z-approximations

Approximation algorithms for NP-hard optimization problems have been widely studied for over three decades. Most of these measure the quality of the solution produced by taking the ratio of the cost of the solution produced by the algorithm to the cost of an optimal solution. In certain cases, this ratio may not be very meaningful-for example, if the ratio of the worst solution to the best solution is at most some constant alpha, then an approximation algorithm with factor alpha may in fact yield the worst solution! To overcome this hurdle (among others), several authors have independently suggested the use of a different measure which we call z-approximation. An algorithm is an alpha z-approximation if it runs in polynomial time and produces a solution whose distance from the optimal one is at most alpha times the distance between the optimal solution and the worst possible solution. The results known so far about z-approximations are either of the inapproximability type or rather straightforward observations. We design polynomial time algorithms for several fundamental discrete optimization problems; in particular we obtain a z-approximation factor of 1/2 for the DIRECTED TRAVELING SALESMAN PROBLEM (TSP) (with no triangle inequality assumption). For the UNDIRECTED TSP this improves to 1/3. We also show that if there is a polynomial time algorithm that for any fixed epsilon > 0 yields an epsilon z-approximation then P = NP. We also present z-approximations for several other problems such as MAX CUT, STACKER CRANE, MAXIMUM ACYCLIC SUBGRAPH, and MINIMUM DISJOINT CYCLE COVER. (C) 2001 Elsevier Science.