A Local Search Algorithm for Balanced Incomplete Block Designs

Abstract

Local search is often able to solve larger problems than systematic backtracking. To apply it to a constraint satisfaction problem, the problem is often treated as an optimization problem in which the search space is the set of total assignments, and the number of constraint violations is to be minimized to zero. Though often successful, this approach is sometimes unsuitable for structured problems with few solutions. An alternative is to explore the set of consistent partial assignments, minimizing the number of unassigned variables to zero. A local search algorithm of this type violates no constraints and can exploit cost and propagation techniques. This paper describes such an algorithm for balanced incomplete block design generation. On a large set of instances it out-performs several backtrackers and a neural network with simulated annealing.