Machine learning the Hohenberg-Kohn map for molecular excited states

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Density-functional theory (DFT) links electron density to system properties.
• Time-Dependent DFT (TDDFT) models electronic excitations but is computationally intensive.
• Approximations in TDDFT limit accuracy and efficiency.

Purpose of the Study:

• To develop a more efficient method for calculating electronic excited states.
• To bypass the computational cost and approximations of TDDFT.
• To enable direct mapping from electron density to excited-state properties using machine learning.

Main Methods:

• Utilizing machine learning to determine density and energy functionals.
• Developing a novel framework for excited-state dynamics simulations.
• Applying machine-learned functionals to malonaldehyde for proton transfer studies.

Main Results:

• Successfully bypassed the equations of TDDFT using machine-learned functionals.
• Performed the first excited-state molecular dynamics simulations with a machine-learned functional.
• Accurately captured the kinetics of excited-state intramolecular proton transfer in malonaldehyde.

Conclusions:

• Machine learning offers a computationally cheaper and direct route to excited-state properties.
• This approach enables highly efficient excited-state dynamics simulations.
• Provides a new tool for understanding and controlling chemical reactions, like proton transfer.