Files
Abstract
Quantum chemistry methods have proven tremendously useful in the prediction of chemical phenomena and the characterization of complex molecular systems. In this work, we explore how recent developments in automatic differentiation, machine learning algorithms, and automatic code generation can be exploited to provide new innovative approaches in applying quantum chemistry techniques to challenging problems. In particular, we present for the first time how automatic differentiation can be leveraged to obtain arbitrary-order energy derivatives for various quantum chemistry methods, which are critical for computing various observables of chemical interest. We then study how machine learning models can be used as representations of molecular potential energy surfaces, and present a novel software package for automating the production of these models. Finally, we explore how joint electron correlation and basis set extrapolation techniques can be used to approach an exact theoretical characterization of challenging open-shell molecules. This is enabled by the recent developments in automated code generation, namely the capability to perform coupled cluster computations with an arbitrary number of excitations.