Ab Initio Molecular Cavity Quantum Electrodynamics Simulations Using Machine Learning Models.

Area of Science:

• Quantum chemistry
• Theoretical chemistry
• Computational physics

Background:

• Molecule-cavity hybrid systems exhibit unique polariton dynamics.
• Simulating non-adiabatic dynamics in these systems is computationally challenging.
• Accurate calculation of molecular properties, like dipole derivatives, is crucial for polaritonics.

Purpose of the Study:

• To present a novel mixed quantum-classical simulation for polariton dynamics in molecule-cavity systems.
• To incorporate machine learning for accurate calculation of dipole derivatives.
• To investigate the photoinduced isomerization of azomethane within an optical cavity.

Main Methods:

• Mixed quantum-classical simulation using trajectory surface hopping.
• Pauli-Fierz quantum electrodynamics Hamiltonian for nuclear gradients.
• On-the-fly complete active space self-consistent field (CASSCF) calculations.
• Kernel ridge regression for machine-learning dipole derivatives.
• Lindblad jump superoperator for cavity loss modeling.

Main Results:

• Validated the accuracy of machine-learned dipoles and their derivatives.
• Successfully simulated polariton dynamics, including non-adiabatic effects.
• Demonstrated that azomethane isomerization is tunable by cavity coupling and light-matter interaction.

Conclusions:

• The developed simulation method accurately captures polariton dynamics in molecule-cavity systems.
• Machine learning is effective for obtaining essential molecular properties in polariton simulations.
• Optical cavities offer a promising route to control chemical reactions like azomethane isomerization.