4096474

Machine learning many-body quantum chemistry of molecules and materials via single-particle Green’s functions

Date
August 20, 2024

We present a machine learning (ML) approach for predicting single-particle Green's functions of molecules and materials, from which various ground- and excited-state electronic properties at quantum many-body level can be extracted. Self-energy matrix elements on compact imaginary frequency grids are predicted from static and dynamical mean-field electronic features through a graph neural network (GNN). A symmetry-adapted intrinsic atomic orbital basis is introduced to enforce rotational invariance in ML features and targets. We demonstrate good transferability and high data efficiency of the proposed ML method across molecular sizes and chemical species by showing accurate predictions of density of states, quasiparticle energies, and density matrices at the levels of many-body perturbation theory (GW) and coupled-cluster theory (CCSD). Extensions to large organic molecules and nanoclusters will also be discussed.

Presenter

Speaker Image for Tianyu Zhu
Yale University

Related Products

Thumbnail for Green's function quantum embedding with local natural orbitals
Green's function quantum embedding with local natural orbitals
We present a new ab initio Green's function embedding method, interacting-bath dynamical embedding theory (ibDET), for simulating charged excitations of molecules…
Thumbnail for Interacting-Bath dynamical embedding for capturing nonlocal electron correlation in solids
Interacting-Bath dynamical embedding for capturing nonlocal electron correlation in solids
Quantitative simulation of electronic structure of bulk solids requires treating local and non-local electron correlations on an equal footing…
Thumbnail for Machine learning Green’s functions for predicting many-body effects in silicon nanocrystals | Poster Board #670
Machine learning Green’s functions for predicting many-body effects in silicon nanocrystals | Poster Board #670
Quantum many-body methods provide a systematic approach for modeling photoemission and optical spectra, which can reveal charge and energy transfer in light-matter interactions…
Thumbnail for Machine learning many-body Green’s function for molecular excitation spectra
Machine learning many-body Green’s function for molecular excitation spectra
The accurate simulation of light-matter interactions in molecular and periodic systems is important to many application areas, including photovoltaics, optoelectronics, and spectroscopy…