Systematic exploration of host-guest potential energy surfaces in metal-organic frameworks

We present a computational framework for the systematic exploration of high-dimensional potential energy surfaces associated with host-guest interactions in porous materials. As case studies, the adsorption of benzyl alcohol, benzene, benzaldehyde, CO, CO₂, H₂S, and methanol in the metal-organic frameworks MFM-300(Sc) and MFM-300(In) was investigated. A stochastic sampling strategy generated thousands of candidate configurations, which were first screened at the semiempirical PM7 level and subsequently refined using dispersion-corrected density functional theory. To identify representative low-energy structures while preserving configurational diversity, a data-driven selection procedure based on outlier detection and unsupervised clustering of energetic and geometric descriptors was implemented. The exploration of approximately 39 000 configurations reveals that the adsorption landscapes are characterized by multiple low-lying minima within narrow energy windows, particularly for weakly interacting guest molecules. This behavior reflects the complex and highly corrugated potential energy surfaces associated with confinement in MOF pores and highlights the limitations of approaches based on chemically intuitive initial guesses or single optimized structures. The multilevel protocol provides a statistically robust and computationally efficient strategy for identifying adsorption sites and representative configurations in porous materials, and offers a transferable framework for exploring configurational landscapes in host-guest systems, heterogeneous interfaces, and other complex condensed-phase environments.