Identifying optimal cycles in quantum thermal machines with reinforcement-learning

The optimal control of open quantum systems is a challenging task but has key role in improving existing information processing technologies. We introduce general framework based on Reinforcement Learning to discover thermodynamic cycles that maximize the power out-of-equilibrium heat engines and refrigerators. apply our method, soft actor-critic algorithm, three systems: benchmark two-level system engine, where we find known cycle; an experimentally realistic refrigerator superconducting qubit generates coherence, non-intuitive sequence outperform previous proposed literature; engine harmonic oscillator, cycle with elaborate structure outperforms optimized Otto cycle. then evaluate corresponding efficiency at maximum power.