New approximation techniques for some linear ordering problems

We describe logarithmic approximation algorithms for the NP-hard graph optimization problems of minimum linear arrangement, minimum containing interval graph, and minimum storage-time product. This improves upon the best previous approximation bounds of Even, Naor, Rao, and Schieber [J. ACM, 47 (2000), pp. 585-616] for these problems by a factor of Ω(log log n). We use the lower bound provided by the volume W of a spreading metric for each of the ordering problems above (as defined by Even et al.) in order to find a solution with cost at most a logarithmic factor times W for these problems. We develop a divide-and-conquer strategy where the cost of a solution to a problem at a recursive level is C plus the cost of a solution to the subproblems at this level, and where the spreading metric volume on the subproblems is less than the original volume by Ω(C/ log n), ensuring that the resulting solution has cost O(log n) times the original spreading metric volume. We note that this is an existentially tight bound on the relationship between the spreading metric volume and the true optimal values for these problems. For planar graphs, we combine a structural theorem of Klein, Plotkin, and Rao [Proceedings of the 25th ACM Symposium on Theory of Computing, 1993, pp. 682-690] with our new recursion technique to show that the spreading metric cost volumes are within an O(log log n) factor of the cost of an optimal solution for the minimum linear arrangement, and the minimum containing interval graph problems.