Ab Initio Multicomponent Methods
An accurate inclusion of nuclear quantum effects is important for a correct description of many chemical phenomena including zero point energy, hydrogen tunneling, and proton-coupled electron transfer. One promising class of methods for including nuclear quantum effects in quantum chemistry calculations are multicomponent self-consistent field methods (SCF), where electrons and select nuclei are both treated fully quantum mechanically by not invoking the Born-Oppenheimer approximation for the select nuclei.
Due to the qualitative failure of multicomponent Hartree-Fock (HF), recent advances in multicomponent SCF methods have focused on multicomponent density functional theory and the development of electron-proton correlation functionals. The qualitative failure of multicomponent HF indicates that the multicomponent SCF formalism is inherently multireference. We seek to investigate the nature of electron-proton correlation in the multicomponent SCF formalism using multicomponent multiconfigurational SCF methods.
Machine Learning for Theoretical Chemical Reaction Dynamics
Image adapted from doi: 10.1063/1.5019779
Machine learning is seeing rapid adoption in chemistry for a variety of applications. Possibly foremost among these applications is the prediction of molecular energies and of potential energy surfaces (PES). In the last year, a class of machine learning called continuous-filter convolutional neural networks have been shown to generate highly accurate PESs for systems near their equilibrium geometry. We seek to use continuous-filter convolutional neural networks to generate PESs for bimolecular reactive systems with a long term goal of performing nonadiabatic mixed quantum-classical surface hopping calculations using machine learning. Additionally, we seek to develop the next generation of neural networks for fitting PESs by leveraging recent machine learning improvements for 3D object recognition.