Efficient charging of driven cavity-Heisenberg spin- chain quantum battery with deep learning methods
Phys. Rev. Applied 25, 054053 – Published 20 May, 2026
DOI: https://doi.org/10.1103/m1gm-f9zy
Abstract
We consider the problem of fast charging a cavity-Heisenberg spin-chain quantum battery, using as control functions an external classical field driving the cavity and the coupling between the cavity and the chain. To find the optimal controls, we employ two deep learning models, a neural network and a recurrent neural network, with recurrent connections that make it suitable for modeling sequential data. In all simulations performed for open and closed quantum system configurations, the recurrent model outperforms the simple neural network. The final stored energy achieved with the recurrent model for the closed system at short times is larger than that obtained in a recent work with reinforcement learning in much longer times. For the various open-system configurations studied, the obtained stored energies for short durations are higher than those achieved in the same work with reinforcement learning for the same durations. Regarding the optimal controls, for the closed system, both the coupling and the drive are modulated during the interval of interest, while for the open system, it seems that the optimal strategy is to fix the drive at its maximum possible absolute value to quickly insert energy into the cavity, which is transferred to the battery by modulating the coupling appropriately. The present work adds the recurrent neural-network model to the arsenal of machine-learning methods used for the control of complex quantum systems. In addition, the proposed setting is more attractive for potential implementation since the cavity field starts from the vacuum state and not an excited Fock state as in other related studies.
