Cooling and work extraction under memory-assisted Markovian thermal processes

Abstract: We investigate the limits on cooling and work extraction via Markovian thermal processes assisted by a finite-dimensional memory. Here the memory is a $d$-dimensional quantum system with trivial Hamiltonian and initially in a maximally mixed state. For cooling a qubit system, we consider two paradigms: cooling under coherent control and cooling under incoherent control. For both paradigms, we derive the optimal ground-state populations under the set of general thermal processes (TP) and the set of Markovian thermal processes (MTP), and we further propose memory-assisted protocols, which bridge the gap between the performances of TP and MTP. For the task of work extraction, we prove that when the target system is a qubit in the excited state the minimum extraction error achieved by TP can be approximated by Markovian thermal processes assisted by a large enough memory. Our results can bridge the performances of TP and MTP in thermodynamic tasks including cooling and work extraction.