Quantum Computers Solve Optimization Faster with New Method

Quantum computers are advancing rapidly, but they often struggle with real-world optimization problems due to inefficiencies in handling constraints. A new approach developed by researchers at IBM Research and Linear Softwares Matemáticos addresses this by creating tailored quantum circuits that only explore valid solutions, avoiding the computational waste of traditional s. This innovation could speed up applications in logistics, resource allocation, and beyond, making quantum computing more practical for everyday s.

The key finding is that these tailored variational forms (TVFs) produce only feasible solutions for specific constraints in linear constrained quadratic binary optimization (LCQBO) problems. Unlike common techniques that use penalties to convert constrained problems into unconstrained ones—often leading to invalid this ensures that every solution generated meets the problem's requirements. For example, in a facility location problem, it guarantees that a customer is only assigned to an open facility, without extra checks or errors.

Ologically, the researchers designed quantum circuits that encode constraints directly into the solution space. They focused on four common constraint types, such as binary comparisons and sum limits, using parameterized rotations and controlled gates on qubits. This approach builds on the Variational Quantum Eigensolver (VQE) algorithm, a hybrid quantum-classical , but simplifies it by reducing the number of gates and parameters. For instance, in the facility location problem test, the TVF used only 4 parameters and 2 CNOT gates, compared to 8 parameters and 3 CNOT gates for a standard 2-Local variational form.

From testing on real quantum computers show significant efficiency gains. In the facility location problem instance with 2 facilities and 1 client, the TVF achieved the optimal solution with an objective function value of 8.0, converging in 45 iterations as shown in Figure 8. The most probable solution was state |1010⟩, indicating facility 1 is open and serves the client. Similarly, for the linear assignment problem with 2 jobs and 2 workers, the TVF found the optimal assignment (state |0110⟩) with an objective value of 15, again in 45 iterations (Figure 11). Table 5 and Table 6 detail the resource usage: for the facility location problem, the TVF had a cost of 30 (using equation 6 from the paper), lower than 44 for 2-Local and 146 for QAOA, due to fewer CNOT gates and parameters.

This matters because optimization problems are everywhere—from deciding where to build hospitals to assigning jobs efficiently—and current quantum s are often too slow or error-prone for practical use. By ensuring feasible solutions and using fewer quantum resources, this could make quantum computers more reliable for industries like supply chain management and healthcare, where accuracy and speed are critical. It also reduces the need for complex parameter tuning, simplifying implementation on noisy intermediate-scale quantum (NISQ) devices.

However, the approach has limitations. It cannot easily combine multiple constraints into a single circuit, as noted in the paper, which restricts its application to problems with isolated constraints. Future work may focus on developing circuits that handle overlapping constraints and reducing gate usage further to improve scalability.