Towards Realistic Correlated Electronic Dynamics: New Developments in the Modeling of Harmonic Bath Models and Many Body Dispersion Interactions

Abstract

In this work we develop and characterize tools to compute realistic environmental models and accurate molecular structures, with the ultimate goal of enabling accurate correlated electronic dynamics. The procedure to compute environmental models starts with {\it ab inito} molecular dynamics, from which an autocorrelation function is computed. In this work we introduce super-resolution as a technique for recovering high resolution bath models from one quarter of the data that the Fourier transform requires. We further characterize the method and the closely related compressed sensing, to better understand transferability to other problems in chemistry. While work remains to improve upon our factor of four undersampling, we are optimistic that super-resolution will provide a path forward to computing accurate environmental models. To accurately model the structure and energetics of many materials, particularly those in condensed phase, it is frequently necessary to include dispersion in density functional theory (DFT) calculations. One of the most accurate techniques for including dispersion in DFT is the many-body dispersion (MBD) model, which models the dispersion energy as the correlation energy of an auxiliary quantum harmonic oscillator system. In this work we present gradients of MBD model with respect to the ions, unit cell parameters, and the charge density, which yield the forces, cell stresses, and dispersion potential respectively. To make the MBD model applicable to a wide range of systems, we develop an efficient implementation of the MBD gradients and energies within the Quantum ESPRESSO, FHI-AIMS, Octopus, and QChem software packages. We present results for gradient and unit cell optimizations for a wide range of systems, and find good agreement with reference values. In addition, we characterize the MBD's dependence on both the application of MBD self consistently and the inclusion of the Hirhsfeld volume gradients and find both are important for accurate forces. We present a framework for combining the MBD model with new exchange correlation functionals and provide this fitting data for twenty-four DFT exchange-correlation functionals. While signficant work still remains in further benchmarking the MBD model, we are encouraged by current results.