Parameter Transfer for Quantum Approximate Optimization of Weighted MaxCut

Finding high-quality parameters is a central obstacle to using the quantum approximate optimization algorithm (QAOA). Previous work partially addresses this issue for QAOA on unweighted MaxCut problems by leveraging similarities in objective landscape among different problem instances. However, we show that more general weighted has significantly modified landscapes, with proliferation of poor local optima. Our main contribution simple rescaling scheme overcomes these deleterious effects weights. We given depth, single “typical” vector can be successfully transferred This transfer leads median decrease approximation ratio only 2.0 percentage points relative considerably expensive direct dataset 34,701 instances up 20 nodes and multiple weight distributions. reduced 1.2 at cost 10 additional circuit evaluations sampled from pretrained metadistribution, or used as starting point run obtain ratios equivalent those achieved exhaustive 96.35% our cases.