Simulating Open Quantum Dynamics with a Neural Network-Enhanced Non-Markovian Stochastic Schrödinger Equation.

Area of Science:

• Quantum mechanics
• Computational physics
• Machine learning in quantum systems

Background:

• The non-Markovian stochastic Schrödinger equation (NMSSE) is valuable for open quantum simulations due to its efficiency.
• However, NMSSE faces convergence issues at low temperatures, demanding extensive computational resources for long-time simulations.

Purpose of the Study:

• To develop an efficient method for accurate long-time quantum evolution using NMSSE, especially at low temperatures.
• To overcome the high computational cost and convergence challenges associated with traditional NMSSE simulations.

Main Methods:

• A novel scheme integrating convolutional neural networks (CNNs) and long short-term memory recurrent neural networks (LSTMs).
• Utilizing iterative attentional feature fusion (iAFF) to extract effective information from simulations.
• Fine-tuning neural networks with short-time data to predict long-term system behavior and mitigate stochastic oscillations.

Main Results:

• The proposed method significantly reduces the number of required stochastic trajectories for long-time simulations.
• Demonstrated substantial improvements in convergence and computational cost reduction, particularly at low temperatures.
• Successfully applied to simulate the dynamics of the spin-boson model and the Fenna-Matthews-Olson (FMO) complex.

Conclusions:

• The neural network-based approach effectively enhances the accuracy and efficiency of NMSSE for quantum simulations.
• This method offers a promising solution for overcoming low-temperature convergence challenges in quantum dynamics.
• The integration of advanced neural network architectures provides a powerful tool for complex quantum system analysis.