Transferring variational parameters in QAOA for weighted MaxCut

The Quantum Approximate Optimization Algorithm (QAOA) is a leading candidate quantum algorithm for solving combinatorial optimization problems. The quality of the solution returned by QAOA depends critically on the quality of the parameters, typically identified by a classical optimizer. Recent results for the unweighted MaxCut and SK model have shown that pre-optimized parameters can be transferred to unseen instances to avoid costly direct optimization. We extend these results to general weighted MaxCut by showing that parameters can be transferred from easier unweighted MaxCut instances to harder weighted ones. We provide theoretical motivation for our parameter transfer scheme. The transferred parameters are competitive with numerically optimized parameters and are robust for a variety of different weight distributions, including negative weights.