QUBO Formulation Using Sequence Pair With Search Space Restriction for Rectangle Packing Problem

The development of quantum annealing has stimulated interest in solving NP-hard problems, including various industrial problems, such as quadratic unconstrained binary optimization (QUBO), with specialized solvers. This study focuses on the rectangle packing problem (RPP), which is an NP-hard problem applicable to industrial applications. The objective of this problem is to place given items in a minimum area without overlap. Naive QUBO representations, such as a direct binary encoding of locations, could require a large number of bits, which limits the problem size, hindering the acquisition of the optimal solution. To circumvent this difficulty, we employ the concept of sequence pairs to efficiently represent the locations of items. We also propose a strategy that uses multiple sampling solutions obtained with a QUBO solver, where we employ new constraints on the number of the rectangles contributing to the placement area for restricting the search space to smoothly bridge the QUBO formulation of the sequence pair and the minimization of area of rectangle placement. We estimate an approximate curve for the optimum number of rectangles contributing to the placement area. Then, we add the penalty term concerning the deviation from the estimated optimal value on the objective function. In numerical experiments with instances with 4 to 10 rectangles, we demonstrate that the proposed sampling-based strategy can efficiently find feasible solutions. This strategy would be useful for dealing with more complex problems, such as rectangle orientation and three-dimensional problems, with a relatively small number of bits using sequence pairs.