Reinforcement Learning Approach to Nonequilibrium Quantum Thermodynamics.

Area of Science:

• Quantum Thermodynamics
• Statistical Mechanics
• Control Theory

Background:

• Controlling quantum systems far from equilibrium is challenging.
• Entropy production quantifies thermodynamic irreversibility.
• Minimizing entropy production is key for efficient quantum processes.

Purpose of the Study:

• To develop an experimentally accessible method for reducing entropy production in closed quantum systems.
• To control nonequilibrium quantum thermodynamics using reinforcement learning.

Main Methods:

• Utilized a reinforcement learning approach with a policy gradient technique.
• Employed an external control Hamiltonian to guide system evolution.
• The method is independent of specific irreversibility measures and system dynamics knowledge.

Main Results:

• Successfully reduced entropy production in single- and two-particle quantum systems.
• Demonstrated an experimentally nondemanding approach to controlling quantum thermodynamics.
• The method does not require tracking the quantum state during evolution.

Conclusions:

• Reinforcement learning offers a powerful and practical tool for controlling quantum thermodynamics.
• The proposed method provides a robust strategy for minimizing irreversibility in driven quantum systems.