Recurrent convolutional neural networks for modeling nonadiabatic dynamics of quantum-classical systems

Area of Science:

• Computational physics
• Quantum mechanics
• Machine learning

Background:

• Recurrent neural networks (RNNs) excel at modeling temporal dependencies in data.
• Hybrid quantum-classical systems involve coupled classical and quantum dynamics.
• Modeling these systems is crucial for understanding complex physical phenomena.

Purpose of the Study:

• To introduce a novel RNN model for simulating nonlinear nonadiabatic dynamics of hybrid quantum-classical systems.
• To apply the model to the one-dimensional semiclassical Holstein model.
• To assess its performance under varying quench conditions.

Main Methods:

• Developed a Physics-Aware Recurrent Convolution (PARC) neural network architecture.
• Incorporated a differentiator-integrator to model spatiotemporal dynamics.
• Utilized convolutional neural networks (CNNs) within the RNN framework.

Main Results:

• The PARC-CNN model accurately captured deterministic dynamics for shallow quenches in the Holstein model.
• Deep quenches induced chaotic evolution, posing challenges for long-term prediction.
• The architecture effectively learned the statistical climate of the system under deep-quench conditions.

Conclusions:

• The PARC-CNN is a powerful tool for modeling complex quantum-classical dynamics.
• It demonstrates adaptability to both deterministic and chaotic regimes.
• This approach advances the application of machine learning in simulating physical systems.