Predicting Atomic Charges in MOFs by Topological Charge Equilibration

Read full text
Authors: Babak Farhadi Jahromi, Sumukh Shankar Sharadaprasad, Rochus Schmid
Cited by: 1
References: 77

Abstract

ABSTRACT An electrostatic model, based on the well‐established charge equilibration method is presented, that allows for a fast prediction of partial atomic charges within non‐reactive force fields for metal–organic frameworks with typical elemental compositions. The required molecular information is inferred in a topological fashion, with generic topological bond lengths, based on experimental covalent radii of the contained elements. The model is parameterized on the basis of atom types, that encode the local atomic coordination environment up the nearest bonded neighbors. Consequently, the model is geometry‐independent, requiring only a graph representation of the molecular bonds, which is easily retrieved due to the explicit bond definitions in non‐reactive force fields. Therefore, by sacrificing the dependence of the atomic charges on changes in the system's geometry, the proposed model avoids the computation of distance matrices, increasing the overall performance. By parameterizing the model on three different atomic encoding levels of increasing parametric resolution, the model's accuracy is systematically increased. To this end, a global optimization scheme is employed, in which the model parameters of atomic width, electronegativity and hardness, are trained to reproduce a set of reference charges. These, were obtained from the QMOF database, which hosts electronic structure properties computed for over 20 000 MOFs and related materials. Curated subsets of the QMOF database were generated depending on the employed atom encoding scheme with a balanced atom type incidence. These subsets were used for the training and testing of the new model, containing atomic reference charges of around 10 000 MOFs for selected metallic species and with comparable relative atom type distributions between the sets. The model is able to predict charges with an accuracy below 0.02 e for the most detailed atom typing scheme for a wide chemical space without the need to to correct for charge neutrality.

Fields of Study & Subjects

Metal-Organic Frameworks: Synthesis and Applications Machine Learning in Materials Science Crystallography and molecular interactions

Publication details

Journal / source
Advanced Functional Materials
Publisher
Wiley
Volume
36
Issue
43
Publication date
7 February 2026
Language
EN