Learning Functional Theories of the 1- and 2-Electron Reduced Density Matrices

In this 52-minute lecture, Michele Pavanello of Rutgers University-Newark presents innovative research on machine learning approaches to quantum chemical models at IPAM's Optimal Transport for Density Operators workshop. Explore how machine learning methods can be effectively applied to information-rich quantities like electron density, density matrices, and wavefunctions rather than single quantities such as energy or dipole. Discover models for one-electron density matrices of small to medium-sized molecules, understand the importance of N-representability conditions, and learn about methods for two-electron density matrices that explicitly account for electronic correlation at reduced computational cost. The presentation addresses the challenge of current quantum chemical models being either too approximate or too computationally expensive, offering machine learning solutions through the open-source QMLearn software (http://qmlearn.rutgers.edu). The lecture references key publications on machine learning electronic structure methods and entropy approximations to electronic correlation energy.