Automatic Evolution of Machine-Learning-Based Quantum Dynamics with Uncertainty Analysis.

Area of Science:

• Quantum Physics
• Computational Science

Background:

• Simulating open quantum systems is crucial for understanding complex quantum phenomena.
• Traditional methods often struggle with long-time quantum dynamics and computational cost.

Purpose of the Study:

• To develop and validate machine learning approaches for simulating the long-time dynamics of open quantum systems.
• To optimize the construction of neural network models for quantum dynamics.
• To quantify the uncertainty in machine learning predictions for quantum simulations.

Main Methods:

• Utilized long short-term memory recurrent neural networks (LSTM-RNNs) for simulating quantum dynamics.
• Employed hyperparameter optimization techniques: simulated annealing, Bayesian optimization, and random search.
• Applied bootstrap sampling and Monte Carlo dropout for uncertainty quantification.

Main Results:

• LSTM-RNN models successfully simulated long-time quantum dynamics based on short-time evolution data.
• Simulated annealing demonstrated superior performance in hyperparameter optimization.
• Uncertainty analysis provided confidence estimates for the predictive accuracy of the models.

Conclusions:

• An effective machine learning framework for simulating open quantum system dynamics has been established.
• An efficient protocol for building optimal neural networks and assessing model reliability was presented.