Multi-parameter optimization of polarization gradient cooling for ⁸⁷Rb atoms based on reinforcement learning

Area of Science:

• Atomic, Molecular, and Optical Physics
• Quantum Information Science
• Machine Learning Applications in Physics

Background:

• Polarization gradient cooling (PGC) is crucial for creating Bose-Einstein condensates (BECs) and cooling single atoms.
• Traditional PGC optimization relies on expertise, limiting efficiency and fine control.
• Challenges exist in optimizing PGC parameters for complex cold atom experiments.

Purpose of the Study:

• To introduce a novel segmented control method for PGC.
• To enhance the efficiency and precision of PGC parameter optimization.
• To enable intelligent preparation of degenerate quantum gases.

Main Methods:

• Expanded PGC experiment parameters from 3 to 30.
• Reformulated timing optimization as a Markov decision process (MDP).
• Utilized a reinforcement learning model for parameter optimization.

Main Results:

• Achieved convergence and effective parameter exploration with the reinforcement learning model.
• Captured approximately 4.3 × 10^8 cold atoms.
• Reached a phase space density of ~7.1 × 10^-4 at ~3.7 µK in ~18.8 minutes.

Conclusions:

• The proposed segmented control and reinforcement learning method significantly improves PGC efficiency.
• This intelligent approach facilitates the preparation of ultracold atoms for quantum applications.
• The method demonstrates powerful parameter exploration and optimization capabilities.