The complexity of approximate counting

  • Authors:
  • Larry Stockmeyer

  • Affiliations:
  • -

  • Venue:
  • STOC '83 Proceedings of the fifteenth annual ACM symposium on Theory of computing
  • Year:
  • 1983

Quantified Score

Hi-index 0.00

Visualization

Abstract

There are several computational problems that can be formulated as problems of counting the number of objects having a certain property. Valiant [22] has introduced the class #P which includes a variety of counting problems such as counting the number of perfect matchings in a graph, computing the permanent of a matrix [22], finding the size of a backtrack search tree [14], and computing the probability that a network remains connected when links can fail with a certain probability [23]. We define and study a class of restricted, but very natural, probabilistic sampling methods motivated by the particular counting problems mentioned above. Instead of “singleton sampling” the algorithm is allowed to sample a large set S ample; U in one step; the information returned from the sample is whether S contains any element having the property being counted. We attempt to classify the complexity of computing approximate solutions to problems in #P. The classification is done in terms of the polynomial-time hierarchy (for short, P-hierarchy) [21]. We give a relativization result that complements a recent result of Sipser and Gaacute;c [19] that BPP is contained in the second level of the P-hierarchy.