Machine-learning-based exchange correlation functional with physical asymptotic constraints

Density functional theory is the standard for computing electronic structure of materials, which based on a that maps electron density to energy. However, rigorous form not known and has been heuristically constructed by interpolating asymptotic constraints extreme situations, such as isolated atoms uniform gas. Recent studies have demonstrated can be effectively approximated using machine learning (ML) approaches. most ML models do satisfy constraints. In this paper, applying model architecture, we demonstrate neural network-based exchange-correlation satisfying physical Calculations reveal trained applicable various materials with an accuracy higher than existing functionals, even whose properties are included in training dataset. Our proposed method thus improves generalization performance ML-based combining advantages analytical modeling.