Understanding Dynamics of Nanocluster-Organic Frameworks and Gas Diffusion from Machine Learning Potential-based Simulations

Nanocluster-organic frameworks (NOFs) have emerged as unique materials with broad applications in sensing, photocatalysis, and optoelectronics. These photofunctional materials have an excellent luminescence switching response to gases such as oxygen (\ce{O2}) and volatile organic compounds. However, the atomistic structural evolution of the NOFs and the mechanism governing small molecule diffusion inside them remain poorly understood. In this study, we developed machine learning potentials to accurately model an experimentally synthesized NOF, \ce{[Ag12(S^{t}Bu)8(CF3COO)4(bpy)4)]_n}, and investigated its structural and dynamic properties using molecular dynamics simulations. Furthermore, we used on-the-fly probability-enhanced sampling (OPES) simulations to capture the pore-to-pore transitions of O$_2$ gas in the NOF and construct the underlying free energy surface (FES). Our results reveal that the \ce{O2} gas predominantly localizes around the bipyridine (bpy) linker, consistent with previous experimental observations. In particular, the nominal pore-to-pore diffusion barrier of $\sim$16-18 kJ/mol for the \ce{O2} transition suggests its feasible diffusion at room temperature. This work presents the first-ever integrated approach for modelling a fully flexible NOF and gas diffusion within it with DFT-level accuracy. Our findings lay the foundation for future investigations on NOFs for potential energy storage, catalysis, and biosensing applications.