Faster Crossing over Quantum Phase Transition Assisted by Reinforcement Learning.

An energy gap develops near quantum critical point (QCP) of quantum phase transition (QPT) in a finite many-body (MB) system, facilitating adiabatic ground state transformation by parameter change. In real application scenarios, however, the efficacy for such adiabatic protocol is compromised by the need to balance finite system life time with adiabaticity, as exemplified in a recent experiment that prepares three-mode balanced Dicke state near deterministically [PNAS {\bf 115}, 6381 (2018)]. Instead of following the instantaneous ground state as unanimously required for most adiabatic crossing, this work reports a faster sweeping policy taking advantage of fast dynamics in the excited levels. It is obtained from deep reinforcement learning (DRL) based on a multi-step training scheme we develop. In the absence of loss, a fidelity $\ge 99\%$ between the prepared and target Dicke state is achieved over a small fraction of the adiabatically required time. When loss is included, training is carried out according to an operational benchmark, the interferometric sensitivity of the prepared state, leading to better sensitivity while using about half of the time previously reported. Implemented in a Bose-Einstein condensate (BEC) of $\sim 10^4$ $^{87}$Rb atoms, the balanced three-mode Dicke state exhibiting an improved number squeezing of $13.02\pm0.20$ dB is observed within 766 ms, highlighting the potential of DRL for quantum dynamic control and quantum state preparation in interacting MB systems.