Holographic current density theorem, neural network bootstrapping algorithm and PPM-GW corrected TDDFT(B)-NEGF method in first-principle calculations

Abstract

Density Functional Theorem (DFT)-based methods play a major role in modern computational chemistry. Their efficiency has been widely acknowledged and well celebrated. However, their accuracy is limited by a series of approximations. In this thesis, the accuracy and applicability of DFT-based methods are improved and augmented from 3 related perspectives. In Chapter 1, the theoretical justification of time-dependent (TD)-DFT in open system is extended into system subject to an arbitrary external magnetic field, opening up more opportunities to future time-dependent current density functional theorem (TDCDFT)-based methods’ developments and applications in open systems. In Chapter 2, DFT calculated heat of formation (HOF) for small molecules are systematically improved in accuracy by a novel method, the Neural-Network Bootstrapping (NNB) method, which is able to give not only a much more accurate HOF prediction, but also a molecule-specific prediction error estimation. In Chapter 3, Time-dependent Density Functional Tight Binding (TDDFTB)-Non-Equilibrium Green’s Function (NEGF) method in molecular device transport calculation are improved by the addition of explicit electron-electron interaction (EEI) effects calculated from the Plasmon-Pole Model (PPM)-GW method, through introduction of another self-energy in addition to the ones from leads’ coupling. Correspondingly, a new set of equation of motions (EOMs) are derived and implemented. All three developments either directly improved (TD)DFT-based methods’ accuracy or extended the range of applications where (TD)DFT-based methods can be applicable and reliable.