Symbolic Reasoning with Weighted and Normalized Decision Diagrams

  • Authors:
  • Jörn Ossowski;Christel Baier

  • Affiliations:
  • Universität Bonn, Institut für Informatik I, Römerstrasse 164, 53117 Bonn, Germany;Universität Bonn, Institut für Informatik I, Römerstrasse 164, 53117 Bonn, Germany

  • Venue:
  • Electronic Notes in Theoretical Computer Science (ENTCS)
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

Several variants of Bryant's ordered binary decision diagrams have been suggested in the literature to reason about discrete functions. In this paper, we introduce a generic notion of weighted decision diagrams that captures many of them and present criteria for canonicity. As a special instance of such weighted diagrams, we introduce a new BDD-variant for real-valued functions, called normalized algebraic decision diagrams. Regarding the number of nodes and arithmetic operations like addition and multiplication, these normalized diagrams are as efficient as factored edge-valued binary decision diagrams, while several other operators, like the calculation of extrema, minimum or maximum of two functions or the switch from real-valued functions to boolean functions through a given threshold, are more efficient for normalized diagrams than for their factored counterpart.