Accurate Deep Learning-Aided Density-Free Strategy for Many-Body Dispersion-Corrected Density Functional Theory

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Accurate modeling of intermolecular forces, particularly dispersion interactions, is crucial for predicting molecular and material properties.
• Traditional many-body dispersion (MBD) models often rely on computationally intensive electron density partitioning.
• Developing efficient and accurate methods for dispersion interactions remains an active area of research.

Purpose of the Study:

• To introduce a transferable, density-free many-body dispersion (DNN-MBD) model utilizing deep neuronal networks.
• To demonstrate the accuracy and computational efficiency of the DNN-MBD model compared to existing methods.
• To extend the applicability of MBD models beyond traditional electronic structure calculations.

Main Methods:

• Training a deep neuronal network (DNN) model on the ANI-1 dataset for small organic molecules.
• Developing a density-free approach that bypasses explicit electron density partitioning.
• Coupling the DNN-MBD model with density functional theory (DFT) and the Stochastic formulation of MBD equations.
• Implementing the DNN-MBD model within the Tinker-HP computational chemistry package.

Main Results:

• The DNN-MBD model achieves accuracy comparable to or exceeding existing MBD methods.
• Significant reduction in computational cost compared to traditional MBD models requiring explicit density partitioning.
• Successful application to large-scale, dispersion-corrected DFT calculations with preserved accuracy.
• Demonstrated applicability to force fields and neural network methodologies.

Conclusions:

• The proposed DNN-MBD model provides an accurate, efficient, and transferable method for calculating dispersion interactions.
• This density-free approach broadens the scope of MBD models, enabling their use in diverse computational chemistry applications.
• The DNN-MBD model represents a significant advancement in the accurate and cost-effective treatment of van der Waals forces in computational studies.