Forecasting nonadiabatic dynamics using hybrid convolutional neural network/long short-term memory network.

Area of Science:

• Quantum dynamics
• Machine learning in chemistry
• Computational physics

Background:

• Modeling long-time nonadiabatic dynamics in complex molecular systems is computationally challenging.
• Existing methods struggle with long-time predictions, limiting understanding of quantum behavior.

Purpose of the Study:

• To develop a novel machine learning scheme for accurate long-time nonadiabatic dynamics prediction.
• To leverage hybrid neural networks for enhanced predictive capabilities in quantum systems.

Main Methods:

• A hybrid convolutional neural network/long short-term memory (CNN-LSTM) framework was employed.
• The model utilizes short-time dynamics to predict long-time quantum behavior.
• Feature fusion of CNN-LSTM models for quantum population and coherence was performed.

Main Results:

• The CNN-LSTM scheme demonstrated high accuracy and robustness in predicting various quantum dynamics models.
• The model successfully predicted dynamics up to 0.3 picoseconds.
• Knowledge transferability between similar systems was observed, enhancing predictive accuracy.

Conclusions:

• The hybrid CNN-LSTM network offers high predictive power for nonadiabatic dynamics.
• This approach is promising for realistic charge and energy transfer processes in photoinduced energy conversion.
• The method advances the computational modeling of complex quantum systems.