Machine-Learned Kohn-Sham Hamiltonian Mapping for Nonadiabatic Molecular Dynamics

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Accurate electronic structure calculations are crucial for understanding material properties and dynamics.
• Traditional methods for solving Kohn-Sham equations can be computationally expensive, especially for large systems.
• Non-self-consistent Hamiltonians offer a faster alternative but lack accuracy for dynamic simulations.

Purpose of the Study:

• To develop a machine-learning (ML) approach for efficient Hamiltonian mapping.
• To create a fast surrogate Hamiltonian calculator for nonadiabatic dynamics simulations.
• To enable accurate simulations of large atomistic systems and explore excitation energy relaxation dynamics.

Main Methods:

• A machine-learning model was trained to map non-self-consistent Kohn-Sham Hamiltonians to nearly self-consistent ones.
• Input and output features were Hamiltonian matrices calculated at different levels of theory.
• The ML model was applied to nonadiabatic dynamics simulations of excitation energy relaxation.

Main Results:

• The ML-based Hamiltonian mapping significantly speeds up calculations (orders of magnitude).
• The method achieves accuracy comparable to conventional calculations for molecular orbitals and energies.
• Simulations of excitation energy relaxation in C60 fullerene and Si75H64 quantum dots provided new insights.

Conclusions:

• The developed ML approach is simple, efficient, scalable, and broadly applicable.
• It enables accurate and fast simulations of complex systems, advancing the study of excitation energy dynamics.
• This method offers a powerful tool for computational materials science and chemistry.