Streamlining Local Search for Spatially Balanced Latin Squares

Streamlined constrained reasoning powerfully boosts the performance of backtrack search methods for finding hard combinatorial objects. We use so-called spatially balanced Latin squares to show how streamlining can also be very effective for local search: Our approach is much faster and generates considerably larger spatially balanced Latin squares than previously reported approaches (up to order 35; the previous best results could only generate solutions up to order 18). We also provide a detailed characterization of our streamliner and solution topology for small orders. We believe that streamlined local search is a general technique suitable for solving a wide range of hard combinatorial design problems.