Answers are grounded in this paper and cite their sources.
Experimental study reveals reverse annealing improves solution quality for complex combinatorial optimization problems, indicating enhanced scalability for quantum hardware.
Quantum annealing is a promising heuristic for combinatorial optimization, but on current hardware its solution quality degrades for larger and more complex problems due to noise and small energy gaps. Reverse annealing has been established as a refinement strategy, yet it remains unclear how problem characteristics, particularly the implemented quadratic unconstrained binary optimization (QUBO) matrix structure, influence refinement success and the optimal parameter regime. We find that combining forward and reverse annealing improves solution quality, with the magnitude of the improvement depending strongly on the problem class and corresponding QUBO size, while efficiency gains are tied to both annealing and problem-instance parameters. The benefits of reverse annealing increase with problem complexity and are strongest in regimes where forward annealing is increasingly limited. In these settings, reverse annealing yields larger efficiency gains than simply extending forward annealing times. We establish these results through a systematic experimental study on a D-Wave Advantage system, benchmarking reverse annealing across Max-Cut, Number Partitioning, and sparse clustering problems while varying reverse distance, pause duration, and annealing time. We identify optimal reverse annealing parameter regimes that depend on QUBO characteristics and are consistent with expected freeze-out behavior and with the energy-level-crossing structure observed for small representative instances. These findings demonstrate that reverse annealing is most valuable for large, high-complexity optimization problems and is likely to gain importance as quantum annealing hardware scales toward more realistic applications.
No takes yet. Share an insight, caveat, or question.
Menger et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: