Year: 2026
Type: preprint
Abstract: The relationship between pore network structure and molecular diffusion in disordered nanoporous materials remains unclear. Here, we revisit two methodological choices that may contribute to inconsistent structure–diffusion relationships in the literature: percolation characterization and inclusion of non-percolating pores in diffusivity calculations. Our rigorous benchmarks show consistent results from Zeo++ and Mofography but reveal biased characterization from PoreBlazer due to limitations in its algorithm. Molecular dynamics simulations further consolidate that initializing adsorbates in percolating pore networks is essential for reliable and robust calculations of molecular self-diffusivity. With these issues addressed, we re-examine structure–diffusion relationships for methane across four amorphous material classes. Symbolic regression identifies the percolating backbone fraction as a key, yet largely overlooked, topology-related descriptor that captures a common diffusion trend across classes. Further analysis supports the added predictive value of this new descriptor beyond conventional geometric descriptors in machine learning tasks. Our work provides a consistent methodological foundation and a complementary descriptor for diffusion prediction in disordered nanoporous materials.
Source: ChemRxiv
Authors: Chao Zheng, Arun Gopalan, Kaihang Shi
Institution: University at Buffalo, State University of New York
Language: English
Subfield: Inorganic Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223660412