Ultrafast dynamics in spatially confined photoisomerization: accelerated simulations through machine learning models

Area of Science:

• Photochemistry
• Computational Chemistry
• Supramolecular Chemistry

Background:

• Photoresponsive host-guest systems involve complex interactions between confined spaces and photosensitive molecules.
• Simulating these systems using nonadiabatic molecular dynamics (NAMD) is computationally intensive.

Purpose of the Study:

• To develop and apply a machine learning (ML) accelerated NAMD approach for efficient simulation of photoisomerization in host-guest systems.
• To investigate the influence of the cucurbit[5]uril host environment on the excited-state dynamics of the benzopyran guest molecule.

Main Methods:

• Utilized quantum mechanics/molecular mechanics (QM/MM) combined with ML-based NAMD simulations.
• Constructed excited-state potential energy surfaces along collective variables.
• Developed an ML-based nonadiabatic dynamics model for comparative analysis.

Main Results:

• Identified key reaction pathways and degrees of freedom leading to conical intersections.
• Compared the excited-state dynamics of benzopyran in the gas phase versus within cucurbit[5]uril.
• Demonstrated ML's effectiveness in cost-effectively simulating trajectory evolution.

Conclusions:

• ML-accelerated NAMD provides an efficient method for studying photochemical reactions in large systems.
• The study elucidates the environmental impact on photoisomerization rates within host-guest complexes.
• This approach has broad applicability for simulating complex photochemical processes.