Variational quantum algorithms offer a practical route to low-energy state preparation on near-term hardware, but performance depends strongly on ansatz design. We introduce the Heat-Exchange (HE) ansatz, a compact circuit family inspired by heat-bath algorithmic cooling and built from a tunable exchange interaction between a system qubit and an auxiliary bath qubit. On superconducting processors the exchange evolution \(U_{\textrm{HE}}(\theta )=\exp [-i\theta (XX+YY)/2]\) is implemented digitally as a shallow compiled gate sequence, requiring no mid-circuit reset, measurement, or feedback. We benchmark HE on two complementary tasks: weighted MaxCut on random complete graphs (noise-free simulation and hardware runs on ibm_strasbourg with readout error mitigation) and a one-dimensional Heisenberg chain with a pinned site within a dissipative-VQE-style workflow (noise-free simulation). HE increases the probability of sampling the best cut compared with a hardware-efficient baseline and QAOA under limited evaluation budgets and yields ground-state energies with sub-percent relative error for the impurity chain. These results highlight exchange-driven cooling blocks as compact, parameter-efficient, hardware-compatible primitives for near-term variational workflows.
