A case study of de-randomization methods for combinatorial approximation algorithms
Andrea Roli, Luca Trevisan · DIMACS series in discrete mathematics and theoretical computer science · 1998
We study three different de-randomization methods that are often applied to approximate combinatorial optimization problems. We analyze the conditional probabilities method in connection with randomized rounding for routing, packing and covering integer linear programming problems. We show extensions of such methods for nonindependent randomized rounding for the assignment problem. The second method, the so called random walks is exemplified with algorithms for dense instances of some NP problems. Another often used method is the bounded independence technique; we explicit this method for the sparsest cut and maximum concurrent flow problems. 1 Introduction Randomized algorithms are often used to solve or to approximate optimization problems related to network design. Theoretical and practical concern about the effective availability of a source of truly random bits, as well as the desire of improved reliability, motivate the search for deterministic versions of such algorithms. While...