QFusion: A Demonstration of Boundary-Aware Fusion Planning and Execution for Large-Scale QUBO Optimization

Large-scale database optimization problems can be encoded as Quadratic Unconstrained Binary Optimization (QUBO) instances for execution on quantum hardware. As the problem size grows, these QUBOs quickly exceed the capacity of current devices and must be decomposed into multiple subproblems. However, decomposition does not eliminate global dependencies: subproblems remain coupled through boundary interfaces, and overall solution quality depends critically on how partial solutions are fused. In this demonstration, we present QFusion , the first boundary-aware fusion planning and execution framework for large-scale QUBO solving that runs on an actual quantum annealer. We propose a DBMS-inspired, cost-based fusion tree that explicitly selects the fusion order and merge strategy. QFusion supports three database optimization tasks: join order optimization, multiple query optimization, and index selection. Conference attendees can interactively control the complete workflow through three designed scenarios: inspecting the construction of QUBOs with their database semantics, adjusting the decomposition granularity to observe the resulting boundary interfaces, and configuring fusion planning to examine the execution results on an actual quantum annealer. Overall, the demonstration highlights that scalable QUBO solving is not only a matter of solving local subproblems, but also a systems problem that requires boundary-aware fusion planning. The video corresponding to this demonstration is available at this link.