A chemically-driven machine learning computational approach that can be applied for a large variety of electronic structure theory methods will be discussed. In our methodology, we learn the electronic wave function from data obtained from small molecules. Different learners have been tested, such as the k-nearest neighbors and deep neural networks, for the prediction of correlation energies of larger molecules. For demonstrating the applicability of CDML, we are examining the convergence of the coupled-cluster singles-and-doubles (CCSD) iterative solver. Features are collected from quantum chemical data, such as orbital energies, one-electron Hamiltonian, Coulomb and exchange terms. Our preliminary data show a remarkable speed-up for the convergence of the CCSD equations. We anticipate extending this approach to electronic structure theory methods that capture the strong correlation of molecules.<br/>

Examining electronic structure theory methods with machine learning

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.