The Art of Avoiding Constraints: A Penalty-free Approach to Constrained Combinatorial Optimization with QAOA (2503.10077v2)
Abstract: The quantum approximate optimization algorithm (QAOA) is designed to determine optimum and near optimum solutions of quadratic (and higher order) unconstrained binary optimization (QUBO or HUBO) problems, which in turn accurately model unconstrained combinatorial optimization problems. While the solution space of an unconstrained combinatorial optimization problem consists of all possible combinations of the objective function's decision variables, for a constrained combinatorial optimization problem, the solution space consists only of solutions that satisfy the problem constraints. While solving such a QUBO problem optimally results in an optimal solution to the underlying constrained problem, setting the right penalty parameters is crucial. Moreover, finding suitable penalties that ensure near-optimal solutions is also challenging. In this article, we introduce our innovative profit-relaxation framework to solve constrained combinatorial optimization problems. Our effective hybrid approach transforms a selected constrained combinatorial optimization problem into an unconstrained profit problem, such that the solution spaces of the two problems are interrelated with respect to their solution qualities. This allows us to determine optimal and near-optimal solutions for the constrained problem. We use three NP-hard problems -- Minimum Vertex Cover, Maximum Independent Set, and Maximum Clique -- to demonstrate the feasibility and effectiveness of our approach. We conducted a detailed performance evaluation of our profit-relaxation framework on two different platforms: Xanadu PennyLane and Argonne QTensor Quantum Circuit Simulator.