Single-atom exploration of optimized nonequilibrium quantum thermodynamics by reinforcement learning

• Published in: Communications physics Vol. 6; no. 1; pp. 286 - 8
• Main Authors: Zhang, Jiawei, Li, Jiachong, Tan, Qing-Shou, Bu, Jintao, Yuan, Wenfei, Wang, Bin, Ding, Geyi, Ding, Wenqiang, Chen, Liang, Yan, Leilei, Su, Shilei, Xiong, Taiping, Zhou, Fei, Feng, Mang
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 07.10.2023; Nature Publishing Group; Nature Portfolio
• Subjects: 639/766/483/481; 639/766/530/951; Accuracy; Algorithms; Calcium isotopes; Consumption; Data processing; Entropy; Evolution; Machine learning; Nonequilibrium thermodynamics; Physics; Physics and Astronomy; Quantum phenomena; Qubits (quantum computing); Spontaneous emission; Thermodynamics
• ISSN: 2399-3650; 2399-3650
• DOI: 10.1038/s42005-023-01408-5

Exploring optimized processes of thermodynamics at microscale is vital to exploitation of quantum advantages relevant to microscopic machines and quantum information processing. Here, we experimentally execute a reinforcement learning strategy, using a single trapped 40 Ca + ion, for engineering quantum state evolution out of thermal equilibrium. We consider a qubit system coupled to classical and quantum baths, respectively, the former of which is achieved by switching on the spontaneous emission relevant to the qubit and the latter of which is made based on a Jaynes-Cummings model involving the qubit and the vibrational degree of freedom of the ion. Our optimized operations make use of the external control on the qubit, designed by the reinforcement learning approach. In comparison to the conventional situation of free evolution subject to the same Hamiltonian of interest, our experimental implementation presents the evolution of the states with higher fidelity while with less consumption of entropy production and work, highlighting the potential of reinforcement learning in accomplishment of optimized nonequilibrium thermodynamic processes at atomic level. The authors use reinforcement learning (RL), an important algorithm in machine learning, to optimize nonequilibrium quantum thermodynamics. They find the optimized evolution of the state with higher fidelity and less consumption of entropy production as well as less work cost than in the case of free evolution, highlighting the potential of RL strategies.