Deep reinforcement learning for universal quantum state preparation via dynamic pulse control

Abstract

Accurate and efficient preparation of quantum state is a core issue in building a quantum computer. In this paper, we investigate how to prepare a certain single- or two-qubit target state from arbitrary initial states in semiconductor double quantum dots with only a few discrete control pulses by leveraging the deep reinforcement learning. Our method is based on the training of the network over numerous preparing tasks. The results show that once the network is well trained, it works for any initial states in the continuous Hilbert space. Thus repeated training for new preparation tasks is avoided. Our scheme outperforms the traditional optimization approaches based on gradient with both the higher efficiency and the preparation quality in discrete control space. Moreover, we find that the control trajectories designed by our scheme are robust against stochastic fluctuations within certain thresholds, such as the charge and nuclear noises.

Preparation of vibrational quasi-bound states of the transition state complex BrHBr from the bihalide ion BrHBr−

“…Although our results show possible limitations of the proposed scheme, due to the presence of BrHBr À eigenstates in the expanded target state that are challenging to reach, we consider that the level of description of the target state one could achieve is good enough, according to the results reported in the literature for these kinds of control approaches. [54][55][56] The results reported in this work, complement our investigation for the bond selective decomposition of the BrHBr transition state complex reported in paper I, in the sense that allow the preparation of the reported quasi-bound vibrational states of BrHBr to control the branching ratio of the product channels involved in the decomposition of BrHBr: Br + HBr and BrH + Br. As in paper I, the results shown assume a collinear configuration during the interaction of BrHBr À with the laser pulses, which neglects the bending modes of the molecule.…”

“…Although our results show possible limitations of the proposed scheme, due to the presence of BrHBr − eigenstates in the expanded target state that are challenging to reach, we consider that the level of description of the target state one could achieve is good enough, according to the results reported in the literature for these kinds of control approaches. 54–56…”

Modularized and scalable compilation for double quantum dot quantum computing

“…Compared to schemes such as [32][33][34] where Ansatz's parameters are updated in a closed-loop style, the application of random states reduces the susceptibility of the optimization to local optima akin to the usage of random samples in the training landscape of classical neural networks [26]. In addition, this utilization of quantum states is more resource efficient than commonly used approaches based on matrices in loss calculation, such as the Hilbert-Schmidt distance or other customized metrics between the associated unitaries [32][33][34].…”

“…A significant surge of interest has recently been focused on nascent tools that are aimed at improving the efficiency of tailoring pulse sequences. For examples, by employing the machine learning [25], Yang et al [24] studies how to directly predict the required pulse sequence with a well-trained neural network; Zhang et al [19] promises to dynamically steer a specific quantum state to another; He et al [26] and Zhang et al [27] could reset an arbitrary quantum state to a target one, and Haug et al [28] shows how to prepare an arbitrary quantum state from |0⟩. Furthermore, high fidelity universal quantum state preparation is also observed in [29] with the revised greedy algorithm.…”

Motion Simulation of Flying Quadruped Robot Based on Deep Reinforcement Learning

“…M Anipulation of quantum systems plays a significant role in many fields such as quantum computing [1][2][3], quantum high-precision measurements [4][5][6], chemical physics [7] and quantum communication networks [8]. To achieve quantum operations with high efficiency, various control approaches [7], such as optimal control [9][10][11], sliding mode control [12], robust and learning control [13][14][15] and Lyapunov control [16][17][18][19] have been proposed to solve this problem.…”