A consistent dynamics view on nanoporous catalysts in action across length and time scales

Abstract

This paper demonstrates the need for a consistent dynamic view of all processes from the nano- to mesoscale with quantum accuracy to design industrial catalysts. This stems from the observation that when a feed of molecules passes through a crystalline nanoporous framework, the following events occur: molecules enter the crystal, diffuse to active sites where they adsorb, desorb, or react, and finally products exit. These events span vastly different time and length scales—from picoseconds to hours and nanometers to micrometers. Length and time phenomena are entangled - the dynamics is altered by modifications at the length scale - leading to spatiotemporal phenomena. Current modeling practices treat reactions and diffusion inconsistently, often combining quantum methods for reactions with classical force fields for diffusion, and frequently neglect true dynamics under operating conditions. Industrial catalysts are highly sensitive to process conditions; some active sites form only at elevated temperatures, and reactive intermediates can change nature under these conditions. First-principles molecular dynamics captures these effects but becomes prohibitively expensive when relying on quantum methods like Density Functional Theory for force evaluations. In this paper, we demonstrate—through the study of key intermediates in CO₂-to-hydrocarbon conversion within small-pore zeolites—that diffusion must also be treated at the quantum level, as molecules can interact with active sites and even react while diffusing. Diffusion may become strongly hindered and compete with reactive events. In later case studies, we present proof-of-concept results for deriving diffusion coefficients in small-pore zeolites using machine learning potentials (MLPs) trained on quantum data. For benzene methylation in large-pore zeolites, we develop reactive MLPs to construct multidimensional free-energy surfaces, revealing competing pathways. Overall, the various cases underscore the need for a consistent dynamic description with quantum accuracy across scales. MLPs offer a promising route to obtain kinetic and transport properties from nano- to mesoscale at reduced cost, however their integration in overall multiscale modeling models and reaction-diffusion frameworks will be necessary to reach industrially relevant scales.