We propose a model called priority branching trees (pBT) for backtracking and dynamic programming algorithms. Our model generalizes both the priority model of Borodin, Nielson and Rackoff, as well as a simple dynamic programming model due to Woeginger, and hence spans a wide spectrum of algorithms. After witnessing the strength of the model, we then show its limitations by providing lower bounds for algorithms in this model for several classical problems such as Interval Scheduling, Knapsack and Satisfiability. 1

Borodin, Nielsen and Rackoff [13] introduced the class of priority algorithms as a framework for modeling deterministic greedy-like algorithms. In this paper we address the effect of randomization in greedy-like algorithms. More specifically, we consider approximation ratios within the context of randomized priority algorithms. As case studies, we prove inapproximation results for two well-studied optimization problems, namely facility location and makespan scheduling.

Abstract. Borodin, Nielsen and Rackoff [5] proposed a framework for abstracting the main properties of greedy-like algorithms with emphasis on scheduling problems, and Davis and Impagliazzo [6] extended it so as to make it applicable to graph optimization problems. In this paper we propose a related model which places certain reasonable restrictions on the power of the greedy-like algorithm. Our goal is to define a model in which it is possible to filter out certain overly powerful algorithms, while still capturing a very rich class of greedy-like algorithms. We argue that this approach better motivates the lower-bound proofs and possibly yields better bounds. To illustrate the techniques involved we apply the model to the well-known problems of (complete) facility location and dominating set.

Greedy-like algorithms have been a popular approach in combinatorial optimization, due to their conceptual simplicity and amenability to analysis. Surprisingly, it was only recently that a formal framework for their study emerged. In particular, Borodin, Nielsen and Rackoff introduced the class of priority algorithms as a model for abstracting the main properties of (deterministic) greedy-like algorithms; they also showed limitations on the approximation power of such algorithms for various scheduling problems. In this thesis we extend and modify the priority-algorithm framework so as to make it applicable to a wider class of optimization problems and settings. More precisely, we first derive strong lower bounds on the approximation ratio of priority algorithms for two well-studied problems, namely facility location and set cover. These are problems for which several greedy-like algorithms with good performance guarantees exist. Subsequently, we address the issue of randomization in priority algorithms, and show how to prove bounds on the power of greedy-like algorithms with access to random bits. Finally, we propose a model for priority algorithms in the context of graph theoretic optimization problems; the later class of problems turns out to be of particular interest, since it poses certain conceptual challenges when studying