Precise characterization of void space in nanoporous materials is critical for advancing adsorption, diffusion, and catalysis applications, but remains challenging with conventional algorithms. We introduce Mofography, a novel computational framework that decomposes the void space in nanoporous materials into a network of accurately defined, non-overlapping pockets. This is achieved by converting a material’s structure into a distance-to-framework field and applying advanced segmentation algorithm, with rigorous treatment of periodic boundary conditions. Mofography constructs a pore graph for the segmented pore network where nodes represent pockets and edges represent pore connections. Characterization of pore graphs allows integration of local descriptors for individual pockets with graph representations, enabling comprehensive annotation of pore regions with geometric, energetic, and chemical properties. We demonstrate the usefulness of Mofography in three applications: (1) detecting pore windows by identifying centroids along shared boundaries; (2) classifying pocket types through comparison of cavity size and van der Waals energy distributions; and (3) rapidly identifying shape-selective pockets for xylene isomers using a Bayesian optimization framework, subsequently validated by grand canonical Monte Carlo simulations. Mofography offers a robust platform for enhanced visualization and analysis of complex pore space, paving the way for accelerated material discovery and design, with potential integration with graph theory and machine learning.
