Elucidating CO2 Dynamics in High-Entropy MOF-74 via Machine Learning Interatomic Potentials

Abstract

High entropy variants of MOF-74 have demonstrated enhanced CO₂ adsorption performance, yet it remains unclear whether homogenous metal mixing alone produces intrinsic enhancements in transport beyond the behavior of single metal parents. We develop a transferable machine-learned interatomic potential and perform large-scale molecular dynamics simulations of CO₂ diffusion in MOF-74 with M = Mg, Co, Cu, Ni, and Zn, together with mixed-metal frameworks including an equimolar high-entropy composition, an experimentally reported ratio, and M-rich 2:1:1:1:1 variants. At 300 K, single-metal axial diffusivities span 0.322 to 1.211 × 10⁻⁸ m²/s, with Mg slowest and Cu fastest. Strikingly, the equimolar high-entropy framework exhibits D_z = × 10⁻⁸ m²/s, reproduced within a few percent by a composition-weighted rule-of-mixtures built from the single-metal parents, and the same rule holds for the experimentally reported ratio and for M-rich mixtures. Temperature-dependent simulations from 300 to 500 K follow Arrhenius behavior with activation energies of 0.044-0.099 eV and prefactors of 6–15 × 10⁻⁸ m²/s, while the high-entropy framework remains within the parent range across the full window. These results establish a clean-limit transport baseline for compositionally complex MOF-74 and identify a simple, predictive rule for CO₂ diffusivity in homogeneous mixed-metal frameworks, providing a quantitative reference for assessing when experimentally relevant factors drive departures from rule-of-mixtures behavior.