1.
Focassio, Bruno; Domina, Michelangelo; Patil, Urvesh; Fazzio, Adalberto; Sanvito, Stefano
Linear Jacobi-Legendre expansion of the charge density for machine learning-accelerated electronic structure calculations Journal Article
Em: npj Comput Mater, vol. 9, não 1, 2023, ISSN: 2057-3960.
@article{Focassio2023,
title = {Linear Jacobi-Legendre expansion of the charge density for machine learning-accelerated electronic structure calculations},
author = {Bruno Focassio and Michelangelo Domina and Urvesh Patil and Adalberto Fazzio and Stefano Sanvito},
doi = {10.1038/s41524-023-01053-0},
issn = {2057-3960},
year = {2023},
date = {2023-12-00},
journal = {npj Comput Mater},
volume = {9},
number = {1},
publisher = {Springer Science and Business Media LLC},
abstract = {AbstractKohn–Sham density functional theory (KS-DFT) is a powerful method to obtain key materials’ properties, but the iterative solution of the KS equations is a numerically intensive task, which limits its application to complex systems. To address this issue, machine learning (ML) models can be used as surrogates to find the ground-state charge density and reduce the computational overheads. We develop a grid-centred structural representation, based on Jacobi and Legendre polynomials combined with a linear regression, to accurately learn the converged DFT charge density. This integrates into a ML pipeline that can return any density-dependent observable, including energy and forces, at the quality of a converged DFT calculation, but at a fraction of the computational cost. Fast scanning of energy landscapes and producing starting densities for the DFT self-consistent cycle are among the applications of our scheme.},
keywords = {Computer Science Applications, General Materials Science, Mechanics of Materials, Modeling and Simulation},
pubstate = {published},
tppubtype = {article}
}
AbstractKohn–Sham density functional theory (KS-DFT) is a powerful method to obtain key materials’ properties, but the iterative solution of the KS equations is a numerically intensive task, which limits its application to complex systems. To address this issue, machine learning (ML) models can be used as surrogates to find the ground-state charge density and reduce the computational overheads. We develop a grid-centred structural representation, based on Jacobi and Legendre polynomials combined with a linear regression, to accurately learn the converged DFT charge density. This integrates into a ML pipeline that can return any density-dependent observable, including energy and forces, at the quality of a converged DFT calculation, but at a fraction of the computational cost. Fast scanning of energy landscapes and producing starting densities for the DFT self-consistent cycle are among the applications of our scheme.
