A Class of Center-Free Resource Allocation Algorithms 1

Abstract This paper presents a class of resource allocation algorithms derived as a projected gradient on a subspace parametrized by a communication constraint. The algorithms have the distinguishing characteristic of converging to the optimal allocation while completely eliminating the need for a coordinating center. Furthermore, during the iteration process the objective function monotonically increases toward its maximum while the allocation always remains feasible. Finally, this paper presents a finite step initialization procedure which yields an initial allocation that has properties required by this class of algorithms.