Preparation of three-atom GHZ states based on deep reinforcement learning

Generally, stimulated Raman adiabatic passage technology has been used to generate the Greenberger–Horne–Zeilinger state. Due to decoherence caused by long operation time, it is almost impossible to implement experimentally. To reduce the operation time, we propose a scheme to construct the shortcut to adiabatic passage based on deep reinforcement learning (DRL). Moreover, in order to facilitate the implementation, we have performed Gaussian fitting on the pulse sequence. Numerical analysis shows that our scheme has better performance than the Gradient Ascent Pulse Engineering and the Genetic Algorithm, and is robust to the leakage of the optical cavity as well as the spontaneous emission of atoms. Besides, we apply the DRL algorithm to another model and give the pulse sequence for the preparation of the three-atom singlet state with high fidelity and robustness.