA), topological analysis, unsupervised machine learning, and machine learning interatomic potentials (MLIPs), specifically the equivariant message passing neural network and atomic cluster expansion (MACE) framework, to efficiently explore this vast landscape. We introduce “redox strain” (R)—the energy cost of structural reorganization upon oxidation—as a primary activity descriptor. Mapping the energy landscape reveals fundamental distinctions between frameworks. The structurally diverse MFI topology provides a high “topological ceiling”, stabilizing distinct sites that maximize redox efficiency. In contrast, higher-symmetry frameworks such as FAU act as thermodynamic sinks, binding oxygen in chemically inert configurations due to a lack of steric strain. We conclude that a high “topological ceiling” may serve as a useful descriptor for identifying high-performance redox catalysts for DeNOx applications.

Cite: Koveza V.A. , Kuziuberdina E.O. , Potapenko O.V.
Common structural motifs of Fe-species at paired Al sites in MFI, FER, and FAU zeolites for DeNOx applications
Molecular Catalysis. 2026. V.599. 116003 :1-19. DOI: 10.1016/j.mcat.2026.116003

Dates:

Submitted:	Feb 27, 2026
Accepted:	Apr 24, 2026
Published online:	May 14, 2026
Published print:	Jun 15, 2026

Identifiers: No identifiers

Altmetrics: