A comparative study for reinforcement learning and traditional algorithms on state transfer problem

While reinforcement learning (RL) has been widely used in quantum control problems, it remains unclear whether RL is the most suitable algorithm when the control has specific constraints. We perform a comparative study on the efficacy of three RL algorithms: tabular Q-learning, deep Q-learning and policy gradient, as well as stochastic gradient descent and Krotov algorithms, in the problem of quantum state preparation. Overall, the deep Q-learning and policy gradient algorithms outperform others when the problem is discretized, e.g. allowing discrete values of control and when the problem scales up. Moreover, the reinforcement learning algorithms can also adaptively reduce the complexity of the control sequences. Our work provides insights into the suitability of reinforcement learning in quantum control problems.