Role of Pore Chemistry and Topology in the CO2 Capture Capabilities of MOFs: From Molecular Simulation to Machine Learning

Open framework materials (OFMs) such as metal–organic frameworks (MOFs) can provide structurally and chemically tailorable nanopores. This exceptional tunability has allowed for careful positioning of optimal adsorption sites within MOF pores to enable selective CO2 physisorption, making these materials promising for energy-efficient CO2 capture. However, given the multitude of features that can be simultaneously altered within the thousands of MOFs synthesized to date, it can be daunting to elucidate the most critical features for boosting CO2 capture capabilities. Here we use a multiscale approachdensity functional theory (DFT), grand canonical Monte Carlo (GCMC), and machine learning (ML)to investigate the role of various pore chemical and topological features in the enhancement of CO2 capture metrics of MOFs. To enable a thorough “sweep” of a target region of MOF structure-space, we used computational synthesis methods to create sets of MOFs encompassing all possible combinations of 16 topologies and 13 functionalized molecular building blocks. The adsorption of pure CO2, and CO2/H2 and CO2/N2 mixtures for the resulting 31 parent MOFs and its derivatives was then simulated, and CO2 capture metrics were calculated. Functionalization with hydroxyl, thiol, cyano, amino, or nitro chemistries was found to often improve CO2 capture metrics of the parent MOFs, but the efficacy of this strategy depended strongly on the pore topology. Decision trees were trained to predict the improvement or decline of CO2 capture metrics upon functionalization of parent MOFs, whereas five additional machine learning algorithms were trained to predict absolute metrics for all MOFs. The training of these algorithms allowed us to determine, without human bias, the relative importance of various pore chemical and structural/topological factors on the CO2 capture capabilities of MOFs.开放骨架材料（Open Framework Materials, OFMs）如金属有机框架（Metal-Organic Frameworks, MOFs），可提供结构与化学性质均可调控的纳米孔道。这种卓越的可调性使得研究人员能够精准排布MOF孔道内的最优吸附位点，实现选择性二氧化碳物理吸附（CO₂ physisorption），从而使这类材料有望应用于节能型二氧化碳捕集。然而，迄今已合成的数千种MOF中，可同时调控的特征繁多，想要阐明提升二氧化碳捕集性能的关键特征极具挑战。本研究采用多尺度方法——密度泛函理论（Density Functional Theory, DFT）、巨正则蒙特卡洛（Grand Canonical Monte Carlo, GCMC）与机器学习（Machine Learning, ML）——探究孔道化学与拓扑结构特征对MOF二氧化碳捕集性能指标的调控作用。为全面遍历MOF结构空间的目标区域，我们采用计算合成方法构建了一系列MOF，涵盖16种拓扑结构与13种功能化分子构筑基元的所有可能组合。随后对所得到的31种母体MOF及其衍生物的纯二氧化碳、CO₂/H₂及CO₂/N₂混合气体吸附行为进行了模拟，并计算了二氧化碳捕集性能指标。研究发现，通过羟基、巯基、氰基、氨基或硝基基团进行功能化修饰，通常可改善母体MOF的二氧化碳捕集性能，但该策略的效果很大程度上取决于孔道拓扑结构。我们训练了决策树模型以预测母体MOF经功能化修饰后二氧化碳捕集性能指标的升降变化，另外还训练了5种机器学习算法来预测所有MOF的绝对性能指标。通过这些算法的训练，我们得以摆脱人为偏见，确定各类孔道化学与结构/拓扑因素对MOF二氧化碳捕集性能的相对重要性。