Solving connected row convex constraints by variable elimination

We propose an algorithm for the class of connected row convex constraints. In this algorithm, we introduce a novel variable elimination method to solve the constraints. This method is simple and able to make use of the sparsity of the problem instances. One of its key operations is the composition of two constraints. We have identified several nice properties of connected row convex constraints. Those properties enable the development of a fast composition algorithm whose complexity is linear to the size of the variable domains. Compared with the existing work including randomized algorithms, the new algorithm has favorable worst case time and working space complexity. Experimental results also show a significant performance margin over the existing consistency based algorithms.