Capturing Excited State Proton Transfer Dynamics with Reactive Machine Learning Potentials.

Area of Science:

• Photochemistry and Molecular Dynamics
• Computational Chemistry and Machine Learning

Background:

• Excited state proton transfer (ESPT) is vital for fluorescence sensing, bioimaging, and optoelectronics.
• Resolving ESPT dynamics is computationally expensive and requires advanced experimental techniques.

Purpose of the Study:

• To develop a machine learning-driven approach for simulating excited state molecular dynamics.
• To accurately map reaction landscapes and capture photorelaxation dynamics of ESPT.

Main Methods:

• An active learning framework combined with enhanced sampling techniques was employed.
• High-quality training sets were generated for excited state machine learning potentials.
• Simulations were performed on 10-hydroxybenzo[h]quinoline as a model system.

Main Results:

• A barrierless excited state proton transfer was observed within approximately 50 femtoseconds.
• A significant emission energy red shift of about 1 eV was detected, matching experimental data.
• Strong coupling between proton transfer and charge redistribution was identified.

Conclusions:

• Machine learning-driven molecular dynamics accurately captures ultrafast photochemical dynamics.
• The developed framework enables efficient large-scale statistical sampling of complex processes.
• This approach significantly advances the study of fundamental photochemical reactions.