The discovery and optimization of metal-organic frameworks (MOFs) for carbon-neutral gas adsorption require a detailed understanding of the interplay between structure, composition, and function. This dissertation presents a hierarchical, data-driven modeling framework to accelerate MOF design and simulation, spanning physics-informed thermal conductivity prediction, the development of accurate interatomic potentials, and machine learning-guided structure recommendation.
We first investigate the thermal transport properties of zeolitic imidazolate frameworks (ZIFs). Through large-scale molecular dynamics simulations of 196 diverse ZIFs, we construct a circuit-based model that captures the influence of topology and ligand functionalization on thermal conductivity. This physics-informed approach enables the rational design of MOFs with tunable thermal properties.
To enable accurate and efficient simulations of gas adsorption in flexible frameworks, we propose a fragment-to-framework training strategy for general MOF potentials. By leveraging the modularity of MOFs, we incorporate configurations of building blocks into the training data, achieving transferability to structures composed of these building blocks. We further identify the chemical environments to which the model can be reliably transferred, guiding the design of training datasets for general machine learning potentials.
Finally, to efficiently explore large chemical spaces, we develop an unsupervised recommendation model that captures structural similarity and functional labels. Given a known MOF or an application-specific performance tag, our model suggests compositionally and structurally relevant candidate materials, serving as a structureâ function mapping engine.
Together, these studies establish a comprehensive strategy for data-guided MOF discovery and modeling, unifying structureâ function insight, recommendation capabilities, and fast, accurate simulation tools for property evaluation, ultimately supporting the design of materials for carbon-neutral technologies.
