We study the problem of decomposing a graph into a weighted sum of a small number of graph matchings. This problem arises in network resource allocation problems such as peer-to-peer energy exchange, and it is challenging to solve with current classical algorithms even for small instances. To address this problem, we propose a hybrid quantum-classical algorithm, E-FCFW, based on the Fully-Corrective Frank-Wolfe (FCFW) algorithm. In particular, E-FCFW extends FCFW by incorporating a matching-sampling subroutine that can be carried out classically or with a quantum approach. We show how to design such a subroutine using QAOA, which aims at solving a constrained discrete optimisation problem approximately to obtain solution-variety. We benchmark our approach on complete, bipartite, and heavy-hex graphs, conducting experiments using the Qiskit Aer state-vector simulator (9-25 qubits), the Qiskit Aer MPS simulator (52-76 qubits) and on IBM Kingston (52-111 qubits), demonstrating performance at a utility-scale quantum hardware level. Our results show that E-FCFW with QAOA consistently yields sparser decompositions (mean and median) than the other methods (random sampling and simulated annealing) for small complete and bipartite graphs. For large heavy-hex graphs with 50 and 70 nodes, E-FCFW with QAOA also outperforms the other methods in terms of approximation error. Our findings highlight a promising role for quantum subroutines in classical algorithms.
Pennington et al. (Fri,) studied this question.