Pareto-optimal cycles for power, efficiency and fluctuations of quantum heat engines using reinforcement learning

  • Paolo Andrea Erdman ,
  • Alberto Rolandi ,
  • Paolo Abiuso ,
  • Martí Perarnau-Llobet ,

arXiv

An ideal quantum heat engine operates at high power, high efficiency, and high stability (i.e. low power fluctuations). However, these three objectives cannot be simultaneously optimized – as indicated by the so-called thermodynamic uncertainty relations – and a systematic approach to finding optimal balances between them has, as yet, been elusive. Here we propose such a general framework to identify Pareto-optimal cycles for driven quantum heat engines that trade-off power, efficiency, and fluctuations. We then employ reinforcement learning to identify the Pareto front of a quantum dot based engine and find abrupt changes in the form of optimal cycles when switching between optimizing two and three objectives. We further derive analytical results in the fast and slow-driving regimes that accurately describe different regions of the Pareto front.