Analysis of the Greedy Approach in Problems of Maximum k-Coverage

In this paper, we consider a general covering problem in which k subsets are to be selected such that their union covers as large a weight of objects from a universal set of elements as possible. Each subset selected must satisfy some structural constraints. We analyze the quality of a k-stage covering algorithm that relies, at each stage, on greedily selecting a subset that gives maximum improvement in terms of overall coverage. We show that such greedily constructed solutions are guaranteed to be within a factor of 1 0 1/e of the optimal solution. In some cases, selecting a best solution at each stage may itself be difficult; we show that if a b-approximate best solution is chosen at each stage, then the overall solution constructed is guaranteed to be within a factor of 1 0 1/e of the optimal. Our results also yield a simple proof that the number of subsets used by the greedy approach to achieve entire coverage of the universal set is within a logarithmic factor of the optimal number of subsets. Examples of problems that fall into the family of general covering problems considered, and for which the algorithmic results apply, are discussed. q 1998 John Wiley & Sons, Inc. Naval Research Logistics 45: 615–627, 1998 INTRODUCTION Covering problems, in which a universal set of elements is to be covered with as few subsets as possible, where each subset must satisfy some structural constraints, arise in many contexts. Many of these problems are known to be NP-complete. The focus of this paper is on developing approximation algorithms for covering problems in which the objective is to select a fixed number of subsets k such that maximum coverage of the universal set is achieved. Since we do not require that all elements of the universal set be covered, it makes sense to assign weights to the individual elements; maximum coverage corresponds to selecting k subsets that cover a set of elements with total weight as great as possible. Because the problem of selecting a minimum number of subsets that cover all elements in Correspondence to: D. S. Hochbaum Contract grant sponsor: Office of Naval Research; contract grant number: N00014-91-J-1241 Contract grant sponsor: National Science Foundation; contract grant number: DMI-9713482 q 1998 by John Wiley & Sons, Inc. CCC 0894-069X/98/060615-13 8m26 991 / 8m26$$0991 07-06-98 11:44:39 nra W: Nav Res 616 Naval Research Logistics, Vol. 45 (1998) Figure 1. Generic k-stage covering algorithm. a universal set is NP-hard, so too is the problem of covering a maximum set of elements with a fixed number of subsets. We derive results for a greedy-like approximation algorithm for such covering problems in a very general setting so that, while the details vary from problem to problem, the results regarding the quality of solution returned apply in a general way. Johnson [17] and Chvátal [5] have analyzed such an algorithm for the related problem of optimally covering all elements in a universal set. In this paper, we have attempted to abstract away as many details of the covering problem we consider as possible, and to make our algorithm as generic as possible, so that our analysis is applicable to a wide variety of problems. The maximum k-coverage problem we consider can be described as follows: INSTANCE: A universal set of elements U , an integer k , and a class R of subsets of U . Each element u √ U has an associated weight w(u) . OPTIMIZATION PROBLEM: Select k subsets, A1 , A2 . . . , Ak , of U , where each subset selected is a member of R, such that the weight of the elements in U lÅ1 Al is maximized. We assume, without loss of generality, that A √ R implies that A* √ R for all A * ⊆ A . So, if there is an element in both Ai √ R and Aj √ R, then it can be removed from Aj say (Aj remains in R) , without changing the quality of the solution to the covering problem. Hence, we can add the restriction to the general covering problem that Ai > Aj Å M, for all 1 ° i õ j ° k ; that is, no two subsets selected have any elements in common. This restriction will make our notation, and subsequent presentation, cleaner; an alternative would have been to specify that, after a subset Ai is selected for the cover, all remaining subsets which have those elements that are also in Ai removed. Beyond this nonrestrictive assumption that the selected subsets are mutually disjoint, it is important to note that there are no conditions that the Ai’s must satisfy with respect to each other. In Section 2, a generic k-stage algorithm that selects the subsets A1 , A2 , . . . , Ak in an iterative manner is given, and performance guarantees are derived for greedy-like implementations of the algorithm. Section 3 develops approximation algorithms for a variety of problems and applications for which the results of Section 2 apply. Problems, taken from a variety of applications, regarding covering graphs with subgraphs, packing, and fixed parameter combinatorial optimization are considered. In Section 4, we extend the analysis of Section 2 in several ways, and then conclude with a summary. 1. A k-STAGE ALGORITHM Consider the generic k-stage covering algorithm, shown in Figure 1, that selects a subset Al at stage l . (Note that, either because there are fewer than k subsets in R or because of 8m26 991 / 8m26$$0991 07-06-98 11:44:39 nra W: Nav Res 617 Hochbaum and Pathria: Problems of Maximum k-Coverage our assumption that selected sets are disjoint, it may be that at some stage l no additional set can be selected. In such a situation, the solution obtained after stage l 0 1 will provide optimal coverage.) Let C(k) Å [C1 , C2 , . . . , Ck] be an optimal solution, where c(k) is the value (weight of elements covered) of the optimal solution. For l Å 1, 2, . . . , k , let A( l) Å [A1 , A2 , . . . , Al] be the solution constructed up to the end of stage l of the generic algorithm; so, A(k) is the final solution returned. We denote the value of Al by al , and the value of A( l) by a( l) : al Å ∑ u√Al w(u) and a( l) Å ∑ l