γ-Al₂O₃ is one of the most widely used catalyst supports in heterogeneous catalysis, yet its catalytic role remains unclear due to intrinsic structural disorder and complex surface species. In a study published in ENG. Chem. Eng. , researchers at Tianjin University employed a neural network-guided global optimization strategy to systematically explore the structural landscape of γ-Al₂O₃ and elucidate how its defect-rich nature influences the anchoring behavior of single-atom Pt and catalytic performance in propane dehydrogenation.
The team developed a high-dimensional machine learning interatomic potential for the Al-O-H system with near-DFT accuracy, using power-type structure descriptors and a feedforward neural network. Combined with the stochastic surface walking global optimization method, this approach enabled efficient exploration of the potential energy surface. Two low-energy configurations were identified: γ-NAV with no aluminum vacancies and γ-AV with aluminum vacancies, which are more stable than the widely used Digne's model by 34.44 and 11.38 meV·(f.u.)⁻¹, respectively.
The γ-NAV and γ-AV models retain an FCC-like oxygen sublattice while exhibiting significant disorder in cation distribution. The fractions of tetrahedrally coordinated Al³⁺ sites are 44.4 % for γ-NAV and 37.5 % for γ-AV, with the latter showing good agreement with theoretical values. Notably, the (100) surfaces of both models expose abundant penta-coordinated Al³⁺ sites, with surface coverage reaching up to approximately 60 %. These sites are widely recognized as strong Lewis acid centers and serve as preferred anchoring sites for Pt atoms.
Simulated X-ray diffraction patterns showed that the γ-AV model exhibits better qualitative agreement with experimental neutron diffraction data compared to the conventional Digne's model and other reported structures, particularly in the positions of major diffraction peaks and the overall distribution of main reflections.
Surface phase diagrams constructed under representative propane dehydrogenation conditions at 600 °C and 1 atm revealed that defect-rich surfaces are thermodynamically stabilized under realistic reaction conditions. For the γ-NAV model, a mixed configuration with both oxygen vacancies and hydroxyl groups is favored, while the γ-AV model is stabilized by a hydroxyl-rich surface. Bader charge analysis confirmed electron accumulation on isolated Pt atoms, indicating charge redistribution from the γ-Al₂O₃ support to Pt and demonstrating effective metal-support interactions.
Propane dehydrogenation reaction pathway calculations showed that both γ-NAV and γ-AV significantly reduce activation barriers compared to the conventional Digne's model, with γ-AV exhibiting the lowest energy barriers. The enhanced catalytic activity is attributed to the preferential stabilization of Pt atoms by surface defects, which provide more favorable active sites for C-H bond activation. The presence of oxygen vacancies and surface hydroxylation modulates the electronic properties of penta-coordinated Al³⁺ sites, strengthening metal-support interactions and facilitating C-H activation.
This work establishes a direct relationship between atomic structure, defect chemistry, and catalytic performance in γ-Al₂O₃, providing a theoretical foundation for the rational design of oxide-supported catalysts.
ENGINEERING Chemical Engineering
Experimental study
Not applicable
Machine learning-guided discovery of γ-Al2O3 as defect-rich structural ensemble governs single-atom Pt catalysis
20-Jul-2026