Accurate and Transferable Pauli Exchange-Repulsion for Molecules with the Anisotropic Valence Density Overlap Model

Abstract

Pauli exchange-repulsion is the dominant short-range intermolecular interaction and it is an essential component of molecular force fields. Current approaches to modeling Pauli repulsion in molecular force fields often rely on over 20 atom types to achieve chemical accuracy. The number of parameters in these approaches hampers the development of force fields with quantum-chemical accuracy that are transferable across many chemical systems. We present the anisotropic valence density overlap (AVDO) model for exchange-repulsion. The model produces sub-kcal/mol accuracy for dimers of organic molecules from the S101x7 dataset, a representative set of the most common biologically relevant intermolecular interactions, and for acene dimers of increasing size. It uses a single universal parameter, related to an atomic cross-sectional area, that is transferable across chemical systems. Given recent progress in machine learning the electronic density, this model offers a promising path toward high-accuracy, next-generation machine-learned force fields.

Figures (3)

• Figure 1: A 2D cross-section of the redistribution of charge due to antisymmetrization for the helium homodimer. Electrons move away from the overlap region and onto each atom.
• Figure 2: The density overlap model (lines) fitted individually for the F$_2$ dimer and the uracil dimer. Squares and circles are the SAPT(DFT) exchange repulsion energies for the parallel or $\pi-\pi$ stacking dimer configurations (circles) and the linear or base-pair dimer configurations (squares). Diamonds mark the minima of the total interaction energy along that trajectory. The models used the all-electron density, the density without orbitals from core electrons, or the density without orbitals corresponding to core electrons and semicore electrons of N,O, and F atoms, which we call the valence density.
• Figure 3: Exchange repulsion energies (kcal/mol) on the training/test dataset for models using one universal $K$ parameter with either the all electron density or the valence density.