(199b) Multi-Layer Connectivity-Based Atom Contribution (m-CBAC) Approach: Fast and Accurate Charge Assignments to MOFs for Adsorption Simulations

Metal-organic frameworks (MOFs) have shown great potential in a variety of energy-related applications such as gas separations and storage. With increasing interest in this emerging class of novel materials, molecular simulations can play a critical role in the efficient exploration of their adsorption properties and thus the MOF discovery. In these calculations, accurate charge assignments to the framework atoms are essential. To facilitate the large-scale computational simulation of MOFs, we develop a multi-layer connectivity-based atom contribution (m-CBAC) approach as an efficient, robust, and accurate approach for charge assignments. Different from the original connectivity-based atom contribution (CBAC) method, which uses 1^st layer connectivity (i.e., directly bonded atoms) of a target atom, we expand on to incorporate multi-layer connectivity from the 0^th layer up to the 2^nd layer. This m-CBAC approach is developed with an automatic procedure of a supervised machine learning for classification, and it also allows further improvements along the time with more data to be included in the training set. In this study, an extensive set of ~2700 MOFs from the CoRE MOFs database with the density derived electrostatic and chemical (DDEC) charges are used to train the m-CBAC model. For the charge predictions, the m-CBAC approach assigns charges by successively searching from the highest-level layer (i.e., 2^nd layer, presumably the most accurate) to the lowest-level connectivity (i.e., 0^th layer, element type). With this approach, it becomes possible to assign charges for essentially all MOFs, with a high accuracy, and within seconds for a single structure. The m-CBAC approach is orders of magnitude computationally less expensive than quantum mechanical methods. The charges assigned using m-CBAC resemble the DDEC charges very well with a Pearson coefficient of as large as 0.988. The accurate charge assignments from m-CBAC are also reflected on their reliable predictions of CO₂ Henry coefficients in MOFs. Besides, a large-scale computation of the CO₂ Henry coefficients for ~12,000 MOFs in the recently released CoRE database has proved the capability of the m-CBAC approach. Overall, the m-CBAC approach can enable fast charge assignments for diverse MOFs with a good accuracy, and the software for m-CBAC charge assignments is made available along with this work.