Cooling performance in Markovian vs non-Markovian quantum collisional models

This paper compares the cooling performance of Markovian and non-Markovian quantum collisional models for nanoscale thermodynamic devices. A minimal setup is considered: a single qubit system repeatedly collides with qubit ancillas acting as a heat bath, while a local cooling “demon” operates on the system after each collision. Two regimes are analyzed: (i) a Markovian protocol where the system interacts with a fresh thermal ancilla at every step and the ancilla is subsequently rethermalized, and (ii) a non-Markovian protocol where the system repeatedly interacts with the same initially thermal environment qubit with no reset, allowing correlations (memory) to build up. A central contribution is an efficiency definition that explicitly includes all external energetic resources, counting both the energy removed by the demon and the otherwise hidden work cost of rethermalizing the ancillas in the Markovian implementation. The demon is modeled as an amplitude damping channel; via an explicit Stinespring dilation, the extracted energy is interpreted as heat dumped into an auxiliary cold reservoir, and the paper discusses how additional demon memory/erasure (Landauer) costs would further reduce operational efficiency. Numerical simulations show that the non-Markovian protocol can cool the system to much lower final energies and achieve substantially higher efficiency than the Markovian protocol across a wide range of collision (coupling) and damping strengths. The enhancement is traced to the finite environment’s initial energy and its capacity to store correlations as usable thermodynamic resources. The results emphasize that bounds and benchmarks derived under Markovian assumptions may significantly underestimate performance in finite, memoryful environments, and that reported efficiencies should be interpreted as upper bounds when demon implementation costs are not fully modeled.